You are on page 1of 292

aefJt^pv=p~

cbcilt

c=b=p~=c
C=q~=p~=p

o=j~~
eKJgKdK=a=

aefJt^pv=de

`=W
k=~===~~=~==I=I==~~======
=~==aefJt^pv=deK
`=EF=NVVUI=OMMNI=OMMOI=OMMRI=OMMTI=OMMV=aefJt^pv=de=_=J=~==K=
aefJt^pv=~=cbcilt=~==~~==aefJt^pv=deK

aefJt^pv=deI===
t~=p~=NMRI==aJNOROS=_I=d~
mW=HQVJPMJST==VV=VUJMI=c~W=HQVJPMJST=VV=VUJVV
bJj~W=~]J~K

==o=j~~

^`hkltibadbjbkq
q= ~= ~= lK= h= Ercw= iF=
= = ~= = = = `~= VI= NMI= NNI
NOI=NPI=NQI=NRI=NS=~=NTK

cbcilt==

==o=j~~

`
m~=f==m~=_~==j
NK=_~=b~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K NP
NKN=
NKO=
NKP=
NKQ=
NKQKN=
NKQKO=
NKQKP=
NKR=
NKS=
NKSKN=
NKSKO=
NKSKP=
NKSKQ=

`~=m =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNP
j~=`~=b~ =K=K=K=K=K=K=K=K=K=KNR
`~=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNU
j=^~=~=p~ K=K=K=K=K=K=K=K=KNV
e~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNV
`~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOM
q~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOP
^J~~=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOQ
b~==p~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOS
c== =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOS
m==K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KPM
f~==EF= =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KPM
^==_ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KPM

NKSKR=
NKT=
NKTKN=
NKTKO=
NKTKP=
NKU=
NKV=
NKNM=
NKNN=

a~==== K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= PN
p~==a=_~~=qK=K=K=K=K=K= PP
`~===~=EPa=~=Oa=E~I=~J
FF K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= PP
`~===~=EOa=~F =K=K=K=K=K= PQ
a~=~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= PR
a=~=s=o=`==eJ
~=`K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= PS
a=~=`=c==q~=b~J
=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= PS
_=^~=~=b =K=K=K=K=K=K= PU
`~~=c=`=m=K=K=K=K=K=K= PV

OK=c=~=q~=m==c=p~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K QP
OKN=
OKO=
OKP=

o=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KQP
e~=EOaF=m==r=^=KQP
c=p~==Pa K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KQR

OKQ=
OKR=

_^pa=q=~=j=Pa=j =K=K=K=K=K=K= QS
^~=`~~=f~=`==aJ
===i~==c=p~ K=K=K=K=K=K=K=K=K= QU

PK=_~=b~==j =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K RN
PKN=
PKO=

d~=c~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KRN
e~=c~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KRP

PKOKN=
PKOKO=

`=~=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= RQ
r=E~F=~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= RR

QK=f~I=_~I=~=`~=` =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K RT
QKN=
QKO=
QKOKN=

f~=`K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KRT
_~=` =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KRT
c=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KRU

QKOKO=
QKOKP=
QKP=

j~=~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= SP
e~=~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ST
`~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= TM

cbcilt==

`
QKPKN=
QKPKO=

c=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KTN
j~=~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KTQ

QKPKP=

e~=~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= TR

RK=^ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= TV
RKN=
RKO=
RKP=
RKPKN=
RKPKO=

m~=a=~=o~K=K=K=K=K=K=K=K=K=K=K=K=K=KTV
qJ~=^J=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KUM
qJ~=^K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KUN
d~=Pa=~=~ K=K=K=K=K=K=K=K=K=K=K=KUN
p~J=Pa=~==p=

RKQ=
RKQKN=
RKQKO=

=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= UO
p~=`~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= UR
^J~~=~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= UR
f K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= UR

SK=o~==c~==cbcilt =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= UT
SKN=
SKO=
SKP=

Pa=c=m~~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KUV
s~==^=Oa=c=m~~=K=K=K=KVO
e~=Oa=c=m~~J =K=K=K=K=K=K=K=K=K=K=K=K=K=KVR

SKQ=
SKR=

j~=q~=m~~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= VT
e~=q~=m~~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= VV

TK=o=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NMP
UK=k~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NMT

m~=ff==^~
VK=q=m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NNT
VKN=
VKO=
VKP=
VKQ=

f=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNNT
p~=p===^=q=E`=^J
F=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNNT
^ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNNT
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNNU

VKR=
VKS=
VKT=
VKU=

^~~=p K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NNU


k~=^~=EcbjFK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NNU
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NOM
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NOO

NMK=l~~=~=_~=mW=NJa=^JaJa~ K=K=K=K=K=K=K=K=K=K= NOP


==o=j~~

`
NMKN=
NMKO=
NMKP=
NMKQ=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNOP
^ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNOP
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNOQ
^~~=p =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNOR

NMKR=
NMKS=
NMKT=

k~=^~=EcbjFK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NOR
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NOS
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NOU

NNK=_=~=p=mW=i~=^ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K NOV


NNKN=
NNKO=
NNKP=

p~=p===^=q=E`=^J
F=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNOV
^ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNPM
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNPM

NNKQ=
NNKR=
NNKS=
NNKT=

^~~=p K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NPO


k~=^~=EcbjFK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NPP
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NPQ
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NPT

NOK=j~=a==c~=m=j~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K NPV


NOKN=
NOKO=
NOKP=
NOKQ=
NOKQKN=
NOKQKO=
NOKQKP=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNPV
^ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNPV
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNQM
^~~=p =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNQN
d~=~= K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNQN
q~======
~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNQN
p~J~= K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNQO

NOKR=
k~=^~=EcbjFK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NQO
NOKRKN= r=~~== K=K=K=K=K=K=K=K=K=K=K=K=K=K= NQO
NOKRKNKN= a~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NQO
NOKRKNKO= o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NQQ
NOKRKO= r=~= K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NQT
NOKRKOKN= a~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NQT
NOKRKOKO= o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NQU
NOKS=
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NQU

NPK=e=~=e~~=bW=q=q=t=mK=K=K=K=K=K=K=K=K=K NQV
NPKN=
NPKO=
NPKP=
NPKQ=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNQV
^ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNRM
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNRM
^~~=p =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNRN

NPKR=
NPKS=
NPKT=

k~=^~=EcbjFK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NRO
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NRQ
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NRT

NQK=b=mW=k~~=` K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K NRV


NQKN=
NQKO=
NQKP=
NQKQ=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNRV
_=^~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNRV
b=_=^~K=K=K=K=K=K=K=K=K=KNSM
d=b~=J=c~==q==eJ
~=e~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNSN

NQKR=
NQKS=
NQKSKN=

b=b~=J=c~==q==p~=
cK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NSQ
k~=^~=fW=`Js~===
j =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NSS
d~==Jcbj =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NTM

cbcilt==

`
NQKSKO=
NQKSKP=
NQKSKQ=
NQKSKR=
NQKSKS=
NQKT=

j==K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNTM
^===~===~ K=KNTM
_~= K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNTM
q==K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNTN
p =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNTN
k~=^~=ffW=a~=b=~=_~J

NQKTKN=
NQKTKO=
NQKTKP=
NQKTKQ=
NQKU=

~=b~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NTR


j= =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NTR
q=~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NTU
c=~=~~=J=_=~~ K= NUP
c~===~=~ =K=K=K=K=K=K=K= NUS
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NVN

NRK=q~=` K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NVP


NRKN=
NRKO=
NRKP=

f=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNVP
p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNVP
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KNVP

NRKQ=
NRKR=
NRKS=

k~=^~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NVR
o=~=a K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= NVT
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OMO

NSK=e=m =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OMP


NSKN=
NSKO=
NSKOKN=
NSKOKO=

f=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOMP
p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOMP
a=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOMP
a==K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOMQ

NSKP=
NSKQ=
NSKR=
NSKS=

d=b~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OMR


k~=^~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OMT
o=~=a K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OMU
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ONN

NTK=p~=a=m=Eevaol`lfk=~=R==NF =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ONP


NTKN=
NTKO=
NTKP=
NTKQ=
NTKQKN=
NTKQKO=
NTKQKP=
NTKR=
NTKRKN=

f=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KONP
p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KONP
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KONP
k~=^~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KONR
j=~=J== K=K=K=K=K=KONR
p=J=~= K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KONR
`~~= K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KONR
o=~=aK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KONT
m~===K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KONT

NTKRKO=
NTKRKP=
NTKRKQ=
NTKRKR=
NTKS=
NTKSKN=
NTKSKO=
NTKT=

a= =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ONT
p~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ONV
aJ= =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OOM
a= =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OOM
q~=b===p~=a=m=K=
=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OON
d=~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OON
`=~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OOQ
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OOR

NUK=d~=mW=r~~=p=`=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OOT
NUKN=
NUKO=
NUKP=

==o=j~~

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOOT
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOOT
^~~=p =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOOU

NUKQ=
NUKR=
NUKS=

k~=^~=EcbjFK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OOV
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OOV
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OPM

`
NVK=`~==~K=mW=r~~=c==~=e=p K=K=K=K=K=K=K=K OPN
NVKN=
NVKO=
NVKP=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOPN
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOPN
k~=^~=EcbjF =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOPO

NVKQ=
NVKR=

o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OPP
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OPQ

OMK=s~=d=mW=r~~=c==~=f=p =K=K OPR


OMKN=
OMKO=
OMKP=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOPR
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOPR
k~=^~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOPS

OMKQ=
OMKR=

o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OPV
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OPV

ONK=s~~==~K=mW=a~~=b K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K OQP


ONKN=
ONKO=
ONKP=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOQP
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOQP
k~=^~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOQR

ONKQ=
ONKR=

o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OQR
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OQT

OOK=t=`~~=_~=m =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K OQV


OOKN=
OOKO=
OOKP=
OOKPKN=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOQV
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KORM
k~=^~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KORM
q=`~~=_~=m =K=K=K=K=K=K=K=K=K=K=K=K=K=K=KORM

OOKPKO=
OOKQ=
OOKR=
OOKS=

^=cbcilt=j K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ORN


o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ORO
k~=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ORR
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ORR

OPK=e=c==~=r~~=p=`K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K ORT
OPKN=
OPKO=
OPKP=
OPKPKN=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KORT
d=b~K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KORT
k~=^~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KORU
i~==~K=m=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KORU

OPKPKO=
OPKQ=
OPKR=
OPKS=

^=cbcilt=j K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= ORV


o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OSM
k~=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OSN
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OSN

OQK=t~=~=e~=bW=p=q~==~=qJJ
~=c~Jj~=pK=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K OSP
OQKN=

p~===m K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOSP

OQKO=

d=b~ K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OSQ

cbcilt==

`
OQKP=
OQKQ=

k~=^~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOSQ
o =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOSS

OQKR=
OQKS=

k~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OSV
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OSV

ORK=qJ~=^=~=m~==d~=t=K=K=K=K= OTN
ORKN=
ORKNKN=
ORKNKO=
ORKO=
ORKOKN=

p~==m=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOTN
b~=NW=p=j==qJi~=`=
^ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOTN
b~=OW=j==kJi~=o~=aJ
=^=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOTO
k~=^~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOTP
b~=N K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=KOTP

ORKOKO=
ORKP=
ORKPKN=
ORKPKO=
ORKQ=
ORKR=
ORKS=

b~=O=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OTR
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OTT
b~=N=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OTT
b~=O=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OTV
` =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OUP
k~ =K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OUP
o K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OUQ

p=f K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OUR


^=f=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K=K= OVN

==o=j~~

m~=f

m~=_~==
j

cbcilt==NN

NO==o=j~~=J=m~=f

_~=b~

_~=b~

NKN `~=m
The core of the mathematical modeling is formed
by the fundamental physical principles of

density function and ( t ) as an intensive balance


quantity. Thus, for an arbitrary volume it gives:
G(t) =

g ( t ) dU

( t ) dU

(1-1)

mass conservation of fluid and solid continua


mass conservation of contaminants and chemical
constituents
momentum conservation of fluid and solid continua
energy conservation (first law of thermodynamics).

All following considerations are related to a spatially fixed Eulerian coordinate system. Accordingly,
the postulate of conservation of property G is stated as:

Mass, motion and energy-related quantities can be


defined in a microscopic (local) volume element
(continuum) denoted by U, for which balance laws are
postulated. Mass, linear momentum and energy represent extensive properties G ( t ) (i.e., quantities are additive over volumes), which are dependent on the volume
of the continuum. On the other hand, intensive properties g ( t ) = ( t ) concern densities of these extensive
properties being independent of the balance volume in
form of mass densities, momentum densities and
energy densities. In this context is introduced as a

where f corresponds to the specific rate of temporary


production (supply) of G and

DG
D--------- ---- dU =
Dt Dt
U

f ( t ) dU

(1-2)

G
D-
----= + vi
xi
Dt t

(1-3)

is the material (substantial) time derivative, in which


G
v i represents the velocity vector of the considered particle associated with the property G.
For all vectorial and tensorial quantities having
cbcilt==NP

NK=_~=b~
repeated indices Einsteins summation convention will
be generally employed throughout the paper, e.g.,
G
G
G
G
v1
+ v2
+ v3
vi
xi
x1
x2
x 3 in the three-dimensional case.

by diffusion laws. Assuming a linear continuous relationship between the diffusive flux of G and the massweighted (barycentric) movement the diffusive flux
can be defined as

The balance law (1-2) can be expressed by using the


Reynolds transport theorem (e.g. Bear4) as follows:
D---- dU =
Dt
U

v i
D ( )
- dU =
- + ------- ---------------x i
Dt

(1-4)

G
-------------( )-
+
( v i ) dU =
xi
t

f dU

where v i represents the barycentric (mass-weighted)


velocity.

j i
-------------( )-
+
( v i ) +
= f
xi
t
xi

Thus

related to a miscroscopic (local) continuum, where


individual balance laws for , j i and f of conserving
mass, momentum and energy are summarized in Tab.
1.1:

Generally, the velocities v i of property G are not


explicitly measurable and have to describe implicitly
Table 1.1 Microscopic balance laws

Mass
Fluid mass
Contaminant mass
Momentum
Energy

NQ==o=j~~=J=m~=f

(1-7)

(1-5)

because the balance expression becomes independent


of the volume U.

Quantity

(1-6)

Introducing (1-6) into (1-5) the conservation principle results in the general form:

G
-------------( )-
+
( v i ) = f
xi
t

j i = ( v i v i )

ji

1
k

0
j ik

0
rk

vi

ij

gi

1
E + --- ( v i v i )
2

ij v j + j iT

gj vj + QT

NKO=j~=`~=b~
Here i, j = 1, , NDM are the indices of the Eulerian
spatial coordinates (summation convention!) with
NDM = 3 for three-dimensional (3D) processes,
k = 1, , N represents the index to indicate a chemical
species (component) with N as the total number of
encountered species, k = C k is the mass fraction
with Ck as the concentration of contaminant species k ,
j ik corresponds to the mass flux vector of the component k , r k is a partial chemical reaction rate, ij is the
stress tensor, g i is the external supply of momentum
(e.g., gravity field), E is the internal energy density, j iT
is the heat flux vector, and Q T is the external supply
due to heat and mass diffusion (for more see
Diersch14,19, Hassanizadeh37, Bear and Bachmat6).

NKO j~=
=b~

`~J

The transformation of the local microscopic balance


equations as formulated in Section 1.1 to a porous or
fractured medium is performed by spatial averaging
procedures referred to the representative elementary
volume (REV) as a heterogeneous medium domain
which is composed by M phases (Fig. 1.1). In passing
from the microscopic to the macroscopic level, often
termed as macroscopization, a number of mathematical
and physical constraints must be satisfied. In the sense
of the ergotic theory the most important assumption
concerns the requirement to the existence of such a
REV. These aspects are thoroughly discussed in
Diersch19. The REV has to be sufficiently large to disappear fluctuations of microscopic properties. On the
other hand, it has to be appropriately small to possess
local variability of particular macroscopic quantities.

Phases = 1, , M are regarded as material subdomains separated by phase interfaces (e.g., solidsolid, fluid-solid, fluid-fluid) where the system properties undergo an abrupt change (discontinuity). Each

phase is composed by N miscible chemical species


characterizing a molecular mixture of throughout distributed several identifiable chemical components

k = 1, , N , , N , for which abrupt interfaces are not


generally present. By definition a chemical species k
only exists in one phase or, if passing through different phases, it is regarded as a phase-pertinent separate
(new) constituent, accordingly it holds N = N (see

also Section 1.3).


The balance equation (1-7) is invalid at a surface of
discontinuity. At an --interface the balance yields:
( ( w i v i ) + j i )

+ ( ( w i v i ) + j i )
, = 1, , M ;

where wi

ni

ni

(1-8)
= 0

is the velocity of the interface and

= n i is the unit normal vector pointing out of


the -phase and into the -phase.
ni

cbcilt==NR

NK=_~=b~

fault zone
porous aquifer

karstic aquifer

fissured rock
fracture matrix
rock mass

multiphase system

microfractures

fractured rock masses

macrofractures

Figure 1.1 Representation of averaging volumes (REV) in the hydrogeologic system (modified from Diersch14
and Kolditz and Zielke50).

As the result, a set of macroscopic balance equations can be derived. After a series of appropriate simplifications (e.g., Diersch19) it leads to the following
individual balance laws:
Mass conservation of a phase:
( ) + ( v ) = Q
i

t
xi

(1-9)

NS==o=j~~=J=m~=f

k ij p


v i + ------------ --------- g j = 0

x j

(1-11)

Energy conservation of a phase:




j
( E )+
( vi E ) +
( )
t
xi
x i iT

Contaminant mass conservation of a phase:





( C )+
( v C )+
( j ) = Rk
x i ik
t k
xi i k

Momentum conservation of a phase:

= QT
(1-10)

(1-12)

NKO=j~=`~=b~
The above balance relationships lead to a closed
system of equations if additional constitutive equations
and constraints are supplemented. Important constitutive relations refer to:
Fickian mass flux governed by the Scheidegger-Bears
dispersion approach (Bear3,4; Scheidegger62):

mol
( D ij

Rk

(1-16)

and density and viscosity relationships according to:

(1-13)

= ( p , Ck , T )

and

(1-17)

= ( p , Ck , T )

as well as balance expressions at interfaces resulting


from the spatial averaging process of the jump condition (1-8) as exemplified for a free surface in form of a
mass conservation of a phase at a macroscopic surface
of discontinuity, viz.:

disp C k
D ij ) ---------x j

= 0

k=1

j ik = j ik ( , v i , C k )
+

v i v j C k

- ---------= ( D d + T V ) ij + ( L T ) ----------
x j
V

( ( wi vi ) )

Fourierian heat flux by a generalized thermodispersion


approach:

ni

n
= Q
i
i

(1-18)

where

j iT =


j iT ( , , v i ,
cond

= ( ij

disp

+ ij

T )

(1-14)

T
) --------x j

v i v j T


- --------= ( + c T V ) ij + c ( L T ) ----------
x j
V

Macroscopic reaction rate for homogeneous and heterogeneous chemical reactions of component k:

het

Rk = rk + Rk

satisfying its balance constraint

(1-15)

volume fraction of -phase, where


0 1 and = 1 ;

= density of -phase;

vi
= velocity vector of -phase;
wi
= macroscopic velocity of phase interface;

Ck
= concentration of component k in the -phase;

k ij
= permeability tensor of -phase;

p
= -phase pressure;

= dynamic viscosity of -phase;


gi
= gravity vector;

E
= internal (thermal) energy of -phase;

Q , Q T = mass and heat supply of -phase,


respectively;

Q
= mass supply through a --interface (positive
i
=

cbcilt==NT

NK=_~=b~
mol
D ij
disp

D ij
ij

Dd
L, T

=
=
=
=
=

V
=

T
=
cond
ij
=
disp

ij

c
=

=
L, T =

rk

het

Rk

=
=

outward from to );
tensor of molecular macrodiffusion of phase;
tensor of mechanical dispersion of -phase;
Kronecker tensor;
molecular medium diffusion of -phase;
longitudinal and transverse dispersivity of
chemical component, respectively;

v i v i absolute flux of -phase;
-phase temperature
conductive part of thermodispersion tensor of
-phase;
dispersive part of thermodispersion tensor of
-phase;
heat capacity of -phase;
heat conductivity of -phase;
longitudinal
and
transverse
thermodispersivity, respectively;
homogeneous reaction of component k in the
-phase;
heterogeneous reaction of component k.

The momentum conservation equation (1-11) appears


here as a generalized form of the well-known Darcy
equation (Bear4).

1r A 1 + 2r A 2 + + Nr A N
(1-19)
( N + 1 )r B ( N + 1 ) + ( N + 2 )r B ( N + 2 ) + + Nr B N

for N < N

( N = number of reactants)

which is related and quantified by the stoichiometric


coefficients kr . The algebraic stoichiometric numbers kr behave:
kr < 0 for 1 k N (reactants)
kr > 0 for N < k N (products).

The chemical equilibrium takes the conservation condition:

k Rk

= 0

(1-20)

where k corresponds to the chemical potential of the


kth species in the -phase. Introducing the chemical
standard potential and defining an useful equilibrium

constant K eq = K eq ( , , T ) the conventional form
of the law of mass action (LMA) applied to the reaction
equation (1-19) for equilibrium states results:
N

NKP `~=
The modeling of chemical reactions needs knowledge about the interacting components (species) participating in chemical reactions, reactants A k , and about
the resulting products B k expressed by the gross reaction equations for a given chemical reaction r in general form:
NU==o=j~~=J=m~=f

K eq =

kr

ak
k=1

[ Bk ]

kr

k = N + 1
= --------------------------------------N

[ Ak ]

(1-21)

kr

k=1

where a k and square bracket [ ] represent activities


of corresponding dissolved species, respectively.

NKQ=j=^~=~=p~
Chemical nonequilibrium reactions are based on the
assumption that in each reaction step a transition state
#
occurs, the so-called activate complex X r , which is in
equilibrium with the reactants A k , viz.:
1r A 1 + 2r A 2 + + Nr A N

#
Xr

(1-22)

( N + 1 )r B ( N + 1 ) + + Nr B N

and can be characterized by an equilibrium constant as


#

[ Xr ]
#
K eq ( r ) = ---------------------------N

[ Ak ]

(1-23)

kr

k=1

The macroscopic bulk rate R k of reaction r may be


expressed by a reaction rate R r :
(r)

Rk Rk

= kr R r

(1-24)

Using Eqs. (1-23) and (1-24) it yields the following


relationship for R r :
N
#

# #

v K eq ( r )

Rr = v [ Xr ] =

b
#

[ Ak ]

kr

(1-25)

k=1
S

where v is the decay frequency and b represents a


reaction rate constant. The latter can be expressed by
the well-known phenomenological Arrhenius equation
given in the form

E
S
b = U S exp -------
RT

(1-26)
#

where U S is the frequency factor and E is known as


activated energy ( R is the universal gas constant).

NKQ j= ^~= ~
p~
For the analysis of three-dimensional and twodimensional flow and transport problems a number of
simplifications in the modeling approach are suited, on
the one hand, to govern mathematically and numerically the complex processes and, on the other hand, to
attend to intrinsic practical needs. The major
approaches and assumptions currently encountered in
the simulation system FEFLOW will be stated in the
following paragraphs to relate the present theoretical
framework to the basic model equations actually
implemented in FEFLOW package and as summarized
in Chapter 3.

NKQKN

e~

The groundwater body (saturated porous medium)


can be described as a 2-phase system (fluid-solid). The
phase indicator ranges then over = s, f , where s indicates solid and f denotes fluid. In the present context
fluid is represented by water.
The phase volume fractions are now defined
according to: = f and s = 1 , where represents the porosity of the porous medium. Additionally,
following relations may be appropriately introduced:
cbcilt==NV

NK=_~=b~
f

The hydraulic head h (piezometric head) related to


f
the reference fluid density o is defined as

= ( h, C, T )

(1-31)

which will be thoroughly discussed in Section 1.6.

p
- + xl
h = -------f
o g

(1-27)

NKQKO

under assuming exclusive action of gravity (cf. Eq. (111))


g j = ge j

(1-28)

where the subscript l indicates the direction of gravity


action in the Cartesian coordinate system, g is the gravitational acceleration, and e j denotes the gravitational
unit vector.
f

The Darcy velocity (flux) q i is related to the fluid


f
phase velocity v i (often referred to as pore, intrinsic or
kinematic velocity) by
f

q i = v i

(1-29)

The hydraulic conductivity tensor related to the reff


f
erence fluid density o and the reference viscosity o
is defined by
f

k ij o g
K ij = -------------f
o
f

(1-30)

The fluid density is regarded as a function of the


hydraulic head h (or the pressure), the concentration C
of one (or more) component(s), and the temperature T:
OM==o=j~~=J=m~=f

`~

Let us focus on a simplified chemical reaction system. On the one hand, reversible binary ion exchange
z
z
reactions between species A 11 and species A 22 take the
form:
z1

z2

z1

z2

1 A 1 ( aq ) + 2 A 2 1 A 1 + 2 A 2 ( aq )

(1-32)

where aq indicates the component of the aqueous phase


f, the overbar symbols designate the sorbed components of solid phase s (exchange phase), and z k is the
charge number of species (ion) k.
On the other hand, each of these components should
be subjected to irreversible decay (disappearance)
reactions according to
A k products

(1-33)

For chemical activities following relations are valid (at


preference of mass-specific quantities rather to molar
concentrations):

fk f

zk
1
------------------------[ Ak ] =
C

mk k
Cq
-------
mq

(1-34)

NKQ=j=^~=~=p~

zk

s
1
1
[ A k ] = ------ k --------------------

mk

Cq
-------
mq

(1-35)

where f k corresponds to the activity coefficients (correcting factors against ideal mixtures), which can be
assumed with f k = 1 (ideal-dilute solution) for the
sorbed phase, and m k is the molecular weight of species k. The mass fractions sorbed on the solid phase k
s
are related to the solid phase concentration C k via the
s
ion exchange capacity C T in form of
s

12
K eq

k = Ck CT

(1-39)

In the case of monovalence 1 = 2 = 1 the sorbed


species may be explicitly expressed by the dissolved
quantities in the fluid phase (restrictedly also for heterovalent chemical equilibrium reactions as more
shown in Diersch19). Using the Eqs. (1-37), (1-38) and
(1-39) the well-known (nonlinear) Langmuir adsorption isotherm results directly from Eq. (1-39):
s
C1

f
f ( C1 )

(1-36)

where

s s

( C1 CT ) ( f2 C2 )
= --------------------------------------------------------2
f 1
s
s s
( f1 C1 ) ( C2 CT )

12

C1
f 1 K eq C T
= -------------------- -----------------------------------------------f2
f
f
f
1 12
C T + ---- K eq 1 C 1
f2

(1-40)

written in the more common form


Ns
s
CT

Ck

k1 C1
s
f
C 1 = f ( C 1 ) = --------------------f1 + k2 C1

(1-37)

12

when N s 2 denotes the total number of sorbed species. Moreover, the total solution normality for the
fluid phase f can be introduced as:
f

(1-41)

f 1 12
f
k 2 = ---- K eq 1 C T
f2

Admitting for low concentrations

Ns

CT =

f 1 K eq C T
-,
k 1 = ------------------f
f2 CT

Ck

(1-38)

k2 C1 1

(1-42)

k
s

Assuming isothermal conditions C T and C T represent


constant (measurable) capacities. The LWA equilibrium condition (1-21) leads after simplification to

the Langmuir isotherm (1-41) immediately reduces to


the well-known linear Henry adsorption isotherm:
s

C 1 = f ( C 1 ) = C 1

k1

(1-43)

cbcilt==ON

NK=_~=b~
where with the Henry sorptivity coefficient is introduced. Otherwise, for high concentrations the sorbed
s
concentration C 1 for the Langmuir isotherm tends to
the limit k 1 k 2 .
In the case of heterovalent equilibrium reactions
1 > 1, 2 > 1 polynomial relationships can be
s
f a
derived C 1 f ( C 1 ) (see Diersch19), where a > 1 repre-

sents a corresponding exponent. An empirical approach


joining such a power character of the exchange reactions provides the so-called Freundlich adsorption isotherm written in the form:
s

f b2

C1 = f ( C1 ) = b1 ( C1 )

(1-44)

where b 1 and b 2 are corresponding coefficients and


exponents, respectively. In case of b 2 = 1 the Freundlich adsorption isotherm reduces to the linear Henry
sorption relationship having b 1 .
Since a chemical component exists only in one
phase by definition, the symbolic formalism for identifying components may be unified as follows:
C1 =

f
C1

C2 =

s
C1

(1-45)

For the fluid phase and the rigid solid phase the following mass balance equations of a chemical component k
hold, respectively:
f
C
qf
( 1) +
( C )+
( j ) = R 1
t
x i i 1 x i i1

OO==o=j~~=J=m~=f

( 1 )C
(
2 ) = ( 1 )R 2
t

(1-47)

Using the above adsorption isotherms (1-41), (1-43)


and (1-44) as well as the definition (1-45) following
transport equation results after summation of Eqs. (146) and (1-47):
f
RC
qf
(
( C )+
( j ) = R 1 + ( 1 )R 2 (1-48)
1) +
t
x i i 1 x i i1

introducing the so-called retardation factor R, which is


defined as
(1 )
R = 1 + ---------------- ( C 1 )

(1-49)

where the adsorption function ( C 1 ) from the equilibrium formulae (1-43), (1-41) and (1-44) with
C 2 = ( C 1 ) C 1 emerges the different adsorption laws
according to:
Henry isotherm:
( C1 ) =

(1-50)

Freundlich isotherm:
( C1 ) = b1 ( C1 )

b2 1

(1-51)

Langmuir isotherm:
(1-46)

k1
( C 1 ) = --------------------1 + k2 C1

(1-52)

NKQ=j=^~=~=p~
Employing Eqs. (1-24) and (1-25) in such a form where
it is assumed the bulk rate R k can be splitted into a
and a zero-order pro-

duction term QCk according to


(r)

R k = R k + Q Ck

(r)

= C k

k = 1, 2

for

(1-54)

where is the decay rate) of component k, and the


)
Referring to radioactive decay processes the decay rate is
frequently expressed in terms of a reaction half-life t 1 2 , which
is a specific solution of the reaction equation

dC k
= C k
dt

applied to a simple chemical batch reaction (without diffusion/


dispersion and advection). Its analytical solution yields
C k = C ok e

(1-55)

Accordingly, the nonequilibrium reaction terms reduce


to R 1 = C1 + QC1 and R2 = C2 + QC2 = ( C1 )C1 + QC2 ,
and the transport equation (1-48) takes finally the form

(1-53)

the term the monomolecular decay reaction (1-33)


leads to a linear expression:
Rk

R k = C k + Q Ck

mol f
disp f C

( RC ) q f
( C)
+
( D ij + D ij ) ------ xi
x j
t
xi i

(1-56)

(r)

higher-order reaction rate R k

bulk rate R k leads to the expression

D ij
+ RC = Q C

kI= = ~
~= = ~=
where C = C1 is the concentration of the essential
= =
transport component in the fluid (aqueous) phase and ~=~J= t 1 2 ==
Q C = Q C1 + ( 1 )Q C2

ln2
(1-57) W= = --------K
t1 2

represents a zero-order production term encompassing


the fluid and solid phase.

NKQKP

q~

The time t for the concentration C k to decrease from the initial


concentration C ok to its half value 1--2- C ok is the half-life t 1 2 .
From above it results
ln2
t 1 2 = -------

where ln2 = 0.693 corresponds to the natural logarithm of 2.


Accordingly, the decay rate can be expressed by
ln2
= --------t1 2

using the half-life t 1 2 as an (alternative) input parameter


related to a (radioactive) component k .

For the fluid phase f and the rigid solid phase s the
individual energy conservation equations are in the following forms, respectively:
f f + f f f
( E)
( vi E )
t f
xi f

(1-58)

disp f T
f f
cond f
(
+ ij ) ------- = f Q T
x j
x i ij

cbcilt==OP

NK=_~=b~
s
T-
s s
s s
conds ------( s E )
ij
= s QT
x j
t
xi

supply expressions, respectively, naturally result:


(1-59)
cond f

ij = ij

cond s

+ ij

disp f

+ ij

(1-63)

Assuming a thermal equilibrium between the phases s


and f it holds:
f f

(T T) = 0

= s, f

(1-60)

qi qj
f
s
f
f f
[ + ( 1 ) ] ij + c T V q ij + ( L T ) -------f
Vq
conductive
dispersive

for

where T is the system temperature. Moreover, it is

assumed the state of internal energy E of the phases


should be primarily a function of temperature (see also
Section 1.6):
dE

E
E


= --------d + ---------- dC k + --------dT c dT

T
k C k

for

(1-61)

where V q =

(1-64)

f f

q i q i designates the absolute Darcy flux.

= s, f

NKR ^J~~=

E
with the heat capacity c = --------- of -phase.

Accordingly, the heat transport equations (1-58) and


(1-59) can be combined (added). As a result, energy
equations for each phase need not be considered.
Rather the total energy conservation equation is preferred. After summation of Eqs. (1-58) and (1-59) it
yields:
s s
f f
f f f
[ E + ( 1 ) E ] +
( qi E )
t
xi
T
ij ------- = Q T
x i x j

(1-62)

In forming the above total energy conservation equation (1-62) the following thermodispersive and heat
OQ==o=j~~=J=m~=f

Q T = Q T + ( 1 ) Q T

b~J

Transport process modeling in aquifers possesses


important special cases, in which the horizontal extent
of a regional flow field much extends the thickness B
of an aquifer. For such conditions vertical variations
can be often neglected to reduce the full 3D equations
to two-dimensional (2D) essentially horizontal relationships. Regarding groundwater hydraulic processes
this procedure is associated with the well-known
Dupuit assumption (Bear4,5).
There are two distinctly different approaches to
integrate vertically the macroscopic balance equations.
A two-step averaging procedure (first step represents a
REV averaging, second step is the vertical averaging)

NKR=^J~~=b~
has been found to be detrimental and inappropriate for
general transport phenomena as discussed in
Diersch19). Alternatively, a one-step aquifer averaging
procedure proposed by Gray36 will be preferred to
develop the representative 2D model equations for horizontal flow and transport processes in a general and
much easier way.
The microscopic balance equations (1-7), Tab. 1.1,
form the starting point. They are spatially averaged for
an aquifer REV (so-called AREV) as schematized in
Fig. 1.2 (see Diersch19) for more details).
dS

Vertically averaged contaminant conservation of a


phase:



( BC ) +
( Bv C ) +
( Bj ) = BR k (1-66)
x i ik
t k
xi i k

Vertically averaged momentum conservation of a


phase:

Bvi

k ij B p

+ ------------ --------- g j = 0

(1-67)

Vertically averaged energy conservation of a phase:


( BE ) + ( Bv E )
i
t
xi

Bj
+
( iT ) = BQ T
xi

(1-68)

For the horizontal equations the following definitions are appropriately introduced:
x3

x2
x1

Depth-integrated Darcy flux:


D
f

As the result the macroscopic horizontal aquifer


balance equations are given as:
Vertically averaged mass conservation of a phase:



( B) +
( Bv i ) = BQ
t
xi

q i = Bq i = Bv i

Figure 1.2 Aquifer averaging volume (AREV).

(1-65)

(1-69)

Transmissivity:
f

k ij B o g
T ij = BK ij = -----------------f
o

(1-70)

Moreover, gravity actions are not operative for horicbcilt==OR

NK=_~=b~
zontal processes. Analogously to the simplifications
done in Section 1.4, the resulting balance equations for
essentially horizontal flow and transport processes are
as follows:
f
f B
f
f f
B ( ) + ------ +
( qi ) = Q
t x i
t

(1-72)

1 f
----f -------
h

dC +

1 f
----f -------
T

( Cs Co )

dT

(1-76)

where is the fluid compressibility, is the fluid density difference ratio, which is normalized by the difference ( Cs C o ) of the maximum concentration Cs and
the reference concentration Co , and is the volumetric thermal fluid expansion coefficient to be introduced
as intensive (volume-independent) material coefficients. If we assume , ( C s C o ) and are constants the integration of Eq. (1-76) immediately leads
to the state equation for the fluid density in the common form
f

f
o

( h h o ) + ----------------------- ( C C o ) ( T T o )
( Cs Co )

(1-77)

o 1 + ( h h o ) + ----------------------- ( C C o ) ( T T o )
( Cs Co )

f
f

Generally, the fluid density should be regarded


OS==o=j~~=J=m~=f

1 f
----f -------
C

(1-74)

The derived balance equations (1-9) to (1-12), (156), (1-62), (1-65) to (1-68) and (1-71) to (1-74) reveal
dependencies upon state variables which have to be
more investigated. In this context the relationships for
f
the fluid density , the porosity , the internal energy

density E , the aquifer thickness B, and the dynamic


f
viscosity of the fluid are of further interest:

c==

dh +

f
f
f
= dh + ----------------------- dC dT
( Cs Co )

NKS b~==p~

NKSKN

(1-73)

+ RBC = Q C
s s
B f f
f f f
[ ( E + ( 1 ) E ) ] +
( qi E )
t
xi
T
B ij ------- = Q T
x j
xi

f
d = ------- dh + ------- dC + ------- dT
h
C
T

disp f C
( RBC ) + q f B D mol f
( C)
( ij + D ij ) ------ xi
x j
t
xi i

(1-75)

The total differential yields then


f

h
= T ij ------x j

= ( h, C, T )

f
qi

(1-71)

as a function of the hydraulic head h, the concentration


C, and the temperature T, viz.,

where suitable reference values appear for the fluid

NKS=b~==p~
f

density o = h o, C o, T o as the reference hydraulic head


h o , the reference concentration C o , and the reference
temperature T o . The above introduced material coefficients may be estimated by appropriate relations as
exemplified for the density difference ratio )
f

= [ ( Cs ) o ] o

(1-78)
f

A first guess of the fluid density at maximum concentration C s can be given by the relation (Baxter and
Wallace1)
f

( C s ) o + 0.7 C s

(1-79)

which leads to an estimation of density difference ratio


in form of
0.7 C s
----------------f
o

in the range between 0 and 100 C .


As a consequence the volumetric thermal fluid
expansion relation cannot generally be considered as
a constant coefficient and a more general approach is
desired. Perrochet57 proposed a formula for computing
a nonlinear temperature-dependent expansion coefficient ( T ) which is based on the physically exact
fluid density curve (Fig. 1.3) measured for temperatures in the range between 0 and 100 C and represents
a correction parameter for the state equation (1-77) by
introducing higher-order terms. This approach has been
implemented in FEFLOW as the option termed as thermally variable fluid density expansion.
Perrochet57 has fit the temperature-dependent fluid
f
density ranging between 0 and 100 C by a 6th
order polynomial with a high accuracy as

kI= =
~J
~~== den-

s=
=

= = ~= ~
~= = =
= = ~= I
= = ~J
~~= = = ~
= ~= J
= ~= = ~
= ~I= ~J
~==~=~=~J
= ~= ~= =
= L~= = Q
C =~~===
f
E~F= = = ~
~= ~= = EKKI=
~= ~=
=e40FK

(1-80)

While a linear relationship for the fluid density difference ratio ( C s ) according to (1-78) and (1-80) is
normally sufficient for the most practical needs the
consideration of the volumetric thermal fluid expansion coefficient as a constant parameter becomes
inappropriate for cases where a larger temperature
range has to be modeled or/and where the 4 C anomf
aly in the fluid (water) density can play a role. This
can be seen in Fig. 1.3 for the physically true water
f
density dependency as a function of temperature T
)

Commonly, > 0 for a heavier than water solution (e.g., saltwater), however, can even be negative if f ( C s ) < fo in the
case of a lighter than water solution (e.g., floating of a dissolved
organic compound). Accordingly, a negative parameter can be
useful and is also accepted by FEFLOW.

cbcilt==OT

NK=_~=b~
1000

b =

c =
d =

990

6.764771 10

3
8.993699 10

9.143518 10

7
8.907391 10

5.291959 10

11
1.359813 10

a = 9.998396 10

e =
f =
g =

(1-82)

980

f [g/l]

where the temperature T is in C and a represents the


fluid density at T = 0 . Taking also into consideration
of concentration and pressure dependencies we can
argue the measured curve (1-81) is related to a reference concentration Co and reference hydraulic head
h o . Indeed, the coefficients (1-82) has been derived for
a freshwater condition C o = 0 . Introducing a reference
temperature T o we can then obtain directly from (1-81)
f
an expression for the reference fluid density o , viz.,

1000

970

999

960

10

variable expansion
linear expansion

950

20

40

60

80

100

[oC]

Figure 1.3 Fluid density as a function of temperature T in the range of 0 and 100 C , close-up view
indicates the density anomaly (from Perrochet57).

o = ( h o, C o, T o )
2
= a h , C + b h , C To + c h , C To +
o
o
o
o
o
o
4
5
6
+ e h , C To + f h , C To + g h , C To
o
o
o
o
o
o

(1-83)
d

3
T
h o, C o o

To find the thermally variable fluid density expansion


( T ) of the fluid density function
2

with the coefficients for water

(1-81)

f
f
= o 1 + ( h h o ) + ----------------------- ( C C o )
( Cs Co )

(T)

( T ) = a + bT + cT + dT + eT + fT + gT
in [g/l]

(1-84)

( T To )

variable expansion

we follow the derivation given by Perrochet57. Let us


consider a Taylor series expansion for the fluid density
OU==o=j~~=J=m~=f

NKS=b~==p~
f

around T o , C o and h o :

f
f
( T, C, h ) = o 1 + ----------------------- ( C C o ) + ( h h o )
( Cs Co )

f
o
2 f

1
( T T o ) + --- ---------22 T
T o, C o, h o

3 f

4 f

1
( T T o ) + ------ ---------424 T
3

T o, C o, h o

T o, C o, h o

( T To ) +
T o, C o, h o

( C C o ) + ------h

T o, C o, h o

T o, C o, h o

f
o ----------------------( Cs Co )

( h ho )

+ ( d + 4eT o + 10fT o + 20gT o ) ( T T o )


2

+ ( e + 5fT o + 15gT o ) ( T T o )
+ ( f + 6gT o ) ( T T o )

T o, C o, h o

+ ( c + 3dT o + 6eT o + 10fT o + 15gT o ) ( T T o )


2

5 f

1
4
( T T o ) + --------- ---------5120 T

( T T o ) + -----C

6 f

1
--------- ---------720 T 6

( T To ) +
T o, C o, h o

1 --------
--- 6 T 3

+ ( b + 2cT o + 3dT o + 4eT o + 5fT o + 6gT o ) ( T T o )


f

f
( T o, C o, h o ) + ------T

( T, C, h ) =

(1-87)

+ g ( T To )

(1-85)

where a 6th order approximation is used for the temperature T while only a linear 1st order approximation
is deemed to be sufficient for C and h dependencies. By
utilizing the Eq. (1-81) we can evaluate the above
n f
n
T at T o , C o and h o according to

From Eqs. (1-87) and (1-84) we finally obtain the


expression for computing the nonlinear thermally variable fluid density expansion to be used in Eq. (1-84)
1
( T ) = ----fo

( b + 2cT o + 3dT o + 4eT o + 5fT o + 6gT o ) +


2

( c + 3dT o + 6eT o + 10fT o + 15gT o ) ( T T o ) +


2

( d + 4eT o + 10fT o + 20gT o ) ( T T o ) +


f

-----T

=
T o, C o, h o

2 f

--------2
T

=
T o, C o, h o

3 f

--------3
T

=
T o, C o, h o

4 f

--------4
T

=
T o, C o, h o

5 f

--------5
T

=
T o, C o, h o

6 f

--------6
T

=
T o, C o, h o

2
3
4
5
b + 2cT o + 3dT o + 4eT o + 5fT o + 6gT o

2
3
4
2c + 6dT o + 12eT o + 20fT o + 30gT o

2
3

6d + 24eT o + 60fT o + 120gT o

24e + 120fT o + 360gT o

120f + 720gT o

720g

( e + 5fT o +

2
15gT o ) ( T

(1-88)

To ) +
4

( f + 6gT o ) ( T T o ) +
g ( T To )

(1-86)

with o computed from (1-83) and the coefficients


( a, b, c, d, e, f, g ) as given in (1-82).
As the consequence from the nonlinear thermally
f
f
variable fluid density expansion the term ( 1 ) T
in Eq. (1-76) is not longer a constant, . Instead we
have

As the result the fluid density (1-85) gives


cbcilt==OV

NK=_~=b~
( T )
( T ) + --------------- ( T T o )
f
1

T
---- ------- = -------------------------------------------------------------------------------------------f T

1 + ----------------------- ( C C o ) ( T ) ( T T o )
( Cs Co )

(1-89)

where here effects of fluid compressibility can be


neglected. The derivation of the nonlinear thermally
variable fluid density expansion in Eq. (1-89) becomes

E
E dT
= --------d + ---------- dC k + --------

T
k C k

(1-93)

( T )
1
--------------- = ----- [ ( c + 3dT o + 6eT 2o + 10fT 3o + 15gT 4o ) +
f
T
o

2 ( d + 4eT o + 10fT o + 20gT o ) ( T T o ) +


2

The internal energy is generally a state function of

the density , the concentration C k , and the tempera


ture T of -phase:

dE

f~==EF= E

NKSKP

(1-90)

3 ( e + 5fT o + 15gT o ) ( T T o ) +

= s, f

for

4 ( f + 6gT o ) ( T T o ) +
5g ( T T o )

NKSKO

Commonly, it is deemed to be sufficient that the inter


nal energy E only represents a state relation with

respect to the temperature T . Thus

m=

In changing the porosity of the porous medium


compression work of the skeleton (deformation) should
be taken into account, where concentration and temperature effects remain disregarded. Thus
= (h)

(1-91)

The differential becomes

(1-92)

1 dh = ( 1 )dh
d = ----- dh = --- ----h
s h s

where represents the coefficient of skeleton compressibility.


PM==o=j~~=J=m~=f

dE

= c dT

for

= s, f

(1-94)

where c corresponds to the heat capacity of -phase.

NKSKQ

^==_

The thickness of an aquifer B = B ( x i, t ) can be


mapped as an interface equation. Since the interface
may be regarded as a material surface of discontinuity
occupied by fluid and solid particles, principles of conservation have to be subjected (Bear and Bachmat6):
----BB
+ w i ------- = 0 ,
x i
t

BB- n
-----= -----i
x i
x i

(1-95)

NKS=b~==p~
where wi is the velocity of surface B and n i denotes
unit normal vector pointing out of B.
Assuming B is a function of hydraulic head h, for
unconfined (phreatic) aquifers the following approach
retains:
B
dB = ------ dh = dh
h

(1-96)

( C ) = o ( 1 + 1.85 4.1 + 44.5 )


with

= C

(1-98)

For geothermal processes the thermal effects in the viscosity have to be necessarily incorporated. An empirical relationship for water is taken from Mercer and
Pinder54 as
3

where 0 1 represents an inhibition factor (ideally


1.0), which corrects either the fillable void space during an aquifer filling period or the drainable void space
during a drainage process in the aquifer (see also Section 1.7). For confined conditions it becomes 0 , i.e.,
the thickness B is stationary.

NKSKR

a~====

Appropriately, the viscosity of the fluid phase is


regarded as a function of concentration C and the temperature T, viz.,
f

= ( C, T )

1 + 0.7063 0.04832
1 -----------= ----------------------------------------------------------f
f
(T)
o
with

Regarding the concentration dependency, especially


for high-concentration saltwater, Lever and Jackson52
as well as Hassanizadeh38 proposed the following relationship:

at T

in C

The reference viscosity o in the constitutive equations (1-98) and (1-99) is related to reference values for
concentration
and
temperature
with
and
C = 0 or = 0 ,
T = 150C or = 0 ,
respectively. A combination of both influences yields
the following expression regarding the reference visf
cosity o :
f

3
o
1 + 0.7063 0.04832
------------------- = -------------------------------------------------------------------f
2
3
1 + 1.85 4.1 + 44.5
( C, T )

(1-97)

The viscosity dependencies are developed by using


empirical polynomial relationships as can be found in
the literature.

( T 150 )
= ----------------------100

(1-99)

(1-100)

Employing an arbitrary reference concentration C o and


reference temperature T o a viscosity relation function
f can be obtained, which is related to these proper reference conditions:
f

o
f = ------------------f
( C, T )
=

2
3
1 + 1.85 ( C = Co ) 4.1 ( C = C o ) + 44.5 ( C = C o )
---------------------------------------------------------------------------------------------------------------2
3

1 + 1.85 4.1 + 44.5

(1-101)
3

1 + 0.7063 0.04832
--------------------------------------------------------------------------------------3
1 + 0.7063 ( T = T o ) 0.04832 ( T = To )

cbcilt==PN

NK=_~=b~
f

where o = Co, To is the reference viscosity with


respect to the reference concentration Co and the reference temperature T o .
The consequences of viscosity and density effects to
the input parameter of hydraulic conductivity are more
discussed below in Section 1.8. For instance, the visf
f
cosity relation f = o ( C, T ) is displayed in Figs.
1.4 and 1.5 for the temperature range T between 0 and
300 C and concentrations C between 0 and 200 g/l in
dependence on different reference temperatures T o for
150 C and 10 C (in each case at reference concentrations of C o = 0 used for freshwater), respectively.
Furthermore, Figs. 1.6 and 1.7 exhibit the development
of the temperature and concentration-dependent viscosf
f
ity ( C, T ) related to the reference viscosity o for
selected reference temperatures and concentrations.

2.0

/l]
[g
0
=
C 50
0
10
0
15
200

1.5

o
------------------- 1.0
f
( C, T )
0.5

0.0

100

200

300

T [ C ]
f

Figure 1.4 Viscosity relation o as function of temperaf


ture T [ C ] and concentration C [ g l ] with o for the reference temperature of T o = 150 C and reference
concentration of C o = 0 (freshwater).
15.0

]
g/ l
0[
= 0
C 5
0
10
0
15

10.0
f
o

200

-------------------f
( C, T )
5.0

0.0

100

200

300

T [ C ]
f

Figure 1.5 Viscosity relation o as function of temperaf


ture T [ C ] and concentration C [ g l ] with o for the reference temperature of T o = 10 C and reference
concentration of C o = 0 (freshwater).

PO==o=j~~=J=m~=f

NKT=p~==a=_~~=q

NKT p~= = aJ
=_~~=q

20.0

The balance equations involve terms in form of


derivatives for the density, the porosity and the thickness which have to be appropriately specified for modeling purposes. For the proper conservation principles
the derivation will be advanced individually:

15.0

( C, T ) 10.0
-------------------f
o

C = 200 [g/l]
C = 150 [g/l]
C = 100 [g/l]
C = 50 [g/l]
C = 0 [g/l]

5.0

0.0

100

200

300

T [ C ]
f

Figure 1.6 Relative viscosity o as function of temf


perature T [ C ] and concentration C [ g l ] with o for
the reference temperature of T o = 150 C and reference
concentration of C o = 0 (freshwater).

NKTKN `~= = = ~= EPa


~=Oa=E~I=~FF
The conservation equation (1-9) for the fluid phase
f, starting from the basic form
f
f
f f
( )+
( qi ) = Q
t
xi

2.5

with

(1-102)

Q = Q

2.0

can be written in more detail:


C = 200 [g/l]
C = 150 [g/l]
C = 100 [g/l]
C = 50 [g/l]
C = 0 [g/l]

0.5

0.0

50

100

150

T [ C ]
f

Figure 1.7 Relative viscosity o as function of temf


perature T [ C ] and concentration C [ g l ] with o for
the reference temperature of T o = 10 C and reference
concentration of C o = 0 (freshwater).

f
f
f
f q i
f

f
------- + ----- + q i ------- + ------- = Q
x i
x i
t
t

1.0

( C, T )------------------f
o

1.5

(1-103)

The first term of Eq.(1-103) can be developed by using


the state equation (1-76) as
f

C
T

f h
------- = ----- + ----------------------- ------ ------
t ( C s C o ) t
t
t

(1-104)

cbcilt==PP

NK=_~=b~

(1-106)

Analogously to relationship (1-104), the third term


yields
f

f
f f h

C
T
q i ------- = q i ------- + ----------------------- ------- -------
x i ( C s C o ) x i
x i
x i

(1-107)

Accordingly, the continuity equation (1-103) leads to


the following form as actually realized in FEFLOW:
f
q i

h
S o ----- + ------- = Q + Q EB ( C, T )
t x i

(1-108)

where
S o = + ( 1 )

(1-109)

corresponds to the specific storage coefficient (comPQ==o=j~~=J=m~=f

x i

NKTKO `~= = = ~= EOa


~F
For essentially horizontal problems density effects
resulting from concentration and/or temperature influences do not play a role. Accordingly, the mass conservation equation (1-73) for the fluid phase f becomes
f
f B
f
f f
B ( ) + ------ +
( qi ) = Q
t x i
t

(1-111)

It can be further developed as follows:


f

f
f
f

f q i
f

B- + f -----B ------- + B ----- + f ----q i - + ------- = Q


x i
t
t
x i
t

f
f
h
----- = ( 1 ) ----t
t

represents an additional term to incorporate massdependent and temperature-dependent compression


effects as required for the extended Boussinesq approximation (see the subsequent Section 1.10 for more),
where compression work of pressure has been
f h
neglected q i ------- 0 .

The second term is provided by using the constitutive equation (1-92)

(1-110)

C
T
f

q i ----------------------- ------- -------


( C s C o ) x i
x i

where is a given constant while ( T ) and ( T ) T


correspond to Eq. (1-88) and (1-90), respectively.

T
C

Q EB ( C, T ) = ----------------------- ------ ------


t
( C s C o ) t

for the constant expansion coefficient

(1-105)

( T )
=
( T ) + --------------- ( T T o )
for the thermally variable
T
----------------------------------------------------------------------------------------- fluid density expansion

1 + -------------------(

(
T
)
(
T

T
)
C
C
o
o

( Cs Co )

pressibility) and

with the modified thermal fluid expansion function :

(1-112)

Admitting the compression effects by pressure, analogously to the 3D mass conservation, the first and second term lead to the following expression

NKT=p~==a=_~~=q
f

Neglecting the fourth term in Eq. (1-112) the horizontal fluid mass conservation equation finally results
h
( S o B + e ) ----- + ------- = Q
t x i

NKTKP

(1-115)

a~=~

k ij
k ij o g
----- = --------------
f
f

o
-----f

K ij

11 -------= K ij f -------f
f
o g
o g

(1-119)

The so-called buoyancy term in the Darcy equation (1118) can be expanded as follows:

The momentum conservation equation (1-11) of the


fluid phase f is to be expressed by the hydraulic head h,
according to Eq. (1-27), the gravitational vector (1-28),
f
the Darcy velocity q i , and the hydraulic conductivity
f
f
K ij . From h = p ( o g ) + x l it yields for the fluid pressure
p = o g ( h xl )

(1-118)

where the constitutive equation (1-101) for the


dynamic viscosity has been incorporated in introducing
the hydraulic conductivity according to
f

f
q i

h o
f
- e j
q i = K ij f ------- + --------------f
x j

(1-114)

where e = at 0 1 is the effective, so-called


drainable or fillable porosity. For confined conditions
in an aquifer the effective porosity vanishes e 0 .

(1-117)

Substituting it into Eq. (1-11) the Darcys law in terms


of hydraulic head and under consideration of density
effects takes finally the form

with the specific storage coefficient S o according to


Eq. (1-109). The third term in Eq. (1-112) becomes,
after applying the state relation (1-96)
f B
f h
f
h
------ = ----- = e ----t
t
t

h
f
p-----= o g ------- e j
x j

x j

(1-113)

f
f
h
B ------- + B ----- = B S o ----t
t
t

(1-116)

o
---------------- = ---------------------( C Co ) ( T To )
f
(
C
s Co )
o

(1-120)

where here effects of pressure compression need not to


be regarded for the groundwater flow problems under
consideration.

and accordingly

cbcilt==PR

NK=_~=b~

NKU a=~=s=oJ
= `=
e~=`
In transforming the pressure-permeability-(p-k)form of the Darcy equation (1-11) to a hydraulic-headconductivity-(h-K)-form of the Darcy equation (1-118)
the hydraulic conductivity has been introduced as
f

k ij o g
K ij = -------------f
o

(1-121)

which depends, among the permeability k ij providing a


property of the porous medium, and the gravitational
acceleration g accepted as a constant for earth conditions, upon two additional reference variables of the
f
fluid phase, viz., the reference density o and the referf
ence viscosity o .
f

Both o and o are basically related to reference


conditions for the hydraulic head h o , the concentration
C o and the temperature T o as expressed by the constif
tutive equation (1-77) for the fluid density and by
the Eq. (1-101) for the viscosity relation function f .
According to Eq. (1-121) the conductivity is to be specified over a range of appropriately referenced magnitudes of head, concentration and temperature, viz.,
K ij =

f f
k ij ( h , C , T ) g
o
o
o
---------------------------------f
( C , T )
o
o

( C , T )
a
o
o
K ij = K ij f = K ij ------------------f
( C, T )

(1-123)

For comparison purposes the viscosity influence


may be explicitly excluded in FEFLOW (it is default),
where the viscosity function f is forced to be f 1
for each case independent of the actually appearing
concentration or temperature relations. Then the effective conductivity corresponds exactly to the input
a
parameter K ij = Kij . This special case is generally
given for pure flow problems.

NKV a= ~= `J
= c= = q~
b~
From the conservation principle (1-7) the following
transport equations have been derived in the above sections for the contaminant mass (cf. Eq. (1-56))
C-
( RC ) + q f D -----+ RC = Q C (1-124)
( C)
x i ij x j
t
xi i

(1-122)

Through the Darcy equation (1-118) formulated with


the conductivity the actual concentration and temperaPS==o=j~~=J=m~=f

ture effects on the fluid density and the fluid viscosf


ity are implied by the buoyancy term (1-120) and
the viscosity function f (1-101), respectively. The viscosity influence provides an effective (intrinsic) cona
ductivity K ij in the form:

and for the heat (cf. Eq. (1-62))

NKV=a=~=`=c==q~=b~
s s
f f
f f f
[ E + ( 1 ) E ] +
( qi E )
t
xi
T
ij ------- = Q T
x i x j

(1-125)

These formulations are called divergence forms due to


the divergent expressions for the properties to be conserved.
The divergence form of the heat transport equation
(1-125) can be completely expressed by the system
temperature T if the equations of state for the internal
f
s
energies E and E for the fluid phase f and solid phase
s, respectively, are applied. According to Eq. (1-94)

with dE = c dT for = s, f the state relation after


integration gives
E

= c ( T To )
f

for

= s, f

(1-126)

s s
f f
f f f
[ ( c + ( 1 ) c )T ] +
( qi c T )
t
xi
T
ij ------- = Q T
x i x j

(1-128)

The mass conservation (continuity) equation (1-9) for


the fluid phase f and the solid phase s, respectively,
( f ) + ( f q f ) = f Q f = f Q
i

t
xi

s s
( 1 ) s
(
) = ( 1 ) Q 0
t

(1-129)

can be utilized to substitute the divergent terms in the


above transport equations (1-124) and (1-125). Notice
s
that the mass source term Q for the solid phase is
assumed to be negligible. The substitution finally
results the following transport equations for the contaminant mass

Since T = T = T according to the thermal equilibrium assumption (1-60) as stated above the internal
energies for the fluid phase f and solid phase s are given
by

f C
C
C-
D -----R d ------ + q i -------
+ ( R + Q )C = Q C (1-130)
t
x i x i ij x j

and for the heat


f
f

E = c ( T To )

s
s
E = c ( T To )

(1-127)

f f
s s T
f f f
[ c + ( 1 ) c ] ------ + c q i
t
T-
f f
-----
+ c Q ( T To )
x i ij x j

T-----x i

(1-131)

= QT

where T o is the reference temperature as introduced in


Section 1.6.

which are called convective forms.

Accordingly, the divergence form of the heat transport equation (1-125) can now be written for the temperature T as

The divergence forms (1-124), (1-128) and the convective forms (1-130), (1-131) of the transport equations are physically completely equivalent. The
cbcilt==PT

NK=_~=b~
difference consists of incorporating explicitly the mass
conservation equation (1-9) in the convective form,
where additional source terms QC = CQ + Q C for the
contaminant mass and Q T = Q T [ f Q f E f + ( 1 )s Qs E s ]
(under neglecting density variations) and a derivative
term for the retardation Rd = ( RC ) C (to be more
described in Section 3.1) appear. Commonly, the convective form of the transport equation is numerically
preferred because simpler boundary-value problems
are accessible. However, FEFLOW is also capable of
tackling the divergence forms by choosing appropriate
options.

NKNM _= ^~J
=~=b
The density-dependent relations in the complete
coupled system of balance equations can be treated in
different manner. This especially depends upon
whether high density contrasts occur and density influences can become important in the proper balance
terms. Of special interests are here phenomena which
are either influenced or totally governed by concentration and/or temperature fields, so-called mixed and free
convection processes, respectively.

For the 2D horizontal transport equations (1-73), (174) written in their divergence forms

For the most density-dependent transport phenomena the so-called Boussinesq approximation (preciously referred to as Oberbeck-Boussinesq

C ( RBC ) + ( q f C ) B ( Dmol + D disp ) -----approximation) is employed (Bird et al.9, Combarnous


+ RBC = Q C
i
ij
ij
xi
x j
t
xi

and Bories13, Nield and Bejan55). On this basis the cou


T
[ B ( f c f + ( 1 ) s c s )T ] + ( f q f c f T ) B -----
Q
=
i
pled nonlinear system of balance equations can be
T
xi
x i ij x j
t

essentially simplified without the elimination of intrin(1-132) sic coupling mechanisms significant for free convection processes. The Boussinesq approximation is valid
equivalent convective forms of the transport equations for small to moderate density variations due to concentration and/or temperature changes. It consists of
can be derived as
neglecting all density dependencies in the balance
terms, except the vital buoyancy term in the momen
mol
disp

-----C

C
f
B ( D ij + D ij ) ------- + ( R + Q )BC = Q C
R d B
+ q i -------
x i x i
x j
t
tum conservation (Darcy-) equation. Accepting the

Boussinesq approximation the following coupled nonT


f f
s s T
f f f T
f f
B [ c + ( 1 ) c ] ------ + c q i ------- B ij ------- + c Q B ( T T o ) = Q T
x i x i
x j
t
linear system of balance equations results:
f

(1-133)

h q i
S o ----- + ------- = Q
t x i

PU==o=j~~=J=m~=f

(1-134)

NKNN=`~~=c=`=m
f

h o
f
- e j
q i = K ij f ------- + --------------f
x j

(1-135)

= K ij f ------- + ----------------------- ( C C o ) ( T T o ) e j

x j ( C s C o )
C
f

q C D ij ------- + RC = Q C
( RC ) +
x j
xi i
t

divergence form

C
C
f C
-----D - + ( R + Q )C = Q C
R d ------ + q i -------
x i x i ij x j
t

convective form

(1-136)
T
f f
s s

f f f ----- - = Q T
[ ( c + ( 1 ) c )T ] +
( qi c T )
t
xi
x i ij x j
divergence form
T
T
f f f T
f f
----- - + c Q ( T T o ) = Q T
[ c + ( 1 ) c ] ------ + c q i -------
x i x i ij x j
t
f f

s s

ature vanish. However, just the evolving features of a


convection process may be thoroughly affected at
higher density contrasts (problems of bifurcation, physical instability and hydrodynamic pattern formation).
The second term Q EB ( C, T ) of (1-138) only vanishes if
the density gradient is essentially orthogonal to the
flow. This is quite often not a tolerable assumption.
FEFLOW also enables the computation by employing the so-called extended Boussinesq approximation.
Here the fluid mass conservation equation (1-134) is
used in a generalized form:
f

h q i
S o ----- + ------- = Q + Q EB ( C, T )
t x i

(1-139)

convective form

(1-137)

Obviously, the Boussinesq approximation becomes


insufficient for large density variations (e.g., at highconcentration brines or high temperature gradients).
The main difference between the Boussinesq approach
and the actual balance quantities is expressed by the
additional term QEB ( C, T ) in the continuity equation
(1-108) according to

C
T- f ---------------------CT
- ----- - ----- ----Q EB ( C, T ) = ---------------------qi
-------
( C s C o ) t t
( C s C o ) x i
x i
1

(1-138)

which is negligible in the Boussinesq approximation.


The first term in Eq. (1-138) becomes only unimportant
if the temporal changes in concentration and/or temper-

The remaining equations correspond to the above formulations (1-135), (1-136) and (1-137).

NKNN `~~= c= `J
=m
Buoyancy-driven transport processes concern phenomena usually referred to as free, natural or cellular
convection. Sometimes such processes are characterized as fingering processes due to their appearance of
fingering and recirculating patterns in the concentration and/or temperature fields along interfaces (fronts)
caused by density-induced flow eddies. Superimposing
free and forced convection it is termed as mixed convection. Usually, the ratio of forces is expressed by the
quotient of Rayleigh and Peclet number Ra Pe . Theoretical and numerical studies regarding free convection
problems can be found in Diersch (et
cbcilt==PV

NK=_~=b~
al.)15,16,17,19,20,21,30,31,32, Kolditz and Diersch49, Kolditz
et al.51, Ackerer et al.1, Younes et al.72, Frolkovic and
De Schepper34 and in the references mentioned there.
The theoretical and analytical basis for free convection
problems associated with the hydrodynamic stability
criteria is presented in the works by Horton and
Rogers46, Wooding70, Elder33, Beck8, Palm et al.56,
Holst and Aziz39, Straus67, Combarnous and Bories13,
Caltagirone10,11, Rubin59, Homsy and Sherwood41,
Robinson and OSullivan58, Rubin and Roth60,
Cheng12, Hsu et al.47, Zebib and Kassoy73, Straus and
Schubert68, Horne and OSullivan42,43, Horne44, Schubert and Straus63, Horne and Caltagirone45, Rudraiah
and Srimani61, Simkins and Blythe65, McKibbin and
OSullivan53, as well as Trevisan and Bejan69.
Density gradients can be caused by either mass or
heat forces. The paradigm of Fig. 1.8 characterizes the
conditions subjected to a porous layer of thickness d
and conductivity K.
As a dimensionless characteristic parameter the Rayleigh number Ra (e.g., Diersch19) is defined
for solutes
--------------------- C K d
( Cs Co )
Ras = ---------------------------------------------------D

(1-140)

Co > C1 (To < T1)


porous medium
K
d
g

C1 (T1)
Figure 1.8 Density-induced buoyancy effects for a
paradigmatic problem of a porous layer subjected
to concentration and/or temperature gradients
between top and bottom.
C
T
K
d
D

=
=
=
=
=
=

C o C 1 , concentration difference;
T 1 T o , temperature difference;
isotropic hydraulic conductivity;
effective height (thickness);
contaminant mass diffusion coefficient;
bulk thermodiffusivity
f
s
f f
( = ( + ( 1 ) ) ( c ) ).

For the paradigm of a porous box as schematized in


Fig. 1.8 the following criteria in form of critical Rayleigh numbers Racrit1, 2 are known from perturbation
and numerical analysis:

and for heat


T K d
Ra t = ------------------------------

where
QM==o=j~~=J=m~=f

(1-141)

Ra < Racrit1 = 4
convection,

diffusion dominates, no free

NKNN=`~~=c=`=m
2

4 Ra < 390 = Ra crit2 onset of free convection,


stationary conditions, stable convective rolls,
Ra Racrit2 390 fluctuating free convection, no
more stationary conditions, transient flow oscillations appear.

cbcilt==QN

NK=_~=b~

QO==o=j~~=J=m~=f

c=~=q~=m==c=p~

c=~=q~=m==c=p~

OKN o
The treatment of free surfaces differs significantly
between 2D and 3D processes. While the formulations
of the basic equations for essentially horizontal problems in an unconfined (or phreatic) aquifer are rather
simple, 3D problems can imply complicate conditions,
especially due to:
existence of multiple (more than one) free surfaces;
effects of location of free surface(s) on the transport process in the depth;
problems arising if parts of the domain of an aquifer system fall dry and/or the water table rises;
boundary and initial conditions in dependence on
the location of free surface(s).
The method usually preferred in hydraulic modeling
is oriented to a free surface computation on fixed grids.
However, such an approach must often fail for transport processes due to following reasons. The main
drawback concerns the tendency of freezing contaminants in unsaturated or dry cells instead of moving

down or upward in dependence on the movement of the


water table.
Alternatively to fixed grid models, FEFLOW provides an advanced method which is based on 3D moving meshes. It allows a more accurate and rigorous
modeling of both the flow and, especially here, the
transport processes.

OKO e~= EOaF= m


=r=^
The modeling is based on the vertically averaged
equations (1-71) to (1-74), which are solved in the following form:
f

h q
( S o B + e ) ----- + -------i = Q
t x i

(2-1)

S
f
h
q i = BK ij ------x j

(2-2)

cbcilt==QP

OK=c=~=q~=m==c=p~
C-
( RBC ) + q f C BD -----ij x + RBC = Q C
xi i
t
j

divergence form

f C
C
C
R d B ------ + q i ------- BD ij ------- + ( RB + Q )C = Q C
x i x i
x j
t

convective form

s s
T-
B f f
f f f B -----[ ( c + ( 1 ) c )T ] +
( qi c T )
= Q T divergence form
t
xi
x i ij x j

f f
s s T
f f f T
T-
f f

-----
B
+
c
Q
(
T

T
)
=
Q
convective
form
B [ c + ( 1 ) c ] ------ + c q i -------

T
o

x i x i ij x j
t

where S is an effective storage coefficient.


The actually discharging aquifer thickness B differs
for unconfined and confined conditions, viz.,
bottom

B = h xl
top
xl

B =

for unconfined condition

bottom
xl

(2-5)

for confined condition

(2-4)

The transition from the unconfined (phreatic) to the


confined aquifer conditions is automatically realized in
dependence on the computed hydraulic head h related
top
to the top and bottom geometry of the aquifer, x l and
bottom
, respectively. The computation of the effective
xl
storage coefficient S at a transition state between confined and unconfined conditions is performed by using
the Heaviside function
S = So B + e ( 1 )

h
1

= --- atan ---------- + ---


2

as exhibited in Fig. 2.1.

(2-3)

(2-6)

top

xl

top

confined
B

unconfined

bottom
xl

Figure 2.1 Unconfined and confined conditions in an


aquifer (vertical cross section).

QQ==o=j~~=J=m~=f

where h = x l h (Fig. 2.1) is the difference


between the aquifer top and the computed head h, and
3
represents a given smoothing parameter.
10
Accordingly, for h 0 unconfined and for h < 0
confined conditions occur.

OKP=c=p~==Pa

OKP c=p~==Pa

free surface(s) the following system of balance equations is to be solved:

Apparently, to date the known and implemented


methods for modeling free-surface problems in 3D are
rather unsatisfactory and suffer from a number of difficulties in practical applications (see also the criticisms
uttered by Segol64, Yeh et al.71, and Knupp48). Enhancements in accuracy, flexibility and extensibility for
transport problems are mostly required.

h q i
S o ----- + ------- = Q + Q EB ( C, T )
t x i
f

(2-7)

h o
f
- e j
q i = K ij f ------- + --------------f
x j

(2-8)

In a 3D domain (Fig. 2.2) involving one or more

(2-9)

T-
[ ( f c f + ( 1 ) s c s )T ] + ( f q f c f T ) -----= Q T divergence form
i
t
xi
x i ij x j

f f
s s T
f f f T

T
f
f
[ c + ( 1 ) c ] ------ + c q i ------- ij ------- + c Q ( T T o ) = Q T convective form
x i x i
x j
t

(2-10)

C-
( RC ) + f C D -----q
ij x + RC = Q C
t
xi i
j

divergence form

f C
C
C
R d ------ + q i ------- D ij ------- + ( R + Q )C = Q C
x i x i
x j
t

convective form

Note, the number of free surfaces is unlimited.


In 3D the incorporation of free surface conditions is
performed via proper boundary conditions. Starting
point represents the discontinuity conditions (1-18) and
(1-95) along the interfacial free surface. For the fluid
phase f it yields:
f

( ( wi vi ) )

with

fs

surface

sf

ni = Po nl

(2-11)

B t
fs
sf
B t
w i n i = w i n i = ------------------ = -------------------B n i
B x i

(2-12)

fs

where n i is the normal pointed outward from the fluid


sf
phase, n i the normal pointed inward, w i the interface
velocity, and P o the so-called groundwater recharge
(or infiltration rate). Along a free (phreatic) surface the
f
pressure is constant p = 0 , accordingly it becomes
bottom
. Then, Eqs. (2-11) with (2h = x l and B = h xl
fs
sf
12), (1-96) and (1-114) (note that n l = n l ) can be
rewritten as:
cbcilt==QR

OK=c=~=q~=m==c=p~
f

fs

qn = vi ni

(2-13)

sf
fs
h sf
B t
= P o n l + --------------------- n l = P o e ----- n l
x l x i
t

upper free surface

Accordingly, along a free surface two boundary conditions are to be satisfied simultaneously:
prescribed flux rate (as an infiltration or, if equal
to zero, then impervious) and
location corresponds to the hydraulic head (constant pressure level)

second free surface

Figure 2.2 Multiple free surfaces in a 3D aquifer system.

It leads to the boundary condition formulation:


P -----h n = q f
n
o e t l

h = xl

(2-14)

where
Po
e
nl
f
qn
xl

=
=
=
=
=

groundwater recharge or infiltration rate;


effective (drainable/fillable) porosity;
normal unit vector (positive outward);
Darcy normal flux of fluid;
elevation.

For 3D modeling of free-surface problems FEFLOW


provides a number of additional extensions for which
the most important features are discussed next.

QS==o=j~~=J=m~=f

OKQ _^pa=q=~=jJ
=Pa=j
The adaptation of a finite element mesh to a changeable free-surface location has the advantage that the
mesh density is maintained in domains which are actually discharged. The conditions (2-14) lead to a nonlinear boundary-value problem due to the a priori
unknown free-surface locations. The process has to be
solved by an appropriate iterative scheme. According
to the computed hydraulic head at a time level the mesh
slices are to be adjusted. Slices of the inner mesh have
to be adapted in accordance with the changed water
table. However, in moving the meshes a number of data
must be adapted to the new locations. For instance, the
movement of a free surface through a layered aquifer
system requires an assignment of conductivities and
storage coefficients at each time of the new free-surface location. Especially at jump conditions in the

OKQ=_^pa=q=~=j=Pa=j
parameter fields typical for aquifer systems, the adaptation of the spatially varying material data to the moving
mesh has to be performed effectively and carefully.
FEFLOW provides here an advanced technique.

bution in the depth without unnecessary mesh refinement and coarsening if ever attainable. Following
criteria are employed to position the material layers:
fit layer surfaces (slices) to surfaces of parameter
discontinuity as soon located within the domain,
distribute remaining slices proportional to layer
thicknesses to attain as best as possible wellbehaved mesh distances in the depth,
join parameter for such elements according to partial volumes and surfaces, which cross and intersect more than one stratigraphic material layer.

This technique termed BASD (Best-Adaptation-toStratigraphic-Data) transforms and joins the model data
containing the stratigraphic initial structure to a moving finite element mesh which is appropriately adapted
to the free-surface locations. In this adaptation process
the mesh slices are aligned in such a manner that the
adjusted mesh is exactly fitted to parameter discontinuities if they ever exist. Remaining slices can be shifted
and repositioned to achieve a well-spaced nodal distri-

K1

finite element
node

K2

K2
K3

t = t1 > to

t = to

initial stratigraphic mesh

alignment

K1

cross element

K3

t = t2 > t1

free surface

K2

K3

K3

K3

K3

K2
K3

Figure 2.3 Moving mesh BASD technique of parameter adaptation applied to 3D free-surface
problems: schematized example for a groundwater rise.

BASD has proven to be a powerful and accurate


computation strategy for complex 3D application with
free surface(s). The principle of the BASD mesh adaptation process is illustrated in Fig. 2.3 for a schematized example. The initial stratigraphy consists here of
three layers having different conductivities. At initial
time t o the water table is on a lower position. The mesh

is accordingly shrunk, where the lower two layers completely fit into the K3 stratigraphy. However, the upper
layer crosses between the K 2 and K 3 stratigraphy and
a special treatment is here required. Such cross elements are admitted only if unavoidable. A proper 3D
interpolation technique has been developed which
allows a data joining for elements intersecting an arbicbcilt==QT

OK=c=~=q~=m==c=p~
trary number of stratigraphic layers (for more see Diersch and Michels28). If the water table ascends (Fig. 2.3
at time t 1 ) the moving mesh totally fits the K2 - K 3
stratum while the remaining slice is used to subdivide
the widest nodal spacing, here in the K3 layer. At later
time t 2 a further rise of the free surface occurs and the
moving mesh slices appear to be well aligned to the
data stratification without any need of interpolation.

OKR ^~= `~~


f~=
`=

a= = = i~J
==c=p~
Commonly, initial conditions in the model domain
are spatially dependent:
C ( xi, 0 ) = C I ( x i )

(2-15)

where C I is the initial concentration distribution.


Regarding problems with free surface(s) such a condition (2-15) is sometimes inappropriate, e.g., in application to groundwater rising or mine flooding.
Apparently, a contaminant distribution prescribed at
initial time can only be active if the water table arrives
at that location. Consequently, new desires in the
numerical modeling of such processes result.
Similarly to the above BASD technique, in
FEFLOW a new method has been implemented which
is capable of activating a contaminant initial distribution in the model domain as soon the particular location
becomes wet. The contaminant initial condition
appears now as an additional function of the hydraulic
QU==o=j~~=J=m~=f

head, viz.,
C ( x i, 0 ) = C I ( x i, h )

(2-16)

A number of special conditions can be raised. The most


important items are:
At initial time the contaminant distribution is
related to the hydraulic head distribution (location
of free surface). When the initial water table does
not meet the location of the contaminants the
actual initial condition of contaminants will correspond to the background value (default is zero).
At changing position of water table it is constantly
checked whether further contaminants have to be
conveyed to the corresponding mesh nodes as initial conditions. It obeys the following criteria:
(1) At a particular location and current time, when
t + t
t
the water table h
> h ascends and the initial
contaminant concentration imposed in the initial
stratigraphic database becomes higher than the
t + t
o
actual one C
( x i ) < C ( x i ) , then the actual concentration should be replaced by the higher initial
value.
(2) When the actual (computed) concentration at a
location is already higher (compared with the
stratigraphic initial concentration) and the water
table further ascends, then the actual concentration is not to be updated at this point.
(3) Descending the water table at a location, also
imaginable as an intermediate event, an overwriting by the initial contaminants is avoided.

OKR=^~=`~~=f~=`==a===i~==c
p~
If water contains a certain dissolved amount of contaminants, the computational procedure provides that
further contaminants can be added as soon the water
table arrives at a location, where higher contaminant
concentrations are charged. In this procedure a chemical equilibrium is assumed. It causes that the contaminants are instantaneously dissolved and conveyed to
the computational nodes as soon contacted with water.
However, if this location possesses a lower concentration in the initial condition, then such a new contribution is disregarded. The advantages of such an
approach are obvious. On the one hand, it controls that
higher contaminant concentrations can always float up
according to a rising water table. On the other hand, if
the water table descends and higher concentrations
occur in the groundwater, contaminants can also reach
deeper regions of the aquifer system which are previously clean (or lowly contaminated).

cbcilt==QV

OK=c=~=q~=m==c=p~

RM==o=j~~=J=m~=f

_~=b~==j

_~=b~==j

PKN d~=c~
The simulation system FEFLOW (see also Diersch
(et al.)22,23,24,25) models flow, contaminant mass and
heat transport processes as coupled or separate phenomena. It is based on the physical conservation principles as stated in Chapter 1.1 for mass, chemical
species, linear momentum and energy in a transient and
three-dimensional (if necessary two-dimensional)
numerical analysis.

For three-dimensional (3D) and two-dimensional


(2D) - vertical and axisymmetric, respectively - processes the representative model equations are as follows (i,j = 1,2,3):
f

h q i
S o ----- + ------- = Q + Q EB ( C, T )
t x i
f

(3-1)

h o
f
- e j
q i = K ij f ------- + --------------f
x j

(3-2)

(3-3)

s s
T-
f f
f f f -----[ ( c + ( 1 ) c )T ] +
( qi c T )
= Q T divergence form
t
xi
x i ij x j

f f
s s T
f f f T
T-
f f
-----

[ c + ( 1 ) c ] ------ + c q i -------
+
c
Q
(
T

T
)
=
Q
convective
form

o
T

t
x i x i ij x j

(3-4)

C
( RC ) + q f C D -----ij x + RC = Q C
xi i
t
j

divergence form

f C
C
C-
D -----R d ------ + q i -------
+ ( R + Q )C = Q C
x i x i ij x j
t

convective form

cbcilt==RN

PK=_~=b~==j
with the constitutive equations:

f
f
= o 1 + ----------------------- ( C C o ) ( T T o )
( Cs Co )

f
f
k ij o g

f
f
f
p
- + xl
= [ ( Cs ) o ] o
K ij = -------------h = -------
f
f

o g
o

2
3
3
o
1 + 1.85 ( C = Co ) 4.1 ( C = Co ) + 44.5 ( C = C o )
1 + 0.7063 0.04832
( T 150 )
C-
---------------------------------------------------------------------------------- = ------------------------------------------------------------------------------------------------------------------------------=
=

f = ------------------3
2
3
f
f
100
1 + 0.7063 ( T = To ) 0.04832 ( T = To )
1 + 1.85 4.1 + 44.5

( C, T )

f f

qi qj
f

D ij = ( D d + T V q ) ij + ( L T ) -------f

Vq

(--------------1 )- ---------------------------[(C) C]
(--------------1 )R = 1+
(C)
Rd = 1 +

cond
disp
f f
s s
ij = ij + ij
Q T = Q T + ( 1 ) Q T

f f

q
q
cond
disp
f
f
s
f f
i j

= [ + ( 1 ) ] ij
ij = c T V q ij + ( L T ) ------- ij

f
Vq

which are to be solved for the four primary variables:


h
f
qi
C
T

, o
s

So
K ij
ej
f

RO==o=j~~=J=m~=f

=
=
=
=

hydraulic head;
Darcy velocity vector of fluid;
concentration of chemical component;
temperature;

R
Rd
D ij

f s
c,c

=
=
=
=
=
=

=
=
=
=
=
=

fluid and reference fluid density, respectively;


solid density;
specific storage coefficient (compressibility);
tensor of hydraulic conductivity;
gravitational unit vector;
constitutive viscosity relation function;

ij
Qx

=
=

where
kI= = ~
~= = ~=
= =
~= = ~J= t 1 2 = ~
= = = ~~J
=K

Q EB

(3-5)

term of extended Boussinesq approximation,


Eq. (1-138);
retardation factor;
derivative term of retardation;
tensor of hydrodynamic dispersion;
decay rate);
porosity;
specific heat capacity of fluid and solid,
respectively;
tensor of hydrodynamic thermodispersion;
source/sink function of fluid ( x = ), of

) Referring to radioactive decay processes the decay rate

can
ln2
- using
be expressed by the reaction half-life t 1 2 , viz., = -------t1 2
t 1 2 as an (alternative) input parameter related to a (radioactive)

component (for more see Section 1.4.2).

PKO=e~=c~

=
C o, T o =
Cs
f
p
g
k ij
f f
, o

=
=
=
=
=

Dd
f
Vq
L, T

=
=
=
=
=

(C )

cond

=
=
=

ij
disp
ij
f s
,

L, T =

contaminant mass ( x = C ) and heat ( x = T );


fluid density difference ratio;
fluid expansion coefficient;
reference concentration and temperature,
respectively;
maximum concentration;
fluid pressure;
gravitational acceleration;
tensor of permeability;
dynamic viscosity and its reference value,
respectively, of fluid;
normalized temperature;
mass fraction;
molecular diffusion coefficient of fluid;
f f
q i q i absolute Darcy fluid flux;
longitudinal and transverse dispersivity,
respectively, of chemical species;
concentration-dependent adsorption function,
Eqs. (1-50) to (1-52);
conductive part of thermodispersion tensor;
dispersive part of thermodispersion tensor;
thermal conductivity for fluid and solid,
respectively;
longitudinal
and
transverse
thermodispersivity, respectively, of fluid.

The relationships of retardation R and R d , respectively, in Eq. (3-5) can be summarized for the different
adsorption laws according to Eqs. (1-50) to (1-52) as
follows:
Henry isotherm:

Freundlich isotherm:
b2 1
(1 )
R = 1 + ---------------- b 1 C

b2 1
(--------------1 )
b1 b2 C
Rd = 1 +

kI= = J
=
(3-7)
K d ==~==
e= =
~= = = = ~J
~=K

Langmuir isotherm:
( 1 ) k1
R = 1 + ---------------- ----------------- 1 + k2 C

k
(1 )
1
R d = 1 + ---------------- -------------------------2-
(1 + k C)

(3-8)

where

b 1, b 2

=
=

k 1, k 2

Henry sorptivity coefficient);


Freundlich sorption coefficient and exponent,
respectively;
Langmuir sorption coefficients.

PKO e~=c~
The essentially horizontal model equations differ
between confined and unconfined conditions. Accord)

The Henry isotherm with an retardation factor of

1
R = 1 + ----------- if often expressed by the distribution coeffi
- s

K d with K d = -----s , where


cient K d in the form7 R = 1 + 1---------

s
is the density of solid. An alternative definition of the distrib

(1 )
R = 1 + ----------------

(1 )
R d = 1 + ----------------

(3-6)

bution coefficient can sometimes be found as R = 1 + -----------d- ,


where b = ( 1 ) s is the bulk density of the porous media
(mass dry media per total volume).

cbcilt==RP

PK=_~=b~==j
PKOKN

ing to the above derivations the vertically averaged balance equations are in summarized form:

`=~

q
S o B -----h + -------i = Q
t x i

(3-9)

So
f
h
q i = T ij ------x j

(3-10)

(3-11)

s s
T
B f f
f f f -----[ ( c + ( 1 ) c )T ] +
( qi c T )
ij - = Q T divergence form
t
xi
x i x j

f f
s s T
f f f T

T
f f

ij ------- + c Q ( T T o ) = Q T convective form


B [ c + ( 1 ) c ] ------ + c q i -------

t
x i x i
x j

(3-12)

C
( RC ) + q f C D -----ij - + RC = Q C
x j
xi i
t

divergence form

f C
C
C-
D -----R d ------ + q i -------
+ ( R + Q )C = Q C
x i x i ij x j
t

convective form

with
top

bottom

f f

qi qj
f

D ij = ( BD d + T V q ) ij + ( L T ) -------
f
Vq

(1 )
( 1 ) [ ( C ) C ]
R = B 1 + ---------------- ( C )
R d = B 1 + ---------------- ---------------------------

cond
disp
f f
s s

ij = ij + ij
Q T = B [ Q T + ( 1 ) Q T ]

f f
qi qj
cond
disp
f
f
s
f f
= B [ + ( 1 ) ] ij
ij = c T V q ij + ( L T ) ------- ij
f
Vq
B = xl

RQ==o=j~~=J=m~=f

xl

(3-13)

PKO=e~=c~
PKOKO

r=E~F=~
f

h q i
( S o B + e ) ----- + ------- = Q
t x i

(3-14)

f
h
q i = BK ij ------x j

(3-15)

(3-16)

f f
s s
T

f f f
-----[ B ( c + ( 1 ) c )T ] +
( qi c T )
ij - = Q T divergence form
t
xi
x i x j

f f
s s T
f f f T

T
f
f

------- + c Q ( T T ) = Q convective form


B [ c + ( 1 ) c ] ------ + c q i -------
ij

T
o

x i x i x j
t

(3-17)

C
( RC ) + q f C D -----ij - + RC = Q C
x j
xi i
t

divergence form

f C
C
C-
D -----R d ------ + q i -------
+ ( R + Q )C = Q C
x i x i ij x j
t

convective form

with
bottom

D ij = B ( D d + T V q ) ij + ( L

(--------------1 )(--------------1 )- ---------------------------[(C) C]


R = B 1+
(C)
Rd = B 1 +

cond
disp
f f
s s

ij = ij + ij
Q T = B [ Q T + ( 1 ) Q T ]

f f
q i q j
cond
disp
f
f
s
f f
= B [ + ( 1 ) ] ij
ij = c T V q ij + ( L T ) ------- ij
f
Vq
B = h xl

f f
qi qj
-------T ) f Vq

(3-18)

cbcilt==RR

PK=_~=b~==j

RS==o=j~~=J=m~=f

f~I=_~I=~=`~=`

f~I=_~I=~=`~=`

The governing model equations summarized in


Chapter 3 have to be supplemented by initial, boundary, and constraint conditions. Their formulations realized in FEFLOW will be described in the following
3
paragraphs. Consider a domain R in the threedimensional space with its boundary . The boundary
consists of disjoint portions i which can be suitably subdivided according to the types of boundary
conditions to be prescribed on outer or inner positions
of the domain.

C ( x i, 0 ) = C I ( x i )

in

(4-2)

in

(4-3)

Heat transport:
T ( x i, 0 ) = T I ( x i )

where h I , C I and T I are known spatially varying functions of initial distribution. Note, for 3D problems with
free surface(s) the initial concentration distribution
(4-2) can also be dependent on the hydraulic head h as
stated in Section 2.5.

QKN f~=`
In the domain following initial conditions are
valid for the flow, contaminant mass and heat transport
processes, respectively:

QKO _~=`
On the boundary closing the domain disjoint
portions
are
appropriately
defined
as
= 1 2 3 4 = 5 6 7 = 8 9 10

Flow:
h ( x i, 0 ) = h I ( x i )

in

(4-1)

for which types of boundary conditions can be separately specified:

Mass transport:

cbcilt==RT

QK=f~I=_~I=~=`~=`
QKOKN

1st kind boundary condition (Dirichlet type):


R

h ( x i, t ) = h 1 ( t )

on

1 t [ 0, )

(4-4)

2nd kind boundary condition (Neumann type):


f
f

h o
R
- e j n i for 3D and 2D vertical
q n ( x i, t ) = q h ( t ) = K ij f ------- + --------------f
h

x j

h
q n ( x i, t ) = q h ( t ) = K ij ------- n i
for 2D horizontal unconfined on 2 t [ 0, )
h

x j

h
q n ( x i, t ) = q h ( t ) = T ij ------- n i
for 2D horizontal confined
h
x j

(4-5)

3rd kind boundary condition (Cauchy type):


R

q n ( x i, t ) = h ( h 2 h )
h

q n ( x i, t ) = h ( h 2 h )
h

for 3D and 2D vertical & unconfined


on 3 t [ 0, )
for 2D horizontal confined

Here the transfer coefficients (or transfer rates) h


and h represent dual directional functions in form of:
in
h
h =
out
h

RU==o=j~~=J=m~=f

for

h2 > h

for

h2 h

in
h
h =
out
h

(4-6)

for

h2 > h

for

h2 h

(4-8)

(4-7)

Accordingly, in specifying two alternate (if necessary temporal) coefficients different transfer conditions
can be input to distinguish between inflowing condi-

QKO=_~=`
tions ( q n h < 0 , e.g. infiltration from a surface water into
the aquifer) and outflowing conditions ( q n h > 0 , e.g.
exfiltrating the aquifer into the surface water). Their
usefulness for river-aquifer interactions is discussed
in
out
further below.
The special case h = h = h or
in
out
h = h = h does not differ between inward and
outward boundary flux, so it becomes directionally
independent.
4th kind boundary condition (single well type
(singular point source)):
w

Q ( x i, t ) =

for

Qm { ( xi xi ) }

(4-9)

for 3D and 2D (horizontal), respectively;


directional coefficient of in-transfer and outtransfer, respectively, for 3D;
in
out
h , h = directional coefficient of in-transfer and outtransfer, respectively, for 2D (horizontal);
w
Q
= well function;
w
Qm
= pumping/injecting rate of single well m;
m
xi
= coordinate of single well m;
Po
= groundwater recharge (infiltration rate);
e
= effective (drainable/fillable) porosity;
ni
= normal unit vector;
xl
= elevation.
in
h ,

out
h =

With respect to boundary conditions of 2nd and 3rd


kind special boundary conditions are available for
problems with free (phreatic) surface(s). They are
referred to as integral boundary conditions and are
defined as follows:

( x i, x i )

Free surface conditions to be satisfied simultaneously:


P -----h n = q
nh
o e t l
on

h = xl

4 t [ 0, )

(4-10)

in which
R

h1 , h2

qn
h
qn

=
=

qh , qh

h, h =

prescribed boundary values of hydraulic head


h;
normal Darcy flux of fluid (positive outward);
vertically averaged normal Darcy flux of fluid
(positive outward);
prescribed normal boundary fluid flux for 3D
and 2D (horizontal), respectively;
fluid transfer coefficient (leakage parameter)

cbcilt==RV

QK=f~I=_~I=~=`~=`

2nd kind integral boundary condition (integral Neumann type):

R
q n ( x i, t ) = q h ( t ) for 2D horizontal-unconfined as depth-integrated flux
h

q n ( x i, t ) = q h ( t ) for 3D related to the initial stratigraphic structure


h

(4-11)

3rd kind integral boundary condition (integral Cauchy type):

R
q n ( x i, t ) = h ( h 2 h ) for 2D horizontal-unconfined as depth-integrated flux
h

q n ( x i, t ) = h ( h 2 h ) for 3D related to the initial stratigraphic structure


h

Using these integral formulations of flux boundary


conditions it is ensured that a given flux rate on their
boundary portions becomes independent of the actually
discharging aquifer thickness and the location of free
surface. This is unlike a default nonintegral boundary
condition where a flux rate is integrated along the
effective aquifer thickness, which depends on the
actual (computed) free-surface position, and accordingly, varying (gross-) discharges may occur through
such boundaries. As a result, it may happen that the
total discharge through such varying boundaries constantly decreases at a descending water table. Consequently, such a flow region can inevitably fall dry and
possibly the problem can collapse with a zero inflow.
Integral boundary conditions prevents such situations
since the gross discharges are not influenced by the
location of free surface. The relation of fluxes, however, is distinguished in 2D and 3D applications due to
reasons of implementation:
SM==o=j~~=J=m~=f

(4-12)

(1) For 2D problems the fluxes have to be assigned as


already depth-integrated. The dimension of these
2 1
fluxes is then L T , similar to a horizontal confined
condition.
(2) For 3D problems the aquifer system is compiled as
an initial stratigraphic layer structure. Boundary conditions of the integral type are related to these initial
structure and accordingly, the integrated gross discharges remain independent of the free-surface location
during the computation with the BASD technique (Section 2.4). Notice, the dimension of these boundary
1
2 1
fluxes is LT (not L T !).
Integral boundary flux conditions have only a distinct meaning for problems with free (movable) surface(s). If no free surfaces exist, they are totally
equivalent to the nonintegral boundary conditions of
2nd and 3rd kind, in accordance with Eqs. (4-5) and (4-

QKO=_~=`
6).

geometric shape.

As natural boundary conditions appear those,


which result automatically for the model without any
specification on boundaries. It is q n h = 0 implicitly
satisfied if no additional assignments occur, i.e., the
boundaries are impervious, independent of their actual

Boundary fluxes (e.g., 2nd kind boundary condition) are defined normal to the boundary and positive outward as illustrated in Fig. 4.1 for both 3D and
2D.
3D

2D

qnh
n

q nh

Figure 4.1 Normal fluxes for 2D and 3D boundary geometries.

The formulation of 3rd kind boundary conditions is


based on a general transfer relation between the referR
ence value h 2 on the boundary portion and the hydraulic head h to be computed at the same place. The
R
reference hydraulic head h 2 can also be time-depenR
R
dent h 2 = h 2 ( t ) . The dual transfer coefficient h possesses the property of a resistance coefficient which
constrains the discharge through the boundary and,
additionally, differs between inflowing and outflowing
in
out
conditions by means of h and h , respectively,
according to Eqs. (4-7) and (4-8). If h 0 the boundary becomes impervious. On the other hand, using a
very large value h the boundary condition of 3rd

kind is reduced to a Dirichlet-type (1st kind) boundary


R
condition approaching to h = h2 .
For flow problems the transfer coefficient h can
be identified as a specific colmation (or leakage) coefficient as outlined in Fig. 4.2 for inflowing (infiltration)
in
R
conditions ( h h (h 2 > h) ). An adjacent river bed
is clogged (colmated) by a layer of thickness d and a
in
conductivity of K o . Normally, the layer conductivity
in
K o is much smaller than the conductivity K 1 of the
aquifer to be modeled. Thereby the model boundary
represents the inner boundary of the colmation layer
3 , where the model domain ends.
cbcilt==SN

QK=f~I=_~I=~=`~=`
For outward directed (exfiltrating) boundary fluxes
according to Fig. 4.3
infiltrating

in

Ko

4
R
h2

exfiltrating

in

K1 Ko

out
Ko

d
R

out

K1 Ko

h2

in

Figure 4.2 Transfer coefficient h ( = h ) as colmation


parameter of a river bed.

The flux through such a colmation layer can be


estimated from the Darcy equation (see Fig. 4.2), viz.:
R

in h
in h 2 h
q n K o ------- = K o --------------h
d
l

in

in

[d ]

(4-14)

For horizontal confined flow problems an inherent


vertical averaging becomes necessary (in the aquifer all
fluxes are integrated over the depth)in resulting in a
depth-integrated transfer coefficient h as:
in

Ko
in
in
h = B h B -------d

SO==o=j~~=J=m~=f

in

[ md ]

(4-15)

out

out

following relationships for h and h


derived, analogously to the above, viz.,

(4-13)

Setting (4-6) equal to (4-13) a simple relationship


in
results for the transfer coefficient h in 3D and 2D
(vertical, horizontal unconfined) cases:
in K o
h -------d

out

Figure 4.3 Transfer coefficient h ( = h ) as colmation parameter of a river bed.

out

out K o
h ---------d
out

in

[d ]

(4-16)

out

Ko
out
= B h B ---------d
in

out

in

can be

[ md ]
in

(4-17)
out

The coefficients h and h (also h and h ) differ if in case of infiltration the conductivities of the
colmation layer become depart from that of the exfilin
out
tration Ko Ko .

QKO=_~=`
QKOKO

j~=~
1st kind boundary condition (Dirichlet type):
R

C ( x i, t ) = C 1 ( t )

on

5 t [ 0, )

(4-18)

2nd kind boundary condition (Neumann type):


for 3D and 2D (vertical and axisymmetric)

convective form: q n ( x i, t ) =
=
C

on
dispersive flux

R
C
R
divergence form: q n ( x i, t ) = q C ( t ) = C 2 q n h D ij ------- n i
C
x j

C
D ij ------- n i
x j

6 t [ 0, )

(4-19)

6 t [ 0, )

(4-20)

R
qC ( t )

total flux

for 2D horizontal (confined and unconfined)

convective form: q n ( x i, t ) =
=
C

on
dispersive flux

R
C
R
divergence form: q n ( x i, t ) = q C ( t ) = C 2 q n h D ij ------- n i
C
x j

C
D ij ------- n i
x j

R
qC ( t )

total flux

cbcilt==SP

QK=f~I=_~I=~=`~=`

3rd kind boundary condition (Cauchy type):


R

q n ( x i, t ) = C ( C 3 C )
C

q n ( x i, t ) = C ( C 3 C )
C

for 3D and 2D vertical

on 7 t [ 0, )
for 2D horizontal

where the transfer coefficients (or transfer rates) C


and C represent dual functions in form of:

in
C
=
out
C

for

R
C3

>C

for

R
C3

Q C ( x i, t ) =

C =
out
C

for

C3 > C

for

R
C3

(4-23)

Accordingly, in specifying two alternate (if necessary temporal) coefficients different mass transfer conditions can be input to distinguish between inflowing
conditions ( q n C < 0 ) and outflowing conditions
in
out
( q nC > 0 ).in Theoutspecial case C = C = C
or
C = C = C does not differ between inward and
outward mass boundary flux.
4th kind boundary condition (single well type
(singular point source)):

SQ==o=j~~=J=m~=f

(4-24)

( x i, x i )

in which
R

in
C

Cm Qm { ( xi xi ) }
m

for
(4-22)

(4-21)

C 1 , C 2 , C 3 = prescribed boundary values of concentration


C;
qn
= normal mass flux (positive outward);
C
qn
= vertically averaged normal mass flux (positive
C
outward);
R R
q C, q C = prescribed normal boundary mass flux of the
dispersive part;
R R
q C , q C = prescribed normal boundary mass flux of the
total (convective + dispersive) part;
C, C = mass
transfer
coefficients
(leaching
parameter);
in
out
C , C = directional coefficient of mass in-transfer and
mass out-transfer, respectively;
in
out
C , C = depth-integrated mass in-transfer and mass
out-transfer, respectively;
w
QC
= mass flux well function;
w
Cm
= concentration of single well m;
w
Qm
= pumping/injecting mass rate of single well m;
m
xi
= coordinate of single well m.

QKO=_~=`
With respect to boundary conditions of 2nd and 3rd
kind special boundary conditions are available for
problems with free (phreatic) surface(s). They are

referred to as integral boundary conditions and are


defined as follows:

2nd kind integral boundary condition (integral Neumann type):

R
q n ( x i, t ) = q C ( t ) for 2D horizontal-unconfined as depth-integrated flux
C

q n ( x i, t ) = q C ( t ) for 3D related to the initial stratigraphic structure


C

(4-25)

3rd kind integral boundary condition (integral Cauchy type):

R
q n ( x i, t ) = C ( C 3 C ) for 2D horizontal-unconfined as depth-integrated flux
C

q n ( x i, t ) = C ( C 3 C ) for 3D related to the initial stratigraphic structure


C

Referred to as natural boundary conditions the relations q n C = 0 and q n C = 0 , respectively, are automatically satisfied in the model if the boundary remains
unspecified. In dependence on the formulation of the
mass transport equation (of convective or divergence
form) the normal mass fluxes on the boundaries are
suppressed for such natural boundary conditions. Here,
it should be considered, however, either the dispersive
or the total fluxes are imposed. The differences are discussed next.
The divergence form is capable of prescribing the
total mass flux along a boundary portion resulting from
R
R
the convective part C2 q nh (load of concentration C 2 in
the fluid flow q nh ) and the dispersive part
D ij ( C x j )n i . However, regarding this formulation
all boundaries have to be specified with such a type of

(4-26)

boundary condition, which can cause a specific handling of such formulations in the case of unknown
R
mass concentration C 2 on outflowing boundaries
R
(rather, C2 is here to be solved). Nevertheless, using
the divergence form FEFLOW is capable of computing
automatically the unknown convective part of the mass
flux at outflowing boundaries independent of the
imposed types of boundary conditions. This is done by
evaluating the fluid flux via a budget analysis in a postprocessing step which is then involved in modifying
the boundary conditions of the mass flux at such portions of boundaries. A more detailed discussion can be
found in Diersch29.
On the other hand, the default convective form does
not require a specific handling associated with formulations on outflowing boundaries and should be usually
cbcilt==SR

QK=f~I=_~I=~=`~=`
preferred. Assigning there q n C = D ij ( C x j )n i 0 as
a natural boundary condition, the mass flux freely
passes through a convectively open boundary section
and the concentration on the boundary automatically
results. Note here, a contaminant mass source, as far as
it should not be modeled via a 1st kind boundary condition, in form of a contaminant mass boundary flux
q n 0 includes only the dispersive part, i.e., the magC
nitude of the flux will result from the gradient of contaminant concentration at the boundary. Thus in
general, the convective form will necessarily produce a
higher concentration gradient to realize the same mass
load q n C through a boundary.
The transfer coefficients C associated with boundary conditions of 3rd kind, Eq. (4-21), can be regarded
as leaching parameters which constrain the mass flux
through the boundary. If C 0 the boundary becomes
impervious. On the other hand, using a very large value
C the boundary condition of 3rd kind is reduced
to a Dirichlet-type (1st kind) boundary condition with
R
C = C 3 . Such a leaching process is displayed in Fig.
4.4 for the example of a flow over a salt dome modeled
R
with a diffusive input condition ( C 3 > C ). Considering
a thickness d for the leaching body and applying the
Fickian law in form of
R

in C
in C 3 C
q n D o -------- = D o ----------------C
d
l

(4-27)

in

the mass transfer coefficient C can be assessed as


in

in D o
C -------d

SS==o=j~~=J=m~=f

in

[ md ]

(4-28)

and analogously to a horizontal problem as


in

Do
in
in
C = B C B -------d

2 1

[m d ]

in

(4-29)

in

where D o denotes a representative diffusion coefficient of the salt body (Fig. 4.4).

7
d

in

Do

C3

in

Figure 4.4 Transfer coefficient C ( = C ) as leaching parameter of a salt dome.


out

out

Analogous assessments for C and C result if


the transition resistance differs between inflowing and
in
out
outflowing (releasing) conditions: D o D o .

QKO=_~=`
QKOKP

e~=~

1st kind boundary condition (Dirichlet type):


R

T ( x i, t ) = T 1 ( t )

on

8 t [ 0, )

(4-30)

2nd kind boundary condition (Neumann type):


for 3D and 2D (vertical and axisymmetric)

on
conductive flux

R
f f
T
R
q n ( x i, t ) = q T ( t ) = c T 2 q n ij ------- n i
h
T
x j

q n ( x i, t ) = q T ( t ) =

T
ij ------- n i
x j

convective form:

(4-31)

9 t [ 0, )

(4-32)

divergence form:

9 t [ 0, )

total flux

for 2D horizontal (confined and unconfined)

T
ij ------- n i
x j

divergence form:

convective form:

q n ( x i, t ) =
=
T

on
conductive flux

f f R
T
R
q n ( x i, t ) = q T ( t ) = c T 2 q n ij ------- n i
h
T
x j

R
qT ( t )

total flux

cbcilt==ST

QK=f~I=_~I=~=`~=`

3rd kind boundary condition (Cauchy type):


R

q n ( x i, t ) = T ( T 3 T )
T

q n ( x i, t ) = T ( T 3 T )
T

for 3D and 2D vertical


on 10 t [ 0, )
for 2D horizontal

in which the transfer coefficients (or transfer rates) T


and T represent dual functions in the form:
in
T
T =
out
T
in
T
T =
out
T

T3 > T

for

T3 T

for

T3 > T

for

T3 T

Tm Qm { ( xi xi ) }

m
m
( x i, x i )

(4-36)

(4-34)

where
R

f f

for

for

(4-35)

Accordingly, in specifying two alternate (if necessary temporal) coefficients different heat transfer conditions can be input to distinguish between inflowing
conditions ( q nT < 0 ) and outflowing conditions
in
out
( q nT > 0 ).in Theoutspecial case T = T = T
or
T = T = T does not differ between inward and
outward heat boundary flux.

4th kind boundary condition (single well type


(singular point source)):

SU==o=j~~=J=m~=f

Q T ( x i, t ) = c

(4-33)

T1 , T2 , T3 =
=
qn
T
qn
=

prescribed boundary values of temperature T;


normal heat flux (positive outward);
vertically averaged normal heat flux (positive
T
outward);
R R
q T , q T = prescribed normal boundary heat flux of the
conductive including thermodispersive part;
R R
q T , q T = prescribed normal boundary heat flux of the
total
(convective
+
conductive
+
thermodispersive) part;
T, T = heat transfer coefficients;
in
out
T , T = directional coefficient of heat in-transfer and
heat out-transfer, respectively;
in
out
T , T = depth-integrated heat in-transfer and heat outtransfer, respectively;
w
QT
= heat flux well function;
w
Tm
= temperature of single well m;
w
Qm
= pumping/injecting heat rate of single well m;
m
xi
= coordinate of single well m.

QKO=_~=`
Integral flux conditions exist for boundary conditions
of 2nd and 3rd kind of free-surface problems, viz.:
2nd kind integral boundary condition (integral Neumann type):

R
q n ( x i, t ) = q T ( t ) for 2D horizontal-unconfined as depth-integrated flux
T

q n ( x i, t ) = q T ( t ) for 3D related to the initial stratigraphic structure


T

(4-37)

3rd kind integral boundary condition (integral Cauchy type):

R
q n ( x i, t ) = T ( T 2 T ) for 2D horizontal-unconfined as depth-integrated flux
T

q n ( x i, t ) = T ( T 2 T ) for 3D related to the initial stratigraphic structure


T

Referred to as natural boundary conditions the relations q n T = 0 and q n T = 0 , respectively, are automatically satisfied in the model.
The heat transfer coefficients T associated with
boundary conditions of 3rd kind, Eq. (4-33), represent
a heat transition resistance. If T 0 the boundary
becomes adiabatic (insulated). On the other hand, using
a very large value T the boundary condition of
3rd kind is reduced to a Dirichlet-type (1st kind)
R
boundary condition with T = T2 . The heat transfer
coefficients can be estimated analogously to the above
transfer coefficients for mass flux. Considering a thickness d for the a heat transition layer and applying FouR
riers law for input condition ( T 2 > T ) in form of:
R

in T
in T 2 T
q n o ------- = o --------------T
d
l

(4-38)

in

the heat transfer coefficient T can be obtained as


in

in o
T ------d

in

2 1 1

[ Jm d K ]

(4-40)

and similarly to a horizontal problem as


in

o
in
in
T = B T B ------d

in

1 1 1

[ Jm d K ]

(4-41)

in

where o represents a heat conduction coefficient of


out
the transition
layer. Analogous relationships for T
out
and T result if the heat transition resistance differs
between inflowing and outflowing conditions:
in
out
o o .

(4-39)

cbcilt==SV

QK=f~I=_~I=~=`~=`

QKP `~
Constraints are limitations for all types of boundary
conditions. They result from the requirement that
boundary conditions should only be valid as long as
minimum and maximum bounds are satisfied. If during
a simulation run the conditions are exceeded or fell
below, the constraints are to be assigned as new intermediate boundary conditions.
Constraints in relation to free-surface problems are
of special significance, for instance in the separation of
a groundwater flow from a surface water if the water
table (or the head difference) falls below a given magnitude and an infiltration from the surface water into
the aquifer has to be constrained by a maximum influx.
Furthermore, it arises in falling below a critical hydraulic head for a well (e.g., below the bottom of a well
screen) operating with a given pumping rate. If attained
a minimum water table has to be held and the pumpage
is to be correspondingly adapted.
Additionally, regarding boundary conditions of
transport equations constraints appear here in a combined manner. Limiting conditions are not only raised
by the contaminant mass or heat regime but should also
be regarded in form of minimum or maximum bounds
for the water table. For instance, a contaminant boundary condition may be related to the current hydraulic
head. If the water table is low, apparently the contaminant boundary condition is yet inactive, a typical case
occurring in rising groundwater tables if (re-)mobilization of contaminants due to an increasing wetting process of a contaminant source can be expected.
Another important applications of constraint condiTM==o=j~~=J=m~=f

tions refer to the case if a component should enter a


boundary (e.g., mimicked by a boundary condition of
1st kind), but only as long as the fluid flows into the
domain (infiltration). However, if it may be expected
that the flow along such a boundary changes its sign
(turning to a outflow condition), then the given concentration boundary condition of 1st kind is obviously
inappropriate and should be dropped for outflowing
conditions.
For free-surface problems it may happen that parts
of the aquifer(s) become dry (water table falls below
the aquifer base). Such a situation referred to as falling
dry can be considered as a natural constraint condition
as described further below in greater detail.
The formulation of constraints is based on the formalism of complementary conditions for a type of
boundary condition. Accordingly, a potential condition
(hydraulic head, concentration or temperature) is constrained by maximum and minimum flux relations
(fluid, mass or heat fluxes). On the other hand, flux
relationships are constrained by complementary potential limits, i.e., the fluid flux is constrained by maximum-minimum heads, the contaminant mass flux by
minimum-maximum concentrations, and the heat flux
by minimum-maximum temperatures. Following formulations are available for flow conditions:

QKP=`~
QKPKN

1st kind

2nd kind

R
h1 ( t )

qh ( t )

3rd kind

h2 ( t )

4th kind

Q ( t )

( t )

if
min 1

R
else Q h = Q h ( t )

max
max
R
R
Max: h < h 2 ( t ) else h = h 2 ( t )
if

min 2
R
Min: h R > h min2 ( t )
else h = h
(t)

Max: Q R < Q max 3 ( t ) else Q R = Q max 3 ( t )


h
h
h
h

if

min 3
min 3
R
R
else Q h = Q h ( t )
Min: Q h > Q h ( t )

max
max
Max: h R < h 4 ( t ) else h R = h 4 ( t )
if

Min: h R > h min 4 ( t ) else h R = h min 4 ( t )

Max: Q R < Q max 1 ( t )


h
h

min 1
R
Min: Q h > Q h ( t )

where Q h = q h dV represents the summed-up, called


lumped balance fluxes at nodal points to which potential boundary values are related. Note, due to technical
reasons the balance quantities are defined positive
inward. These fluxes are computed in a balance analysis during the simulation. The minimum-maximum
R
R
bounds for the fluxes Q h and the potentials h themselves can be time-dependent functions again. Accordingly, it is possible to consider time-dependent
variations in the existence and influence of boundary
and constraint conditions. For instance, these temporary capabilities of constraints are very useful in modeling the temporarily varying occurrence of sealing or
drainage activities over a restricted time period, or in
simulating time-constrained boundary conditions (e.g.,
1st kind) which are associated with certain construction
or remedial actions arising only at given times. Typical
applications of such constraint conditions formulated

else

max 1

Qh = Qh

(4-42)

by Eqs. (4-42) are shown in sketches of Fig. 4.5.


While constraints (4-42) are applied to active flow
boundary conditions there exists one important exception, which includes all points of the domain and the
boundaries regardless of acting boundary and constraint conditions there: The control of the simulated
water levels for problems with free surface by using a
natural constraint in case of falling dry of aquifers and
a system of aquifers, respectively, in points, in subdomains, or in the whole area given by the following criterion:

h(t)

if

h > x bottom

and

h > h min i

or

confined

else

bottom

h = xl

(4-43)

cbcilt==TN

QK=f~I=_~I=~=`~=`

a)

b)

Q-(t)

infiltrating

Q+(t)

exfiltrating
h(t)

Q max
Q(t)
surface

hmax(t)
Q max(t)
wate

hmin(t)

r tab

le

hR2 h

Figure 4.5 Examples of using constraints for flow problems:


a) constraining a single well by an allowable drawdown in form of a minimum well head and by an allowable injection water level in form of a maximum well head,
b) flow separation in infiltration from surface water due to constraining the maximum seepage through the river bed.

kI=
=
P=

`~J=
~= =
~= = ~
= ~= = ~
= =
= t= m~= J= sK= f
= b= J
~= = ~=
`~J= EP= F
~= =
cbciltF= =
=cbcilt=~=QKVK

The constraint (4-43) implies that the hydraulic head


h ( t ) is valid as long as the water table lies above the
bottom
aquifer bottom x l
(bottom elevation in case of 2D
or fixed aquifer slices in case of 3D), unless other
stronger conditions for the potential field are to be
subjected due to proper constraints with respect to a
min
minimum water table h i for boundary conditions
i = 1, 2, 3, 4 according to Eq. (4-42). Furthermore,
such a constraint is neither relevant in areas having
fully confined conditions nor in areas where the water
table rises above possible aquifer confinements (mixed
conditions). Violating this constraint condition the minimum water table is intermediately set equal to the
bottom
elevation at current time as depicted in Fig. 4.6.
xl
Obviously, the criterion (4-43) have to take prior over
the constraint condition (4-42) because the prescribed

TO==o=j~~=J=m~=f

constraints are subordinated to the natural constraint


conditions in order to avoid generally negative aquifer thicknesses for free-surface problems (even at
wrong boundary conditions too). It is important to
note that the constraint (4-43) is optional that means it
can be set inoperative whenever an auxiliary hydraulic
head boundary condition at the bottom cannot be
accepted due to the possible occurrence of inappropriate water balance conditions (constraints introduce
additional fluxes). Under such conditions the hydraulic
head can fall below the bottom in an unlimited manner
while the effective thickness of the aquifer is hold at a
minimum (as default 1 mm). As the result, it mimics an
unsaturated condition where the pressure head
becomes a negative suction and the minimum thickness
implies a residual water content of an element.

QKP=`~
a)
Q

h = xbottom

h = xbottom

confinement

falling dry
falling dry
aquifer bottom

confined

area of testing unconfined conditions

b)
Q

upper area of testing unconfined conditions

1st aquifer
falling dry

unconfined

confined
1st aquitard

2nd aquifer
confined
unconfined

falling dry

2nd aquitard

confined

lower area of testing unconfined conditions


3rd aquifer

fully confined

Figure 4.6 Natural minimum constraints regarding aquifer bottom encountered in free-surface problems:
a) unconfined test area and areas of falling dry in 2D,
b) complex situation for testing unconfined conditions in a 3D aquifer system involving multiple free surfaces
at juxtaposed unconfined-confined relations.

The constraints for the transport equations provide


more complex criteria resulting from the desired com-

bining features as mentioned above. In more general


form they yield:
cbcilt==TP

QK=f~I=_~I=~=`~=`
QKPKO

j~=~

1st kind C 1 ( t )

if

max 1
R

QC < QC ( t )
Q R = Q max1 ( t ) as long as h min 1 h R h max 1

C
C
Max:
else
and
min 1
R
R

or h R > h max1
Q C = 0 if h < h

h min 1 h R h max1

min 1

R
QC > QC ( t )

Q R = Q min 1 ( t ) as long as h min 1 h R h max1

C
C
Min:
else

and
R

Q
=
0
if hR < h min1 or hR > h max1
C
h min1 h R h max1

max

2nd kind q C ( t )

3rd kind C 3 ( t )

if

if

2
R

C <C
(t)
C R = C max2 ( t ) as long as h min 2 h R h max 2

Max:
else R
and
min 2
R

or h R > hmax2
Q C = 0 if h < h
min
max
R

2
2
h

min 2
R
C >C
(t)

C R = C min2 ( t ) as long as h min 2 h R h max 2

else R
and
Min:
min 2
R

or hR > h max2
Q C = 0 if h < h
min
max
R
2
2
h

h h

max3
R

QC < QC ( t )
Q R = Q max3 ( t ) as long as h min 3 h R h max 3

C
C
Max:
else

and
R

Q
=
0
if h R < h min3 or h R > h max3
C

h min 3 h R h max3

min 3

R
QC > QC ( t )

Q R = Q min 3 ( t ) as long as h min3 h R h max3

C
C
else
Min:
and
R

Q
=
0
if hR < h min3 or h R > hmax3

C
h min 3 h R h max3

max

4th kind Q C ( t )

TQ==o=j~~=J=m~=f

if

4
R

C <C
(t)
C R = C max4 ( t ) as long as h min 4 h R h max 4

Max:
else R
and
min 4
R

or h R > hmax4
Q C = 0 if h < h

h min 4 h R h max 4

min 4
R
C >C
(t)

C R = C min4 ( t ) as long as h min 4 h R h max 4

else R
and
Min:
min 4
R

or hR > h max4
Q C = 0 if h < h
h min 4 h R h max4

(4-44)

QKP=`~
QKPKP

e~=~

1st kind T 1 ( t )

if

max 1
R

QT < QT ( t )
Q R = Q max 1 ( t ) as long as h min1 h R h max1

T
T
Max:
else
and
min 1
R
R

or h R > hmax1
Q T = 0 if h < h

h min 1 h R h max 1

min 1

R
QT > QT ( t )

Q R = Q min1 ( t ) as long as h min 1 h R h max 1

T
T
Min:
else

and
R

Q
=
0
if h R < h min1 or h R > h max1
T
h min 1 h R h max 1

max

2nd kind q T ( t )

3rd kind T 2 ( t )

if

if

2
R

T <T
(t)
T R = T max2 ( t ) as long as

Max:
else R
and
min 2
R

or
Q T = 0 if h < h
min
max
R

2
2
h

min 2
R
T >T
(t)

T R = T min 2 ( t ) as long as

else R
and
Min:
min 2
R

or
Q T = 0 if h < h
min
max
R
2
2
h

h h

min 2

max 2

h >h
min 2

h h

max2

max 2

h h

max2

h >h

max3
R

QT < QT ( t )
Q R = Q max 3 ( t ) as long as h min3 h R h max3

T
T
Max:
else

and
R

Q
=
0
if h R < hmin3 or h R > hmax3
T

h min3 h R h max3

min 3

R
QT > QT ( t )

Q R = Q min 3 ( t ) as long as h min 3 h R h max 3

T
T
else
Min:
and
R

Q
=
0
if hR < h min3 or hR > h max3

T
h min 3 h R h max 3

max

4th kind Q T ( t )

if

4
R

T <T
(t)
T R = T max 4 ( t ) as long as

Max:
else R
and
min 4
R

or
Q T = 0 if h < h

h min 4 h R h max 4

min 4
R
T >T
(t)

T R = T min 4 ( t ) as long as

else R
and
Min:
min 4
R

or
Q T = 0 if h < h
h min4 h R h max4

min 4

h h

h >h
min 4

h h

max 4

max4

max4

h >h

max 4

(4-45)

cbcilt==TR

QK=f~I=_~I=~=`~=`
min

The fluxes Q C, Q T represent again summed-up,


called lumped mass balance fluxes at nodal points,
analogously to the flow problem and defined positive
inward, viz.,
Q C = q C dV

R
R
Q T = q T dV

(4-46)

Naturally, the specific balance fluxes q C, q T are


composed by their convective and dispersive parts (see
here also the relationships (4-19) and (4-20), and (4-31)
and (4-32), respectively) according to

convective

TS==o=j~~=J=m~=f

C
D ij ------- n i
x j

R R

C qh

dispersive

convective

T
ij ------- n i
x j

(4-47)

(4-48)

R
f f
c T qh

conductive

In practice, it has shown to be inappropriate to


include the total (convective plus dispersive/conductive) flux into the procedure of controlling the constraint conditions (4-44) and , because the direction of
the dispersive/conductive fluxes is ambiguous (e.g., the
dispersive spreading also occurs against the convective
flow direction). Accordingly, the balance-based evaluation of fluxes is exclusively related to the convective
mass and heat fluxes, respectively:
R

R R

R R

Q C = q C dV ( C q h ) dV

to which potential boundary values (concentrations and


temperatures, respectively) are assigned. The miniR
R
mum-maximum bounds for the fluxes Q C, Q T and the
R R
potentials C , T can be again time-dependent functions allowing very comfortable rules for constraints.

qC =

qT =

max

where ( ) typ and ( ) typ denote the prescribed


maximum and minimum bounds, respectively, for the
R
R
R
corresponding type of boundary condition, Qh , Q C, Q T
the computed balance fluxes of fluid, contaminant
R
R R
mass and heat, respectively, and h , C , T the computed potential values of hydraulic head, concentration
and temperature, respectively, on the boundary.

Q T = q T dV ( T q h ) dV

(4-49)

presenting unambiguously directional balance quantities.


The transport constraints essentially consist of two
parts for the individual types of boundary conditions:
(1) A minimum-maximum bound complementary
for the type of boundary condition (e.g., a concentration boundary (1st or 3rd kind) is controlled by an allowable minimum-maximum
boundary mass flux, or a mass flux boundary
magnitude (2nd or 4th kind) is controlled by an
allowable minimum-maximum boundary concentration).
(2) A permitted range for boundary conditions
within tolerable limits of hydraulic head (ranging

QKP=`~
min

max

between h i and h i ) is imposed. If the simulated water table lies outside this range the nodal
boundary conditions (all types, 1st to 4th kind)
are suppressed. This can easily be realized by
R
assigning intermediately a zero flux ( Q C = 0
R
and Q T = 0 , respectively), i.e., no additional
mass or heat flux then occurs at such nodes and
the boundary conditions become intermediately
hidden).

intermittent pumping regime encountered in remedial


activities for a contaminant plume. The wells inject
R
clean water ( C = C1 ) as long as an injecting pumpage
w
occurs ( Q < 0 ) and pump up contaminated groundwater in a time-given pumping operation with a mixing
concentration to be determined. Both wells possess a
concentration boundary condition of 1st kind for freshR
water ( C = C1 ) and a minimum mass flux constraint
min 1
of zero Q C 0 .

Typical applications of transport constraints are outlined in Figs. 4.7 a-c. Figure 4.7a describes the case of
a density-coupled saltwater intrusion problem (flow
over a salt dome) having a boundary on which alternate
boundary concentrations appear in dependence on the
dynamic process: As long as water enters the domain it
should have a prescribed concentration of freshwater.
However, if the water releases the domain (along the
same boundary) the concentration on the boundary is
unknown and should be automatically computed. Such
a description can be easily realized if the entire boundary section is assigned by a freshwater boundary condiR
tion of 1st kind ( C = C1 ), and at the same time, the
boundary will be imposed by a constraint condition in
min
form of a null minimum mass flux Q C 1 0 (more
constraints are not necessarily to be specified). Such an
arrangement provides that the freshwater condition
remains valid as long as the convective flux, being concentration-dependent due to the density coupling phenomena, points into the domain:

The third example shown in Fig. 4.7c describes an


application in modeling a contaminant spreading process from a deposit associated with rising groundwater
in a phreatic aquifer (referred to as flooding problem).
The contaminant boundary condition (e.g., modeled as
a 1st kind type) should be active only when the water
table reaches the contaminant deposit (wetting case),
min
i.e., a constraint in form of h 1 is prescribed representing the bottom of the contaminant deposit. More
constraints are not necessarily required in such a case.

R R

Q C = ( C q h ) dV
since

qh < 0

>

min 1

QC

= 0

f~= W
^= ~= J
= ~=
= E C R1 0 F= = ~J
~= = = = ~~J
~=~==
= ~= ~= J
~= =
K= f= ~= ~= ~
=~=~=
R
Q C = ~= = bK= EQJRMF
===~=
min
R
QC = 0 QC = 0 >
^I= ~= =
= = = = = ~
~==~=~
E~= C Cs 108 FK
1

(4-50)

for inflow

A similar example is schematized in Fig. 4.7b for an


cbcilt==TT

QK=f~I=_~I=~=`~=`

b)

a)
freshwater C = 0

releasing unknown concentration C


t

=0
well 1
C=0
min
QC 1 = 0

mi
xin
g

wa
te

min1

QC

aquifer

contaminant plume

saltwater

C=0
min
QC 1 = 0

well 2
salt dome

c)

C
t
t

contaminant deposit C = C

wetting

hmin
groundwater rise, flooding

TU==o=j~~=J=m~=f

Figure 4.7 Application of transport constraints:


a) Saltwater intrusion by flowing groundwater
over a salt dome,
b) intermittend pumping regime for remediating
a contaminant plume,
c) wetting and activating a contaminant deposit
during a groundwater rise (flooding).

RKN m~= a= ~
o~
For the tensors of conductivity Kij and transmissivity T ij anisotropy is to be taken into account. It is
assumed they have orthotropic properties, i.e., the conm
ductivities are given along their principal directions x i
(i = 1,2,3), which are turned against the global Cartesian coordinate system x i (i = 1,2,3) by a rigid-body
rotation (Fig. 5.1), see e.g. Goldstein et al.35.
x3m

x3

x2
x2m

Accordingly, the conductivity along the principal direcm


tions x i represents a diagonal matrix
m

K1 0
m
( K ij )

0
0

m
K2

0
0

(5-1)

0 K3

which can be mapped by using the transformation in


form of the rotation matrix a ij
a 11 a 12 a 13
( a ij ) = a 21 a 22 a 23

(5-2)

a 31 a 32 a 33

x1

onto the global Cartesian coordinate system x i , viz.:

x1m

Figure 5.1 Definition of a 3D anisotropic conductivity.

cbcilt==TV

RK=^
After rotation, the tensor K ij in the global Cartesian
coordinates results, which describes an anisotropic
conductivity in the form:

K ij = a li K lk a kj
m

a 11 a 21 a 31 K 1 0 0 a 11 a 12 a 13
K 11 K 12 K 13
( K ij ) = K 21 K 22 K 23 = a 12 a 22 a 32 0 K m 0 a 21 a 22 a 23
2
m a 31 a 32 a 33
K 31 K 32 K 33
a 13 a 23 a 33
0 0 K3

a 11 K 1 + a 21 K 2 + a 31 K 3

(5-3)

( K ij ) =

a 12 K 1 + a 22 K 2 + a 32 K 3

The elements of the rotation matrix a ij , Eq. (5-2),


represent the directional cosines of the rotation angles
appearing in transforming from the Cartesian coordim
nate system x i to the principal direction system x i .
They are referred to as Eulerian angles. In 3D three
angles ( , , ) result, while in 2D only one Eulerian
angle has to be specified to perform an unique principal axis rotation. As the result, anisotropy for threedimensional applications needs formally six parameters
m
m
m
while
three
parameters
( K 1 , K 2 , K 3 , , , ) ,
m
m
( K 1 , K 2 , ) are required to handle a two-dimensional
anisotropy. Considering the effort and the appropriateness in computing the anisotropic conductivity (5-4),
different strategies are employed for 3D and 2D cases
as described subsequently.

RKO qJ~=

a 11 a 13 K 1 + a 21 a 23 K 2 + a 31 a 33 K 3

(5-4)

a 12 a 13 K 1 + a 22 a 23 K 2 + a 32 a 33 K 3
2
m
a 13 K 1

symm.

2
m
a 23 K 2

2
m
a 33 K 3

Cartesian coordinate system x 1, x 2 by a given angle


m
between the first principal axis of K ij and the x 1 -axis
(Fig. 5.2). An anisotropic conductivity is then uniquely
defined by their maximum and minimum conductivim
m
ties, K max K 1 and K min K2 , respectively, acting
along the principal axes, and by the rotation angle , as
depicted in Fig. 5.2.
m

x2

K min

x2

K
m

x1

K max
x1

^J

Two-dimensional anisotropy represents a special


case of the principal direction transformation as introduced above. It rotates the principal axes of the conm
ductivity tensor K ij into the axes of the global
UM==o=j~~=J=m~=f

a 11 a 12 K 1 + a 21 a 22 K 2 + a 31 a 32 K 3

Figure 5.2 Definition of a 2D anisotropic conductivity.

The components of the conductivity tensor K ij in


the global coordinate system x 1, x 2 are determined by

RKP=qJ~=^
using the rotation formula (5-3) for the two-dimensional case. Expressing the directional cosines a ij
( a ij ) =

cos sin
sin cos

(5-5)

by the rotation angle it yields:


m

(5-6)

K ij = a li K lk a kj

K 11 K 12
K 21 K 22

cos sin K max 0


cos sin
sin cos
0 K min sin cos

The components of the two-dimensional anisotropic


conductivity Kij lead finally to:
K 11
K 22
K 12

= K max cos + K min sin

2
2

= K max sin + K min cos

= K 21 = ( K max K min ) sin cos

missivity Tij , Eq. (1-70), applied to 2D problems of


confined aquifers.

RKP qJ~
^
RKPKN

d~=Pa=~=~J

The three-dimensional rotation of the principal axes


needs the knowledge of the three Eulerian angles
( , , ) , which are defined in Fig. 5.3 (Note that
, , and , , represent intermediate stages of a
sequential rotation of the axes).
x3,

(5-7)

x2

x1

x3

In FEFLOW for each discretized (finite) element following quantities are input:
(1) the maximum conductivity K max ,
(2) a factor of anisotropy defined as
aniso = K min K max

with

'

'

x3

x3m

x1m

x2
x1

x1
'

'

x2m

x2

Line of nodes

K max > 0 , (5-8)

being unity for isotropic relations, and


(3) the rotation angle .
Analogous relationships exist for the tensor of trans-

Figure 5.3 The rotations defining the Eulerian angles in


3D (modified from35).

If the Eulerian angles are given, the directional


cosines a ij of the rotation matrix (5-2) can be immedicbcilt==UN

RK=^
ately expressed as35:
cos cos cos sin sin
( a ij ) = sin cos cos sin cos
sin sin

cos sin + cos cos sin


sin sin + cos cos cos
sin cos

This represents the most general formulation of anisotropy in 3D. This case is optionally available in
FEFLOW, where the user can directly input the six
m
m
m
parameters ( K1 , K 2 , K 3 , , , ) .
The general rotation matrix (5-9) contains the following three important special cases of axis rotation:
(a) Rotation about the x 3 -axis only:
cos sin 0
( a ij ) = sin cos 0
0
0 1

at

= = 0 (5-10)

(b) Rotation about the x 2 -axis only:


( a ij ) =

cos 0 sin
0 1 0
sin 0 cos

at

= 90
= 90

(5-11)

(c) Rotation about the x 1 -axis only:


1 0
0
( a ij ) = 0 cos sin
0 sin cos

UO==o=j~~=J=m~=f

at

= = 0 (5-12)

sin sin
cos sin
cos

(5-9)

RKPKO p~J= Pa= ~


=p=
Applying the rotation matrix (5-9) to the transformation (5-4) it leads to relatively expensive formulae
for the resulting conductivity matrix K ij in the threedimensional anisotropic case.
Unlike the 2D case, where the additional effort in
prescribing only one Eulerian angle for the definition
of an anisotropic property of conductivity is still justifiable, in 3D the Eulerian angles ( , , ) are often not
utilized as a direct input because, in practice, they are
generally not explicitly known and not given.
Rather, one can make use of the fact that the principal axes generally correlate with the geological layer
structure and, accordingly, provided that the threedimensional shape of the layers is known, the spatial
rotation of the principal directions is automatically
accomplished by computation. The conductivity in a
geologically layered structure is often orthotropic inasm
m
much the higher conductivity K1 = K2 is parallel to
m
the layering and a lower value K 3 is normal to the
stratigraphy. It leads to a computational method for
transforming the principal axes of each finite element if
three-dimensional prismatic elements are utilized,
which are capable of fitting their top and bottom faces

RKP=qJ~=^
to the actual stratigraphic layering (Fig. 5.4a).
For three-dimensional prismatic finite elements a
method of shape-derived principal directions has been
proposed by Springer66 which is implemented in
FEFLOW to compute 3D anisotropy. Using quadrilateral or triangular prismatic elements the two opposite
top and bottom faces of each element represent layer
boundaries. A finite element is usually characterized by
its local distorted coordinate system ( , , ) as shown
in Fig. 5.4b. The coordinate transformation from the

local system ( , , ) to the global Cartesian system


( x 1, x 2, x 3 ) is achieved by the Jacobian matrix J :

x 1, x 2, x 3,
J = x 1, x 2, x 3,
x 1, x 2, x 3,

x 1
-------

x1
= ------
x
-------1

x 2
-------
x 2
-------
x 2
-------

x
-------3
x 3
-------
x 3
-------

(5-13)

a)

to p l

b)

a ye r

m
tto
bo

u3 K3

to p f

er
lay

ace

u1 K1

u2 K2
x3
x2

m
tto
bo

x1

e
fac

Figure 5.4 Layer-oriented principal directions of anisotropy for a prismatic finite element.

The idea behind Springers method of shape-derived


principal directions is in relating appropriately the
three principal axes of the conductivity anisotropy to
the local finite-element coordinate system ( , , ) .
Basically, the directions of the mutually orthogonal
principal axes are identified by the following normalized three vectors u i (Fig. 5.4b):

x 1,
u 1 = x 2,

(parallel to direction)

(5-14)

x 3,

cbcilt==UP

RK=^
x 1,

x 1,

x 1,

u 2 = m 11 x 2, m 12 x 2,
x 3,

(5-15)

u 3 = x 2, x 2,

x 3,

x 3,

x 1,

x 1,
(5-16)

x 3,
x 1,

x 1,

(perpenticular to the area spanned by x 2, and x 2, )

(orthogonalization of x 2, with respect to x 2, )


x 3,

x 1,

x 3,

x 3,

x 3,

where
2

x 1, + x 2, + x 3,

x 1, x 1, + x 2, x 2, + x 3, x 3,

( m ij ) = J J =

2
x 1,

2
x 2,

2
x 3,

x 1, x 1, + x 2, x 2, + x 3, x 3,
x 1, x 1, + x 2, x 2, + x 3, x 3,
2
x 1,

symm.

2
x 2,

(5-17)

2
x 3,

As the result, the directional cosines can be simply expressed by


kI=
~= = J
~===_^pa
= E= p= OKQF
= ~= =
~= ~= ERJNVF
~===~=J
==~===~
EF=
~~
=K

a ij

T
1
0
0
ui ej
-------------- , with the basis vectors: e 1 = 0 , e 2 = 1 , e 3 = 0
= cos ( u i, e j ) =
ui ej
0
0
1

(5-18)

to compute the rotation matrix a ij . Finally, it yields:


x 2,
-----------m 11

x 1,
-----------m 11

( a ij ) =

m 11 x 1, m 12 x 1,
--------------------------------------------2
2
m 11 m 22 m 11 m 12
x 2, x 3, x 2, x 3,
--------------------------------------------2
m 11 m 22 m 12

m 11 x 2, m 12 x 2,
--------------------------------------------2
2
m 11 m 22 m 11 m 12
x 3, x 1, x 3, x 1,
-------------------------------------------2
m 11 m 22 m 12

In the practical simulation, FEFLOW computes the


rotation matrix by using Eq. (5-19) at each Gaussian
integration point of a finite element according to the
UQ==o=j~~=J=m~=f

x 3,
-----------m 11
m 11 x 3, m 12 x 3,
--------------------------------------------2
2
m 11 m 22 m 11 m 12

(5-19)

x 1, x 2, x 1, x 2,
-------------------------------------------2
m 11 m 22 m 12

3D stratigraphic layer structure. In this way, the anisotropic three-dimensional tensor of conductivity K ij is
automatically computed by applying Eq. (5-4).

RKQ=p~=`~
The key advantages of this method concern that
only the conductivities of the three principal directions
along the layer structure need to be input for each discretized element, viz.:
m

(1) the conductivity K 1 parallel to layering,


(2) the conductivity

m
K2

K ii

K1 0

parallel to layering, and

(3) the conductivity K 3 normal to layering,


and the 3D shape of the layer geometry is embodied in
the anisotropy relations with a relatively high accuracy.

This case is to be referred to as axis-parallel anisotropy. Here, different conductivities can be prescribed
in each direction of the x i -axes:
m

For 3D: K11 = K 1 , K 22 = K2 and K 33 = K 3 .


For 2D: Kmax = K11 and aniso = K 22 K 11 with
= 0.

RKQ p~=`~
In specifying the conductivity matrix K ij two special cases are important:

RKQKN

( K ij ) = K 21 K 22 K 23 = 0 K m 0

K 31 K 32 K 33
m
(5-20)
0 0 K3

or

m
= K i and K ij = 0 for i j at i = 1,2,3
K 11 K 12 K 13

^J~~=~
m

The system of principal axes x i is parallel to the


Cartesian coordinate system x i . Accordingly, the Eulerian angles become zero: = = = 0 and the conductivity matrix Kij reduces to the diagonal matrix
m
K ij , Eq. (5-1), viz.:

RKQKO

The directional independency of the conductivity


leads to an isotropic conductivity matrix Kij in the
form
K 11 K 12 K 13
( K ij ) = K 21 K 22 K 23 =
K 31 K 32 K 33
or
K ij = K ij

K 0 0
0 K 0

0 0 K

(5-21)

where K corresponds to an isotropic conductivity coefficient. Here, the following is to be input:


cbcilt==UR

RK=^
For 3D: K 11 = K22 = K 33 = K .
For 2D: K max = K and aniso = 1 with = 0 .

US==o=j~~=J=m~=f

o~==c~==cbcilt

o~==c~==cbcilt

The following tables relate the constitutive and


material relationships used in the FEFLOW package to
the balance equations and constraint conditions derived
above. The relations depend on both the problem class
(flow, mass transport, heat transport, thermohaline process) and the dimension (3D, 2D (vertical, axisymmetric, horizontal)) of the chosen problem.

o~=O=J=q~=~~W

The heat capacities c for = s, f appear in the governing balance equations as products of the densities
s s
f f
c and c . It is acceptable over a large temperature range that these products, termed as volumetric
heat capacities, are approximately constant:

c const

o~=N=J=^J~~=~=~=W
Axis-parallel anisotropic conductivity (or transmissivity) implies (Section 1.4) that principal axes of conductivity coincide with the axes of the global Cartesian
system x 1, x 2, ( x 3 ) . Accordingly, it holds:
2D: K max = K 11, aniso = K 22 K 11 and = 0, as well as

m
and K ij = 0 for i j at i = 1,2,3

3D: K ii = K i

(6-1)

If permissible the default option axis-parallel anisotropy should be selected because, especially for 3D
problems, the computational burden in the coefficient
matrix processing can be reduced. Obviously, isotropy
represents a special case of axis-parallel anisotropy
according to Eq. (1-17).

for

= s, f .

(6-2)

o~= P= J= a= = ~= Co
~==~= T o W
For mass and/or heat transport problems the conductivf
ity K ij depending on the reference density o and the
f
reference viscosity o , Eq. (1-122), correlates with the
values of the reference concentration C o and reference
temperature To . The reference concentration Co and
the reference temperature T o can be specified in particular menus of FEFLOW. The suitable ranges of values
are summarized in Tab. 6.13. It is taken into consideration that for mass transport the common reference
value of Co = 0 is suited, where (measured) conductivities are automatically related to fresh (clean) water
conditions. For the heat or thermohaline transport is not
possible to predefine a general default value for the reference temperature T o . However, it is rather important
cbcilt==UT

SK=o~==c~==cbcilt
for heat transport to specify a temperature reference
frame, because effects of thermal viscosity can become
large (e.g., between 10 C and 150 C the difference is
given by a factor of 7). It is to be noted that T o is to be
settled in a temperature range as much as possible
which is covered by the initial or boundary conditions
between maximum and minimum temperatures
(entrance temperature, background temperature, etc.)
as indicated in Tab. 6.13.
o~=Q=J=aJ~=~=~=~W
The description of the geometric relations in the depth
of an aquifer and an aquifer system differs for 2D and
3D applications.
For (essentially) horizontal unconfined 2D conditop
tions the top and bottom geometries of the aquifer, x l
bottom
and x l
, respectively, are assigned within the flow
material attributing procedures. Accordingly, their elevations are handled as areal properties which are
related to the centroid of a finite element. The relation
of these elevations to nodal points of the finite-element
mesh is performed program-internally in considering
the elevations of elements belonging to the node via a
patch evaluation. For confined 2D problems, vertical or
axisymmetric, the specification of depth relation is not
required. If both confined and unconfined conditions
simultaneously occur the task is to be globally declared
as an unconfined problem and the top elevation of the
top
aquifer x l is appropriately specified as stated in Sectop
tion 2.2. For instance, using a numerically large x l value (say 1000 m) it always provides a free water
top
table for corresponding hydraulic conditions h < x l .
Three-dimensional problems generally need the
overall specification of the depth coordinates (elevaUU==o=j~~=J=m~=f

tions) within the layering structure of the problem to be


solved. Principally, they are input at nodal points on
each mesh slice. In order to do it very effectively, specific editing tools are available in FEFLOW.
Take into account, due to the different procedures in
relating depth coordinates in the 2D case (area-related)
and in the 3D case (node-related) small deviations can
occur if an unconfined 2D problem is expanded to a 3D
problem and the elemental (centroid) elevations are
automatically adapted to nodal quantities via a patch
evaluation. If such derivations should not be tolerable it
is recommended to redo the assignment process for the
elevations in FEFLOWs problem attribute editor
based on the primary elevation data (measured triplets).

SKN=Pa=c=m~~

SKN Pa=c=m~~
Table 6.1 Material conditions for 3D flow (with and without mass transport)
Item

Symbol

Unit

Default

Equations

Axis-parallel anisotropy:
Conductivity [Kxx]
Conductivity [Kyy]
Conductivity [Kzz]

K 11

10 ms

5-20, 3-5, 1-30

K 22

10 ms

5-20, 3-5, 1-30

K 33

10 ms

5-20, 3-5, 1-30

General anisotropy (optional):


Conductivity [K1m]

K1

10 ms

5-1, 5-4

Conductivity [K2m]

m
K2
m
K3

10 ms

5-1, 5-4

5-1, 5-4

Conductivity [K3m]

10 ms

5-9 (Fig. 5.3)

5-9 (Fig. 5.3)

5-9 (Fig. 5.3)

2-14

1-76, 1-78, 1-80,


3-5

0.2

1-114

In(+)/out(-)flow on top/bottom

Po

10 md
4

Density ratio

10

Storativity (drain/fillable)

Storage compressibility

So

Source(+)/sink(-)

10 d

In-transfer rate
Out-transfer rate

in
h
out
h

10

1-109, 3-1

4 1

1-102, 3-1

4 1

4-7, 4-14

4 1

4-7, 4-16

10 d
10 d

cbcilt==UV

SK=o~==c~==cbcilt

Table 6.2 Material conditions for 3D flow (at heat transport)


Item

Symbol

Unit

Default

Equations

Axis-parallel anisotropy:
Conductivity [Kxx]
Conductivity [Kyy]
Conductivity [Kzz]

K 11

10 ms

5-20, 3-5, 1-30

K 22

10 ms

5-20, 3-5, 1-30

K 33

10 ms

5-20, 3-5, 1-30

General anisotropy (optional):


Conductivity [K1m]

K1

10 ms

5-1, 5-4

Conductivity [K2m]

m
K2
m
K3

10 ms

5-1, 5-4

5-1, 5-4

Conductivity [K3m]

10 ms

5-9 (Fig. 5.3)

5-9 (Fig. 5.3)

5-9 (Fig. 5.3)

2-14

1-76, 1-77, 3-5

1-88, 1-82

0.2

1-114

In(+)/out(-)flow on top/bottom

Po

10 md

10 K

(T)

10 K

Constant thermal fluid density expansion:


Expansion coefficient (constant)

4 1

Variable thermal fluid density expansion:


Expansion coefficient (variable)
Storativity (drain/fillable)
Storage compressibility
Source(+)/sink(-)
In-transfer rate
Out-transfer rate

VM==o=j~~=J=m~=f

4 1

So

4 1

1-102, 3-1

4 1

4-7, 4-14

4 1

4-7, 4-16

in
h
out
h

10 d
10 d
10 d

10

1-109, 3-1

SKN=Pa=c=m~~

Table 6.3 Material conditions for 3D flow (at thermohaline transport)


Item

Symbol

Unit

Default

Equations

Axis-parallele anisotropy:
Conductivity [Kxx]
Conductivity [Kyy]
Conductivity [Kzz]

K 11

10 ms

5-20, 3-5, 1-30

K 22

10 ms

5-20, 3-5, 1-30

K 33

10 ms

5-20, 3-5, 1-30

General anisotropy (optional):


Conductivity [K1m]

K1

10 ms

5-1, 5-4

Conductivity [K2m]

m
K2
m
K3

10 ms

5-1, 5-4

5-1, 5-4

Conductivity [K3m]

10 ms

5-9 (Fig. 5.3)

5-9 (Fig. 5.3)

5-9 (Fig. 5.3)

2-14

1-76, 1-80, 3-5

1-76, 1-77, 3-5

1-88, 1-82

In(+)/out(-)flow on top/bottom
Density ratio

Po

10 md

10

10 K

(T )

10 K

Constant thermal fluid density expansion:


Expansion coefficient (constant)

4 1

Variable thermal fluid density expansion:


Expansion coefficient (variable)
Storage compressibility
Source(+)/sink(-)
In-transfer rate
Out-transfer rate

4 1
1

1-109, 3-1

So

4 1

1-102, 3-1

4 1

4-7, 4-14

4 1

4-7, 4-16

in
h
out
h

10 d
10 d
10 d

10

cbcilt==VN

SK=o~==c~==cbcilt

SKO s~= = ^
Oa=c=m~~
Table 6.4 Material conditions for vertical 2D flow (with and without mass transport)
Item

Symbol

Unit
4

Equations

5-7, 5-20

Conductivity [Kmax]

K max

Anisotropy factor [Kmin/Kmax]

aniso

5-8, 5-20

Angle from +x-axis to Kmax

10 ms

Default

5-6, Fig. 5.2

1-76, 1-78, 1-80,


3-5

10

Density ratio

10

Storage compressibility

So

Source(+)/sink(-)

4 1

10 d

1-102, 3-1

In-transfer rate

in

10 d

4 1

4-7, 4-14

Out-transfer rate

out
h

4 1

4-7, 4-16

VO==o=j~~=J=m~=f

10 d

1-109, 3-1

SKO=s~==^=Oa=c=m~~

Table 6.5 Material conditions fr vertical 2D flow (at heat transport)


Item

Symbol

Unit
4

Equations

5-7, 5-20

Conductivity [Kmax]

K max

Anisotropy factor [Kmin/Kmax]

aniso

5-8, 5-20

5-6, Fig. 5.2

1-76, 1-77, 3-5

1-88, 1-82

Angle from +x-axis to Kmax

10 ms

Default

Constant thermal expansion of fluid density:


Expansion coefficient (constant)

4 1

10 K

(T )

10 K

Variable thermal expansion of fluid density:


Expansion coefficient (variable)

4 1
1

Storage compressibility

So

Source(+)/sink(-)

10 d

In-transfer rate
Out-transfer rate

in
h
out
h

10

1-109, 3-1

4 1

1-102, 3-1

4 1

4-7, 4-14

4 1

4-7, 4-16

10 d
10 d

cbcilt==VP

SK=o~==c~==cbcilt

Table 6.6 Material conditions for vertical 2D flow (at thermohaline transport)
Item

Symbol

Unit
4

Equations

5-7, 5-20

Conductivity [Kmax]

K max

Anisotropy factor [Kmin/Kmax]

aniso

5-8, 5-20

5-6, Fig. 5.2

1-76, 1-78, 1-80,


3-5

1-76, 1-77, 3-5

1-88, 1-82

Angle from +x-axis to Kmax


Density ratio

10 ms

Default

10

10 K

(T)

10 K

Constant thermal expansion of fluid density:


Expansion coefficient (constant)

4 1

Variable thermal expansion of fluid density:


Expansion coefficient (variable)
Storage compressibility
Source(+)/sink(-)
In-transfer rate
Out-transfer rate

VQ==o=j~~=J=m~=f

4 1
1

So

4 1

1-102, 3-1

4 1

4-7, 4-14

4 1

4-7, 4-16

in
h
out
h

10 d
10 d
10 d

10

1-109, 3-1

SKP=e~=Oa=c=m~~J

SKP e~= Oa= c= m~~J

Table 6.7 Material conditions for horizontal 2D flow (confined)


Item

Symbol

Unit
4

2 1

Equations

1-70, 3-10, 5-7

Transmissivity [Tmax]

T max

Anisotropy factor [Tmin/Tmax]

aniso

5-8

5-6, Fig. 5.2

So ( = So B )

10

Angle from +x-axis to Tmax


Storage compressibility

10 m s

Default

1-109, 3-9

Source(+)/sink(-)

10 md

3-9

In-transfer rate

in
h

10 md

4-8, 4-15

Out-transfer rate

out
h

10 md

4-8, 4-17

cbcilt==VR

SK=o~==c~==cbcilt

Table 6.8 Material conditions for horizontal 2D flow (unconfined)


Item

Symbol

Unit
4

Equations

5-7, 5-20

Conductivity [Kmax]

K max

Anisotropy factor [Kmin/Kmax]

aniso

5-8, 5-20

Angle from +x-axis to Kmax


Aquifer bottom elevation
Aquifer top elevation
Storativity (drain/fillable)
Storage compressibility
Source(+)/sink(-)
In-transfer rate
Out-transfer rate

VS==o=j~~=J=m~=f

10 ms

Default

5-6, Fig. 5.2

bottom
xl
top
xl

0.

2-5, Fig. 2.1

1000.

2-5, Fig. 2.1

0.2

1-114

So

in
h
out
h

10

1-109, 3-14

3-14

4 1 a

4-7, 4-14

10 md
10 d

4 1 b

10 d

0
4

4-7, 4-16
1

a. In case of integral

boundary conditions of 3rd kind the unit for the In-transfer rate is: 10 md

b. In case of integral

boundary conditions of 3rd kind the unit for the Out-transfer rate is: 10 md

SKQ=j~=q~=m~~

SKQ j~= q~= m~~J

Table 6.9 Material conditions for 3D, vertical or axisymmetric mass transport
Item

Symbol

Unit

Default

Equations

0.3

1-91, 3-5

1-50, 3-5, 3-6

Sorption coefficient

b1

( mg l )

1-51, 3-5, 3-7

Sorption exponent

b2

1-51, 3-5, 3-7

Sorption coefficient (numerator)

k1

1-52, 3-5, 3-8

Sorption exponent (denominator)

k2

l mg

1-52, 3-5, 3-8

Molecular diffusion

Dd

1-13, 3-5

Longitudinal dispersivity

1-13, 3-5

Transverse dispersivity

0.5

1-13, 3-5

4 1

1-54, 3-5

3 1

gm d

1-57, 3-3

md

4-22, 4-28

md

Porosity
Henry sorption:
Sorption coefficient
Freundlich sorption:

1 b2

Langmuir sorption:

Decay

ratea

Source(+)/sink(-)
In-transfer rate
Out-transfer rate

QC
in
C
out
C

2 1

10 m s

10 s

a.For radioactive decay processes the decay rate can be alternatively expressed by the half-life t 1 2
Section 1.4.2).

4-22
ln2
in form of = --------- (see
t1 2

cbcilt==VT

SK=o~==c~==cbcilt
Table 6.10 Material conditions for 2D horizontal mass transport (confined and unconfined)
Item

Symbol

Unit

Default

Equations

Aquifer thickness
(only relevant to confined conditions)

1.0

2-5, Fig. 2.1

Porosity

0.3

1-91, 3-13, 3-18

1-50, 3-5, 3-6

Sorption coefficient

b1

( mg l )

1-51, 3-5, 3-7

Sorption exponent

b2

1-51, 3-5, 3-7

Sorption coefficient (numerator)

k1

1-52, 3-5, 3-8

Sorption exponent (denominator)

k2

l mg

1-52, 3-5, 3-8

Molecular diffusion

Dd

1-13, 3-13, 3-18

Longitudinal dispersivity

1-13, 3-13, 3-18

Transverse dispersivity

0.5

1-13, 3-13, 3-18

4 1

1-54, 3-13, 3-18

2 1

1-57, 3-11, 3-17

md

1 b

4-23, 4-29

md

1 c

4-23

Henry sorption:
Sorption coefficient
Freundlich sorption:
1 b2

Langmuir sorption:

Decay ratea
Source(+)/sink(-)
In-transfer rate
Out-transfer rate

QC
in
C
out
C

2 1

10 m s

10 s

gm d

ln2

a. For radioactive decay processes the decay rate can be alternatively expressed by the half-life t 1 2 in form of = --------- (see
t1 2
Section 1.4.2).
b. For confined conditions and for unconfined conditions at simultaneous application of integral
boundary conditions of 3rd
2 1
kind the unit for the In-transfer rate is: m d
c. For confined conditions and for unconfined conditions at simultaneous application of integral
boundary conditions of 3rd
2 1
kind the unit for the Out-transfer rate is: m d

VU==o=j~~=J=m~=f

SKR=e~=q~=m~~

SKR e~= q~= m~~J

Table 6.11 Material conditions for 3D, vertical or axisymmetric heat transport
Item

Symbol

Porosity

Volumetric heat capacity of fluid

f f

Unit

Default

Equations

0.3

1-91, 3-5

3 1

10 Jm K

4.2

1-61, 1-94, 3-4

1000kgm
f

c 4200Jkg K

Volumetric heat capacity of solid

s s

3 1

10 Jm K

2.52

1-61, 1-94, 3-4

2650kgm
s

c 950Jkg K

Heat conductivity of fluid

Heat conductivity of solid

Longitudinal dispersivity

Transverse dispersivity

Source(+)/sink(-) of fluid
Source(+)/sink(-) of solid
In-transfer rate
Out-transfer rate

f
QT
s s
QT
in
T
out
T
f

1 1 1

0.65

1-14, 1-63, 3-5

1 1 1

1-14, 1-63, 3-5

1-14, 1-63, 3-5

Jm s K
Jm s K

0.5

1-14, 1-63, 3-5

3 1

1-64, 3-5

3 1

1-64, 3-5

2 1 1

4-34, 4-40

2 1 1

4-34

Jm d
Jm d

Jm d K
Jm d K

cbcilt==VV

SK=o~==c~==cbcilt

Table 6.12 Material conditions for 2D horizontal heat transport (confined and unconfined)
Item

Symbol

Unit

Default

Equations

Aquifer thickness
(only relevant to confined conditions)

1.0

2-5, Fig. 2.1

Porosity

0.3

1-91, 3-13, 3-18

Volumetric heat capacity of fluid

f f

3 1

10 Jm K

4.2
f

1000kgm
f

c 4200Jkg K

Volumetric heat capacity of solid

s s

3 1

10 Jm K

c 950Jkg K

Heat conductivity of fluid

1-61, 1-94, 3-12,


3-17

1 1 1

0.65

1-14, 1-63, 3-13

1 1 1

1-14, 1-63, 3-13

Jm s K

2.52
s

2650kgm

1-61, 1-94, 3-12,


3-17

Heat conductivity of solid

Longitudinal dispersivity

1-14, 1-63, 3-13

Transverse dispersivity

0.5

1-14, 1-63, 3-13

2 1

1-64, 3-13, 3-18

2 1

1-64, 3-13, 3-18

2 1 1 a

4-35, 4-41

2 1 1 b

4-35

Jm s K

Source(+)/sink(-) of fluid

B Q T

Source(+)/sink(-) of solid

s
B Q T
in
T
out
T

In-transfer rate
Out-transfer rate

Jm d
Jm d

Jm d K
Jm d K

a. For confined conditions and for unconfined conditions at simultaneous application of integral
1 1 1
kind the unit for the In-transfer rate is: J m d K
b. For confined conditions and for unconfined conditions at simulatneous application of integral
1 1 1
kind the unit for the Out-transfer rate is: J m d K

NMM==o=j~~=J=m~=f

boundary conditions of 3rd


boundary conditions of 3rd

SKR=e~=q~=m~~

Table 6.13 Reference concentration C o and reference temperature T o


Reference condition

Symbol

Unit

Default value

Concentration

Co

mg l

0.0
(freshwater condition)

Temperature

To

unspecified (150.0)
to be chosen in relation to the
boundary and initial condiR
tions, e.g., T o = T 1 or
To = TI

cbcilt==NMN

SK=o~==c~==cbcilt

NMO==o=j~~=J=m~=f

1. Ackerer, P., Younes, A. and Mos, R. (1999) Modeling variable


density flow and solute transport in porous medium: 1. Numerical
model and verfication. Transport in Porous Media Vol. 35, No. 3,
345-373.
2. Baxter, G.P. and Wallace, C.C. (1916) Changes in volume upon
solution in water of the halogen salts of the alkali metals. J. Amer.
Chem. Soc. Vol. 38, 70.
3. Bear, J. (1961) On the tensor form of dispersion. J. Geophys.
Research Vol. 66, No. 4, 1185-1197.
4. Bear, J. (1972) Dynamics of fluids in porous media. American
Elsevier, New York.
5. Bear, J. (1977) On the aquifers integrated balance equations.
Advances in Water Resources Vol. 1, 15-23.
6. Bear, J. and Bachmat, Y. (1991) Introduction to modeling of
transport phenomena in porous media. Kluwer Academic Publishers, Dordrecht.
7. Bear, J. and Verruijt, A. (1987) Modeling groundwater flow and
pollution. D. Reidel Publ., Dordrecht.
8. Beck, J.L. (1972) Convection in a box of porous material saturated with fluid. Phys. Fluids, Vol. 15, No. 8, 1377-1383.
9. Bird, R.B., Stewart, W.E. and Lightfoot, E.N. (1960) Transport
phenomena. John Wiley, New York.
10. Caltagirone, J.-P. (1974) Convection naturelle fluctuante en
milieu poreux. C. R. Acad. Sc. Paris, t. 278, Serie B, 259-262.
11. Caltagirone, J.-P. (1975) Thermoconvective instabilities in a horizontal porous layer. J. Fluid Mech. Vol. 72, Part 2, 269-287.
12. Cheng, P. (1977) Similarity solutions for mixed convection from
horizontal impermeable surfaces in saturated porous media. Int. J.
Heat Mass Transfer Vol. 20, 893-898.

13. Combarnous, M.A. and Bories, S.A. (1975) Hydrothermal convection in saturated porous media. Advances in Hydroscience, Ed.
by Ven Te Chow, Vol. 10, Academic Press, New York, 231-307.
14. Diersch, H.-J. (1980) Mehrphasiger Stoff-, Impuls- und Wrmetransport im ungesttigten und gesttigten porsen Medium
(Multiphase mass, momentum, and heat transport in an unsaturated and saturated porous medium). Acta Hydrophysica Vol. 25,
No. 4, 351-407.
15. Diersch. H.-J. (1981) Primitive variables finite element solutions
of free convection flows in porous media. Zeitschr. Angew. Math.
Mech. (ZAMM) Vol. 61, No. 7, 325-337.
16. Diersch, H.-J. (1981) Study of free convective flows in porous
media and effects of mechanical dispersion using finite element
methods. Gerlands Beitrge zur Geophysik Vol. 90, No. 6, 489506.
17. Diersch, H.-J. (1983) On finite element upwinding and its numerical performance in simulating coupled convective transport processes. Zeitschr. Angew. Math. Mech. Vol. 63, No. 10, 479-488.
18. Diersch, H.-J., Prochnow, D. and Thiele, M. (1984) Finite element analysis of dispersion-affected saltwater upconing below a
pumping well. Appl. Math. Modelling Vol. 8, No. 5, 305-312.
19. Diersch, H.-J. (1985) Modellierung und numerische Simulation
geohydrodynamischer Transportprozesse (Modeling and numerical simulation of geohydrodynamic transport processes). Habilitation (Dissertation B), Academy of Sciences of GDR, Berlin
1985, 267 p., Reprint WASY Ltd., Berlin 1991.
20. Diersch, H.-J. (1988) Finite element modelling of recirculating
density-driven saltwater intrusion processes in groundwater.
Advances in Water Resources Vol. 11, No. 1, 25-43.

cbcilt==NMP

TK=o
21. Diersch, H.-J., Kolditz, O. and Jesse, J. (1989) Finite element
analysis of geothermal circulation processes in hot dry rock fractures. Zeitschr. Angew. Math. Mech. Vol. 69, No. 3, 139-153.
22. Diersch, H.-J. (1992) Interactive, graphics-based finite element
simulation of groundwater contamination processes. Adv. Engineering Software Vol. 15, 1-13.
23. Diersch, H.-J., Grndler, R, Kaden, S, and Michels. I. (1992)
Toward GIS-based 3D/2D groundwater contamination modeling
using FEM. In: Computational Methods in Water Resources IX,
Vol. 1., Proc. IX. Intern. Conf. on Comp. Methods in Water
Resources, June 1992, Denver, Comp. Mech. Publ., Southampton, 749-760.
24. Diersch, H.-J. (1993) GIS-based groundwater flow and contaminant transport modeling - the simulation system FEFLOW. In:
Praxis der Umweltinformatik Vol. 4 (Rechnergesttzte Ermittlung, Bewertung und Bearbeitung von Altlasten), ed. F. Ossing,
Metropolis-Verlag, Marburg, 187-208.
25. Diersch, H.-J. (1994) Computational aspects in developing an
interactive 3D groundwater transport simulator using FEM and
GIS. Intern. Conf. on Groundwater Quality Management (GQM
93), Tallinn, Estonia, 6-9 September 1993, IAHS Publication No.
220, 313-326.
26. Diersch, H.-J., Voigt, R. and Grndler, R. (1994) Visiometric
techniques in a 3D groundwater transport code. In: Computational Methods in Water Resources X, Vol. 2, X. Intern. Conf. on
Comp. Methods in Water Resources, July 1994, Heidelberg, Kluwer Academic Publishers, 1449-1456.
27. Diersch, H.-J.G. (1996) Finite element analysis of three-dimensional transient free convection processes in porous media. In:
Three-Dimensional Complex Flows, (Proc. Intern. IMACS-COST
Conf. Lausanne, September 13-15, Switzerland, 1995), ed. by M.
Deville et al., Notes on Numerical Fluid Mechanics Vol. 53,
Vieweg & Sohn, Braunschweig/Wiesbaden, 1996, 38-44.
28. Diersch, H.-J.G. and Michels, I. (1996) Moving finite element
meshes for simulating 3D transient free surface groundwater flow
and transport processes. In: Computational Methods in Water
Resources XI, Vol. 1 Computational Methods in Subsurface Flow
and Transport Problems, ed. A.A. Aldama et al., Comp. Mech.
Publ., Southampton, 1996, 85-92.
29. Diersch, H.-J.G. (2002) About the difference between the convective form and the divergence form of the transport equation.
FEFLOW White Papers - Vol. I, WASY Ltd, Berlin.

NMQ==o=j~~=J=m~=f

30. Diersch, H.-J.G. and Kolditz, O. (1998) Coupled groundwater


flow and transport: 2. Thermohaline and 3D convection systems.
Advances in Water Resources Vol. 21, No. 5, 401-425.
31. Diersch, H.-J.G. (2000) Note to the opposing flow regime at
mixed convection around a heated cylinder in a porous medium.
Transport in Porous Media Vol. 38, No. 3, 345-352.
32. Diersch, H.-J.G. and Kolditz, O. (2002) Variable-density flow and
transport in porous media: approaches and challenges. Advances
in Water Resources Vol. 25, Nos. 8-12 (Special issue), 899-944.
33. Elder, J.W. (1967) Transient convection in a porous medium. J.
Fluid Mech. Vol. 27, Part 3, 609-623.
34. Frolkovic, P. and De Schepper, H. (2000) Numerical modelling of
convection dominated transport with density driven flow in
porous media. Advances in Water Resources Vol. 24, No. 1, 6372.
35. Goldstein H., Poole, C. and Safko, J. (2002) Classical mechanics.
3rd edition, Addison Wesly, San Francisco.
36. Gray, W.G. (1982) Derivation of vertically averaged equations
describing multiphase flow in porous media. Water Resources
Research Vol. 18, No. 6, 1705-1712.
37. Hassanizadeh, S.M. (1986) Derivation of basic equations of mass
transport in porous media. Part 1. Macroscopic balance laws, Part
2. Generalized Darcys and Ficks laws. Advances in Water
Resources Vol. 9, 196-222.
38. Hassanizadeh, S.M. (1988) Modeling species transport by concentrated brine in aggregated porous media. Transport in Porous
Media Vol. 3, 299-318.
39. Holst, P.H. and Aziz, K. (1972) Transient three-dimensional natural convection in confined porous media. Int. J. Heat Mass Transfer Vol. 15, 73-90.
40. Holzbecher, E. (1998) Modeling density-driven flow in porous
media. Springer-Verlag, Berlin.
41. Homsy, G.M. and Sherwood, A.E. (1976) Convective instabilities
in porous media with through flow. AIChE Journal Vol. 22, No.1,
168-174.
42. Horne, R.N. and OSullivan, M.J. (1974) Oscillatory convection
in a porous medium heated from below. J. Fluid Mech. Vol. 66,
Part 2, 339-352.
43. Horne, R.N. and OSullivan, M.J. (1978) Origin of oscillatory
convection in a porous medium heated from below. Phys. Fluids
Vol. 21, No. 8, 1260-1264.

44. Horne, R.N. (1979) Three-dimensional natural convection in a


confined porous medium heated from below. J. Fluid Mech. Vol.
92, Part 4, 751-766.
45. Horne, R.N. and Caltagirone, J.-P. (1980) On the evolution of
thermal disturbances during natural convection in a porous
medium. J. Fluid Mech. Vol. 100, Part 2, 385-395.
46. Horton, C.W. and Rogers Jr., F.T. (1945) Convection currents on
a porous medium. Vol. 16, J. Appl. Phys., 367-369.
47. Hsu, C.T., Cheng, P. and Homsy, G.M. (1978) Instability of free
convection flow over a horizontal impermeable surface in a
porous medium. Int. J. Heat Mass Transfer Vol. 21, 1221-1228.
48. Knupp, P. (1996) A moving mesh algorithm for 3-D regional
groundwater flow with water table and seepage face. Advances in
Water Resources Vol. 19, No. 2, 83-95.
49. Kolditz, O. and Diersch, H.-J. (1993) Quasi-steady-state strategy
for numerical simulation of geothermal circulation in hot dry rock
fractures. Int. J. Non-Linear Mechanics, Vol. 28, No. 4, 467-481.
50. Kolditz, O. and Zielke, W. (1994) Modell des geklfteten porsen
Mediums fr Festgesteinsaquifere (Model of the fractured porous
medium for rock aquifers). Preprint, Institute for Flow Mechanics, Hannover, 57 p.
51. Kolditz, O., Ratke, R., Diersch, H.-J.G. and Zielke, W. (1998)
Coupled groundwater flow and transport: 1. Verification of density variable flow and transport models. Advances in Water
Resources Vol. 21, No. 1, 27-46.
52. Lever, D.A. and Jackson, C.P. (1985) On the equations for the
flow of concentrated salt solution through a porous medium. U.K.
DOE Report No. DOE/RW/85.100.
53. McKibbin, R. and OSullivan, M.J. (1980) Onset of convection in
a layered porous medium heated from below. J. Fluid Mech. Vol.
96, Part 2, 375-393.
54. Mercer, J.W. und Pinder, G.F. (1974) Finite element analysis of
hydrothermal systems. Finite Element Methods in Flow Problems
(Proc. 1st Symp., Swansea, 1974), (ed. Oden, J. T. et al.) Univ. of
Alabama Press, 401-414.
55. Nield, D. A. and Bejan, A. (1992) Convection in porous media.
Springer Verlag, New York Inc.
56. Palm, E., Weber, J.E. and Kvernvold, O. (1972) On steady convection in a porous medium. J. Fluid Mech., Vol. 54, Part 1, 153161.
57. Perrochet, P. (1996) Personal communication, EPFL Lausanne,
GEOLEP Laboratoire de gologie, Lausanne (Switzerland).

58. Robinson, J.L. and OSullivan, M.J. (1976) A boundary-layer


model of flow in a porous medium at high Rayleigh number. J.
Fluid Mech. Vol. 75, Part 3, 459-467.
59. Rubin, H. (1975) On the analysis of cellular convection in porous
media. Int. J. Heat Mass Transfer Vol. 18, 1483-1486.
60. Rubin, H. and Roth, C. (1978) Instability of horizontal thermohaline flow in a porous medium layer. Israel Journal of Technology
Vol. 16, No. 5/6, 216-223.
61. Rudraiah, N. and Srimani, P.K. (1980) Finite-amplitude cellular
convection in a fluid-saturated porous layer. Proc. R. Soc. Lond.,
A 373, 199-222.
62. Scheidegger, A.E. (1961) General theory of dispersion in porous
media. J. Geophys. Res. Vol. 66, No. 10, 3273-3278.
63. Schubert, G. and Straus, J.M. (1979) Three-dimensional and multicellular steady and unsteady convection in fluid-saturated
porous media at high Rayleigh numbers. J. Fluid Mech. Vol. 94,
Part 1, 25-38.
64. Segol, G. (1994) Classic groundwater simulations: proving and
improving numerical models. PTR Prentice Hall, Englewood
Cliffs, New Jersey, 531p.
65. Simkins, P.G. and Blythe, P.A. (1980) Convection in a porous
layer. Int. J. Heat Mass Transfer Vol. 23, 881-887.
66. Springer, J. (1996) Shape-derived anisotropy directions in quadrangle and brick finite elements. Communications in Numerical
Methods in Engineering Vol. 12, 351-357.
67. Straus, J.M. (1974) Large amplitude convection in porous media.
J. Fluid Mech. Vol. 64, Part 1, 51-63.
68. Straus, J.M. and Schubert, G. (1979) Three-dimensional convection in a cubic box of fluid-saturated porous material. J. Fluid
Mech. Vol. 91, Part 1, 155-165.
69. Trevisan, O.V. and Bejan, A. (1985) Natural convection with
combined heat and mass transfer buoyancy effects in a porous
medium. Int. J. Heat Mass Transfer Vol. 28, No. 8, 1597-1611.
70. Wooding, R.A. (1962) Free convection of fluid in a vertical tube
filled with porous material. J. Fluid Mech. Vol. 13, 126-144.
71. Yeh, G.-T., Cheng, J.-R. and Bensabat, J.A. (1994) A threedimensional finite-element model of transient free surface flow in
aquifers. In: Computational Methods in Water Resources X, Vol.
1, X. Intern. Conf. on Comp. Methods in Water Resources, July
1994, Heidelberg, Kluwer Academic Publishers, 131-138.
72. Younes, A., Ackerer, P., and Mos, R. (1999) Modeling variable
density flow and solute transport in porous medium: 1. Re-evalu-

cbcilt==NMR

TK=o
ation of the salt dome flow problem. Transport in Porous Media
Vol. 35, No. 3, 375-394.
73. Zebib, A. and Kassoy, D.R. (1978) Three-dimensional natural
convection motion in a confined porous medium. Phys. Fluids
Vol. 21, No. 1, 1-3.

NMS==o=j~~=J=m~=f

k~

k~

The present list of notations encompasses dimensions of the quantities and relations to relevant equations of the text. Symbols without dimensions possess
only qualitative information, though sometimes quantitative properties are possible to be related to, as for
instance in chemical reaction equations, where corresponding object quantities (stoichiometric coefficients)
and object masses (molar masses of reactants and products) can be balanced. All chemical activities and
chemical equilibrium constants are generally defined
by a dimension of unity. Notations of molar quantities
are usually not given.

Latin symbols
Ak
a ij
ak
B
Bk
b 1, b 2
b2

reactant
1
rotation matrix, (5-2), (5-9), (5-19)
1
activity of species k, (1-21)
L
aquifer thickness, (2-5), Fig. 2.1
chemical product
3 1 b2
( ML )
Freundlich sorption coefficient,
(1-44), (1-51), (3-7)
1
Freundlich sorption exponent, (1-

3 1

b
C

ML T
3
ML
R

C 1 , C 2 , C 3 ML

CI

ML

Ck

ML

Cm

ML

Co
Cs

ML
3
ML

CT

ML

ML

L T

CT
c

Dd

2 2
2 1

L T
2 1
L T

44), (1-51), (3-7)


reaction rate constant, (1-25)
concentration of essential transport
component in the fluid (aqueous)
phase, (1-56)
boundary values of concentration,
(4-18), (4-19), (4-20), (4-21)
initial condition of concentration,
(4-2)
concentration of component k of phase
concentration of single well m, (424)
reference concentration, (1-77)
maximum concentration, (1-76), (178), (1-79), (1-80)
total solution normality of fluid
phase, (1-38)
ion exchange capacity of solid
phase, (1-37)
specific heat capacity of -phase,
(1-93), (1-127)
diffusion constant, (1-140)
molecular medium diffusion of phase, (1-13), (3-5), (3-13), (3-18)

cbcilt==NMT

UK=k~
2 1

D ij

L T

D ij

L T

mol

D ij

disp
D ij

d
E

3 1

2 1

L T

2 1

L T

L
2 2
L T
#

2 2

E
ei
f

L T
1

fk
f
G
g
gi
h
R R
h1 , h2

1
1

hI

ho
J
ji

j ik

L
L

LT
2
LT
L
L

2 1

ML T

j iT

MT

LT

NMU==o=j~~=J=m~=f

mol f

disp f

= D ij + D ij
tensor
of
hydrodynamic dispersion of fluid
phase, (1-56)
depth-integrated
tensor
of
hydrodynamic dispersion, (3-13),
(3-18)
tensor of molecular macrodiffusion
of -phase, (1-13)
tensor of mechanical dispersion of
-phase, (1-13)
thickness, Figs. 1.8, 4.2, 4.3
internal (thermal) energy density,
Tab. 1.1
activated energy, (1-26)
gravitational unit vector, (1-28)
specific
rate
of
temporary
production, (1-2)
activity coefficient, (1-34)
viscosity relation function, (1-101)
extensive property, (1-1)
gravitational acceleration, (1-28)
gravity vector, (1-28)
hydraulic head, (1-27)
prescribed boundary values of
hydraulic head h, (4-4), (4-6)
initial condition of hydraulic head h,
(4-1)
reference hydraulic head, (1-77)
Jacobian matrix, (5-13)
diffusive flux vector, (1-6)
Fickian mass flux vector of phase, (1-13)
Fourierian heat flux vector of phase, (1-14)
coefficient of isotropic hydraulic

Kd
K eq
K ij
a

K ij
m

Ki

K max, K min
k 1, k 2

k ij
M
mk
N
N
Ns
ni
Po
p

p
Q EB

conductivity, (1-140), (1-141), (521)


1 3
s
M L
= distribution coefficient,
(Chapter 3)
1
equilibrium constant, (1-21)
1
LT
tensor of hydraulic conductivity of
fluid phase, (1-30), (1-122)
1
LT
effective (intrinsic) conductivity of
fluid phase, (1-123)
1
LT
hydraulic conductivity along the
principal directions, (5-1)
1
LT
maximum
and
minimum
conductivity, respectively, (5-6), (57)
1 3
1 and M L
Langmuir numerator and
Langmuir denominator sorption
coefficient, respectively, (1-41), (152), (3-8)
2
L
permeability tensor of -phase, (111)
number of phases
1
Mmol
molecular weight of species k, (134), (1-35)
number of chemical components
(species)
number of reactants
number of sorbed species, (1-37)
1
normal unit vector (positive
outward), Fig. 4.1
1
LT
groundwater recharge (infiltration
rate), (2-11), (4-10)
1 2
ML T
thermodynamic pressure of phase, (1-11)
1 2
ML T
fluid pressure, (1-116)
1
T
term of extended Boussinesq
approximation, (1-110), (1-138)

QC
QC
R

QC
w

QC
R
Qh
w

Qm
Q
Q
w


Q
i

QT
QT

QT
R
QT

QT
R R
q C, q C

3 1

ML T
mass source term, (1-57)
2 1
ML T
= B Q C depth-integrated mass
source term, (2-3), (3-11), (3-16)
1
MT
lumped balance flux of contaminant
mass (positive inward), (4-44), (446), (4-49), (4-50)
1
MT
mass flux well function, (4-24)
3 1
L T
lumped balance flux of fluid
(positive inward), (4-42)
3 1
L T
pumping/injecting rate of single
well m, (4-9)
1
T
specific sink/source rate of fluid, (1102)
1
LT
= B Q depth-integrated specific
sink/source rate of fluid, (1-108), (21), (3-9), (3-14)
3 1
L T
well function, (4-9)
1
T
specific mass supply of -phase,
(1-9)
2 1
ML T
mass supply through a --interface
(positive outward from to ), (118)
1 3
ML T
source/sink function of heat, (1-64),
(3-4)
3
MT
= B Q T depth-integrated source/
sink function of heat, (2-4), (3-12),
(3-13), (3-17), (3-18)
2 3
L T
heat supply of -phase, (1-64)
2 3
ML T
lumped balance flux of heat
(positive inward), (4-45), (4-46), (449)
2 3
ML T
heat flux well function, (4-36)
2 1
ML T
prescribed normal boundary mass
flux of the dispersive and the total
(convective plus dispersive) part,
respectively, (4-19), (4-25)

q C, q C

1 1

ML T

LT

L T

LT
2 1
L T

qh

2 1

qh

qi
f
qi

2 1

qn

qn

qn

qn

qn

qn

ML T

1 1

ML T
LT

2 1

L T

MT

MLT

qT, qT

qT, qT

MT

MLT

depth-integrated prescribed normal


boundary mass flux of the dispersive
and the total (convective plus
dispersive) part, respectively, (420), (4-25)
normal boundary fluid flux (positive
outward), (4-5), (4-11)
depth-integrated normal boundary
fluid flux (positive outward), (4-5),
(4-11)
Darcy velocity of fluid phase, (1-29)
depth-integrated Darcy velocity of
fluid phase, (1-69)
normal
mass
flux
(positive
outward), (4-19), (4-21), (4-25), (426)
depth-integrated normal mass flux
(positive outward), (4-20), (4-21)
normal Darcy flux of fluid (positive
outward), (4-5), (4-6), (4-11), (4-12)
depth-integrated normal Darcy flux
of fluid (positive outward), (4-5), (46)
normal heat flux (positive outward),
(4-31), (4-33), (4-37), (4-38)
depth-integrated normal heat flux
(positive outward), (4-32), (4-33)
prescribed normal boundary heat
flux of the conductive plus
thermodispersive part and the total
(convective plus conductive plus
thermodispersive) part, respectively,
(4-31), (4-37)
depth-integrated normal boundary
heat flux of the conductive plus
thermodispersive part and the total
(convective plus conductive plus

cbcilt==NMV

UK=k~

2 2

R
R

L T
1

Rd

Rd

Ra s
Ra t
Rk

1
1
3 1
ML T

het

3 1

Rk

ML T

3 1

rk

ML T

S
So

1
1
L

R
T1 ,

R
T2 ,

R
T3

T
TI

T ij

L T

2 1
2 1

T max, T min L T
w

Tm

NNM==o=j~~=J=m~=f

thermodispersive) part, respectively,


(4-32), (4-37)
universal gas constant, (1-26)
retardation factor, (1-49), (3-5), (36), (3-7), (3-8)
derivative term of retardation, (3-5),
(3-6), (3-7), (3-8)
= BR
depth-integrated
retardation factor, (3-13), (3-18)
= B Rd
depth-integrated
derivative term of retardation, (313), (3-18)
solutal Rayleigh number, (1-140)
thermal Rayleigh number, (1-141)
macroscopic reaction rate of
component k, (1-15), (1-16)
heterogeneous
reaction
of
component k, (1-15)
homogeneous
reaction
of
component k in the -phase, (1-15)
storage coefficient, (2-1), (2-6)
specific
storage
coefficient
(compressibility), (1-109)
system (equilibrium) temperature,
(1-60)
prescribed boundary values of
temperature T, (4-30), (4-31), (432), (4-33)
temperature of -phase
initial condition of temperature T,
(4-3)
tensor of transmissivity of fluid, (170)
maximum
and
minimum
transmissivity, respectively
temperature of single well m, (4-36)

To
t
t1 2

T
T

U
US
ui

ML T

V
f
Vq
f

3 1

LT
1
LT
2 1

Vq

L T

vi

vi
G
vi
#
v
wi

LT
1
LT
1
LT
3 1
ML T
1
LT

xi

L
L

bottom

L
L
1

xi
xi
xl

xl
top
xl
zk

reference temperature, (1-77)


time
reaction half-life, footnote in
Section 1.4.2
microscopic volume element
frequency factor, (1-26)
vectors of computed principal axes,
(5-14), (5-15), (5-16), Fig. 5.4

v i v i absolute flux of -phase
f f
q i q i absolute Darcy flux of fluid
phase
f f
q i q i absolute depth-integrated
Darcy flux of fluid phase
barycentric velocity, (1-6), Tab. 1.1
velocity vector of -phase, (1-11)
particle velocity of property G, (1-3)
decay frequency, (1-25)
macroscopic velocity of interface,
(1-8), (2-12)
Cartesian coordinates, Eulerian
spatial coordinate vector
coordinate vector of single well m,
(4-9), (4-24), (4-36)
principal axes, Figs. 5.1, 5.2
elevation; Cartesian coordinate
along acting of gravity, (2-14)
bottom geometry of aquifer, Fig. 2.1
top geometry of aquifer, Fig. 2.1
charge number of species (ion) k, (132)

Greek symbols

fluid density difference ratio, (1-76),


(1-78), (1-80)

L, T

L, T
, i

C
h
T
()
ij

ij
ij

longitudinal
and
transverse
thermodispersivity, respectively, (114)
1

volumetric thermal fluid expansion


coefficient (constant), (1-76)
1

modified thermal fluid expansion


function (variable), (1-105)
L
longitudinal
and
transverse
dispersivity,
respectively,
of
chemical component, (1-13)
total boundary and boundary
portion, respectively
1
L
fluid compressibility, (1-76)
3
ML
concentration difference, (1-140)
L
difference of hydraulic head, (2-6)

temperature difference, (1-141)


Dirac delta function
Kronecker tensor ( ij = 1 for
i = j , else ij = 0 for i j )
1
porosity (= volume fraction of fluid
phase)
1
volume fraction of -phase
1
effective porosity, (1-114)
local coordinate
local coordinate
Eulerian angle
1
T
chemical decay rate, (1-54)
1
Henry sorptivity coefficient, (1-43),
(1-50), (3-6)
2 1
f
s
f f
L T
= ( + ( 1 ) ) ( c )
thermodiffusivity, (1-141)
3 1
MLT tensor
of
hydrodynamic
thermodispersion, (1-63)
2 3 1
ML T depth-integrated
tensor
of
hydrodynamic
thermodispersion,
(3-13), (3-18)

cond

ij

disp

ij

k
f

kr
aniso

f
f

ij

MLT

thermal conductivity of -phase,


(1-14)
3 1
MLT conductive part of thermodispersion
tensor of -phase, (1-14)
3 1
MLT dispersive part of thermodispersion
tensor of -phase, (1-14)
1 1
ML T
dynamic viscosity of -phase, (117), (1-98), (1-99)
2 2
L T
chemical potential of the kth species
in the -phase, (1-20)
1 1
ML T
reference viscosity of the fluid phase
related
to
the
reference
concentration C o and the reference
temperature T o , (1-101)
1 1
ML T
reference viscosity of the fluid phase
related
to
the
reference
concentration C = 0 and the
reference temperature T = 150C ,
(1-98), (1-99)
stoichiometric number of species k
and reaction r
1
factor of anisotropy, (5-8)
1
inhibition factor, (1-96)
local coordinate, Fig. 4.7
3
ML
density function
3
s
ML
= ( 1 ) bulk density of porous
medium, (Chapter 3)
3
ML
density of fluid phase, (1-31), (177), (1-84), (1-87)
3
ML
reference density of fluid phase
related to the reference head h o , the
reference concentration C o , and the
reference temperature T o , (1-27),
(1-77)
3
ML
density of solid phase
1 2
ML T
stress tensor, Tab. 1.1

cbcilt==NNN

UK=k~

C
C
in

out

in

out

C , C
C , C

h
h
in

out

in

out

h , h
h , h

T
T
in

out

in

out

T , T
T , T

NNO==o=j~~=J=m~=f

normalized temperature, (1-99), (35)


1
L
coefficient
of
skeleton
compressibility, (1-92)
1
Heaviside function, (2-6)
1
LT
mass transfer coefficient, leaching
parameter, (4-21), (4-22), (4-26)
2 1
L T
vertically averaged mass transfer
coefficient,
aquifer
leaching
parameter, (4-21), (4-23), (4-26)
1
LT
directional coefficient of mass intransfer and mass out-transfer,
respectively, (4-22), (4-28)
2 1
L T
vertically averaged directional
coefficient of mass in-transfer and
mass out-transfer, respectively,(423), (4-29)
1
T
fluid transfer coefficient, colmation
parameter, (4-6), (4-7), (4-12)
1
LT
vertically averaged fluid transfer
coefficient,
aquifer
colmation
parameter, (4-6), (4-8), (4-12)
1
T
directional coefficient of fluid intransfer and fluid out-transfer,
respectively, (4-7), (4-14), (4-16)
1
LT
vertically averaged directional
coefficient of fluid in-transfer and
fluid out-transfer, respectively, (48), (4-15), (4-17)
3 1
MT
heat transfer coefficient, (4-33), (434), (4-38)
3 1
MLT vertically averaged heat transfer
coefficient, (4-33), (4-35), (4-38)
3 1
MT
directional coefficient of heat intransfer and heat out-transfer,
respectively, (4-34), (4-40)
3 1
MLT vertically averaged directional

(C)

coefficient of heat in-transfer and


heat out-transfer, respectively, (435), (4-41)
Eulerian angle, rotation angle, Fig.
5.2
adsorption function, (1-50), (1-51),
(1-52)
intensive balance quantity, (1-2),
Tab. 1.1
Eulerian angle
mass fraction of component k, Tab.
1.1
mass fraction of species k sorbed on
the solid phase, (1-36)
domain

Indices
,
, ,
f
I
i, j
k
l
n
q
R
r
s
w

phase indicators
local coordinate directions
fluid (water) phase
initial condition
spatial Eulerian coordinate
chemical component (species)
gravity acting direction in the
Cartesian coordinate system
normal direction
chemical component (species)
boundary
chemical reaction
solid phase
pumping/injecting well

Abbreviations
AREV
BASD
FEFLOW
FEM
LMA
REV
WASY
2D
3D

aquifer representative elementary


volume
best adaptation to stratigraphic data
finite element subsurface flow
system
finite element method
law of mass action
representative elementary volume
institute for water resources
planning and systems research
two-dimensional
three-dimensional

cbcilt==NNP

UK=k~

NNQ==o=j~~=J=m~=f

m~=ff

^~

cbcilt==NNR

NNS==o=j~~=J=m~=ff

q=m

q=m

VKN f
Theis problem examines the transient lowering of
the water table induced by a pumping well. Theis fundamental insight was to recognize that Darcys law is
analogous to the law of the flow of heat by conduction,
hydraulic pressure being analogous to temperature,
pressure-gradient to thermal gradient. Therefore, the
mathematical theory of heat conduction developed by
Fourier and subsequent writers (see Carslaw and
Jaeger1) is largely applicable to hydraulic theory.
(cited from Theis5).

VKP ^
The following assumptions must be introduced to
obtain an analytical solution:
aquifer: confined, infinite areal extend, homogeneous,
isotropic, uniform thickness, horizontal piezometric
surface;
well:constant discharge rate, well penetrates the entire
thickness, well storage effects can be neglected.

VKO p~= p= =
^= q= E`
^F
Figure 9.1 shows a schematical sketch of the considered aquifer type.
Figure 9.1 Aquifer type (from Kruseman and Ridder2).

cbcilt==NNT

VK=q=m

VKR ^~~=p
VKQ d=b~

The analytical solution of Eq. (9-1) with the conditions Eq. (9-2) and Eq. (9-3) derived by Theis is:

The governing equations of the Theis problem are:

Q
h ( t, r ) = h 0 ---------- W ( u )
4T

h 1--- h
--S- -----h -------
= 0
T t r 2 r r

(9-1)

with the well function:

initial conditions:

inf

h ( 0, r ) = h 0

(9-2)

boundary conditions:
lim r

r 0

W ( u ) = Ei ( u ) =

lim h ( t, r ) = h 0

r inf

(9-3)

e ( )

- d
---------

(9-5)

W ( u ) = 0.5772 ln ( u ) +

h
Q
= --------- r
2T

(9-4)

inf

ui

( 1 ) i 1 ---------i i!

(9-6)

i=1

r2S
u = -------4Tt

with:
Table 9.1 Model parameters
Sym
bol

NNU==o=j~~=J=m~=ff

Property

Unit

hydraulic head

radial distance from the well

storage coefficient

time

transmissivity

L2 T-1

constant discharge rate

L3 T-1

VKS k~=^~=EcbjF
For the numerical analysis of Theis problem we
use a case study by Segol4, who referred to results from
Pinder and Frind3. They tested different finite element
meshes and element types (cf. Figure 9.2). Table 9.2
summarizes the parameters for simulation. Figure 9.2
shows the finite element grid used by Pinder and
Frind3. These finite elements are used as superelements
in our following investigation. The refined mesh is presented in Figure 9.3.

VKS=k~=^~=EcbjF

Figure 9.3 Refined finite element grid (FEFLOW).


Figure 9.2 Coarse finite element grid (from Pinder and
Frind3).

Table 9.2 Simulation parameters


Parameters

Pinder and Frind3

FEFLOW

h(0,r) = 0

h(0,r) = 0, h(0,rb) = 0.01

Q = 0.5 ft3s-1 = 1.22329x103 m3d-1


rb = 1 ft (wellbore radius)

q = Q/2 rb = 638.75425 m2d-1


rb = 0.3048 m

h(t,1000 ft.) = 0

h(t,304.8 m) = 0

T = 0.01ft2s- 1 = 9.2903x10-4 m2s-1

T = 9.2903 [x10-4] m2s-1

S = 0.001

S = 0.001

unsteady flow
flow initials
flow boundaries:
well pumping rate
no drawdown at the
outer boundary
flow materials:
transmissivity
storage coefficient
storage compressibility

So= 0

FEM:
mesh

Figure 9.2

Figure 9.3

cbcilt==NNV

VK=q=m
Table 9.2 Simulation parameters (continued)
Parameters

Pinder and Frind3

time stepping regime

variable

FEFLOW
1000 x 1.15741x10-2 d
automatic (FE/BE, AB/TR)
t0 = 10-5 d (169 steps at all)

solver

FEFLOW provides two temporal discretization


techniques: (1) implicit or Crank-Nicolson time stepping with predefined time steps and (2) automatically
controlled time stepping based on predictor-corrector
techniques of first (FE/BE) and second order (AB/TR)
accuracy, which both are applied in the following
study.

CG

VKT o
Figure 9.5 and Table 9.3 show the comparison of
analytically and numerically calculated drawdown of
hydraulic head versus time at the distance of r = 9.639
m from the well.

Figure 9.4 Calculated drawdowns at a distance of 31.6228 ft. from the well (from Pinder
and Frind3).

NOM==o=j~~=J=m~=ff

VKT=o

Figure 9.5 Calculated drawdowns at a distance of 9.639 m from the well (FEFLOW).

Table 9.3 Comparison of analytically and numerically calculated drawdown


Time in [s]

Analytical

10

Constant time
steps

FE/BE

AB/TR

0.0375

0.0312

0.0254

20

0.214

0.170

0.167

50

0.679

0.648

0.660

100

1.291

1.218

1.240

200

2.029

1.907

1.930

cbcilt==NON

VK=q=m
Table 9.3 Comparison of analytically and numerically calculated drawdown (continued)
Time in [s]

Analytical

500

2.993

1000

3.829

2000

Constant time
steps

FE/BE

AB/TR

2.917

2.951

3.715

3.747

4.690

4.523

4.561

5000

5.732

5.613

5.642

10000

6.594

6.442

6.474

20000

7.469

7.257

7.298

50000

8.519

8.052

8.083

100000 = 1.16 d

9.384

8.237

8.239

200000

10.260

8.257

8.257

500000

11.311

8.258

8.258

1000000 = 11.6 d

12.176

8.258

8.258

Up to one day the agreement of analytical and


numerical results is quite well. Afterwards, the numerical solution is already influenced by the constant head
condition at the outer boundary. In contrary, the corresponding boundary condition for the analytical solution, following to Eq. (9-3), is constant head at infinite
distance from the well. To improve the agreement of
the drawdown curves for later times the calculation
domain must be enlarged.

NOO==o=j~~=J=m~=ff

3.165

6.421

8.248

8.258

VKU o
1. Carslaw, H.S. and Jaeger, J.C. (1959) Conduction of heat in solids. 2nd ed. Oxford: Oxford University Press.
2. Kruseman, G.P. and de Ridder, N.A. (1990) Analysis and evaluation of pumping test data, ILRI publication 47, ISBN 90 70754
207.
3. Pinder, G.F. and Frind, E.O. (1972) Application of Galerkins procedure to aquifer analysis. Water Resources Research 8(1), pp.
108-120.
4. Segol, G. (1994) Classic groundwater simulation: proving and
improving numerical models. PRT Prentice Hall, Englewood
Cliffs, New Jersey.
5. Theis, C.V. (1935) The relation between lowering of the piezometric surface and the rate and duration of discharge of a well
using groundwater storage. Trans. Amer. Geophys. Union 16, pp.
519-524.

l~~=~=_~=mW=NJa=^JaJ
Ja~

NM

l~~=~=_~=mW=NJa=^JaJa~

NMKN p~===m
Figure 10.1 shows typical field situations, which
permit a reduction of the dimensionality following to
an 1-D problem

NM

NMKO ^
The following assumptions must be introduced to
obtain an analytical solution:
aquifer:confined, infinite areal extend, homogeneous,
isotropic, uniform thickness, horizontal piezometric
surface
well gallery:constant discharge rate, well penetrates the
entire thickness, one-dimensional

Figure 10.1 Typical field situations for the application of 1-D models (from Sauty3 and Kinzelbach1).

cbcilt==NOP

NMK=l~~=~=_~=mW=NJa=^JaJa~

NMKP d=b~

c ( 0, x ) = 0

The governing equations of the considered problem


are:

boundary conditions:
c ( t, 0 ) = 1,

c
c
c
R ----- + q x D xx -------2- + Rc = 0
x
t
x

(10-1)

(10-2)

lim c ( t, x ) = 0

x inf

with:

initial condition:
Table 10.1 Model parameters
Symbol
c
Dd

Name

Unit

contaminant concentration

ML-3

molecular diffusion coefficient

L2 T-1

D xx = D d + L q x = D Scheidegger-Bear hydrodynamic dispersion


tensor
q x = v x = q = v
(1 )
R = 1 + ----------------

t1 2
vx = v

L T-1

retardation factor

half-life of radioactive decay

pore velocity

LT-1

coordinate along the porous strip

coefficient of longitudinal dispersivity

Henry sorption coefficient

= ln2 t 1 2

NOQ==o=j~~=J=m~=ff

Darcy flux, specific discharge

L2 T-1

decay rate (cf. Section 1.4.2)


kinematic porosity

T-1
1

(10-3)

NMKQ=^~~=p

NMKQ ^~~=p

and theEq. (10-4) reduces to:

The analytical solution of Eq. (10-1) with the conditions Eq. (10-2) and Eq. (10-3) derived by Ogata and
Banks2 is the following:
c1
x( 1 )
x vt/R
---= --- exp ------------------- erfc -----------------------c0
2
2 L
2 v t/R
x(1 + )
x + vt/R
+ exp ------------------- erfc ------------------------
2 L
2 v L t/R

(10-4)

c1
x(1 )
x vt/R
---= --- exp ------------------- erfc ------------------------
c0
2
2 L
2 v t/R

(10-9)

NMKR k~=^~=EcbjF
The Figure 10.2 shows the finite element grid used
for the numerical analysis.

with
=

1 + 4R L /v

(10-5)

and the complementary error function

Figure 10.2 Finite element grid used in FEFLOW.

2 2
erfc ( u ) = 1 ------- e d

(10-6)

Table 10.2 Simulation parameters

Neglecting decay effects, = 0, = 1, the


Eq. (10-4) simplifies:
x
c1
x vt/R
x + vt/R
---= --- erfc ------------------------ + exp ------ erfc ------------------------
L
c0
2
2 v t/R
2 v t/R
L

x( 1 + )
x + v t/R
lim exp ------------------- erfc ------------------------ = 0
2 L
2 v L t/R

FEFLOW

steady flow
(10-7)

Considering migration of ideal tracer we can use


R = 1. For estimating the long-term behavior we can
apply the limes relation:
x

Parameters

(10-8)

flow boundaries:
constant head at the
injection point

h(t,x=0m) = 0

constant flux at the


outer boundary

q(t,x=100m) = 0.1 m2d-1

flow materials:
transmissivity

T = 1.0 [x10-4] m2s-1

cbcilt==NOR

NMK=l~~=~=_~=mW=NJa=^JaJa~
Table 10.2 Simulation parameters (continued)
Parameters

FEFLOW

unsteady transport
transport initials:
homogeneous concentration

c(0,x) = 0

transport boundaries:
constant concentration
at the injection point

c(t,0) = c 0

transport materials:
aquifer thickness

M = 1m

porosity

= 0.2

sorption coefficient
molecular diffusivity
longitudinal, transverse dispersivity
decay rate

NMKS o
Figure 10.3 and Table 10.3 show the comparison of
analytically and numerically calculated breakthrough
curves of contaminants at different time steps including
all transport mechanisms due to Eq. (10-1). We achieve
very good agreement of the analytical and numerical
results. Figure 10.4 and Table 10.4 illustrate the comparison of analytically and numerically calculated
breakthrough curves of contaminants at different times
excluding decay according to Eq. (10-7). For long-term
calculations it is difficult to compute the second term in
Eq. (10-4) with sufficient accuracy using series approximation. Therefore, we use the asymptotic solution corresponding to Eq. (10-9), which differs slightly from
the numerical one.

= 0, therefore, R = 1.0
D d = 0 [10-9 m2s-1]
L = 5m
= 0.002 x [10-4s-1]

FEM
mesh

Figure 10.2

time stepping regime

constant: t = 0.1-1 d

solver

PCG (ORTHOMIN)
Figure 10.3 Calculated breakthrough curves at different
times (advective-dispersive-decay transport).

NOS==o=j~~=J=m~=ff

NMKS=o

Table 10.3 Comparison of analytical and numerical results (including decay)


distance
in [m]

t = 10 d

t = 50 d

t = 500 d

analytical

numerical

analytical

numerical

analytical

numerical

0.94030

0.93971

0.96938

0.96940

0.97040

0.97043

0.87637

0.87547

0.93944

0.93942

0.94168

0.94167

0.66660

0.66460

0.85320

0.85317

0.86051

0.86053

10

0.32769

0.32585

0.71906

0.71889

0.74048

0.74039

20

0.02430

0.02559

0.47540

0.47518

0.54831

0.54842

40

0.11803

0.11827

0.30064

0.30071

60

0.00677

0.00898

0.16484

0.16471

80

0.00012

0.00016

0.09038

0.09047

100

0.05107

0.05662

Table 10.4 Comparison of analytical and numerical results (excluding decay)


distance
in [m]

t = 10 d

t = 50 d

t = 100 d

analytical

numerical

analytical

numerical

analytical

numerical

0.95612

0.95135

0.99661

0.99636

0.99956

0.99953

0.90465

0.89528

0.99254

0.99205

0.99904

0.99897

0.71379

0.69249

0.97558

0.97398

0.99681

0.99658

10

0.36498

0.35029

0.92783

0.92376

0.99012

0.98945

20

0.02806

0.03995

0.74493

0.73530

0.95778

0.95504

cbcilt==NOT

NMK=l~~=~=_~=mW=NJa=^JaJa~
Table 10.4 Comparison of analytical and numerical results (excluding decay) (continued)
distance
in [m]

t = 10 d

t = 50 d

t = 100 d

analytical

numerical

analytical

numerical

analytical

numerical

30

0.00036

0.00184

0.47763

0.47109

0.88437

0.87892

40

0.23014

0.23350

0.75761

0.74965

50

0.08022

0.08716

0.58544

0.57883

60

0.01810

0.02479

0.39915

0.39614

80

0.00094

0.08986

0.12411

100

0.01421

0.0304

NMKT o
1. Kinzelbach, W. (1992) Numerische Methoden zur Modellierung
des Transports von Schadstoffen im Grundwasser. R. Oldenbourg
Verlag, 2. Auflage, ISBN 3-486-26347-1.
2. Ogata, A. and Banks, R.B. (1961) A solution of the differential
equation of longitudinal dispersion in porous media. Professional
paper No. 411-A, U. S. Geological Survey.
3. Sauty, J.P. (1980) An analysis of hydrodispersive transfer in aquifers. Water Resources Research 16(1): 145-158.

Figure 10.4 Calculated breakthrough curves at different


times (excluding decay).

NOU==o=j~~=J=m~=ff

_=~=p=mW=i~=^

NN

_=~=p=mW=i~=^

NNKN p~= p= =
^= q= E`
^F

NN

ered aquifer situation. The layered aquifer consists of a


more permeable upper part (e.g. gravels, quartz or
sands) and a less permeable lower part (e.g. glimmer
sands). Therefore, differential velocities may develop
within such layered structures.

Figure 11.1 shows a schematic sketch of the consid-

Figure 11.1 (a) Schematic sketch of the considered aquifer situation and (b) its idealization for modeling purposes (from Thiele and Diersch2).

cbcilt==NOV

NNK=_=~=p=mW=i~=^
Due to different initial situations of the aquifers
contamination, two basic cases are considered:
(A) by Bruch and Street1
Transient pollution of an initially uncontaminated
semi-infinite layered aquifer:
Freshwater and saltwater intrude through the left
boundary into the model domain. This situation
may be caused e.g. by a well, which is cased in the
upper aquifer section.
(B) by Thiele and Diersch2
Transient pollution of an initially precontaminated
finite layered aquifer:
Due to the different permeabilities within the layers the stagnant lower part of the aquifer may be
even more contaminated then the upper part.
Bruch and Street as well as Thiele and Diersch demonstrated that transversal dispersive mixing may cause
significant saltwater upconing despite the horizontal
flow within the layers.

NNKP d=b~
The governing equations of the considered problem
are:
R

c
c
c
c
+ q x D xx -------2- D yy -------2- = 0
t
x
x
y

(11-1)

(A) Bruch and Street considered the following initial


and boundary conditions:
homogeneous initial condition:
c ( 0, x, y ) = c 2

(11-2)

boundary conditions:
c1 , 0 y < E
c ( t, 0, y ) =
c2 , E y < B
lim c ( t, x, y ) = c 2

c ( t, x , 0 ) = 0

c ( t, x, B ) = 0

(11-3)

x inf

NNKO ^
The following assumptions must be introduced to
obtain an analytical solution:
aquifer:confined, layered, isotropic, uniform thickness
flow:constant discharge rate within each layer, onedimensional flow
transport: absorption and decay are neglected

(B) Thiele and Diersch extended the initial and boundary conditions:
inhomogeneous initial condition:

NPM==o=j~~=J=m~=ff

(11-4)

NNKP=d=b~
c1 , 0 y < E
c ( 0, x, y ) =
c2 , E y < B

c ( t, L, y ) = 0
x

c ( t, x, 0 ) = 0

c ( t, x , B ) = 0

(11-5)

boundary conditions:
c1 , 0 y < E
c ( t, 0, y ) =
c2 , E y < B

(11-7)

(11-6)

with the following model parameters:


Table 11.1 Model parameters

Symbol

Name

Unit

aquifer thickness

contaminant concentration

ML-3

molecular diffusion coefficient

L2 T-1

Dd

D xx = D d + L q x = D Scheidegger-Bear hydrodynamic dispersion


tensor (lateral component)

L2 T-1

D yy = D d + T q x = D Scheidegger-Bear hydrodynamic dispersion


tensor (transversal component)

L2 T-1

q x = v x = q = v
(1 )
R = 1 + ----------------

vx = v
x
L , T
1 = v1 v2

Darcy flux, specific discharge


retardation factor

pore velocity

L T-1
1

LT-1

coordinate along the porous strip

coefficient of longitudinal and transverse dispersivity

velocity relation

kinematic porosity

cbcilt==NPN

NNK=_=~=p=mW=i~=^
initial conditions Eq. (11-2) and boundary conditions (Eq.
(11-3) - Eq. (11-4)) derived by Bruch and Street1 is the following:

NNKQ ^~~=p

The analytical solution of Eq. (11-1) with homogeneous


c c2
E-
x qt/
x
x + qt/
--------------- = -----erfc ------------------------ + exp ------ erfc ------------------------ +
c1 c2
2B
L
2 q t/
2 q t/
L

iE

iy

x I i q t/

- sin --------- cos -------- exp --------- ( 1 I i ) erfc ----------------------- --- B


2 L

i B
2 q t/

i=1

(11-8)

iE

iy

x + I i qt/

- sin --------- cos -------- exp --------- ( 1 + I i ) erfc ----------------------- --- B


2 L

i B
2 q t/
L

i=1

with
2 2

Ii =

4i
1 + ------------L T
2
B

(11-9)

Thiele and Diersch2 extended the analytical solution


ofEq. (11-1) for inhomogeneous initial conditions Eq.
(11-5) and boundary conditions (Eq. (11-6) - Eq. (11-7)):

c c2
--------------- = E
--- +
c1 c2
B

x I i q t/
x + I i qt/
y
x
x
1- E
+ ---sin i --- cos i --- exp --------- ( 1 I i ) erfc ------------------------ + exp --------- ( 1 + I i ) erfc ------------------------
B
B
2 L
2 L
i
2 q t/
2 q t/
i=1


x + I i qt/
x I i q t/
2 2 q T t/ 1
x
y
2- E
- exp -----
- 1 --- erfc ----------------------- erfc ----------------------+ ---sin i --- cos i --- exp i --------------2

L
B
B
i
2 q L t/
2 q L t/

B 2
i = 1

A more general analytical solution permitting different


velocities within the layers is given by Thiele and
Diersch2.
NPO==o=j~~=J=m~=ff

(11-10)

NNKR=k~=^~=EcbjF

NNKR k~=^~=EcbjF
Figure 11.2 shows the finite element mesh used for
the numerical analysis.

Table 11.2 Simulation parameters (continued)


Parameters

FEFLOW

flow materials:
transmissivity of the
lower layer

T1 = q1/q2 x T2

transmissivity of the
upper layer

T2 = 2.025463 [x10-4] m2s-1 x

unsteady transport
transport initials:
homogeneous concentration

c(0,x,y) = 0

transport boundaries:
Figure 11.2 Finite element mesh used.

constant concentration
on the left side at the
lower layer

c(t,0,0<y<E) = c1

Table 11.2 Simulation parameters

constant concentration
on the left side at the
upper layer

c(t,0,E<y<B) = c2

no flux condition at the


right side

Eq. (11-7)

Parameters

FEFLOW

steady flow
flow initials

h(0,x,y) = 0 m

flow boundaries:

transport materials:
aquifer thickness

constant head on the


left side

h(t,0,y) = 100 m

constant head on the


right side

h(t,3500,y) = 0 m

porosity
sorption coefficient
molecular diffusivity

M = 1m

= 0, therefore, R = 1.0
Dd = 0 [10-9 m2s-1]

cbcilt==NPP

NNK=_=~=p=mW=i~=^
Table 11.2 Simulation parameters (continued)
Parameters
longitudinal, transverse dispersivity
decay rate

FEFLOW
L = 10 m, T = 1 m
= 0 [10-4s-1]

FEM:
mesh
Figure 11.2

Figure 11.2
B = 30 m, E = 12 m, L = 3500 m

xmin = 5m, ymin = .5m


time stepping regime

constant: t = 1-10 d

solver

direct
PCG (ORTHOMIN)

NNKS o
Figure 11.3 describes the transient saltwater development for the case (B) after Thiele and Diersch2 at an
observation point (x,y) = (200m,20m) in the upper
layer. Furthermore, Figure 11.3 answers the question,
how the mixing process is influenced by varying velocity ratios v 1 /v 2 = 1 . As can be seen, there is a significant dependence of the saltwater breakthrough
behavior on this velocity relation. Figure 11.3 and
Figure 11.4 compare analytical (Thiele and Diersch2)
and numerical (FEFLOW) results for verification purposes. They reveal a good visual agreement between
the results of both methods.

NPQ==o=j~~=J=m~=ff

NNKS=o

Figure 11.3 Predicted saltwater breakthrough and their dependence on the


velocity ratio at the observation point (200m, 20m) - analytical solution by
Thiele and Diersch2.

cbcilt==NPR

NNK=_=~=p=mW=i~=^

Figure 11.4 Predicted saltwater breakthrough and their dependence on the


velocity ratio at the observation point (200m, 20m) - numerical FE solution.

Figure 11.5 Effect of homogeneous (Bruch and Street1) and inhomogeneous initial conditions (Thiele and Diersch2) within the aquifer.

NPS==o=j~~=J=m~=ff

NNKT=o
Figure 11.5 illustrates clearly the differences
between the models (A) by Bruch and Street and (B)
Thiele and Diersch. In the first case (A) of a homogeneous initial concentration distribution a common
monotonic breakthrough behavior of the contaminants
is attained. On the other hand for a stratified initial condition of the saltwater concentration typical overshooting characteristics develop, which depend on the
velocity ratio. Sooner or later the breakthrough curves
of both models have to meet at the steady-state solutions. The temporal impact of the initial contaminant

distribution vanishes with proceeding time.


Figure 11.6 shows the temporal contaminant breakthrough at different observation points (model B):
1-3 - within the initially contaminated lower zone
4 - even at the layer interface
5-6 - within the initially uncontaminated upper zone
In this case the seepage velocity within the upper layer
is twice as this within the lower one

Figure 11.6 Temporal contaminant breakthrough at different observation points.

NNKT o
1. Bruch, J.C. and Street, R.L. (1967) Two-dimensional dispersion.
J. Sanit. Eng. Div., ASCE, 93(SA6): 17-39.

2. Thiele, M. and Diersch, H.-J. (1986) Overshooting effects due


to hydrodispersive mixing of saltwater layers in aquifers. Adv.
Water Resources 9: 24-33.

cbcilt==NPT

NNK=_=~=p=mW=i~=^

NPU==o=j~~=J=m~=ff

j~=a==c~=m=j~

NO

NO

j~=a==c~=m=j~

NOKN p~===m
The objective of this chapter is to describe the processes important in the transport of solutes through
fractured media. The role of diffusion in solute transport through geologic media has been recognized in
many areas of research. Grisack and Pickens2 reported
that quantitative analysis of solute transport through
fractures with diffusion into a porous matrix can be
considered in several hydrogeologic context. These
include (i) groundwater recharge and associated solute
flux to the water table, (ii) evolution of groundwater
chemistry, secondary fracture mineralogy, and matrix
diagenesis in fractured media, (iii) contaminant transport, (iv) tracer test, (v) groundwater aging, and (vi) ore
deposition in hydrodynamic systems.
Figure 12.1 shows a schematical sketch of the considered aquifer situation for investigating the contaminant transport along a discrete fracture embedded in a
saturated porous rock.

Figure 12.1 Schematical sketch of the fracture-matrix


system (from Tang et al.3).

NOKO ^
The following assumptions are introduced relating
cbcilt==NPV

NOK=j~=a==c~=m=j~
the geometrical and hydraulic properties of the fracture-matrix-system:
fracture: rigid, uniform thickness, width << length,
complete mixing across the fracture width by transverse diffusion and dispersion
matrix:permeability is negligible
fracture flow: constant discharge rate, one-dimensional flow
The following transport mechanisms are taken into
account:
fracture: advection, longitudinal mechanical dispersion, molecular diffusion, radioactive decay
fracture-matrix: molecular diffusion from the fracture
into the matrix, adsorption onto the fracture surface
matrix: adsorption, molecular diffusion, radioactive
decay.

2
D' yy
c
c
c
+ q x D xx -------2- + Rc ---------- c'
t
x
b y
x

c ( 0,x ) = 0

c ( t,0 ) = c 0

lim c ( t,x ) = 0 (12-2)

The governing equation for the matrix domain


( b y ) is:
2

R c' D' yy
c' + R''c' = 0
2
t
y

c' ( t,x,b ) = c ( t,x )

lim c' ( t,x,y ) = 0

with the following parameters:

Table 12.1 Model parameters


Name
half of fracture width

contaminant concentration

ML-3

molecular diffusion coefficient

L2 T-1

D xx = D d + L q x = D Scheidegger-Bear hydrodynamic dispersion


tensor for the fracture

NQM==o=j~~=J=m~=ff

Unit

Dd

(12-3)

with the following initial and boundary conditions:

NOKP d=b~

Symbol

(12-1)

with the following initial and boundary conditions:

c' ( 0,x,y ) = 0

The governing equation balancing the mass of contaminant in the fracture domain ( 0 y b ) is:

= 0
x=b

L2 T-1

(12-4)

NOKQ=^~~=p
Table 12.1 Model parameters (continued)
Symbol
D' yy = 'D' d = D'

Name
Scheidegger-Bear hydrodynamic dispersion
tensor for the matrix

Unit
L2 T-1

Kf

distribution coefficient for the fracture

Km

distribution coefficient for the matrix

q x = q = v x

Darcy flux, specific discharge

L T-1

R = ( 1 + K f /b )

specific retardation factor for the fracture

R' = ' + 'K m

specific retardation factor for the matrix

t 1/2

half-life

vx

pore velocity

coordinate along the fracture

coordinate perpendicular to the fracture

coefficient of longitudinal dispersivity

kinematic porosity

= ln2/ t 1/2

LT-1

decay rate (cf. Section 1.4.2)

NOKQ ^~~=p
NOKQKN d~=~=
The general solution of the coupled system of equations (Eq. (12-1) - Eq. (12-3)) was derived by Tang et
al.3 using the Laplace-transform method. This solution
takes the form of an integral which must be evaluated
by numerical quadrature (Tang et al. used Gaussian
quadrature).

kI= = J
=
~~
~= R = ~= R'
= = ~=
= = = ~~
~= = ~= =
bK=(3-6)==(3-8).

T-1

NOKQKO q~= =
===~
Neglecting dispersion within the fracture D = 0 , a
simpler solution can be obtained.

cbcilt==NQN

NOK=j~=a==c~=m=j~
contaminant distribution within the fracture:
c1
Rx
---= --- exp ----------
q
c0
2

R'D'
R'D'
x erfc ------------------------------------- x
{exp--------------------
2bvR t xR/q
bq

t xR/q +

R'D'
R'D'
exp --------------------- x erfc -------------------------------------- x + t xR/q
bq

2bqR t xR/q

(12-5)

contaminant distribution within the matrix:


1
Rx
c---= --- exp ----------
2
q
c0
R'D'
x+
{exp--------------------bq

R'D' yb R'- ( y b ) erfc -----------------------------------R'- ----------------------------x + --- t xR/q +

2bvR t xR/q

D'
D' 2 t xR/q

(12-6)

R'D'
R'D' yb R'- ( y b ) erfc -----------------------------------R'- -------------------------
exp --------------------- x + ---x
---
2bvR t xR/q + D' 2 t xR/q + t xR/q
bq
D'

NOKQKP p~J~=
Another simplified solution is available for steady
state
conditions,
if
it
is
assumed
that
c/ t 1, c'/ t 1

steady-state contaminant distribution within the fracture:


2
q
cq
+ D'/b
---= exp ------- + ---------2- + ------------------------------- x
c0
2D

2D
4D

(12-7)

2
q
q
+ D'/b
c'- ( y b )
---= exp ------- + ---------2- + -------------------------------- x exp --- D'

2D
2D
c0


4D

(12-8)

A steady-state solutions most useful for the prediction


of ultimate penetrations distances.

NOKR k~=^~=EcbjF
NOKRKN r=~~==
NOKRKNKN a~

steady-state contaminant distribution within the matrix:


NQO==o=j~~=J=m~=ff

Figure 12.2 shows the finite element mesh used for

NOKR=k~=^~=EcbjF
the numerical analysis. It is formed by two-dimensional quadrilateral finite elements for the discretization of both the porous matrix and the fracture.

Table 12.2 Simulation parameters (continued)


Parameters

FEFLOW

flow materials:
fracture transmissivity

T = 1 [x10-4] m2s-1

rock transmissivity

T = 0 [x10-4] m2s-1

unsteady transport
transport initials:
homogeneous concentration

c(0,x,y) = 0

transport boundaries:
constant concentration
at the left fracture
entrance

c(t,0,y<b) = c0

no flux condition at the


right fracture exit
Figure 12.2 Finite element mesh (vertical exaggeration 30 : 1) of the half-space fracture-matrix domain.

Table 12.2 Simulation parameters


Parameters

FEFLOW

h(0,x,y) = 0 m

flow boundaries:
flux at the left fracture
entrance
constant head on the
right fracture exit

transport materials:
aquifer thickness
porosity

steady flow
flow initials

no flux condition at the


matrix boundary

sorption coefficient
molecular diffusivity

q(t,0,y<b) = 0.75 md-1


h(t,x=3m,y<b) = 0 m

longitudinal, transverse dispersivity

M=1m
= 1, ' = 0.35
= ' = 0,
therefore, R = , R = '
D d = 0 [10-9 m2s-1]
D' d = 0.1-0.00001
[10-9 m2s-1]
L = 0.76 m, T = 0 m

cbcilt==NQP

NOK=j~=a==c~=m=j~
Table 12.2 Simulation parameters (continued)
Parameters

FEFLOW
-4 -1

= 0 [10 s ]

decay rate
FEM:
mesh (cf. Figure 12.2)

np = 2562, ne = 2460

xmin = 6 10-2 m ,
ymin = 6 10-5 m
time stepping regime

constant: t = 0.04 d

solver

direct
PCG (ORTHOMIN)

NOKRKNKO o
The following figures illustrate the effect of matrix
diffusion on the concentration distribution within the
fracture resulting from the permanent contamination at
the left fracture entrance. The range of diffusion coefficients used is representative of most geological media.
Figure 12.3 plots analytical profiles by Tang et al.3 and
numerical profiles by Grisack and Pickens2 at a calculation time of 4 days. Depending on the diffusion values the possible concentration profile spreads over the
whole range. Figure 12.4 shows the corresponding
FEFLOW results.

Figure 12.3 Effect of matrix diffusion on concentration profiles in the fracture (from Tang et al.3).

NQQ==o=j~~=J=m~=ff

NOKR=k~=^~=EcbjF

Figure 12.4 Effect of matrix diffusion on concentration profiles


in the fracture - FEFLOW results compared with analytical by
Tang et al.3.

Figure 12.5 Effect of matrix diffusion on temporal solute breakthrough at a distance 0.76 m from the source point (from Tang et
al.3).

cbcilt==NQR

NOK=j~=a==c~=m=j~

Figure 12.6 Effect of matrix diffusion on temporal solute breakthrough;


FEFLOW results compared with analytical by Tang et al.3.

Figure 12.5 presents analytical and numerical concentration history curves at a distance 0.76 m from the
contaminant source for different values of matrix diffusivity obtained by Tang et al.3 and Grisack and
Pickens2, respectively. Especially in the mid range of
the diffusivity values considerable discrepancy appears
in breakthrough curves, which result from insufficient
discretization near the fracture interface.
Figure 12.6 shows temporal breakthrough curves
for the same case study as above computed by
FEFLOW at a distance 0.76 m from the source point.
First, the effect of the parabolic velocity profile is demNQS==o=j~~=J=m~=ff

onstrated, y = 0 corresponds to the fracture centre, y = b


corresponds to the fracture surface. For large matrix
diffusivities the concentration values at the fracture
surface and at the fracture centre (area of maximum
velocity) are identically. Decreasing the influence of
matrix diffusion differences between concentrations at
the fracture centre and at the fracture surface become
obvious. The concentration values differ up to 5 percent. The FEFLOW results for the fracture centre,
using the FE mesh illustrated in Figure 12.2, agree
good with the analytical findings by Tang et al.3, which
based on the concept of average velocity within the
fracture.

NOKR=k~=^~=EcbjF
NOKRKO r=~=
NOKRKOKN a~
Instead of discretizing the fracture by standard areal
finite elements as used in the above section 12.5.1,
there is the possibility to discretize the fracture by onedimensional discrete feature elements, so-called fracture elements. Their theoretical basis is given in the
White Papers1.
a)

matrix system is shown in Figure 12.7, where


Figure 12.7a depicts the standard meshing without
fracture elements (used for comparison purposes) and
Figure 12.7b depicts the alternative mesh including 50
one-dimensional fracture elements. For the fractures a
Hagen-Poiseuille law1 of the flow motion is assumed.
Table 12.3 summarizes the parameters used for the
fracture elements.
Table 12.3 Fracture parameters
Parameters

FEFLOW

flow materials:
flow law

Hagen-Poiseuille
5

cross section area

6 10

hydraulic aperture
(=2b)

1.2 10

m2

transport materials:

b)

molecular diffusivity
longitudinal dispersivity
decay rate
1D fracture elements

Figure 12.7 Used finite element meshes (vertical exaggeration 300 : 1) of the half-space fracture-matrix domain: a)
standard quadrilateral elements (no fracture elements), b)
one-dimensional fracture elements combined with twodimensional porous matrix elements.

D d = 0 [10-9 m2s-1]
L = 0.76 m
= 0 [10-4s-1]

FEM (cf. Figure 12.7b):


2D porous matrix elements

np = 1326, ne =1250

1D fracture elements

50

time stepping regime

constant: t = 0.04 d

solver

PCG (BiCGSTABP)

The used discretizations of the half-space fracturecbcilt==NQT

NOK=j~=a==c~=m=j~
NOKRKOKO o
Figure 12.8 compares the simulated solute breakthrough curves for the cases with and without using
fracture elements. The agreement is well. While the
velocity at the facture-matrix contact is averaged for
the case without fracture elements, such a transition in
the velocity profile from the fracture to the porous
matrix does not occur if using fracture elements. This
results in differences in the breakthrough behaviors as

revealed in Figure 12.8. As expected, the differences


increase with decreasing matrix diffusion. The accuracy of the results obtained with fracture elements is
improved if compared to the analytical solutions as
shown in Figure 12.7.
It can be concluded that the application of fracture
elements increases accuracy and efficiency of the solution.

100

without fracture elements


with fracture elements

90

D d = 10

80

concentration c/c0 [per cent]

14

m s

13

m s

D d = 10

70

D d = 10

60

12

m s

50
40

D d = 10

11

m s

30
20

D d = 10

10
0

10

m s

time [days]

Figure 12.8 Simulated solute breakthrough in the fracture at a distance of 0.76 m from the
source point; comparison of solutions with and without using fracture elements.

NOKS o
1. Diersch, H.-J.G. (2002) Discrete feature modeling of flow, mass
and heat transport processes by using FEFLOW. White Papers Vol. I, Chapter 9, WASY Ltd., 147-190.

NQU==o=j~~=J=m~=ff

2. Grisak, G.E. and Pickens, J.F. (1980) Solute transport through


fractured media: 1. The effect of matrix diffusion. Water
Resources Research 16(4): 719-730.
3. Tang, D.H., Frind E.O. and Sudicky, E.A. (1981) Contaminant
transport in fractured porous media: Analytical solution for a single fracture. Water Resources Research 17(3): 555-564.

e= ~= e~~= bW= q= q
t=m

NP

NP

e=~=e~~=bW=q=q=t=m

NPKN p~===m
Hoopes and Harlemann1 conducted a number of
experiments using a sand model to measure the distribution of a solute between recharging and discharging
wells. A plane view of their laboratory sand model representing a horizontal, confined aquifer is illustrated in
Figure 13.1. Furthermore, Hoopes and Harlemann1
developed an analytical framework to reproduce the
experimental results.
Tracer data from wellbore doublets can be used for
testing the transport characteristics of real aquifers.
Therefore, the two well problem and its analytical
modeling by Hoopes and Harlemann have proved to be
an important benchmark for checking the validity of
numerical models (Segol3).

Figure 13.1 Plan view of Hoopes and Harlemanns sand


model (from Hoopes and Harlemann1).

Figure 13.2 indicates the principal problem of data


interpretation using simplified models such as a homogeneous porous medium (cf. Table 13.1). Basically, the
validity of this conceptual model depends on the existence of a unique parameter set which must fit the overall solute distribution.
cbcilt==NQV

NPK=e=~=e~~=bW=q=q=t=m

Figure 13.2 Measured and analytically calculated concentration breakthrough at


two observation points on the line of symmetry between the wells (from Hoopes
and Harlemann1).

NPKO ^

NPKP d=b~

The model assumptions are summarized in Table


13.1.

The governing equation balancing the mass of contaminant can be written as

Table 13.1 Model assumptions


aquifer

homogeneous, isotropic, constant thickness

wells

doublet, penetrating entire aquifer thickness

flow

isothermal, steady state, 2-D horizontal, constant discharge/ recharge

transport

advective, dispersive (along streamlines), diffusive

NRM==o=j~~=J=m~=ff

2
c
2 c
c
+ v D --------2- = 0
t

(13-1)

with the quantities:


q
Q
2M
2M
v = --- = ------------------ cosh ------------ + cos ------------

2Md
Q
Q

(13-2)

NPKQ=^~~=p
2

Q
2yd
-
= ------------ arctan --------------------------2
2M
2
2
x +y d
Q
(x + d) + y = ------------ ln -----------------------------4M ( x + d ) 2 y 2

c ( 0,, ) = 0

c ( t,(-d,y),(-d,y) ) = c 0

c ( t,0, ) = 0

(13-3)

(13-4)

and following parameters and quantities:


and the initial and boundary conditions:
Table 13.2 Model parameters
Symbol

Name

Unit

contaminant concentration

ML-3

half of well separation

Dd
D = L + Dd
M

molecular diffusion coefficient

L2 T-1

Scheidegger-Bear hydrodynamic dispersion


coefficient along the streamline

L2 T-1

aquifer thickness

Darcy flux, specific discharge

L T-1

volumetric flow rate

L3 T-1

pore velocity along streamline, Eq. (13-2)

velocity potential, Eq. (13-3)

L2 T-1

stream function, Eq. (13-3)

L2 T-1

coefficient of longitudinal dispersivity

kinematic porosity

qx = q

NPKQ ^~~=p
Hoopes and Harlemann1 provide an analytical solution to Eq. (13-1) - Eq. (13-4). According to Eq. (13-1))

LT-1

this solution neglects dispersion and diffusion across


the streamlines. Eq. (13-1) corresponds to a formulation of the advection-dispersion-equation in velocity
potential-streamline notation. This coordinate transformation supplies in fact a 1-D formulation with the
cbcilt==NRN

NPK=e=~=e~~=bW=q=q=t=m
velocity potential playing the part of the local variable.
However, due to Eq. (13-2) the velocity depends on
both, the velocity potential and stream function. The
analytical solution of Eq. (13-1) through Eq. (13-4)
obtained by Hoopes and Harlemann is
cD

I D t D
c
1
= ----- = --- erfc ----------------
c0
2
4J

(13-5)

Qt
t D = --------------------2
2Md

d D

I D = ---------2

v
D

DD

J D = ------d
D
4

vD

numerical methods (Huyakorn et al.2). The quality of


numerical solutions depends essentially on an appropriate discretization of the calculation domain because
of the steep concentration increase (cf. Figure 13.2).
We use different meshes for the FE-analysis
(Figure 13.3, Figure 13.4) and observe strong mesh
effects in the following. The wellbores are represented
by small circles, which are cut off around the borehole
centers (Figure 13.5)

(13-6)

with the dimensionless quantities:


2
xd
2
2
x+d
- ------------------------------
v D = d -----------------------------2
2
2
2

(x + d) + y (x d) + y
2
y
y
- -----------------------------
+ -----------------------------2
2
2
2
(x + d) + y (x d) + y
2

1 (x + d) + y D = --- ln -----------------------------2 ( x + d )2 y2

Figure 13.3 Finite element grid (mesh A).


(13-7)

2yd D = arctan --------------------------(13-8)


2
2
2
x +y d

NPKR k~=^~=EcbjF
Hoopes and Harlemanns experiments was analyzed
by several authors in the literature using analytical and
NRO==o=j~~=J=m~=ff

Figure 13.4 Finite element grid (mesh B).

NPKR=k~=^~=EcbjF
Table 13.3 Simulation parameters (continued)
Parameters

FEFLOW

aquifer transmissivity

T = 1 [x10-4] m2s-1

unsteady transport
transport initials:
homogeneous concentration

c ( t=0, x, y ) = 0

transport boundaries:
constant concentration
at the injection well

c ( t, r B ) = c 0

transport materials:
Figure 13.5 Well discretization (detail of supermesh B).

The simulation parameters are summarized in the following Table 13.3.


Table 13.3 Simulation parameters
Parameters

FEFLOW

steady flow

aquifer thickness
porosity

M = 0.089 m
= 0.374

sorption coefficient

= 0, therefore, R = 1

molecular diffusion

D d = 0 [10-9 m2s-1]

longitudinal, transverse dispersivity

L = 0.0015 m/ 0.00294 m
T = 0 m

decay rate

= 0 [10-4s-1]

FEM:

flow boundaries:
flux at the injection
well

q = Q ( 2r B ) = 6.4327m2d-1

wellbore radius, outer


boundary

r B = 0.005 m, R o = 1.45 m

constant head at the


production well

h =0m

mesh A (Figure 13.3)


triangles

np = 1782, ne = 952

flow materials:

lmin = 1.55 10-3 m

cbcilt==NRP

NPK=e=~=e~~=bW=q=q=t=m
Table 13.3 Simulation parameters (continued)
Parameters

FEFLOW

mesh B (Figure 13.4)

np = 6391, ne = 6200

-3

lmin = 1.55 10 m
time stepping regime
initial time step
solver

automatic time step control


constant: t 0 = 10-5 d
direct
PCG (ORTHOMIN)

NPKS o
Huyakorn et al.2 use a finite-element formulation
based on curvilinear coordinates along and normal to
the streamlines. This approach was intended specially
for the analysis of injection/withdrawal tracer test,
which is a common method for describing the transport
characteristics of aquifers.

Figure 13.6 Concentration breakthrough at two observation points between the


wells (results from Huyakorn et al.2).

Because of the extremely small dispersivities within


the sand-box model, a sharp increase of tracer concentration takes place at the observation points. From this
NRQ==o=j~~=J=m~=ff

advection-dominated transport well-known problems


arise concerning oscillating numerical solutions. Huyakorn et al.2 applied the streamline-upwinding method

NPKS=o
for damping such oscillations. Unfortunately, from
Figure 13.6 it is not clear whether or not oscillations
occur before and after tracer breakthrough.
With help of Figure 13.7 we discuss some numerical aspects of modeling with the sparse mesh A (cf.
Figure 13.3). The Galerkin-FEM produces oscillating
solutions. The oscillations grow with decreasing dispersion (increasing Peclet-numbers, which characterize
the relation of advective and dispersive transport frac-

tions). Corresponding to the analytical model by


Hoopes and Harlemann, two different dispersion
parameters - L = 0.0015 m, L = 0.00294 m - are used.
Applying the streamline upwind method the oscillations disappear. Due to the introduction of additional
(non-physical) dispersion the concentration fronts are
smoothed. Figure 13.7 provides an idea, how the
upwind FEM could change the results.

Figure 13.7 Numerical aspects of modeling advection-dominated processes.

The discussion of numerical aspects is continued in


Figure 13.8. A refined mesh B (Figure 13.4) was introduced, which is constructed on the basis of physical

considerations. The velocity potential and stream function produced by a well doublet are given by Eq. (13-3).
The general idea is to refine the mesh in domains of
cbcilt==NRR

NPK=e=~=e~~=bW=q=q=t=m
higher velocities (dominating advection) in the vicinity
of the well and otherwise to spare grid points in areas
of smaller velocities.
Figure 13.8 documents the success of this operation.
At first, non-oscillating solutions can be obtained even
by the Galerkin-FEM. Secondly, using mesh B we can
improve the agreement between numerical and analytical results. However, it must recognized that the
numerical approximation of extremely steep fronts
requires very expensive local discretizations.

The remaining differences between analytical and


numerical results may be caused by different geometrical features. The analytical approach treats the wells as
point-shaped sources/sinks, whereas the numerical
model use finite wells corresponding to the laboratory
situation (see Figure 13.5). Furthermore, the analytical
method assumes an infinitely extended sand-box
region, whereas the numerical model reflects the finite
geometry of the sand-box laboratory.

Figure 13.8 Mesh effects and best fit by use of mesh B.

NRS==o=j~~=J=m~=ff

NPKT=o

NPKT o
1. Hoopes, J.A. and Harlemann, D.R.F. (1967) Wastewater recharge
and dispersion in porous media. ASCE Journal of the Hydraulics
Division 93 (HY5): 51-71.
2. Huyakorn, P.S., Andersen, P.F., Gven, O. and Molz, F.J. (1986)
A curvilinear finite element model for simulating two-well tracer
tests and transport in stratified aquifers. Water Resources
Research 22(5): 663-678,
3. Segol, G. (1994) Classic groundwater simulation: proving and
improving numerical models. PRT Prentice Hall, Englewood
Cliffs, New Jersey.

cbcilt==NRT

NPK=e=~=e~~=bW=q=q=t=m

NRU==o=j~~=J=m~=ff

NQ

b=mW=k~~=`

NQ

b=mW=k~~=`

NQKN p~===m
Elder3,4 presents experimental and numerical studies concerning the flow produced by heating the base
of a porous layer. The experiments are performed in
Hele-Shaw cells. The coupled system of equations
describing free convection flows and corresponding
heat transfer is non-linear because of the advection of
heat and is time-dependent through the thermal capacity of the medium. Elder conducts these studies mainly
to verify the finite-difference model which he used for
the numerical analysis of thermally driven convection.
Furthermore, he obtained criteria for preventing
numerical instabilities. Elders problem can be used as a
thermal analog for the saltwater intrusion into uncontaminated aquifers by density driven convection
(Figure 14.1).
The numerical results of Elder for this problem of
complex natural convection are frequently recommended for the verification of simulators (e.g.
Diersch7, Voss and Souza14, Kolditz et al.13, Diersch
and Kolditz9, Ackerer et al.1, Frolkovic and De
Schepper6). Diersch et al.8, Kolditz and Diersch11

investigate numerical details for the simulation of


mixed convection in geothermal fractures including
thermoelastic effects.

Figure 14.1 Definition of Elders problem (from Voss and


Souza14).

NQKO _=

^~J

Commonly, the so-called Boussinesq approximation


is used for describing steady flow in hydrogeologic
systems. Basically, the Boussinesq approxi-mation subcbcilt==NRV

NQK=b=mW=k~~=`
stitutes the conservation of mass by the conservation of
fluid volume only. It takes the form (Bear1):
q = 0

(14-1)

The Boussinesq approximation is applicable in many


groundwater problems in which the considered fluid is
incompressible and its density is only weakly dependent on temperature and solute concentration.
This approach is inappropriate when significant
density fluctuations arise from temperature and/or concentration variations. In this case the following more
general form of the continuity equation must be used.
( q ) = 0

(14-2)

The mathematical difference between the two formulations Eq. (14-1) and Eq. (14-2) can be discovered by
rearranging the previous equation.
0 c c
q
q = ----------- = q ------------------

(14-3)

Therefore, the term on the right hand side in Eq. (14-3)


vanishes only if the specific discharge is orthogonal to
the density gradient or, in other words, if the streamlines follow the density isolines.
Evans and Raffensperger5 study the limitation of the
Boussinesq approximation comparing the results of
both formulations. For a similar case study of saltwater
intrusion they obtained differences up to 9%.

NSM==o=j~~=J=m~=ff

The model assumptions are summarized in Table 14.1.


Table 14.1 Model assumptions
aquifer

homogeneous, isotropic, constant


thickness

flow

steady state, incompressible,


2-D vertical
Boussinesq approximation is not
recommended

transport

advective, diffusive

NQKP b=
_
^~
For the simulation of density-variable groundwater
flow FEFLOW can optionally provide an extended
Boussinesq approximation which accounts for the local
density variations within inhomogeneous fluids. Starting from the more general form of the continuity equation,

c
+ ( q ) = Q h
c t

(14-4)

different weak formulations can be constructed dependent on how the fluid density variations were incorporated. If substituting the Darcy flux into the continuity
equation of fluid mass an unsymmetrical formulation
for the hydraulic head results. We like to avoid this and
use alternatively the following weak formulation of the
continuity equation

NQKQ=d=b~=J=c~==q==e~=e~
0 w h

d
Kij ---- xi xj

f
0
c
w -----f q i c
x

d
i

w
0
K ------ ( c c 0 )e j d
x i ij c

f
0 c
w -----f c d
t

transport
(14-5)

+ wQ h d

= 0 ( 1 + c ( c c0 ) )

= 0 ( 1 T ( T T0 ) )

1
c = ----0 c

wq n d

1
T = ----0 c
2

(14-9)

= 0 ( 1 + 1.85 4.1 + 44.5 )

The second and the third terms on the right side of


Eq. (14-5) appear due to the extended formulation of
the Boussinesq approximation. The third term will be
activated only for predictor-corrector time stepping
schemes.

NQKQ d= b~= J
c~= = q=
e~=e~
To show the formal analogy between mass and heat
transfer we develop in this chapter both equations of
mass and heat conservation, respectively. The conservation of fluid mass is already represented by Eq. (142). The generalized Darcy law describes the momentum balance for groundwater systems.
0 g
q = K h --------------- -----

0 g

(14-6)

kg 0
K = -----------

(14-7)

(14-8)

(14-10)

0
= -----------------------------------------------------------21 + 0.7063 0.04832
T 150
= -----------------150

= 10 c

(14-11)

The governing equations balancing the mass of contaminant and the heat, respectively, can be written as

f f

c
+ qc D d c = 0
t
s s

( c + ( 1 )c )
f

T
f f
+ c qT
t

(14-12)

(14-13)

( + ( 1 ) )T = 0

The acting initial and boundary conditions for the case


of solute transport are illustrated in Figure 14.1.
We transform the Eq. (14-6), Eq. (14-12), Eq. (1413) to dimensionless forms by introducing the following change of variables,

with the constitutive equations for the density and the


viscosity of the fluid, respectively for solute and heat
cbcilt==NSN

NQK=b=mW=k~~=`
L
q D = ---- q

x
x D = --L

c
c D = ----c0

T
T D = ----T0

h
h D = ---------------L T T 0

t D = -----2 t
L

(14-14)

with the dimensionless quantities


KL T T 0
Ra = -------------------

(14-18)

c c0
N = -----------T T0

(14-19)

D d
Le = --------

(14-20)

which yield the following dimensionless Darcy- and


transport equations
g
q D = Ra h D N ( c D 1 ) -----
g
g

q D = Ra h D + ( T D 1 ) -----

(14-15)

where
f

c D

+ q D c D Lec D = 0
tD

(14-16)

c
+ ( 1 )c - T D
--------------------------------------------+ q D T D T D = 0
f f
tD
c

(14-17)

f f

s s

+ ( 1 )
= ----------------------------------f f
c

(14-21)

The following table summarizes the model parameters


and its values used in the present study.

Table 14.2 Model parameters


Symbol

Value

Unit
M L-3

contaminant concentration

f f

thermal capacity of the fluid phase

4.2 x 106 J m-3 K-1

ML-1T-2K-1

s s

thermal capacity of the solid phase

ML-1T-2K-1

Dd

molecular diffusion coefficient

3.565 x 10-6 m2 s-1

L2 T-1

9.81 m s-2

L T-2

NSO==o=j~~=J=m~=ff

Name

gravity acceleration vector

piezometric head

permeability of the porous medium

hydraulic conductivity of the porous medium

L
4.845 x

10-13

m2

4.752945 x 10-6 m s-1

L2
L T-1

NQKQ=d=b~=J=c~==q==e~=e~
Table 14.2 Model parameters (continued)
Symbol

Name

Value

Unit

150 m

Height of the cell

Le

Lewis number

aquifer thickness

Pe

Peclet number

q
Ra

Rayleigh number

Sh

Stroughal number

basic function

density ratio

volumetric thermal expansion coefficient

kinematic porosity

bulk heat diffusivity, Eq. (14-21)

400

1
1
1

0.2 | c 0 |

m3

kg-1

L3

M-1

0.2 | T 0 | K-1

K-1

0.1

1
L2 T-1

thermal conductivity of the fluid phase

0.65 J m-1 s-1 K-1

M L T-3 K-1

thermal conductivity of the solid phase

1.5914444 J m-1 s-1 K-1

M L T-3 K-1

L T-1

Darcy flux, specific discharge

fluid dynamic viscosity

fluid dynamic viscosity, where c = c 0

calculation domain

fluid density

fluid density, where c = c 0

stream function

vorticity

M L-1 T-1
10-3 kg m-1 s-1

M L-1 T-1
L2
M L-3

103 kg m-3

M L-3
L2 T-1
T-1

cbcilt==NSP

NQK=b=mW=k~~=`

NQKR b=b~=J=cJ
~==q==p~
c
In modeling groundwater motion it is often useful to
formulate the flow equations in terms of stream
functions10. For example, analytical solutions of 2-D
flow problems can be obtained using the stream function. The stream function provides a scalar description
of the velocity field (existence of potential). In this way
the contours of the stream function represent the flow
paths. Another advantage of the formulation in terms of
stream functions is the inherent conservation of mass.
On the other hand the use of stream functions implicitly assumes steady state flow and invokes the Boussinesq approximation. As discussed before, the
Boussinesq approximation can provide erroneous
results for density-variable flow. Evans and
Raffensperger5 introduce mass-based stream functions
to avoid the Boussinesq approximation.
Elder4 uses the Boussinesq approximation to simplify the field equations of free convection in porous
media. The Boussinesq approximation implies the
existence of stream functions. For 2-D flow, the specific discharge defined in terms of stream functions is
expressed as
qx =

qz =

(14-22)

From the definition of vorticity one can obtain immediately (Elder omits the factor 1/2 commonly used for the
vorticity)

= q = e z

(14-23)

Non-vanishing vorticity characterizes rotational flow.


Substituting the generalized Darcy law Eq. (14-6) into
the vorticity equation Eq. (14-23) we achieve
= K T

T
e
x z

(14-24)

Combining the Eq. (14-23) and Eq. (14-24)) we get


finally
= K T

T
x

(14-25)

which is the flow equation Elder used in a dimensionless form


D = Ra

T D
xD

(14-26)

with the dimensionless quantities:

D = ---

T
T D = ----T0

x
x D = --L

(14-27)

Starting from Eq. (14-13) Elder develops the dimensionless equation of heat balance
T D
D D
= T D ( D ,T D ) = T D
,
T D (14-28)
tD
zD xD

The following table summarizes the data of the therNSQ==o=j~~=J=m~=ff

NQKR=b=b~=J=c~==q==p~=c
mal experiment by Elder4. Unfortunately, Elder reports
only the concentrated dimensionless quantities, and the
back-transformation to physical quantities is complicated. No data can be reconstructed for the thermal

properties of the silicon-oil Elder uses in his experiments. Therefore, we add some material data for silicon-oil from the literature.

Table 14.3 Original data from Elders thermal experiment


Symbol

Name

Value

Unit

cell width (channel aperture)

0.004 m

cell length

0.20 m

cell height

0.05 m

gravity acceleration vector

9.81 m s-2

L T-2

1.333333 x 10-6 m2

L2

0.1308 m s-1

L T-1

400

6.372 x 10-6 / | T 0 | K-1

K-1

1.041667 x 10-7 m2 s-1

L2 T-1

10-4 m2 s-1

L2 T

k = b 12

permeability of the cell

hydraulic conductivity of the cell

Ra

Rayleigh number

volumetric thermal expansion coefficient

kinematic porosity

heat diffusivity of the fluid

kinematic viscosity of the fluid


material properties for silicon-oil (syltherm)

thermal capacity

thermal conductivity

corresponding heat diffusivity

at temperature 20C
1.39425 x 106 J m-3 K-1
0.106 W m-1 K-1

M L T-3 K-1

7.602654 x 10-8 m2 s-1

L2 T-1

cbcilt==NSR

NQK=b=mW=k~~=`

NQKS k~= ^~= fW


`Js~=

=j
Elder4 uses a finite-difference representation for the
numerical model. The equations are solved by repeated
application of Leibmanns extrapolated method (overrelaxation) with alternating directions of scan (Elder3).
Forward temporal differences are employed (explicit
scheme). The time step regime is restricted by the Courant criterion. The regularly finite-difference grid consists of 81 nodes laterally and 21 nodes vertically. A
corresponding finite-element mesh containing the same
number of grid points is shown in Figure 14.2.

Voss and Souza14 simulated the Elder problem with


the finite-element code SUTRA, using a mesh coarser
than Elders. Their mesh contains 44 elements horizontally and 25 elements vertically (Figure 14.3).
The simulation parameters we use in the present
study are summarized in the following Table 14.4.
Table 14.4 Simulation parameters
Parameters

FEFLOW

problem class:
steady flow / transient
transport
vertical problem projection
temporal and control
data:
time step regime

Figure 14.2 Finite-element mesh A (number of elements ne = 1600, number of grid points - np = 1701).

Galerkin- /upwindFEM
steady flow
flow boundaries:

Figure 14.3 Finite-element mesh B (number of elements ne = 1100, number of grid points - np = 1170).

NSS==o=j~~=J=m~=ff

head at the upper corners

h =0m

no-flux along the


boundary, except upper
corners

q n = 0 m d-1

NQKS=k~=^~=fW=`Js~===j
Table 14.4 Simulation parameters (continued)
Parameters

FEFLOW

density ratio
thermal expansion
coefficient

Parameters

FEFLOW

aquifer thickness

flow materials:
hydraulic conductivity

Table 14.4 Simulation parameters (continued)

K = 4.752945x10-6 ms-1
c = 2000 x [10-4]
T = 2000 x [10-4 K-1]

M =1m
= 0.1

porosity
sorption coefficient

= 0, therefore, R = 1.0

molecular diffusivity

D d = 3565 x [10-9 m2s-1]


L = 0.0, T = 0 m

longitudinal, transverse dispersivity

transient transport

= 0 [10-4s-1]

decay rate
mass transport initials:
homogeneous concentration

c ( t=0, x, y ) = 0

T ( t=0, x, y ) = 0

c ( t, y=0 ) = 0
c ( t, y=L ) = 1

heat transport boundaries:


const. temperature at
the cell top
const. temperature at a
part of the bottom

=0

porosity

mass transport
boundaries:
const. concentration at
the cell bottom
const. concentration at
a part of the top

M =1m

aquifer thickness

heat transport initials:


homogeneous temperature

heat transport materials:

T ( t, y=0 ) = 0
T ( t, y=L ) = 1

s s

thermal capacity of the


solid phase

c = 0.42 x 106 J m-3 K-1

thermal conductivity of
the solid phase

= 1.4973 J m-1 s-1 K-1

We start the discussion presenting former results by


Elder4 and Voss and Souza14, which provide the basic
solutions we refer on for comparison with our results.
Figure 14.4 shows the results obtained by Elder at
0.005, 0.01, 0.02, 0.05, 0.075 and 0.1 units of dimensionless time, and the results obtained by Voss and
Souza for 1, 2, 4, 10, 15 and 20 years, respectively.

mass transport materials:

cbcilt==NST

NQK=b=mW=k~~=`

Figure 14.4 Comparison of nondimensional results for the Elder problem, showing 20 and 60% of
maximum concentration for SUTRA (solid curves) and of maximum temperature from Elder (dashed
curves) (from Voss and Souza14).

It is important to point out that Elder simulated thermal convection using a finite difference scheme,
whereas Voss and Souza considered the analogous case
for solute movement in a thin cavity using the finiteelement method. Despite of the different numerical
methods applied, we feel both solutions agree qualitatively well. However, it is important to realize at the
beginning of our study, that one cannot expect perfect
NSU==o=j~~=J=m~=ff

agreement between numerical solutions derived by different methods and codes for complex, nonlinear problems.
The following table should provide an overview on
the simulation runs presented in order to show the
impact of distinct model assumptions and numerical

NQKS=k~=^~=fW=`Js~===j
aspects.
Table 14.5 Simulation overview
Topic

Realization

models

saltwater intrusion c 0
thermal convection T 0

T = 0
c = 0

couplings

flow and solute transport, flow and heat transport

assumptions

Boussinesq approximation (BA), extended


Boussinesq approximation (ext. BA)

constitutive laws

density, viscosity

FEM-schemes

standard Galerkin, streamline upwinding

temporal schemes semi-implicit, fully implicit, predictor-corrector


(FE/BE, AB/TR)
iteration methods Picard iteration, Newton iteration
meshes

A, B

The next table summarizes the simulations carried out


in order to compare our results with former findings by
Elder as well as Voss and Souza. Up to a simulation

time of 10 years all results agree quite well with those


of Elder as well as Voss and Souza.

Table 14.6 Comparing results


Fig.

Ass.

FEM

time

14.6 a-f

BAa)

Galerkin fully implicit

14.6 A-F

BA

upwind

14.7 a-f

ext. BA Galerkin fully implicit

14.7 A-F

ext. BA upwind

14.8 a-f

ext. BA Galerkin FE/BE pre-cor

14.8 A-F

ext. BA upwind

t = 4 yr
Elder

fully implicit

Elder
Voss

Elder

fully implicit

FE/BE pre-cor

t = 10 yr t = 15 yr t = 20 yr

Elder

Voss
Elder

Elder

Elder
Elder

Voss

Elder

Voss

Elder

both

Elder

cbcilt==NSV

NQK=b=mW=k~~=`
Table 14.6 Comparing results (continued)
Fig.

Ass.

FEM

time

14.9 a-f

ext. BA Galerkin AB/TR pre-cor

14.9 A-F

ext. BA upwind

14.10 a-f

ext. BA Galerkin Crank-Nicolson

14.10 A-F

ext. BA upwind

a) BA

Elder

AB/TR pre-cor

t = 10 yr t = 15 yr t = 20 yr
Elder
Elder

Elder

Elder

Elder
Elder

Elder
Voss

Crank-Nicolson

- Boussinesq approximation

In our numerical investigation of the Elder problem we


pay intention to the following numerical and physical
aspects:

NQKSKN d~==Jcbj
As expected the upwind-FEM produces in general
smoother contours of concentrations. Furthermore, it
provides better agreement with Elders results at later
simulation times, e.g. from 15 through 20 years.

NQKSKO j=
Comparing the Figure 14.5 a-f with Figure 14.6 a-f
and Figure 14.5 A-F with Figure 14.6 A-F one can see,
that the salt distribution depends on the finite-element
discretization. It seems that changes of mesh density
give rise for changes of the convection pattern.

NQKSKP ^= = = ~J
===~
Comparing Figure 14.6 through Figure 14.8 it
becomes obvious that differences from the Boussinesq
NTM==o=j~~=J=m~=ff

t = 4 yr

approximation (BA) and the extended BA appear at


later simulation times, e.g. after 10 years. As seen
above only one new term is introduced using constant
time stepping due to local density variations by the
extension of the BA. This formulation encourages
stronger circulation (Figure 14.7).
Applying the extended BA in connection with predictor/corrector time stepping methods the circulation
pattern for later times return to those of the BA (cf.
Figure 14.6 F with Figure 14.8 F and Figure 14.9 F).
These results compare well with those obtained with
BA using the mesh A (Figure 14.5).

NQKSKQ _~=
Elder as well as Voss and Souza used different
boundary conditions. Elder prescribed zero temperatures on the top and the sides of the convection cell.
Whereas Voss and Souza defined the condition of no
mass flux through the boundary beside the bottom and
the part of the top (see Figure 14.1). These different
boundary condition have no obvious influence on the
results.

NQKS=k~=^~=fW=`Js~===j
NQKSKR q==
We use semi-implicit (Crank-Nicolson), fully
implicit and predictor-corrector time stepping schemes.
It was found that the results for later simulation times
(after 15 years) depend strongly on the time stepping
scheme. This observation is similar to that obtained for
the different local discretization (e.g. mesh A and mesh
B).

NQKSKS p
Figure 14.11 shows results which are derived by the
iterative (PCG) and the direct solver (Gaussian).

FEM: standard Galerkin


time: fully implicit, const. step

FEM: streamline upwind


time: fully implicit, const. step

Figure 14.5 Model: Boussinesq approximation, mesh A


(Figure 14.2).

cbcilt==NTN

NQK=b=mW=k~~=`

FEM: standard Galerkin


time: fully implicit, const. step

FEM: streamline upwind


time: fully implicit, const. step

Figure 14.6 Model: Boussinesq approximation, mesh B


(Figure 14.3).

NTO==o=j~~=J=m~=ff

FEM: standard Galerkin


time: fully implicit, const. step

FEM: streamline upwind


time: fully implicit, const. step

Figure 14.7 Model: extended Boussinesq approximation,


mesh B (Figure 14.3).

NQKS=k~=^~=fW=`Js~===j
a

FEM: standard Galerkin


time: FE/BE predictor-corrector

FEM: streamline upwind


time: FE/BE predictor-corrector

Figure 14.8 Model: extended Boussinesq approximation,


mesh B (Figure 14.3).

FEM: standard Galerkin


time: AB/TR predictor-corrector

FEM: streamline upwind


time: AB/TR predictor-corrector

Figure 14.9 Model: extended Boussinesq approximation,


mesh B (Figure 14.3).

cbcilt==NTP

NQK=b=mW=k~~=`

FEM: standard Galerkin


time: semi-implicit CrankNicolson

FEM: streamline upwind


time: semi-implicit CrankNicolson

Figure 14.10 Model: extended Boussinesq approximation.

NTQ==o=j~~=J=m~=ff

solver: iterative

solver: direct

Figure 14.11 Model: extended Boussinesq approximation.

NQKT=k~=^~=ffW=a~=b=~=_~~=b~

NQKT k~=^~=ffW=aJ
~= b= ~
_~~=b~
NQKTKN j=
The previous Section 14.6 was dedicated to a crossverification to compare different numerical models by
Elder4 as well as Voss and Souza14 for the simulation of
the Elder problem. To this end we had to carry out our
numerical simulations using identical discretizations.
However, the slightly varying meshes used by Elder4
(Figure 14.2) and Voss & Souza14 (Figure 14.3) produces already different results (Figure 14.4). This fact
indicates that a convergent approximate solution is not
reached yet. Therefore, we consecutively refine the
meshes (Figure 14.12, Figure 14.13) to obtain convergence. The convergent solution will change no more if
the mesh is further refined.

Figure 14.13 Finite-element mesh D (number of elements ne = 9900, number of grid points - np = 10108).

Figure 14.4 shows sequences of temporal salinity


developments resulting by using the meshes B, C, D
shown in Figure 14.3, Figure 14.12 and Figure 14.13,
respectively. The finer meshes reveal equal results
which are considered as the convergent approximate
solution due to the local discretization.
Table 14.7 and Table 14.8 provide detailed information on the meshes and numerical models. For the sake
of accuracy we apply the Galerkin and the semiimplicit Crank-Nicolson schemes for local and temporal approximation.
Table 14.7 Simulation overview
Topic

Figure 14.12 Finite-element mesh C (number of elements ne = 4400, number of grid points - np = 4539).

Realization

salt mass balance convective form of transport equation


fluid mass balance

extended Boussinesq approximation

FEM scheme

Galerkin-FEM

temporal scheme

semi-implicit Crank-Nicolson scheme,


time step = 30.416667 d

cbcilt==NTR

NQK=b=mW=k~~=`
Inspecting Table 14.8 we realize that fulfilling the
standard criteria for the Courant and Peclet numbers
does not guarantee accuracy for this type of non-linear
differential equations. Therefore, for the investigation
of stability and accuracy of these non-linear problems,
consecutive refinement of local and temporal discretizations is strongly recommended.
Table 14.8 Mesh characteristics
Topic

Mesh B

Mesh C

Mesh D

elements

ne = 1100
44 x 25

ne = 4400
88 x 50

ne = 9900
132 x 75

grid points

np = 1170

np = 4539

np = 10108

Courantnumber

Crx = 0.089
Crz = 0.203

Crx = 0.178
Crz = 0.406

Crx = 0.268
Crz = 0.608

Peclet-number

Pgx = 1.771
Pgz = 0.779

Pgx = 0.885
Pgz = 0.390

Pgx = 0.590
Pgz = 0.260

The previous study of mesh effects reveals that convergent solutions due to the local discretization can be
obtained by use of mesh C, which provides the basis
for the following considerations.
Now we continue the discussion in 14.6 concerning
the time discretization schemes and the extended
Boussinesq approximation for the fluid mass balance.
Furthermore, we investigate effects resulting from the
numerical realization of the different formulations of
the transport equation: the convective and the divergence forms of the salt mass balance.
Table 14.9 provides an overview on the presented
simulations. The abbreviations are: BA - Boussinesq
approximation, EBA - extended Boussinesq approximation, CFTE - convective form of transport equation,
DFTE - divergence form of transport equation.

Table 14.9 Simulation guide


transport equation
time discretization
fully implicit

CFTE
BA

DFTE
EBA

Figure 14.19
a-f

A-F

semi-implicit
Crank-Nicolson

Figure 14.20

predictor-corrector
AB/TR

Figure 14.21

a-f
a-f

A-F
A-F

fluid mass balance

NTS==o=j~~=J=m~=ff

BA

EBA

Figure 14.22
a-f

A-F

Figure 14.23
a-f

A-F

Figure 14.24
a-f

A-F

fluid mass balance

NQKT=k~=^~=ffW=a~=b=~=_~~=b~

t = 10 years

t = 15 years

t = 20 years

mesh B

mesh C

mesh D

t = 4 years

Figure 14.14 Mesh effects: salinity distributions corresponding to meshes B, C and D.

cbcilt==NTT

NQK=b=mW=k~~=`

NQKTKO q=~
Figure 14.15 through Figure 14.18 show the effect
of different temporal discretizations on the salinity evolutions.The fully implicit scheme works with first order
accuracy 0( t ) whereas the Crank-Nicolson- and the
predictor-corrector AB/TR schemes produce results at
2
higher accuracy order of 0( t ).
As expected, the first and the second order schemes
produce different results. But we find that the impact of
the various methods on the salinity distributions
decreases with proceeding simulation time. The longterm results of all temporal discretizations agree well.
The higher order schemes (CN, AB/TR) reveal nearly
identical results over the whole simulation time. However, the predictor-corrector method failed for the convective formulation of the transport equation.
The predictor-corrector method with automatic time
step control proceeds with time steps of about 10 days.
The fixed time step for implicit schemes is t=30.417
days. In contrast, the coarser meshes used in Section
14.6 reveal quite different results for the specific temporal approximation. This is a further indication that
we really found a convergent solution for the Elder
problem.

NTU==o=j~~=J=m~=ff

NQKT=k~=^~=ffW=a~=b=~=_~~=b~

t = 10 years

t = 15 years

t = 20 years

fully implicit

CN

AB/TR

t = 4 years

Figure 14.15 Effect of temporal discretization: fully implicit, CN and AB/TR schemes; basic
case: CFTE and BA.

cbcilt==NTV

NQK=b=mW=k~~=`

t = 10 years

t = 15 years

t = 20 years

fully implicit

CN

AB/TR

t = 4 years

Figure 14.16 Effect of temporal discretization: fully implicit, CN and AB/TR schemes; basic
case: CFTE and EBA.

NUM==o=j~~=J=m~=ff

NQKT=k~=^~=ffW=a~=b=~=_~~=b~

t = 10 years

t = 15 years

t = 20 years

fully implicit

CN

AB/TR

t = 4 years

Figure 14.17 Effect of temporal discretization: fully implicit, CN and AB/TR schemes; basic
case: DFTE and BA.

cbcilt==NUN

NQK=b=mW=k~~=`

t = 10 years

t = 15 years

t = 20 years

fully implicit

CN

AB/TR

t = 4 years

Figure 14.18 Effect of temporal discretization: fully implicit, CN and AB/TR schemes; basic
case: DFTE and EBA.

NUO==o=j~~=J=m~=ff

NQKT=k~=^~=ffW=a~=b=~=_~~=b~
NQKTKP c=~=~~=J=_
~~
As denoted in Section 14.6.3 the implicit schemes
with constant time stepping realize the first step extension of the Boussinesq approximation, whereas the predictor-corrector schemes calculate the complete
extension of the Boussinesq approximation.
CFTE (Figure 14.19 - Figure 14.21)
Differences occur in the short-time scale simulations. The long-term results agree. However, the
predictor-corrector method produces non-symmetrical salinity distribution and finally failed for
later calculation times.
DFTE (Figure 14.22 - Figure 14.24)
Using the divergence formulation of the transport
equation we obtain nearly identical results over
the whole simulation time for the Boussinesq
approximation and its extension. This agreement
of the results for the different approaches for the
fluid mass balance is a little bit surprisingly,
because we consider high-saline brines with maximum density of 1200 kgm-3.

fluid mass balance: BA

f===J
= ~~=
~=
J
===Z=NI=OI=QI=NMI=NR
~=OM=~=~=
E===FK

fluid mass balance: EBA

Figure 14.19 Basic case: CFTE, fully implicit time


scheme, Galerkin-FEM, iterative solver, mesh C.

cbcilt==NUP

NQK=b=mW=k~~=`

fluid mass balance: BA

fluid mass balance: EBA

Figure 14.20 Basic case: CFTE, Crank-Nicolson time


scheme, Galerkin-FEM, iterative solver, mesh C.

NUQ==o=j~~=J=m~=ff

fluid mass balance: BA

fluid mass balance: EBA

Figure 14.21 CFTE, predictor-corrector AB/TR time


scheme, Galerkin-FEM, iterative solver, mesh C.

NQKT=k~=^~=ffW=a~=b=~=_~~=b~

fluid mass balance: BA

fluid mass balance: EBA

Figure 14.22 Basic case: DFTE, fully implicit time


scheme, Galerkin-FEM, iterative solver, mesh C.

fluid mass balance: BA

fluid mass balance: EBA

Figure 14.23 Basic case: DFTE, Crank-Nicolson time


scheme, Galerkin-FEM, iterative solver, mesh C.

cbcilt==NUR

NQK=b=mW=k~~=`
a

C
+ qC D d C = 0
t

where the fluid mass balance is explicitly included, and


the divergence form of the transport equation (DFTE),
( C ) qC D
+ (

d C ) = 0
t

fluid mass balance: BA

(14-30)

fluid mass balance: EBA

Figure 14.24 Basic case: DFTE, predictor-corrector AB/


TR time scheme, Galerkin-FEM, iterative solver, mesh C.

NQKTKQ c~= = = ~
~
FEFLOW permits the computation of two alternative formulations of the contaminant transport equation
termed as the convective form of the transport equation
(CFTE),
NUS==o=j~~=J=m~=ff

(14-29)

The divergence form of the transport equation has a


distinct advantage to conserve the total (dispersive +
advective) contaminant flux through boundaries. However, the numerical handling of these boundary terms is
complicated.
Developing the transport equation (Eq. (14-30)) by
use of the extended Boussinesq approximation we get
finally

C
C
+ qC D d C = C ------ + q C
t
Cs t

(14-31)

This means, both formulations for the transport equation are equivalent only if the Boussinesq approximation is consequently employed.
Comparing the salinity evolutions in the following
figures computed by the different formulations of the
transport equation we find that the DFTE forces the
saltwater intrusion process. In this case the isohalines
propagate more rapidly into the depth of the porous
cavity.

NQKT=k~=^~=ffW=a~=b=~=_~~=b~
a

convective form of transport


equation (CFTE)

divergence form of transport


equation (DFTE)

Figure 14.25 Basic case: BA, fully implicit time scheme,


Galerkin-FEM, iterative solver, mesh C.

convective form of transport


equation (CFTE)

divergence form of transport


equation (DFTE)

Figure 14.26 Basic case: BA, Crank-Nicolson time


scheme, Galerkin-FEM, iterative solver, mesh C.

cbcilt==NUT

NQK=b=mW=k~~=`
a

convective form of transport


equation (CFTE)

divergence form of transport


equation (DFTE)

Figure 14.27 Basic case: BA, predictor-corrector AB/TR


time scheme, Galerkin-FEM, iterative solver, mesh C.

NUU==o=j~~=J=m~=ff

convective form of transport


equation (CFTE)

divergence form of transport


equation (DFTE)

Figure 14.28 Basic case: EBA, fully implicit time scheme,


Galerkin-FEM, iterative solver, mesh C.

NQKT=k~=^~=ffW=a~=b=~=_~~=b~
a

convective form of transport


equation (CFTE)

divergence form of transport


equation (DFTE)

Figure 14.29 Basic case: EBA, Crank-Nicolson time


scheme, Galerkin-FEM, iterative solver, mesh C.

convective form of transport


equation (CFTE)

divergence form of transport


equation (DFTE)

Figure 14.30 Basic case: EBA, predictor-corrector AB/TR


time scheme, Galerkin-FEM, iterative solver, mesh C.

Finally, we present in this Section two figures which


illustrate the distributions of streamlines, hydraulic
heads, and velocities for the Elder problem.

cbcilt==NUV

NQK=b=mW=k~~=`

Figure 14.31 Streamlines and velocities at t = 20 years for the Elder problem.

Figure 14.32 Hydraulic heads and velocities at t = 20 years for the Elder problem.

NVM==o=j~~=J=m~=ff

NQKU=o

NQKU o

freshwater-saltwater transition zone. Water Resources Research


23(10), 1851-1866.

1. Ackerer, P., Younes, A. and Mos, R. (1999) Modeling variable


density flow and solute transport in porous medium: 1. Numerical
model and verfication. Transport in Porous Media 35(3), 345373.
2. Bear, J. (1972) Dynamics of fluids in porous media. Elsevier,
New York, 764pp.
3. Elder, J.W. (1966) Numerical experiments with a free convection
in a vertical slot. J. Fluid Mechanics 24, 823-843.
4. Elder, J.W. (1967) Transient convection in a porous medium. J.
Fluid Mechanics 27(3), 609-623.
5. Evans, D.G. and Raffensperger, J.P. (1992) On the stream function for variable-density groundwater flow. Water Resources
Research 28(8), 2141-2145.
6. Frolkovic, P. and De Schepper, H. (2000) Numerical modelling of
convection dominated transport with density driven flow in
porous media. Advances in Water Resources 24(1), 63-72.
7. Diersch, H.-J. (1981) Primitive variables finite element solutions
of free convection flows in porous media Zeitschr. Angew. Math.
Mech. (ZAMM) 61(7), 325-337.
8. Diersch, H.-J., Kolditz, O. and Jesse, J. (1989) Finite element
analysis of geothermal circulation processes in hot dry rock fractures. Zeitschr. Angew. Math. Mech. (ZAMM) 69(3), 139-153.
9. Diersch, H.-J.G. and Kolditz, O. (1998) Coupled groundwater
flow and transport: 2. Thermohaline and 3D convection systems.
Advances in Water Resources 21(5), 401-425.
10. Holzbecher, E. (1998) Modeling density-driven flow in porous
media. Springer-Verlag, Berlin.
11. Kolditz, O. and Diersch, H.-J. (1993) Quasi steady-state strategy
for numerical simulation of geothermal circulation in hot dry rock
fractures. Int. J. Non-Linear Mechanics 28(4), 467-481.
12. Kolditz, O. (1997) Strmung, Stoff- und Wrmetransport im
Kluftgestein (flow, contaminant and heat transport in fractured
rock). Gebr. Borntraeger, Berlin.
13. Kolditz, O., Ratke, R., Diersch, H.-J.G. and Zielke, W. (1998)
Coupled groundwater flow and transport: 1. Verification of density variable flow and transport models. Advances in Water
Resources 21(1), 27-46.
14. Voss, C.I. and Souza, W.R. (1987) Variable density flow and solute transport simulation of regional aquifers containing a narrow

cbcilt==NVN

NQK=b=mW=k~~=`

NVO==o=j~~=J=m~=ff

NR

q~=`

NR

q~=`

NRKN f
Thermohaline processes in groundwater systems are
connected with the presence of heterogeneous temperature and concentration fields. Thus, complex convections can arise from heat and salinity gradients acting
simultaneously (Murray and Chen4, Shen5, Angirasa
and Srinivasan1). Geophysical applications of thermohaline models can be found e.g. in the field of geothermics and waste disposal in salt formations (Evans and
Nunn3). In this way thermohaline effects gain importance for the production of mineralized thermal water
and groundwater movement near salt domes.

Figure 15.1 Definition of the thermohaline Elder problem


(adapted from Voss and Souza6).

NRKO p~===m NRKP d=b~


Referring to the above study of Elders problem
(see Chapter 14) we define its thermohaline expansion.
In addition to the salinity differences a thermal gradient
is superposed to the aquifer while a permanently heating of the base of the rectangular aquifer is maintained
(Figure 15.1). A detailed discussion of the thermohaline problems can be found in Diersch and Kolditz2.

The mathematical model describing density dependent groundwater systems was developed in the previous Chapter 14. In these models the densities are
altered either from concentration or from temperature
variations. Now we condense the equations for the
thermohaline problem, where the fluid density depends
on both the salinity concentration and the temperature.
cbcilt==NVP

NRK=q~=`
Furthermore, in this situation mass and heat transfer
have to be computed simultaneously. The following
equations are denoted for the balances of fluid mass,
solute mass, linear momentum, and heat.
C
T

q = ------ T q ------ C T T
Cs

Cs t
t

(15-1)

( C ) + ( qC DC ) = 0
t

(15-2)

( C, T ) 0 g
q = K h ------------------------------ -----
0
g

(15-3)

f f

s s

( c + ( 1 )c )
f

f f
T
+ c qT
t

with the constitutive equation for the density,

= 0 1 + ------ ( C C 0 ) T ( T T 0 )
Cs

The dimensionless analysis provides three numbers


that control the coupled problem (Eq. (14-18) to Eq.
(14-20) of Chapter 14). The acting initial and boundary
conditions for the case of solute transport are illustrated
in Figure 15.1. The following table summarizes the
model parameters and its values used in the present
study.

(15-4)

( + ( 1 ) )T = 0

Table 15.1 Model parameters


Symbol

NVQ==o=j~~=J=m~=ff

Name

contaminant concentration

Cs

maximum contaminant concentration

f f

Value

Unit
M L-3

M L-3

thermal capacity of the fluid phase

4.2 x 106 J m-3 K-1

ML-1T-2K-1

s s

thermal capacity of the solid phase

ML-1T-2K-1

Dd

molecular diffusion coefficient

3.565 x 10-6 m2 s-1

L2 T-1

9.81 m s-2

L T-2

gravity acceleration vector

piezometric head

(15-5)

permeability of the porous medium

hydraulic conductivity of the porous


medium

L
4.845 x

10-13

m2

4.752945 x 10-6 m s-1

L2
L T-1

NRKQ=k~=^~
Table 15.1 Model parameters (continued)
Symbol

Name

Height of the cell

Le

Lewis number

buoyancy ratio, Eq. (14-19)

Darcy flux, specific discharge

Ra

Rayleigh number

temperature

Cs

density ratio

Value

Unit

150 m

2,3,4,5,10,

0.04 m d-1

L T-1

400

1
K

volumetric thermal expansion coefficient

kinematic porosity

heat diffusivity

0.2

0.2 ... 0 | T 0 | K-1

K-1

0.1

3.565 x 10-7 m2 s-1

L2 T-1

thermal conductivity of the fluid phase

0.65 J m-1 s-1 K-1

M L T-3 K-1

thermal conductivity of the solid phase

1.5914444 J m-1 s-1


K-1

M L T-3 K-1

fluid dynamic viscosity, where C = C 0

10-3 kg m-1 s-1

M L-1 T-1

fluid density

fluid density, where C = C0

NRKQ k~=^~

M L-3
103 kg m-3

M L-3

The simulation parameters we use for the computations are summarized in Table 15.2.

The finite element mesh which has proved to obtain


a convergent solution for the Elder problem due to the
spatial resolution is shown in Figure 14.12 and is therefore applied for the following investigations.
cbcilt==NVR

NRK=q~=`
Table 15.2 Simulation parameters
Parameters
problem class:

FEFLOW
thermohaline problem

steady flow / transient


transport

FEFLOW

transient transport

homogeneous concentration

C ( t=0, x, y ) = 0

heat transport initials:

temporal and control


data:
predictor-corrector AB/TR

homogeneous temperature

T ( t=0, x, y ) = 0

mass transport
boundaries:

Galerkin-FEM
error tolerance for iterative solver

Parameters

mass transport initials:

vertical problem projection

time step regime

Table 15.2 Simulation parameters (continued)

10-3

steady flow
flow boundaries:

const. concentration at
the cell bottom
const. concentration at
a part of the top

C ( t, y=0 ) = 0
C ( t, y=L ) = 1

heat transport boundaries:

head at the upper corners

h =0m

no-flux along the


boundary,
except upper corners

q n = 0 m d-1

const. temperature at
the cell top
const. temperature at a
part of the bottom

T ( t, y=0 ) = 0
T ( t, y=L ) = 1

mass transport materials:

flow materials:
hydraulic conductivity
density ratio
thermal expansion
coefficient

NVS==o=j~~=J=m~=ff

K = 4.752945x10-6 ms-1
= 2000 x [10-4]
T = 0 ... x [10-4 K-1]

aquifer thickness
porosity

M =1m
= 0.1

sorption coefficient

= 0, therefore, R = 1.0

molecular diffusivity

D d = 3565 x [10-9 m2s-1]

NRKR=o=~=a
Table 15.2 Simulation parameters (continued)
Parameters

FEFLOW
L = 0.0, T = 0.0 m

longitudinal, transverse dispersivity

= 0 [10-4s-1]

decay rate
heat transport materials:

M =1m

aquifer thickness

= 0.1

porosity
s s

thermal capacity of the


solid phase

c = 0.0 x 106 J m-3 K-1

thermal conductivity of
the solid phase

= 1.5914444 J m-1 s-1 K-1

NRKR o=~=a
The case of pure saltwater intrusion presented in
Figure 14.18 provides the basis for the comparison of
the results of the thermohaline simulations. This case
corresponds to the best available numerical approximation: divergence formulation of the transport equations,
extended Boussinesq approximation, Galerkin-FEM,
predictor-corrector forward Adams-Bashforth/backward trapezoid Rule time integration scheme.
The series of Figure 15.2 show the salinity intrusion
for different buoyancy ratios N = , 10, 5, 4. Up to
N = 5 we detect only small changes in the salinity evolutions. This agreement of the results for higher buoyancy ratios where density alterations are dominated by

the salinity variations (we calculate also the cases


N = 1000, 100) with the asymptotic (N = ) pure saltwater Elder problem could be used for the plausibility
of the thermohaline models.
The next series of Figure 15.3 is dedicated to investigate the onset of the thermohaline convection more in
detail. We consider the results for decreasing buoyancy
ratios N = 5,4,3,2. Beginning with N = 4 the double-diffusive convection system is dramatically influenced by
the superposing thermal convection. There are not
monotonic changes in the salinity pattern. Surprisingly
we find non-symmetrical salinity pattern for the longterm situations if the influence of thermal convective
currents becomes stronger (N = 2). We suspect numerical effects. Therefore, again we use a finer mesh consisting of 9900 elements and 10108 nodes (Figure
14.13). The long-term salinity patterns for small buoyancy ratios are illustrated in Figure 15.4. Now we
derive a symmetric salinity distribution for N = 2. But
there is a quite different salinity evolution with respect
to N = 3,4. The flow system suffers further qualitative
changes. For equilibrated solute and thermal buoyancy
effects (N = 1) again non-symmetric salinity pattern are
revealed. Because of exhausting computation time we
have not tried further mesh refinements grid convergence.
It seems, that the numerical solution reflects the
physical instabilities which the thermohaline flow systems encounters if solute and thermal effects are nearly
equilibrated. We understand that modeling of such
instable thermohaline systems will be very expensive
especially in three dimensions. Finally, Figure 15.5
presents the temperature and the salinity distributions
for the case N = 5.
cbcilt==NVT

N=

N = 10

N=5

N=4

NRK=q~=`

Figure 15.2 Salinity evolutions (t = 1,2,4,10,15,20 yr) for different buoyancy ratios N starting from the
pure saltwater intrusion Elder problem (from bottom to top: N = ,10,5,4).

NVU==o=j~~=J=m~=ff

N= 5

N=4

N=3

N=2

NRKR=o=~=a

Figure 15.3 Salinity evolutions (t = 1,2,4,10,15,20 yr) for different buoyancy ratios (from bottom to
top: N = 5,4,3,2), onset of thermohaline effects.

cbcilt==NVV

N= 4

N=2

N=1

NRK=q~=`

Figure 15.4 Salinity evolutions (t = 10,15,20 yr) for small buoyancy ratios (from bottom to top:
N = 4,2,1), computed with the very fine mesh (Figure 14.13).

OMM==o=j~~=J=m~=ff

NRKR=o=~=a
salinity C

temperature T

q= ~
~= ~= J
= ~= J
~= = = Z= NI= OI= QI= NMI
NR= ~= OM= ~= E=
=FK=

Figure 15.5 Salinity and temperature distributions for N = 5.

cbcilt==OMN

NRK=q~=`

NRKS o
1. Angirasa, D. and Srinivasan, J. (1989) Natural convection flows
due to the combined buoyancy of heat and mass diffusion in a
thermally stratified medium. J. Heat Transfer 111, 657-663.
2. Diersch, H.-J.G. and Kolditz, O. (1998) Coupled groundwater
flow and transport: 2. Thermohaline and 3D convection systems.
Advances in Water Resources 21(5), 401-425.
3. Evans, G.E. and Nunn, J.A. (1989) Tree thermohaline convection
in sediments surrounding a salt column. J. Geophysical Research
94(B9), 12413-12422.
4. Murray, B.T. and Chen, C.F. (1989) Double-diffusive convection
in a porous medium. J. Fluid Mechanics 201, 147-166.
5. Shen, C.Y. (1989) The evolution of the double-diffusive instabilities: Salt fingers. Phys. Fluids A1(5).
6. Voss, C.I. and Souza, W.R. (1987) Variable density flow and solute transport simulation of regional aquifers containing a narrow
freshwater-saltwater transition zone. Water Resources Research
23(10), 1851-1866.

OMO==o=j~~=J=m~=ff

NS

e=m

NS

e=m

NSKN f
Henrys problem describes the advance of a saltwater front in a confined aquifer which is initially charged
with uncontaminated freshwater. Henry3 develops a
solution technique to this problem which can calculate
the steady-state solute distribution. To this end he
applies the Boussinesq approximation implying the
existence of stream functions.
Henry3 derives analytical expressions for the stream
function and the concentration in the form of Fourier
series. The resulting algebraic equations for the determination of the coefficients of the Fourier series must
be obtained by numerical techniques. The mystery of
Henrys solution is, that no numerical model so far has
been reproduced closely his semi-analytical model (cf.
Segol5). Nevertheless, because of the absence of any
other non-numerical technique for this kind of non-linear problems, Henrys solutions has become one of the
standard tests of variable density groundwater models.

NSKO p~===m
NSKOKN a
The idealized aquifer for the simulation of Henrys
problem is shown in Figure 16.1. The boundary conditions for flow consist of impermeable boundaries along
the top and the bottom. Hydrostatic pressure is
assumed along the vertical boundary of the sea side.
The aquifer is charged at a constant flux with freshwater. At the inland side the concentration is zero which
corresponds to freshwater. At the coastal side concentration of brine is set (Eq. (16-1).
The steady-state flow pattern and the concentration
distribution derived by Henry3 are shown in the
Figure 16.2 and Figure 16.3 Additionally the sharpinterface solution (Ghyben-Herzberg interface) is
superposed on the isochlors.

cbcilt==OMP

NSK=e=m

NSKOKO a==
Table 16.1 provides the quantities which are used
for the description of density variable flows: mass density, mass concentration, and mass fraction. Mass concentration is mass of species (water or salt) per m3 of
solution. The values correspond to the considered situation of sea water intrusion.
Figure 16.1 Definition of the Henry problem (from
Krhn4).

Table 16.1 Density, concentration and mass fraction


Mass density
[kg m-3]

Figure 16.2 Stream functions obtained by Henry from his


semi-analytical solution (from Henry3).

C = 988.398

C = 36.592

water

= 1000

salt

= 1700

brine

Mass
Mass fraction
concentration
[kg m-3]
[1]

= 1024.99

= 0.9643

= 0.0357

=1

C = 1024.99

The varying density of the solution, depending on


the content of dissolved salt, is denoted by the equation
w w

s s

= + = + ( )

(16-1)

Density variations of binary solutions (water and


salt) are further illustrated in the Figure 16.4. The more
salt is dissolved, the less water is contained in a cubic
meter of solution.

Figure 16.3 Concentration distribution obtained by Henry


from his semi-analytical solution (from Henry3).

OMQ==o=j~~=J=m~=ff

NSKP=d=b~

NSKP d=b~
We shortly summarize the basic equations for saltwater intrusion problems. For transient flow problems
the fluid mass is balanced by
S0

h
C
+ q = Q ------ + qC

t
Cs t

(16-4)

Inserting the expressions for the specific discharge it


yields
S0
Figure 16.4 Mass concentrations of salt, water and solution.

To denote the dependence of the solution density


from the salt content frequently a linear function is
applied,
C
= 0 1 + ------
Cs

(16-2)

h
C
+ -----= Q + Kh
t
Cs t

(16-5)

+ K h + ------ ( C C 0 )e z ------ C

Cs
Cs

For steady flow Eq. (16-4) simplifies to


C
q = Q ------ + qC
Cs t

(16-6)

The equation of continuity of salt concentration is


which is based on a Taylor series approximation,

1
= 0 +
( C C 0 ) = 0 1 + ----- ( C C 0 )
C
0 C

(16-3)

The density ratio coefficient must therefore correspond to the mass quantity to be conserved: mass concentration, mass fraction, or its normalized quantities.

C
+ qC DC = 0
t

(16-7)

Darcys law formulates the balance of linear momentum as


( C ) 0
q = K h + ------------------------ e z
0

(16-8)

= K h + ------ ( C C 0 )e z

Cs

cbcilt==OMR

NSK=e=m
The following table summarizes the model parameters and its values used in the present study.
Table 16.2 Model parameters
Symbol
C

contaminant concentration

Dd

molecular diffusion coefficient

gravity acceleration vector

piezometric head

permeability of the porous medium

hydraulic conductivity of the porous medium

height of the cell

Darcy flux, specific discharge

Ra

density ratio

L, T

longitudinal and transverse


dispersivities

kinematic porosity

fluid dynamic viscosity,


where C = C0

fluid density (density of the solution)

fluid density,
where C = C0

Value

Unit
M L-3

6.6 or 18.86 x 10-6 m2 s-1

L2 T-1

9.81 m s-2

L T-2
L

-9

1.019368 x 10

m2

density of brine (max density of the solution)

L2

10-2 m s-1

L T-1

1m

40 m d-1

L T-1

Rayleigh number

Cs

OMS==o=j~~=J=m~=ff

Name

1
0.025

0.035, 0.035

0.35

10-3 kg m-1 s-1

M L-1 T-1
M L-3

103 kg m-3

M L-3

1.025 x 103 kg m-3

M L-3

NSKQ=k~=^~

NSKQ k~=^~
Figure 16.5 and Figure 16.6 present the finite element meshes for the following computations. The
coarse mesh A corresponds to one that is frequently
used in the literature for verification purposes. However, this mesh produces a lot of numerical dispersion
because of its grid Peclet number equal to 20. The finer
mesh B is applied to obtain more accurate results.

The simulation parameters we use for the computations are summarized in Table 16.3.
Table 16.3 Simulation parameters
Parameters
problem class:

FEFLOW
flow and mass transport

steady flow / transient


transport
vertical problem projection
temporal and control
data:
time step regime

predictor-corrector AB/TR

Galerkin-FEM
Figure 16.5 Finite element mesh A (finite elements
ne = 200, nodal points np = 231).

flow data (steady-state)


flow boundaries:
hydrostatic heads at the
coast boundary
constant flux at the
inland boundary
no-flux through the top
and the bottom

h = z
q n = -5.7024 m d-1
q n = 0 m d-1

flow materials:
Figure 16.6 Finite element mesh B (finite elements
ne = 5000, nodal points np = 5151).

hydraulic conductivity
density ratio

K = 100 x [10-4] ms-1


= 250 x [10-4]

cbcilt==OMT

NSK=e=m
Table 16.3 Simulation parameters (continued)
Parameters

FEFLOW

transport data (transient)


mass transport initials:
homogeneous concentration

C ( t=0, x, z ) = 0

mass transport
boundaries:
brine concentration at
the coast boundary
zero concentration gradient at the coast
boundary
freshwater concentration at the inland
boundary

C ( t, z 0.5m ) = 1
C n ( z 0.5m ) = 0
C =0

mass transport materials:


aquifer thickness

M =1m

porosity

= 0.35

sorption coefficient

= 0, therefore, R = 1.0

molecular diffusivity

D d = 6600 x [10-9 m2s-1]

longitudinal, transverse dispersivity


decay rate

OMU==o=j~~=J=m~=ff

L = 0.035, T = 0.035 m
= 0 [10-4s-1]

NSKR o=~=a
We start the study to compare our results with previous findings by Voss and Souza6 and Galeati et al.2
obtained with the coarse mesh A. Figure 16.7 a-c
shows the normalized salt concentrations for different
dispersion models.
The computed advances of the isochlors into the
aquifer compare very well. Some slight differences
between the salinity distributions arise near the coastal
boundary. This is caused by different boundary conditions used by Voss and Souza6. They applied a switching boundary condition, e.g. setting brine concentration
if the flow is inward and setting zero concentration gradient if the flow is outward the domain. The fine mesh
B reveals quite different results as can be seen in
Figure 16.8 a-c.
The results presented on the left hand side are calculated assuming the Boussinesq approximation, whereas
the salt distributions on the right hand side correspond
to the extended Boussinesq approximation. Furthermore, on the right hand side salinity patterns for different boundary conditions for the salt concentration at
the coastal side are presented. The ranges of Dirichlet
( C = 1 ) and Neumann conditions ( C n = 0 ) vary
between z = 0.5 - 0.7 m. However, these changes have
no impact on the advance of the saltwater front.

NSKR=o=~=a

a)

b)

c)

Figure 16.7 Positions of isochlors at equilibrium (coarse mesh results) - (a) const. dispersion D d = 6.6 x 10-6 m2s-1,
(b) const. dispersion D d = 18.86 x 10-6 m2s-1, and (c) velocity-dependent dispersion L = 0.035, T = 0.035 m; left
hand side: FEFLOW results, right hand side: (a)+(b) Voss and Souza6, (c) Galeati et al.2.

cbcilt==OMV

NSK=e=m

a)

b)

c)

Figure 16.8 Positions of isochlors at equilibrium (fine mesh results) - (a) const. dispersion D d = 6.6 x 10-6 m2s-1, (b)
const. dispersion D d = 18.86 x 10-6 m2s-1, and (c) velocity-dependent dispersion L = 0.035, T = 0.035 m.

ONM==o=j~~=J=m~=ff

NSKS=o
In Figure 16.9 and Figure 16.10 finally, we present
the streamline and hydraulic head distributions computed with the fine mesh B.

2. Galeati, G., Gambolati, G. and Neumann, S. (1992) Coupled and


partially coupled Eulerian-Lagrangian model of freshwater-seawater mixing. Water Resources Research 28(1): 149-165.
3. Henry, H.R. (1964) Effects of dispersion on salt enrichment in
coastal aquifers. In: U.S. Geological Survey Water Supply Paper
1613-C, Sea water in coastal aquifers: C70-C84.
4. Krhn, K.-P. (1991) Simulation von Transportvorgngen im
klftigen Gestein mit der Methode der Finiten-Elemente. Bericht
Nr. 29/1991, Institut fr Strmungsmechanik, Universitt Hannover, Dissertation.
5. Segol, G. (1994) Classic groundwater simulation: proving and
improving numerical models. PRT Prentice Hall, Englewood
Cliffs, New Jersey.
6. Voss, C.I. and Souza, W.R. (1987) Variable density flow and solute transport simulation of regional aquifers containing a narrow
freshwater-saltwater transition zone. Water Resources Research
23(10), 1851-1866.

Figure 16.9 Stream functions - fine mesh results.

Figure 16.10 Head distribution - fine mesh results.

NSKS o
1. Frind, E.O. (1982) Simulation of long-term transient densitydependent transport in groundwater. Advances in Water
Resources 5: 73-88.

cbcilt==ONN

NSK=e=m

ONO==o=j~~=J=m~=ff

p~=a=m=Eevaol`lfk=~=R==NF

NT

NT

p~=a=m=Eevaol`lfk=~=R==NF

NTKN f
In this Section we discuss one of the test cases that
was proposed for study by the participants of the international HYDROCOIN project (Swedish Nuclear
Power Inspectorate 1984) for the verification of
groundwater models. This test case is addressed to
model density variable groundwater flow over a hypothetical salt dome.

NTKO p~===m
The considered test case is an idealization of flow
over a salt dome, where the geometry is greatly simplified. The attention is directed to the physics of flow of
an non-homogeneous fluid, whose density is depending
on the salt concentration. The geometry and boundary
conditions used are shown in Figure 17.1. The cross
section of the model extends horizontally to 900 m and
vertically to 300 m. The aquifer is considered to be
homogeneous and isotropic.

Figure 17.1 Definition of the salt dome problem (from


Herbert et al.1).

The hydraulic head varies linearly on the top of the


aquifer. All the other sides are impervious for flow. The
concentration on the top is taken equal zero on the
inflow domain ( 0m x 200m ). The middle section of
the base represents the top of the salt dome with relative salt concentration to equal unity. All the remaining
parts of the boundary are closed for diffusive fluxes.

NTKP d=b~
The governing equations of balancing fluid mass,
salt mass and linear momentum as well as the equation
cbcilt==ONP

NTK=p~=a=m=Eevaol`lfk=~=R==NF
of state for the solution density are already described in
detail in Chapters 14 through 16.

of the corresponding densities. More recently, the salt


dome problem has been studied by Younes et al.5.

Herbert et al.1 introduced a slightly different formulation by using mass fractions of fluid and salt instead

The following table summarizes the model parameters and its values used in the present study.

Table 17.1 Model parameters


Symbol
C

contaminant concentration

Dd

molecular diffusion coefficient

gravity acceleration vector

piezometric head

permeability of the porous medium

hydraulic conductivity of the porous medium

Height of the cell

Darcy flux, specific discharge

Ra s
Cs

L, T

Value

Unit
M L-3

50,500,5000 x 10-9 m2 s-1

L2 T-1

9.80665 m s-2

L T-2
L

10-12

m2

L2

1.0985252 x 10-5 m s-1

L T-1

300 m

L
L T-1

salinity Rayleigh number

269 - 6710

density ratio

0.2036108

1
K-1

thermal expansion coefficient


longitudinal and transverse dispersivities

20 m, 2 m

0.2

8.9 x 10-4 kg m-1 s-1

M L-1 T-1

kinematic porosity

fluid dynamic viscosity, where C = C 0

fluid density (density of the solution)

fluid density, where C = C 0

997 kg m-3

M L-3

density of brine (max density of the solution)

1200 kg m-3

M L-3

ONQ==o=j~~=J=m~=ff

Name

M L-3

NTKQ=k~=^~

NTKQ k~=^~

NTKQKO p=J=~=

Herbert et al.1 solve the resulting algebraic equaNTKQKN j= ~= J= J


tions by use of direct methods (fast direct frontal
=
al.1

Souza4

Herbert et
as well as Voss and
modify
the standard Galerkin finite element method by using
mixed finite element interpolations for the field variables - pressure and concentration. In this way they
model concentrations on each element by a polynomial
of lower degree than that for pressure. In particular,
they use linear elements for concentration and quadratic elements for pressure. This was done to obtain
the same order of variation (linear) of the discretized
pressure gradients and gravitational terms, too. Otherwise, wiggles can arise in regions where the vertical
pressure gradient (upward) and the gravitational term
(downward) are nearly balanced. By this technique a
much better point-wise balance of pressure and gravitational forces can be achieved within the finite elements.
Furthermore, Herbert et al.1 computed continuous
velocities by applying a finite-element scheme for the
velocity function instead of substituting the finite element values for pressure gradient and gravitational
term into the Darcy equation which leads to discontinuous velocity fields.

Figure 17.2 Finite element mesh A (finite elements


ne = 1920, nodal points np = 2013).

solver). Additionally, a Newton-Raphson scheme is


implemented for the iterative solution of the non-linear
mass transport equation. The non-linearity arises from
concentration dependent density variations.

Starting from an initial distribution of zero concentration this iterative scheme works for high diffusivity
situations, such as D = 5 x 10-6 m2s-1. This provides a
basic solution. The basic solution is used as an initial
guess for less diffusivity cases. During the iterative
scheme the diffusivities or dispersion lengths are then
reduced by small steps. These parameter increments
must be small enough to obtain satisfactorily convergence of the Newton-Raphson scheme
Kolditz and Diersch3 successfully prove this kind of
quasi steady-state solution procedure in connection
with parameter stepping for highly non-linear hydrothermal convection problems.

NTKQKP `~~=
In contrast to Herbert et al.1 we calculate the fully
transient evolution of the convection system. We use
the standard Galerkin FEM in connection with the second order AB/TR rule for time stepping with initial
increment of t = 10-5d. This computational procedure
has proved already for the simulation of Elders problem (Chapter 14). Furthermore, the extended Boussinesq approximation for the fluid mass balance and the
convective formulation of the transport equation is calculated. The finite element mesh we employed is
cbcilt==ONR

NTK=p~=a=m=Eevaol`lfk=~=R==NF
shown in Figure 17.2. The actual simulation parameters are summarized in Table 17.2.

Parameters

Table 17.2 Simulation parameters


Parameters
problem class

flow and mass transport

temporal and control


data:
predictor-corrector AB/TR

Galerkin-FEM
-3

1, 0.1, 0.01 x 10

flow data (steady-state)

h = 20.455691 - 0.0113642 x

no-flux through the


sides and the bottom

q n = 0 m d-1

ONS==o=j~~=J=m~=ff

C =1

freshwater concentration at the inflow


region of the top

C =0

zero concentration gradient at the remaining


parts of the boundary

C n = 0

porosity
molecular diffusivity
longitudinal, transverse dispersivity

flow materials:

transport data (transient)

brine concentration at
the middle section of
the bottom

aquifer thickness

linear head distribution at the top

density ratio

C ( t=0, x, z ) = 0

mass transport materials:

flow boundaries:

hydraulic conductivity

homogeneous concentration
mass transport
boundaries:

vertical problem projection

error tolerance

FEFLOW

mass transport initials:

FEFLOW

steady flow / transient


transport

time step regime

Table 17.2 Simulation parameters (continued)

K = 0.10985252 x [10-4] ms-1


= 2036.108 x [10-4]

decay rate

M =1m
= 0.2
D d = 0,50,500,5000 x [10-9
m2s-1]
L = 20 m, T = 2 m
= 0 [10-4s-1]

NTKR=o=~=a

NTKR o=~=a
In our study we calculate different models: diffusive, dispersive, and diffusive-dispersive ones (Table
17.3). The data are due to the HYDROCOIN test case

which has been tackled by several authors and is well


described in the literature (Herbert et al.1, Oldenburg
and Pruess2).

Table 17.3 Simulation overview


Model

diffusive models

Diffusivity

Long. dispersivity

Trans. dispersivity

5 x 10-6 m2 s-1

0m

0m

-7

2 -1

5 x 10 m s

0m

0m

5 x 10-8 m2 s-1

0m

0m

20 m

2m

1.39 x 10-8 m2 s-1

20 m

2m

20 m

2m

5x
diffusive-dispersive models
dispersive models

10-8

m2 s-1

NTKRKN m~==

NTKRKO a=

In all simulation we detect stratified systems (Figure 17.3 - Figure 17.10). There is a freshwater region in
the upper part with higher velocities where flow is
driven by the superposed head gradient on the surface.
And there is a brine pool along the bottom with recirculating flow at lower velocities. The outflow of the salt
is focussed on the upper right-hand corner. With
decreasing diffusivities the freshwater zone is dipping
more into the depth of the reservoir or otherwise, the
saltwater rising is more limited.

Figure 17.3 presents the findings for the steadystate salt distribution by Herbert et al.1 and our results
for the diffusive models.

cbcilt==ONT

NTK=p~=a=m=Eevaol`lfk=~=R==NF

NAMMU results (Herbert et al.1)

FEFLOW results

Figure 17.3 Comparison of different diffusive models: D d = 5 x 10-6, 5 x 10-7m2s-1 (from top to bottom).

Herbert et al.1 begin their computations with molecular diffusivity of D = 5 x 10-6 m2s-1 which corresponds to the largest principal component of the
dispersion tensor occurring within the model region.
The largest velocities arise near the top where the flow
is driven purely by the specified head gradient. Starting
from the high-diffusivity basic case, they applied
parameter stepping for reducing the diffusivity down to
D = 5 x 10-8 m2s-1.
Figure 17.4 presents results for the case with small
diffusivity ( D = 5 x 10-8 m2s-1). Oldenburg and
Pruess2 compute a quite different salt distribution - they
called it swept forward. They argued that for large
diffusive fluxes ( D > 2 x 10-7 m2s-1) the amount of
brine entering rising from the bottom source is sufficient to cause a clockwise recirculation against the
ONU==o=j~~=J=m~=ff

overall counterclockwise flow imposed by the head


differences along the top. In contrast, for small diffusive fluxes ( D < 10-7 m2s-1) the convective forces are
insufficient to overcome the pressure-driven flow. As a
consequence, the brine is swept forward from left to
right with the overall flow.
However, the results must be very careful treated by
physical interpretation. The considered flow system is
highly sensitive to diffusion. Therefore, especially for
small physical diffusivities the flow regime may be significantly affected by numerical diffusion. The presented findings by the several authors are quite similar
but a perfect agreement could not be achieved (Figure
17.5). We feel that the results strongly depend on the
specific numerical schemes.

NTKR=o=~=a

NAMMU results (Herbert et al.1)

TOUGH2 results (Oldenburg and Pruess2)

Figure 17.4 Comparison of salt distributions for the diffusive model: D d = 5 x 10-8 m2s-1.

TOUGH results (Oldenburg and Pruess2)

FEFLOW results

Figure 17.5 Comparison of salt distributions for the diffusive model: D d = 5 x 10-7 m2s-1.

NTKRKP p~
We observe a strong influence of the error tolerance
for the iterative solving procedure (ORTHOMIN). For
the large diffusivity case D d = 5 x 10-6 m2s-1 an error
tolerance = 10-3 was sufficient to obtain convergence. The intermediate diffusivity case D d = 5 x 10-7
m2s-1 required more accurate iterations ( = 10-4). For
small diffusivities Dd = 5 x 10-8 m2s-1 we reduce the

error tolerance further down to = 10-6 but we detect


only instable solutions. This is connected with the corresponding large grid Peclet numbers (Eq. (17-1),
Table 17.4). Finer meshes would be required for this
case.
x
Kh
Pg x = ------ = ----------L
D d

hK 0
y
Pg y = ------ = ---------------------------- (17-1)
D d
L

cbcilt==ONV

NTK=p~=a=m=Eevaol`lfk=~=R==NF

Table 17.4 Grid Peclet numbers


Model

Diffusivity

Pg x

Pgy

D d = 5 x 10-6 m2 s-1 1.872 1.098


diffusive models

D d = 5 x 10-7 m2 s-1 18.72 10.98


D d = 5 x 10-8 m2 s-1 187.2 109.8

dispersive models L = 20 m, T = 2 m

0.75

0.25

NAMMU results (Herbert et al.1)

The failing of the divergence formulation of the


transport equation is related to the outflow of saltwater
on the right-hand side of the upper surface.

NTKRKQ aJ=
Figure 17.6 compares the findings by Herbert et al.1
and our results for the diffusive-dispersive models. We
observe a quite optimistic agreement.
FEFLOW results

Figure 17.6 Comparison of different diffusive-dispersive models: L =20 m, T =2 m; D d = 5 x 10-8 m2s-1 and 1.39 x
10-8 m2s-1 (from top to bottom).

NTKRKR a=
Oldenburg and Pruess2 recently reported on disperOOM==o=j~~=J=m~=ff

sive transport dynamics in strongly coupled groundwater-brine flow systems. In contrast to previous
simulations of the HYDROCOIN case they found socalled swept forward regimes (see above) for small diffusivities (Figure 17.4) and also for the dispersive
model (Figure 17.7).

NTKS=q~=b===p~=a=m

TOUGH results (Oldenburg and Pruess2)

FEFLOW results

Figure 17.7 Comparison of salt distributions for dispersive models: L =20 m, T =2 m.

Finally, we plot the contours of hydraulic head,


streamlines, velocity field, and salt concentrations for
the dispersive model: L = 20 m, T = 2 m (Figure
17.8). Stream functions are computed by the finite element method. Contour plots of streamlines give an
excellent image of the flow system.
Streamlines point out the stratification of the flow system:
(i) a freshwater region in the upper part flow is driven by the superposed head gradient on the surface
(higher velocities) and
(ii) a recirculating brine pool along the bottom with
two convection cells (lower velocities)

NTKS q~= b
==p~=a=m
NTKSKN d=~

Elder problem (see Chapter 14) are derived by use of


thermal Rayleigh numbers,
KLT
Ra t = ------------------

(17-2)

where
f

+ ( 1 )
= ----------------------------------f f
c

(17-3)

Frequently salinity Rayleigh numbers are introduced to


characterize solute convections. Starting from Darcys
law which describes the momentum balance for
groundwater systems,
0 g
q = K h --------------- -----

0 g

(17-4)

we transform the momentum balance Eq. (17-3) to


dimensionless forms by introducing the following
change of variables

The dimensionless equations for the thermohaline


cbcilt==OON

NTK=p~=a=m=Eevaol`lfk=~=R==NF
g
q D = Ra s h D ( C D 1 ) -----
g

(17-6)

with the salinity Rayleigh number,

KL ------ C
Cs
L
Ra s = ---------------------- = -----D
T

(17-7)

Finally, we present results for the thermohaline


extension of the salt dome problem (dispersive model).
To this extend the coupled mass and heat transfer has to
be simulated. Similar to the thermohaline Elder problem we superpose a thermal gradient by fixing normalized temperatures at the bottom (T = 1) and the surface
(T = 0). Figure 17.9 shows the salinity and temperature
distributions at t = 100 yr for different buoyancy ratios
(Table 17.5).
Table 17.5 Buoyancy ratios and Rayleigh numbers

Figure 17.8 Hydraulic head, streamlines, velocity field, and


salinity distribution at t = 100 yr for the diffusive-dispersive
model: L = 20 m, T = 2 m; D d = 1.39 x 10-8 m2s-1 (from
top to bottom).
L
q D = ------- q
D

x
x D = --L
C
C D = -------C

h
h D = -----------------------------L ( C s )C

(17-5)

which yield the following dimensionless Darcy equation,


OOO==o=j~~=J=m~=ff

Buoyancy
ratio N
Eq. (14-19)

Ra s

Ra t

Eq. (17-7)

Eq. (17-2)

150

1114

150

557

150

371

150

223

NTKS=q~=b===p~=a=m

salinity

temperature

Figure 17.9 Salinity and temperature distributions at t = 100 yr for different buoyancy ratios N = 5, 3, 2, 1 (from top to
bottom) as well as velocity fields for the thermohaline salt dome problem.

cbcilt==OOP

NTK=p~=a=m=Eevaol`lfk=~=R==NF

salinity

temperature

Figure 17.10 Salinity and temperature distributions at t = 100 yr for different buoyancy ratios N = 10, 5, 1 (from top to
bottom) as well as velocity fields for the thermohaline salt dome problem. Thermal Rayleigh number is Ra t = 22.

The results presented in Figure 17.10 correspond to


a smaller thermal Rayleigh number of around Rat =
22, which is due to a thermal expansion coefficient of
= 5 x 10-4 K-1 and a geothermal gradient of T L =
0.33 Km-1.

OOQ==o=j~~=J=m~=ff

NTKSKO `=~
In the previous sections we discussed three prominent examples of variable density groundwater systems: Elders problem, Henrys problem, and now the
salt dome problem. We learn a lot of their physics and
how to simulate numerically this kind of non-linear
problem. There are still numerous questions. We sus-

NTKT=o
pect that the stage of verification of variable density
models is not finished yet. Even modeling benchmarks
which are simple in comparison to field situations is
still a challenge for hydrologists. Comparison of the
results computed by different simulators seems to be
the right way to provide more confidence in the numerical simulations.

NTKT o
1. Herbert, A.W., Jackson, C.P. and Lever, D.A. (1988) Coupled
groundwater flow and solute transport with fluid density strongly
dependent upon concentration. Water Resources Research 24(10):
1781-1794.
2. Oldenburg, C.M. and Pruess, K. (1995) Dispersive transport
dynamics in a strongly groundwater-brine flow system. Water
Resources Research 31(2): 289-302.
3. Kolditz, O. and Diersch, H.-J. (1993) Quasi steady-state strategy
for numerical simulation of geothermal circulation in hot dry rock
fractures. Int. J. Non-Linear Mechanics 28(4), 467-481.
4. Voss, C.I. and Souza, W.R. (1987) Variable density flow and solute transport simulation of regional aquifers containing a narrow
freshwater-saltwater transition zone. Water Resources Research
23(10), 1851-1866.
5. Younes, A., Ackerer, P., and Mos, R. (1999) Modeling variable
density flow and solute transport in porous medium: 1. Re-evaluation of the salt dome flow problem. Transport in Porous Media
35(3), 375-394.

cbcilt==OOR

NTK=p~=a=m=Eevaol`lfk=~=R==NF

OOS==o=j~~=J=m~=ff

NU

d~=mW=r~~=p=`

NU

d~=mW=r~~=p=`

solved in two and three dimensions:

NUKN p~===m
An analytical solution for the steady-state capillary
head distribution above the water table in a soil column
(Figure 18.1) has been presented by Gardner2. This
analytical solution is compared with a steady-state simulation in FEFLOW.

vadose zone
L

NUKO d=b~
The governing equations for the unsaturated-saturated flow, contaminant mass and heat transport are
derived from the macroscopic phase-related conversation principles for mass, linear momentum and energy.
d
Let t and ( 0, ) be the spatial and temporal
domains, respectively, where d is the number of space
dimensions (2 or 3), and let t denote the boundary of
t . The subscript t implies the time dependence of the
spatial domain (if the domain becomes time-invariant:
t and t ). The spatial and temporal coordinates are denoted by x t and t ( 0, ) , The following nonlinear system finally results which has to be

water table

Figure 18.1 Sketch of the unsaturated solution


domain above the water table.
f
h
[ S o s ( ) + C ( ) ] ------ + q = Q h
t
f

(18-1)

f
- e
q = K r ( s )Kf h + --------------f

(18-2)

cbcilt==OOT

NUK=d~=mW=r~~=p=`
f
C
s ( )R d ( C ) ------- + q C
t

(18-3)

qq
D = ( L T ) ------------- + T q I
q

(18-5h)

[ ( s ( )D d I + D ) C ]
f

f f

+ [ s ( )R ( C ) + Q h ]C = s ( )Q C

(18-5i)
s

Q T ( ) = s ( )Q T + ( 1 )Q T

f
f f
s s T
f f
[ s ( ) c + ( 1 ) c ] ------ + c q T
t
f

= c D + s ( ) I

(18-5j)

(18-4)

[ ( + ( 1 ) I ) T ]
f f

+ c Qh ( T To ) = QT ( )

with the definitions and constitutive relationships


f

To solve the nonlinear equations Eq. (18-1) to Eq.


(18-4) for h , q , C , and T under unsaturated-saturated
conditions constitutive relationships for the saturationdependent moisture capacity C ( ) and hydraulic conf
ductivity Kr ( s ) have to be specified in form of empirical relationships for the capillary pressure headf
f
saturation ( s ) , with its inverse s ( ) , and for the relf
ative conductivities Kr ( s ) .

p
h = --------- + z = + z
o g

(18-5a)

S o = + ( 1 )

(18-5b)

NUKP ^~~=p

(18-5c)

Assuming the relative conductivity K r ( ) relationship as given by Philip1

s ( )
C ( ) = ---------------
f

k o g
K = ----------f
0

(18-5d)


Kr ( ) = e
1

for < 0
for 0

(18-6)

0
f = ------------------f
( C, T )

(18-5e)

= o [ 1 + ( C Co ) ( T To ) ]

(18-5f)

(1 )
R ( C ) = 1 + ---------------- ( C )

(--------------1 )- ---------------------------[ ( C ) C -]
Rd ( C ) = 1 +

OOU==o=j~~=J=m~=ff

(18-5g)

where is the sorptive number and the capillary


pressure head, an analytical solution for the steadystate capillary head distribution above the water table
(Figure 18.1) has been presented by Gardner2 in the
form
( L z )
1
1
( z ) = --- ln ------ ( v + ( K s v ) e
)
Ks

(18-7)

NUKQ=k~=^~=EcbjF
where z is vertical coordinate (positive in the gravitational direction g), v is the infiltration/exfiltration rate
(also positive in the gravitational direction g), Ks is the
saturated conductivity, L is the height of the column,
and ln represents the natural logarithm.
Table 18.1 Used parameters
Symbol

Value

Unit
4

8.64 10

Ks

10

1/m

m/d
m/s

NUKQ k~=^~=EcbjF
The comparison with the analytical solution is performed by a simulation run using the following parameters as summarized in Table 18.2.

Table 18.2 FEFLOW parameters (continued)


Parameters
constant infiltration
rate (flux boundary)

FEFLOW
v = 8.64 10

m/d

flow materials:
saturated hydraulic
conductivity

4
0.001. 10 m/s

Gardner model
parameters
curve-fitting parameter

= 1 m-1

porosity

= default

residual saturation
maximum saturation

s r = default
f

s s = default

Table 18.3 Mesh parameters


Parameters

Table 18.2 FEFLOW parameters

FEFLOW

mesh

Parameters

FEFLOW

steady state

number of nodes in zdirection

flow initials:
initial pressure head
flow boundaries:

depth

L=1m
101

0 = default

NUKR o
The comparison of capillary pressure heads for the
cbcilt==OOV

NUK=d~=mW=r~~=p=`
case of constant infiltration is shown in the next table:
.
Table 18.4 Comparison of the capillary head for
the case of a constant infiltration v
Elevation above
water table
(L-z) [m]

analytical
[m]

FEFLOW
[m]

.95

-.802813

-.802814

.85

-.724280

-.724281

.75

-.644110

-.644111

.65

-.562398

-.562399

.55

-.479238

-.479240

.45

-.394725

-.394726

.35

-.308948

-.308948

.25

-.221993

-.221994

.15

-.133946

-.133946

.05

-.044886

-.044886

The agreement is quite perfect. The same accuracy is


4
achieved for the exfiltration case v = 8.64 10 m/d
as shown in Table 18.5.

Table 18.5 Comparison of the capillary head for


the case of a constant exfiltration - v
Elevation above
water table
(L-z) [m]

analytical
[m]

FEFLOW
[m]

.95

-1.122654

-1.122649

.85

-.993830

-.993826

.75

-.868446

-.868443

.65

-.746020

-.746018

.55

-.626153

-.626151

.45

-.508510

-.508509

.35

-.392810

-.392809

.25

-.278814

-.278813

.15

-.166316

-.166316

.05

-.055140

-.055140

NUKS o
1. Philip, J.R., Theory of infiltration. Adv. Hydroscience 5 (1969),
215-296.
2. Gardner, W.R., Some steady-state solutions of the unsaturated
moisture flow equation with application to evaporation from a
water table. Soil Sci. 35 (1958) 4, 228-232.

OPM==o=j~~=J=m~=ff

`~= = ~K= mW= r~~= c= = ~


e=p

NV

`~==~K=mW=r~~=c==~=e=p=

NVKN p~===m
The objective of this chapter is to describe the processes of true unsaturated flows in homogeneous soils
to show the capabilities and the requirements in simulating processes with variable saturation. The example
refers to Celia et al.s water infiltration problem3 to
benchmark the present solution of the unsaturated-saturated modeling approach for a strong infiltration front
development. Celia et al. preferred a mixed -form
of the governing Richards equation to achieve accurate
solution and sufficient mass conservation. Unlike, the
present model is h-based and embedded in a second
order predictor-corrector scheme for tackling the nonlinear solution process by an error-controlled time step
adaptation. More details can be found in Diersch and
Perrochet2.
A detailed problem description of the Celia et al.s
problem is given elsewhere3,4 and only the major characteristics are summarized: homogeneous soil column
with a length of 1 m, van Genuchten-Mualem parametric model (19-1), (19-2) in using n = 2, (m = 0.5), A =
f
f
3.35 1/m, = 0.368, s r = 0.277, and s s = 1.0, saturated

NV

4
hydraulic conductivity of 0.922. 10 m/s, boundary
condition of constant head = -0.75 m at the top and
= -1.0 m at the bottom, and initial pressure head 0
of -1.0 m. The same spatial discretization characteristics are applied as given in Celia et al., who used z =
0.5 cm (dense grid) and z = 2.5 cm (coarse grid).
Otherwise, Celia et al. used a constant time step length
of t = 60 s for the dense-grid simulation.

NVKO d=b~
The governing equations for the unsaturated-saturated flow, contaminant mass and heat transport are
derived from the macroscopic phase-related conversation principles for mass, linear momentum and energy.
d
Let t and ( 0, ) be the spatial and temporal
domains respectively, where d is the number of space
dimensions (2 or 3), and let t denote the boundary of
t . The subscript t implies the time dependence of the
spatial domain (if the domain becomes time-invariant:
t and t ). The spatial and temporal coordinates are denoted by x t and t ( 0, ) , The following nonlinear system finally results which has to be
solved in two and three dimensions (Eq. (18-1) to Eq.
cbcilt==OPN

NVK=`~==~K=mW=r~~=c==~=e=p
(18-4)) with the definitions and constitutive relationships of Eq. (18-5a) to Eq. (18-5j).
To solve the nonlinear equations Eq. (18-1) to Eq.
(18-4) for h , q , C , and T under unsaturated-saturated
conditions constitutive relationships for the saturationdependent moisture capacity C ( ) and hydraulic conf
ductivity K r ( s ) have to be specified in form of empirical relationships for the capillary pressure headf
f
saturation ( s ) , with its inverse s ( ) , and for the relf
ative conductivities Kr ( s ) . Following parametric
model will be applied1:
van Genuchten - Mualem model:
1
----------------------------------
f
m
se = [ 1 + ( A )n ]

Kr =

1
---
f 2
( se ) 1

for

<0

for

2
1- m
--
f m
( se )

(19-1)

The model has been created in FEFLOW using the following parameters:
Table 19.1 FEFLOW parameters
FEFLOW

transient flow
flow initials:
initial pressure head

(19-2)


s sr
f
- = ---------------rs e = ------------f
f
f
f
ss sr
s r

(19-3)

The above strongly nonlinear parametric curves


f
describe the fluid saturation s ( ) and the relative
f
hydraulic conductivity K r ( s ( ) ) as a function of the
pressure head in an unique manner. They are continuous over the entire range of which is an useful feaOPO==o=j~~=J=m~=ff

NVKP k~=^~=EcbjF

Parameters

with the effective saturation of fluid


f

ture for the numerical implementation. Under saturated


f
conditions s 1 the nonlinearities in the parametric
relationships vanish, however, for free-surface conditions a nonlinear boundary-value problem remains to
be solved.

0 = -1.0 m

flow boundaries:
constant head at top of
column

= -0.75 m

constant head at the


bottom

= -10.0 m

flow materials:
saturated hydraulic
conductivity

4
0.922. 10 m/s

van Genuchten model


parameters
curve-fitting parameter

n = 2, (m = 0.5)

NVKQ=o
Table 19.1 FEFLOW parameters (continued)
Parameters
curve-fitting parameter

FEFLOW
A = 3.35 1/m

porosity

= 0.368

residual saturation

s r = 0.277

maximum saturation

s s = 1.0

The fine grid has a element width of 0.5 cm, the


coarse grid of 2.5 cm. Both finite element models are 1
m long and 1cm wide (Figure 19.1).
Table 19.2 FEM parameters
Parameter
mesh
width, depth

FEFLOW
Figure 19.1
B = 0.01 m, L = 1 m

dense mesh:
x = 0.01 m, y = 0.005 m
coarse mesh:
x = 0.01 m, y = 0.025 m
time stepping regime
error measure
initial time step length
final time

AB/TR time integration


scheme
10e-3
1e-5 [d]
1 [d]

NVKQ o
A comparison with Celia et al.s results gives very
good agreements. Figure 19.2 shows the computed
pressure profiles at a time of 1 day for the case of the
dense grid (notice, Celia et al.s results are picked from
a table presented in4, where only selected sample points
are listed, however). Celia et al. used both finite difference and finite element techniques. For the -based
Richards equation (which should be equivalent to a hbased form) they found nonoscillatory results only for
finite differences, while finite elements echoed wiggles
at the moisture front once coarse meshes were applied.
They could improve the results if resorting to a modified Picard iteration technique based on a mixed form of the governing Richards equation.
These difficulties as reported in3 are completely prevented in the present predictor-corrector technique
even for coarse meshes. Figure 19.3 displays the pressure profiles for different spatial discretizations. It
reveals three things: Firstly, the proposed numerical
technique gives wiggle-free solutions. The loss of
accuracy against the dense grid solution is similar to
that as observed by Celia et al. (for more details see3).
Secondly, the accuracy of the solution is significantly
influenced by the spatial (and temporal) discretization.
That means, the simulation of unsaturated flow processes requires sufficiently refined meshes. Thirdly, a
50% moisture profile centre point would not necessarily serve as an accurate identification of a possible free
surface location. As evidenced in Figure 19.3, in
dependence on the mesh refinement a leading or lagging phase error in the moisture profiles can occur, but,
unfortunately, the effect of phase lag or lead does not
appear as a simple function of grid spacing (cf. also3).

Figure 19.1 Coarse


and fine finite element grid used.

cbcilt==OPP

NVK=`~==~K=mW=r~~=c==~=e=p
0

-20

-20

Pressure [kPa]

Pressure [kPa]

-40

-60
Celia et al.
Present

-80

-100
0.0

-40

-60

-80

0.2

0.4
0.6
Depth [m]

0.8

1.0

Figure 19.2 Pressure profiles at t = 1 day for the dense


grid: Celia et al.s -based solution3,4 using a mixedform Picard iteration versus the present predictor-corrector h-based solution using an adaptive time stepping.

-100
0.0

z = 0.50 cm
z = 2.50 cm
z = 16.67 cm

0.2

0.4
0.6
Depth [m]

0.8

1.0

Figure 19.3 Pressure profiles at t = 1 day for different


grid spacing z computed by the present predictor-corrector technique with a h-based form of the Richards
equation.

NVKR o
1. Bear, J. and Bachmat, Y., Introduction to modeling of transport
phenomena in porous media. Kluwer Academic Publ., Dordrecht,
1991.
2. Diersch, H.-J.G. and Perrochet, P., On the primary variable
switching technique for simulating unsaturated-saturated flows,
Advances in Water Resources 23 (1999) 3, 271-301.
3. Celia, M.A., Bouloutas, E.T. and Zarba, R.L., A general massconservative numerical solution for the unsaturated flow equation. Water Resour. Res., 26 (1990) 7, 1483-1496.
4. Segol, G., Classic groundwater simulations: proving and improving numerical models. PTR Prentice Hall, Englewood Cliffs, New
Jersey, 1994.

OPQ==o=j~~=J=m~=ff

s~=d=mW=r~~=c==~
f=p

OM

s~=d=mW=r~~=c==~=f=p

OMKN p~===m

Depth (cm)

The objective of
0
Clay Loam (1)
this chapter is to
describe the pro25
Loamy Sand (3)
cesses of true unsat(4)
(5)
urated
flows
in
Changing to
layered soils to show
(6)
the capabilities and
(7)
Sand
(8)
the requirements in
75
Dense
Layer
(2)
simulating processes
87
with variable saturation. The example
(9)
Sand
concerns the simulation of the moisture
movement in a layered soil as studied by
Node
van Genuchten1. A
170
soil column with a
length of 170 cm
Figure 20.1 Layered soil proincludes 4 layers:
file.
clay loam (0-25 cm),
loamy sand (25-75 cm), dense material (75-87 cm), and
sand (87-170 cm), where the loamy-sand layers prop-

OM

erties change gradually with depth. The initial conditions for the flow are given by 0 = -3.5 m. A constant
h
flux is at the surface q n = -0.25 m/d at t 1 day (infilh
tration) and q n = 0.005 m/d at t > 1 day (evaporation).
At the bottom a drainage gradient-type BC of = 4 m/d
is imposed, accordingly, the bottom boundary can
freely drain. The parameter in the constitutive relations
(van Genuchten-Mualem model) can be found in4.

OMKO d=b~
The governing equations for the unsaturated-saturated flow, contaminant mass and heat transport are
derived from the macroscopic phase-related conversation principles for mass, linear momentum and energy.
d
Let t and ( 0, ) be the spatial and temporal
domains respectively, where d is the number of space
dimensions (2 or 3), and let t denote the boundary of
t . The subscript t implies the time dependence of the
spatial domain (if the domain becomes time-invariant:
t and t ). The spatial and temporal coordinates are denoted by x t and t ( 0, ) , The following nonlinear system finally results which has to be
solved in two and three dimensions:
cbcilt==OPR

OMK=s~=d=mW=r~~=c==~=f=p
f
h
[ S o s ( ) + C ( ) ] ------ + q = Q h
t
f
o

f
- e
q = K r ( s )Kf h + --------------f

o
f
C
s ( )R d ( C ) ------- + q C
t

(20-1)

(20-2)

(20-3)

1
----------------------------------
f
m
se = [ 1 + ( A )n ]

Kr =

1
--f 2
( se ) 1

[ ( s ( )D d I + D ) C ]
f

+ [ s ( )R ( C ) + Q h ]C = s ( )Q C
f
f f
s s T
f f
[ s ( ) c + ( 1 ) c ] ------ + c q T
t
f

<0

for

(20-5)

for

2
1 m
---
f m
( se )

(20-6)

with the effective saturation of fluid


f

s sr

f
- = ---------------rs e = ------------f
f
f
f
ss sr
s r

(20-4)

[ ( + ( 1 ) I ) T ]

(20-7)

f f

+ c Qh ( T To ) = QT ( )

with the definitions and constitutive relationships of


Eq. (18-5a) to Eq. (18-5j).
To solve the nonlinear equations Eq. (20-1) to Eq.
(20-4) for h , q , C , and T under unsaturated-saturated
conditions constitutive relationships for the saturationdependent moisture capacity C ( ) and hydraulic conf
ductivity K r ( s ) have to be specified in form of empirical relationships for the capillary pressure headf
f
saturation ( s ) , with its inverse s ( ) , and for the relf
ative conductivities Kr ( s ) . Following parametric
model will be preferred3:
van Genuchten - Mualem model:

OMKP k~=^~
The parameter settings of the van Genuchten model
are summarized in the following tables.
Table 20.1 Initial and boundary conditions used by
van Genuchten1
Parameters
transient flow
flow initials:
initial pressure head

0 = -3.5 m

flow boundaries:
constant flux at surface at t 1 day

OPS==o=j~~=J=m~=ff

Value

q n = -0.25 m/d

OMKP=k~=^~
Table 20.1 Initial and boundary conditions used by
van Genuchten1 (continued)
Parameters

Value

constant flux at surface at t > 1 day

q n = 0.005 m/d

drainage at bottom

The conversion of van Genuchtens example to a


FEFLOW model requires some changes in the units of
the parameters. The finite element model is 1.70 m
long and 0.01 m wide (Figure 20.2). The grid spacing
is 0.01 m.

h
q n = 4 m/d

Table 20.2 Material properties1,4


f

Soil

Type

(cm/cm)

(cm/cm)

A
(cm-1)

Ks

(cm/day)

So

(cm-1)

clay loam

0.2

0.54

0.008

1.8

25.

4e-7

dense layer

0.25

0.40

0.009

3.0

10.

5e-8

loamy sand

0.17

0.47

0.010

2.0

75.

1e-7

0.1611

0.4611

0.01306

2.178

132.8

1e-7

0.15

0.45

0.0108

2.4

205.

1e-7

0.14

0.44

0.0112

2.6

270.

1e-7

0.1311

0.4311

0.01156

2.778

327.8

1e-7

0.1244

0.4244

0.01182

2.911

371.1

1e-7

0.12

0.42

0.012

3.0

400.

1e-7

sand

cbcilt==OPT

OMK=s~=d=mW=r~~=c==~=f=p
Table 20.3 Initial and boundary conditions applied
in FEFLOW
Parameters

FEFLOW

transient flow
flow initials:
initial pressure head

0 = -3.5 m

Table 20.3 Initial and boundary conditions applied


in FEFLOW (continued)
Parameters

FEFLOW

constant flux at surface at t 1 day

q n = -0.25 m/d

constant flux at surface at t > 1 day

q n = 0.005 m/d

drainage at bottom

q n = 4 m/d

flow boundaries:

Table 20.4 Material properties applied in FEFLOW


f

Soil

OPU==o=j~~=J=m~=ff

(cm/cm)

(cm/cm)

A
(m-1)

Ks

(10-4m/s)

So

(m-1)

clay loam

0.2000

0.5400

0.800

1.8

0.029

4e-5

dense layer

0.2500

0.4000

0.900

3.0

0.012

5e-6

loamy sand

0.1700

0.4700

1.000

2.0

0.087

1e-5

0.1611

0.4611

1.306

2.178

0.154

1e-5

0.1500

0.4500

1.08

2.4

0.237

1e-5

0.1400

0.4400

1.12

2.6

0.313

1e-5

0.1311

0.4311

1.156

2.778

0.379

1e-5

0.1244

0.4244

1.182

2.911

0.430

1e-5

0.1200

0.4200

1.200

3.0

0.463

1e-5

9
Figure 20.2 Finite
element grid used.

Type

sand

OMKQ=o

OMKR o
Table 20.5 FEM parameters
Parameter
mesh
width, depth

FEFLOW
Figure 20.2
B = 0.01 m, L = 1.70 m

grid spacing:
x = 0.01m, y = 0.01m
time stepping regime

AB/TR time integration


scheme

initial time step length

1e-5 [d]

final time

1. van Genuchten, M.Th., Mass transport in saturated-unsaturated


media: One-dimensional solutions. Research Rep. No. 87-WR11, Water Resources Program, Princeton Univ., 1978.
2. Diersch, H.-J.G., FEFLOW-Physical basis of modeling. WASY
Institute for Water Resources Planning and Systems Research
Ltd., Berlin, 1997.
3. Bear, J. and Bachmat, Y., Introduction to modeling of transport
phenomena in porous media. Kluwer Academic Publ., Dordrecht,
1991.
4. Segol, G., Classic groundwater simulations: proving and improving numerical models. PTR Prentice Hall, Englewood Cliffs, New
Jersey, 1994.

8 [d]

OMKQ o
A comparison between the present solutions and the
results obtained by van Genuchten1 who used a Hermitian-finite element approach is exhibited in Figure 20.3
and Figure 20.4 displays the simulated moisture-content profiles during the infiltration period at t 1 day.
The moisture-content histories during the redistribution
phase at t > 1 day of the soil column are seen in
Figure 20.4. As shown the agreement of the results is
nearly perfect.

cbcilt==OPV

OMK=s~=d=mW=r~~=c==~=f=p

0.00
t=0

0.25

.05
.15

fr
.25

Depth [m]

0.50

.35
.50

fs

0.75

1.00

.65

.80

1.25

.95

1.50

0.0

0.1

0.2

0.3

0.4

0.5

0.6

Moisture content f [1]


Figure 0.1
20.3Simulated
Simulated moisture-content
moisture-content profiles
profiles during
during infiltration:
infiltration: present
present solutions
solutions (left)
(left) and van Genuchtens
results1, 4 (right), time in days.

OQM==o=j~~=J=m~=ff

OMKR=o

0.00

0.25

fr

Depth [m]

0.50

1.5

4.0

8.0

1.0

fs

0.75

1.00
1.5

1.0

1.25
4.0

1.50

0.0

8.0

0.1

0.2

0.3

0.4

0.5

0.6

Moisture content [1]


Figure 20.4 Simulated moisture-content profiles during redistribution: present solutions (left) and van Genuchtens
results1, 4 (right), time in days.

cbcilt==OQN

OMK=s~=d=mW=r~~=c==~=f=p

OQO==o=j~~=J=m~=ff

s~~==~K=mW=a~~=b

ON

s~~==~K=mW=a~~=b

ONKN p~===m
Vauchaud et al.2 reported experimental results
which referred to a ditch-drained soil problem. Their
results are useful for proving and comparing numerical
schemes applied to a typical drainage problem as
already done by Gureghian3, Ntzmann4 and Ngyuen5.
A half drain-spacing with a length of 3 m and a height
of 2 m is considered (Figure 21.1). Initially, the water
level in the box is at z = h 0 and the system is under
hydrostatic equilibrium 0 = h 0 z . The soil is
assumed to be isotropic with a saturated conductivity
4
of K = 1.11 10 m/s.
The Haverkamp parametric model Eq. (21-5), Eq.
(21-6) is used for the unsaturated soil with
f
4
= 0.063396 m, A = 3.6 10 m, = 0.3 , s s = 1 and
f
s r = 0 . The initial head h 0 is given by 1.45 m. The
water level of the ditch h w is 0.75 m. The magnitude
h s represents the elevation of the seepage face which is
h s = h 0 at t = 0 and has to be determined in the solution process. In Vauchaud et al.s experiment the drainage process has been performed without any
infiltration ( v = 0 on top, Figure 21.1). Accordingly,

ON

the water table descends continuously up to reaching


the water level h w of the ditch.
h

qn = v
unsaturated zone
h

qn = 0

initial water table


transient water table
h

qn = 0
= 0

h0
saturated zone

hs

= hw z

hw

qn = 0

Figure 21.1 Sketch of Vauchaud et al.s drainage


experiment2: geometry and boundary conditions.

ONKO d=b~
The governing equations for the unsaturated-saturated flow, contaminant mass and heat transport are
derived from the macroscopic phase-related conversation principles for mass, linear momentum and energy
cbcilt==OQP

ONK=s~~==~K=mW=a~~=b
(see Part I: Physical Basis of Modeling, p. 11 and fold
lowing.) Let t and ( 0, ) be the spatial and temporal domains respectively, where d is the number of
space dimensions (2 or 3), and let t denote the boundary of t . The subscript t implies the time dependence
of the spatial domain (if the domain becomes timeinvariant: t and t ). The spatial and temporal coordinates are denoted by x t and t ( 0, ) ,
The following nonlinear system finally results which
has to be solved in two and three dimensions:
f
h
[ S o s ( ) + C ( ) ] ------ + q = Q h
t

f
C
s ( )R d ( C ) ------- + q C
t

tionships for the capillary pressure head-saturation


f
f
( s ) , with its inverse s ( ) , and for the relative conf
ductivities Kr ( s ) . Following parametric model will be
preferred1:
Haverkamp model:

f
se

(21-1)

- e
q = K r ( s )Kf h + --------------f

o
f

K r ( s ) have to be specified in form of empirical rela-

(21-2)


-------------------= +

A
-------------------Kr = A + B

for

<0

for

for

<0

for

(21-5)

(21-6)

(21-3)

[ ( s ( )D d I + D ) C ]
f

q = -0.384 m/d

+ [ s ( )R ( C ) + Q h ]C = s ( )Q C
f
f f
s s T
f f
[ s ( ) c + ( 1 ) c ] ------ + c q T
t
f

(21-4)

h =1.45 m

[ ( + ( 1 ) I ) T ]
f f

h =0.75 m

+ c Qh ( T To ) = QT ( )

with the definitions and constitutive relationships Eq.


(18-5a) to Eq. (18-5j).
To solve the nonlinear equations (21-1) to (21-4) for
h , q , C , and T under unsaturated-saturated conditions

constitutive relationships for the saturation-dependent


moisture capacity C ( ) and hydraulic conductivity
OQQ==o=j~~=J=m~=ff

h =0.75 m
Figure 21.2 FEFLOW finite element grid with boundary
conditions.

ONKP=k~=^~

ONKP k~=^~
The modeling of the experiment has been realized
by using the following parameters:
Table 21.1 Used initial and boundary conditions
Parameters

Table 21.2 Used material parameters (continued)


Parameters

FEFLOW

curve-fitting parameter

curve-fitting parameter

A = 3.6 10-4 [m]

FEFLOW

= 0.063396 [m]

Table 21.3 FEM parameters

transient flow

Parameter
flow initials:
initial pressure head

0 = 1.45 [m]

flow boundaries:
inflow on top

h
qn

mesh
width, depth

FEFLOW
Figure 21.1
B = 3 m, L = 2 m

grid spacing:
= -0.384 [m/d]

hydraulic head at ditch


from 1.45 m to 0.75 m

h=z

hydraulic head at ditch


from 0.75 m to 0 m

h = 0.75 [m]

x = 0.15 m,
y = 0.0583 m - 0.06875 m
time stepping regime

FE/BE time integration


scheme

initial time step length

1e-3 [d]

final time

0.208334 [d]

Table 21.2 Used material parameters


Parameters

FEFLOW

flow materials:
conductivity

1.11[10e-4 m/s]
f

total saturation

ss = 1

residual saturation

sr = 0

porosity

= 0.3

ONKQ o
Numerical results are given by Gureghian3. However, his solutions are based on a non-zero infiltration
rate v . Accordingly, the water table descends continuously up to reaching the water level h w of the ditch.
Figure 21.3 compare the present numerical results with
Vauchaud et al.s experimental data. As seen the agreement is quite well. A comparison of the hydraulic head
cbcilt==OQR

ONK=s~~==~K=mW=a~~=b
contours, the water table and capillary fringe at time of
t = 1 hour is presented in Figure 21.4 between the
present solutions and Gureghians results. The agreement is satisfactory. Differences appear for the upper

head contours which obviously results from different


description of the infiltration boundary condition.

t=0
0.1
1.0
5.0

Figure 21.3 Descending water table of the drainage experiment: simulated free-surface locations (left) and water tables
measured by Vauchaud et al.2 (right), times in hours.

Figure 21.4 Hydraulic head contour and water table location at t = 1 hour: present results (left) and Gureghians
solutions3,5 (right).

OQS==o=j~~=J=m~=ff

ONKR=o

ONKR o
1. Bear, J. and Bachmat, Y., Introduction to modeling of transport
phenomena in porous media. Kluwer Academic Publ., Dordrecht,
1991.
2. Vauchaud, G., Vauclin, M. and Khanji, D., tude exprimentale
des transferts bidimensionels dans la zone non-sature - Application a ltude du drainage dune nappe a surface libre. Houille
Blanche 1 (1973), 65-74.
3. Gureghian, A.B., A two-dimensional finite element solution
scheme for the saturated-unsaturated flow with applications to
flow through ditch-drained soils. J. Hydrol. 50 (1981), 333-353.
4. Ntzmann, G., Eine Galerkin-finite-element-Methode zur Simulation instationrer zweidimensionaler ungesttigter und gesttigter Wasserstrmungen im Boden. Acta Hydrophysica XXVIII
(1983) 1/2, 37-107.
5. Nguyen, H., WHAT-FEM - A two-dimensional finite element
model for simulating water flow in variably saturated soils. Technical Note, No. I.96.125, JRC, Ispra Institute, July, 1996.

cbcilt==OQT

ONK=s~~==~K=mW=a~~=b

OQU==o=j~~=J=m~=ff

OO

t=`~~=_~=m

OO

t=`~~=_~=m

OOKN p~===m
A capillary barrier arises in unsaturated flow conditions at the contact of a fine layer of soil overlying a
coarse layer of soil1,2. If such a contact is tilted, infiltrating moisture in the fine layer will be diverted and
flow down the contact as capillary diversion
(Figure 22.1). In practical applications, a capillary barrier can be built by placing a fine layer (e.g., fine sand)
over a coarse layer (e.g., gravel) along a sloping contact. It is used to protect underground regions from
wetting by downward infiltration (protective cover for
waste disposals)3.
Ross1 developed an analytical relationship to calculate the steady-state diversion length of a tilted fine
over coarse capillary barrier with constant infiltration.
Oldenburg and Pruess2 presented a first numerical
study of a 2D tilted fine over coarse capillary barrier by
using the finite difference TOUGH2 code5. To find
qualitatively reasonable results Oldenburg and Pruess
had used an upstream weighting scheme. From the
quantitative point of view the agreement of their results
with Roos predictions became generally poor. More

recently, Webb4 presented improved results obtained by


the TOUGH2 simulator which compare now well with
Ross analytical solution.
infiltrati
on

g
fine

coarse

Figure 22.1 Schematic of a capillary barrier (from Webb4).

The present paper aims at a comparison of Ross


capillary barrier diversion formula with numerical simulations by using the finite-element simulator
FEFLOW. The numerical investigations refer to a capillary barrier problem using van Genuchten characteristic curves as recently studied by Webb4. The
simulations are performed for both 2D and axisymmetcbcilt==OQV

OOK=t=`~~=_~=m

OOKP k~=^~

ric geometries.

OOKPKN q=`~~=_~=m

OOKO d=b~
The van Genuchten-Mualem parametric model is
described by the characteristic curves (symbols are
summarized under Notation)
1
-----------------------------
m
s e = [ 1 + n ]

1
1

for

<0

for

(22-1)

A fine over coarse configuration has been selected


with a total thickness of 1 m. The fine and coarse layers
are both 0.5 m thick, and the dip of the layers is 5%
(2.86). The van Genuchten-Mualem parametric model
for the saturation-dependent capillary pressure head
and the relative conductivity are applied. The layer
properties are summarized in Table 22.2.
Table 22.1 Geometric, initial and boundary
conditions

1 m 2

------

2
m
Kr = se 1 1 se

Quantity

(22-2)

Value

dip of the layers, %

with the effective saturation


r
s sr
s e = -------------- = ---------------- .
ss sr
s r

(22-3)

As summarized in Table 22.1 the assumed infiltration rate is 0.0048 m/d, and the initial saturation is
assumed equal to the residual values for each layer. In
the lateral direction the model domain is 100 m in the
length. At the bottom and the right side boundaries a
potential boundary condition is prescribed to realize a
freely drain whenever water reaches these parts of the
domain.

thickness of fine layer, m

0.5

thickness of coarse layer, m

0.5

lateral domain dimension, m

100.

initial saturation

sr

infiltration rate, m/d

0.0048

Table 22.2 Material properties


Parameter
porosity
saturated conductivity K s , m/s
residual saturation s r

ORM==o=j~~=J=m~=ff

Upper
layer
(fine)

Lower
layer
(coarse)

0.39

0.42

2.1 10

0.3949

0.1
0.0286

OOKP=k~=^~
Table 22.2 Material properties (continued)
maximum saturation s s
van Genuchten exponent n
(m = 1 1 n )
van Genuchten parameter , 1/m

1.

1.

5.74
(0.826)

2.19
(0.543)

3.9

490.

Table 22.3 FEFLOW initial and boundary


conditions
Quantity
initial conditions:
initial saturation

OOKPKO ^=cbcilt=j

sr

boundary conditions:
infiltration rate, m/d

FEFLOW solves the governing Richards equation


by the finite-element method. A predictor-corrector
technique combined with an error-controlled automatic
time stepping is used to tackle the strongly nonlinear
equations. For the present problem the second order
accurate Adams-Bashforth/trapezoid rule (AB/TR)
scheme is preferred. For the error criterion a relative
3
maximum norm of 10 is applied. FEFLOW uses a
mass-conservative solution procedure based on a chord
slope approximation for the moisture capacity term and
a mass lumping scheme. The predictor-corrector technique has proven very powerful also for the present
problem class of unsaturated-saturated flow6. Unlike
the TOUGH2 code, no upstream weighting schemes
become necessary for the present unsaturated problem
if using FEFLOW. For the 2D and axisymmetric problem both quadrilateral and triangular meshes of linear
finite elements are performed. Mesh refinements are
used along the fine-coarse material interface and at the
bottom of the coarse layer. The used meshes are shown
in Figure 22.2.

Value

0.0048

Table 22.4 FEFLOW material properties


Upper
layer
(fine)

Lower
layer
(coarse)

porosity

0.39

0.42

saturated conductivity, 10e-4 m/s

2.1

1000

0.3949

0.0286

1.

1.

5.74
(0.826)

2.19
(0.543)

3.9

490.

Parameter

residual saturation
maximum saturation
van Genuchten exponent
van Genuchten parameter, 1/m

cbcilt==ORN

OOK=t=`~~=_~=m

a)
1m

Table 22.5 FEM parameter

mesh

FEFLOW
Figure 22.2

quadrilateral mesh

1472 bi-linear elements, 1551


nodes

triangular mesh

2944 bi-linear elements, 1551


nodes

dip of the layers, %

thickness of fine layer, m

0.5

thickness of coarse layer, m

0.5

lateral domain dimension,


m

100.

time stepping regime


relative maximum error
norm
solution procedure
approximation for the moisture capacity term

5m

Parameter

100 m
b)

AB/TR time integration


scheme
10

chord slope
mass lumping scheme

Figure 22.2 Used finite element meshes: a) quadrilateral


mesh (1472 bi-linear elements, 1551 nodes), b) triangular
mesh (2944 bi-linear elements, 1551 nodes), exaggeration
10 : 1.

OOKQ o
The simulations have been considered as steadystate if an equilibrium of the flow budget has been
achieved. Practically, it appears after about 100 days.
For the 2D cross-sectional problem Figure 22.3 exhibORO==o=j~~=J=m~=ff

OOKQ=o
its the computed streamline pattern. It clearly reveals
the capillary diversion effect in the form of a moisture
movement downdip above the contact of the fine and
coarse layer. The diversion is maintained up to a certain
distance, called the diversion length, from where an
amount of water equal to the infiltration flow enters the
coarse layer. The corresponding distribution of the saturation can be seen in Figure 22.4.
For the axisymmetric problem the length of diversion increases compared with 2D as displayed in Figure 22.5 for the computed streamlines. Up to now, the
axisymmetric problem was not considered by other
authors. Ross formula is only applicable for 2D capil-

lary barrier schemes.


A comparison of the computed diversion for the 2D
capillary barrier with Ross analytical formula and
TOUGH2 numerical results presented recently by
Webb4 can be expressed as a function of leakage/infiltration ratio. The theoretical value of the diversion
length from Ross formula for the presented van Genuchten parameter is given by 33.2 m (cf. Webb4). As
shown in Figure 22.6 there is both a good qualitative
and quantitative agreement between the analytical
results and the numerical simulations performed by
TOUGH2 and FEFLOW.

Figure 22.3 Computed streamline pattern for the 2D problem, exaggeration 10 : 1.

cbcilt==ORP

OOK=t=`~~=_~=m

Figure 22.4 Computed distribution of saturation for the 2D


problem, exaggeration 10 : 1.

Figure 22.5 Computed streamline pattern for the axisymmetric problem, exaggeration 10 : 1.

ORQ==o=j~~=J=m~=ff

OOKR=k~

Leakage/infiltration

1.5

1.0

Ross' formula
present results
Webb's results

0.5

0.0

10

20

30

40

50

60

70

80

90

100

Distance [m]

Figure 22.6 Computed barrier diversion in form of the


leakage/infiltration ratio at the lower coarse layer compared with Ross analytical formula and Webbs numerical
results for the 2D capillary barrier problem.

OOKR k~
g
Kr

LT
1

Ks
m

LT
1

gravitational acceleration.
relative hydraulic conductivity
( 0 < K r 1 , K r = 1 if saturated at
s = 1 ).
saturated hydraulic conductivity.
1 1 n curve-fitting parameter
(Mualem assumption).
n > 1 , van Genuchten curve-fitting
exponent (pore size distribution
index).
saturation ( 0 < s 1 , s = 1 if

se
sr
ss

1
1
1
1
L

1
1

medium is saturated).
effective saturation.
residual saturation.
maximum saturation.
van
Genuchten
curve-fitting
parameter.
porosity ( 0 < 1 ).
s , volumetric moisture content
( 0 < , = if medium is
saturated).
residual
volumetric
moisture
content.
saturated
volumetric
moisture
content.
angle of the fine-coarse interface
with respect to the horizontal.
pressure head ( > 0 saturated
0
medium,
unsaturated
medium).

OOKS o
1. Ross, B., The diversion capacity of capillary barriers. Water
Resour. Res. 26 (1990) 10, 2625-2629.
2. Oldenburg, C.M. and Pruess, K., On numerical modeling of capillary barriers. Water Resour. Res. 29 (1993) 4, 1045-1056.
3. Selker, J., Design of interface shape for protective capillary barriers. Water Resour. Res. 33 (1997) 1, 259-260.
4. Webb, S.W., Generalization of Ross tilted capillary barrier diversion formula for different two-phase characteristic curves. Water
Resour. Res. 33 (1997) 8, 1855-1859.
5. Pruess, K., TOUGH2 - A general-purpose numerical simulator
for multiphase fluid and heat flow. LBL-29400, Lawrence Berkeley Lab., Berkeley, Calif., May 1991.
6. Diersch, H.-J.G., Treatment of free surfaces in 2D and 3D groundwater modeling. Math. Geologie Vol. 2, 17-43, 1998.

cbcilt==ORR

OOK=t=`~~=_~=m

ORS==o=j~~=J=m~=ff

OP

e=c==~=r~~=p=`

OP

e=c==~=r~~=p=`

1
-----------------------------
m
s e = [ 1 + n ]

OPKN p~===m
Modeling of hysteretic flow processes in unsaturated porous media requires knowledge of the main wetting and drying retention curves plus the hydraulic
conductivity-water saturation relation of the medium.
The importance of hysteresis has been shown in laboratory studies. Lenhard et al.3 conducted an air-water
flow experiment in a 72-cm vertical soil column, where
the water table is fluctuated. This column experiment
having been previously simulated by Clausnitzer1 using
a one-dimensional finite element method, the corresponding FEFLOW solution can be compared with
independently obtained numerical results for both hysteretic and nonhysteretic conditions.

Kr = se

The relations of saturation vs pressure head (s-)


and hydraulic conductivity vs saturation (K-s) are
described by the van Genuchten parametric model
(symbols are summarized under Notation):

<0

for

(23-1)

1 m 2

----

m
1 1 se

(23-2)

with the effective saturation


s sr
r
s e = -------------- = ---------------- .
ss sr
s r

(23-3)

Any hysteresis loop is confined by the main drying and


wetting retention curves which can be expressed
according to (23-1). The s- retention relation s ( ) is
d

OPKO d=b~

1
--2

for

denoted by s ( ) for the main drying curve and s ( )


for the main wetting curve. The main hysteresis loop is
then
described
by
the
parameter
vector
d d
d d
d
( s s , s r , , n , m ) for the main drying curve and
w

( s s , s r , , n , m ) for the main wetting curve. The

hysteresis loop is closed at residual and maximum


d
w
d
w
saturation, i.e., s r = s r = s r and s s = s s = s s .
cbcilt==ORT

OPK=e=c==~=r~~=p=`
For the prediction of the scanning curves in the
s ( ) relation Scotts model4 is used, in which the drying scanning curves are scaled from the main drying
curve and wetting scanning curves from the main wetting curves so as to pass through a reversal point
(Figure 23.1). If the reversal is from wetting to drying,
the main drying curve is scaled by changing the maximum saturation value for that curve such that the reversal point falls on the resulting curve. Analogously, if
the reversal is in the opposite direction, the main wetting curve is scaled by changing the minimum (i.e.,
residual) saturation value. The scanning curves thus
computed are maintained for all nodes.

OPKP k~=^~
OPKPKN i~==~K=m
A soil column with a length of 0.72 m is considered
The column is filled by a homogeneous sandy material
with a saturated conductivity of 1.19 m/h and porosity
of 0.36. The complete set of material parameters is
given in Table 23.1.
Table 23.1 Material properties
Parameter

Value

porosity

0.36
3.3056 10

saturated conductivity K s , m/s


d

100

residual saturation s r = s r = s r

0.17

maximum saturation
d
w
ss = ss = ss

1.0

capillary head, -y [m]

van Genuchten exponent n of


drying curve
d
d
(m = 1 1 n )

main drying curve


10-1

van Genuchten exponent n


wetting curve
w
w
(m = 1 1 n )

main wetting curve


scanning curves

10

van Genuchten parameter


wetting curve, 1/m
0.2

0.4
0.6
saturation, s [1]

0.8

(0.8095)
5.25

of

4.2
8.4

1.0

Figure 23.1 The main characteristic curves for wetting,


drying, and one set of scanning curves in a hysteresis loop.

ORU==o=j~~=J=m~=ff

5.25

(0.8095)

van Genuchten parameter of


drying curve, 1/m

-2

0.0

of

Initially, most of the column is fully saturated, with the


water table positioned at 0.695 m (25 mm below the
top of the column). At the bottom of the soil column
the pressure head is varied in time. Its boundary

OPKP=k~=^~
curve is listed in Table 23.2.
Table 23.2 Time-dependent pressure head ( t ) at
the bottom of the soil column
Time t
[min]

( t ) a)
[m]

0.695

125

0.070

175

0.070

245

0.420

295

0.420

345

0.170

395

0.170

505

0.720

600

0.720

a)

changes of pressure head between the


time stages are linear.

Results will be compared for five observation points


located at (P1) 0.69 m, (P2) 0.59 m, (P3) 0.49 m, (P4)
0.39 m, and (P5) 0.29 m, measured from the soil bottom. The geometric, initial, and boundary conditions
are summarized in Table 23.3.

Table 23.3 Geometric, initial and boundary


conditions
Value
[m]

Quantity
column height

0.720

initial hydraulic head h = + z ,


z = 0 at bottom

0.695

time-dependent pressure head ( t )


at bottom of soil column

Table
23.2

OPKPKO ^=cbcilt=j
Using FEFLOW the column is discretized by 200
quadrilateral bi-linear finite elements with an element
width of z = 3.6 mm in the vertical direction. A variable time stepping technique is applied that is based on
the FE/BE predictor corrector scheme with an initial
5
4
time step of 10 d and an RMS error criterion of 10 .
The standard form of the Richards equation with a
Picard iteration strategy is used. The simulation control
parameters are summarized in Table 23.4.
Table 23.4 FEM parameters
Parameter
mesh
width

FEFLOW
200 quadrilateral bi-linear
elements
z = 3.6 mm

cbcilt==ORV

OPK=e=c==~=r~~=p=`
Table 23.4 FEM parameters (continued)

time stepping regime


error measure (RMS)
initial time step length

FEFLOW
FE/BE time integration
scheme, automatic time steps
10
10

[d]

0.3

water content [1]

Parameter

0.4

P5

P4
0.2

P3
P2

0.1

final time

0.416667 [d], (10 [h])

P1

0.0

A comparison of the FEFLOW results with the


solutions obtained by Clausnitzer1 gives a good agreement as revealed in Figure 23.2. It is noted that Clausnitzer used a coarser mesh with a width of z = 1 cm
and a fully implicit scheme with variable time steps.
The hysteresis effect on the water content ( t ) for the
present problem can be seen in comparing the solutions
for the hysteretic case (Figure 23.2) with the solutions
for the nonhysteretic case (Figure 23.3). The results for
nonhysteretic conditions are obtained by using the
parameters
of
the
drying
main
curve
d
d
d
d
d
= , n = n , m = m , sr = sr , ss = ss .

10

time [h]

Figure 23.2 History of water content simulated at different vertical locations P1 through P5 in the column for
hysteretic conditions: comparison between FEFLOW and
Clausnitzers1 results.
0.4

P3
0.3

water content [1]

OPKQ o

Clausnitzer
FEFLOW

P5
P2
P4

0.2

P1
0.1

Clausnitzer
FEFLOW
0.0

10

time [h]

Figure 23.3 History of water content simulated at different vertical locations P1 through P5 in the column for
nonhysteretic conditions: comparison between FEFLOW
and Clausnitzers1 results.

OSM==o=j~~=J=m~=ff

OPKR=k~

OPKR k~
h
Kr

L
1

Ks
m

LT
1

m ,m

1
1

n ,n

se
sr
d w
sr , sr

1
1
1

ss
d w
ss , ss

1
1

L
1
L
d

L
1
1

+ z , hydraulic head.
relative hydraulic conductivity
( 0 < K r 1 , K r = 1 if saturated at
s = 1 ).
saturated hydraulic conductivity.
1 1 n curve-fitting parameter
(Mualem assumption).
m -value of the drying and wetting
curve, respectively.
n > 1 , van Genuchten curve-fitting
exponent (pore size distribution
index).
n -value of the drying and wetting
curve, respectively.
saturation ( 0 < s 1 , s = 1 if
medium is saturated).
effective saturation.
residual saturation.
s r -value of the drying and wetting
curve, respectively.
maximum saturation.
s s -value of the drying and wetting
curve, respectively.
vertical coordinate.
van
Genuchten
curve-fitting
parameter.
-value of the drying and wetting
curve, respectively.
porosity ( 0 < 1 ).
s , volumetric moisture content
( 0 < , = if medium is
saturated).

residual
volumetric
moisture
content.
saturated
volumetric
moisture
content.
pressure head ( > 0 saturated
0
medium,
unsaturated
medium).

OPKS o
1. Clausnitzer, V., Beitrag zur Bildung und Verifikation von Parametermodellen der Mehrphasenstrmung in porsen Medien
(Contribution to the development and verification of parametric
models for multi-phase flow in porous media). M. S. Thesis,
Technical University Dresden, Germany, 1991.
2. Kool, J.B. and Parker, J.C., Development and evaluation of
closed-form expressions for hysteretic soil hydraulic properties.
Water Resour. Res. 23 (1987) 1, 105-114.
3. Lenhard, R.J., Parker, J.C. and Kaluarachchi, J.J, Comparing simulated and experimental hysteretic two-phase transient fluid flow
phenomena. Water Resour. Res. 27 (1991) 8, 2113-2124.
4. Scott P.S., Farquhar, G.J. and Kouwen, N., Hysteretic effects on
net infiltration. Advances in Infiltration, Publ. 11-83, Am. Soc.
Agric. Eng., St. Joseph, Mich., 1983, pp. 163-170.

cbcilt==OSN

OPK=e=c==~=r~~=p=`

OSO==o=j~~=J=m~=ff

t~=~=e~=bW=p
q~==~=qJ~=c~Jj~
p

OQKN p~===m

The fracture with a mean aperture of > 350 m divides


the block into two parts. The geometric details of the
fracture plane are shown in Figure 24.2.

Wendland and Himmelsbach3 conducted laboratory


experiments and numerical computations of solute
transport in a fractured sandstone block. The sandstone
block has a length of 24 cm, a width of 21 cm and a
height of 24 cm (Figure 24.1).

40 mm

90 mm

outlet

sealed
areas

240 mm

outlet

35 mm

40 mm

35 mm

t~=~=e~=bW=p=q~==~=qJ~=c~Jj~=p

100 mm

OQ

OQ

240 mm

sealed
areas
ure
fract

injection
100 mm
210 mm
120
mm
420
mm

injection
120
mm

210

mm

Figure 24.1 Schematic representation of the fractured


sandstone block.

Figure 24.2 Geometry of the fracture plane.

Water is pumped from below flowing upwards at a constant rate of Q = 4.57 ml/h through the fracture plane.
cbcilt==OSP

OQK= t~= ~= e~= bW= p= q~= = ~= qJJ


~=c~Jj~=p
A multi-tracer experiment with pyranine and cadmium
as solutes was performed. The tracer is injected in the
fracture at the bottom and the tracer breakthrough is
observed at the outlet on the top of the fracture (Figures
24.1 and 24.2).

q ( D d + T q )I + ( L T ) q-----------porous matrix
q

D =
(24-5)
q D I + q-----------fracture
L q
d

and

OQKO d=b~

=
1

porous matrix
fracture

(24-6)

The porous and fracture flow is assumed to be at


steady-state. Accordingly, the conservation of mass and
momentum can be described by the following
equations1 (symbols are summarized under Notation):

v
q =
v

porous matrix
fracture

(24-7)

q = Q

(24-1)

q = Kh

(24-2)

with
K

K = b 2 g
------------ I
12

Darcy law in the porous matrix


Hagen-Poiseuille law in the fracture

(24-3)

The solute transport of a conservative tracer in the


porous matrix and the fracture is governed by the
advection-dispersion equation (solute mass balance) in
the following form:
C
------- + q C ( D C ) = Q c
t

with the tensor of hydrodynamic dispersion

OSQ==o=j~~=J=m~=ff

(24-4)

OQKP k~=^~
In the tracer experiment the injection of the solutes
is considered as a pulse directly into the fracture. The
injection of the total tracer mass of 32.2 g lasted less
than 1 min. For modeling purposes, Wendland and
Himmelsbach3 smoothed the pulse injection over a
time interval of 6 min, which is still small relative to
the duration of the tracer experiment.
Geometric, initial and boundary conditions for the
present simulation are summarized in Table 24.1.
Material parameters are listed in Table 24.2.

OQKP=k~=^~
Table 24.2 Material parameters
Table 24.1 Geometric, initial and boundary
condition

Quantity
flow:

Quantity

Value

height of sandstone block

0.240 m

fluid density

length of sandstone block

0.240 m

dynamic viscosity

width of sandstone block

0.210 m

hydraulic fracture aperture b

507 m

isotropic matrix conductivity


K

geometry:

2.535 10

fracture thickness of the


half-space

matrix porosity
m

reference head at outlet point

diffusion in the fracture D d


12

4.57 ml/h
5
= 5.484 10
3 1
m d )
h = 0

impervious outer boundaries, except injection and outlet


points
transient solute transport:
initial solute concentration
pulse injection of tracer mass
(input concentration over a
time period of 6 min)
gradient-type condition at
outlet

diffusion in the matrix D d


dispersivities in the matrix

(Q

0 mg/l
32.2 g
(70.46 mg/l)

C
------- = 0
n

1 10

ms

10 kgm
1.3 10

3
1 1

kgm s

transport:

steady-state flow:
flux Q at injection point
(discharge in the half-space)

Value

longitudinal dispersivity in
the fracture L

0.085
5 10

11

2 1

m s

L = T = 0
2 1

0m s

6 mm

Wendland and Himmelsbach3 simulated the tracer


experiment by using the SICK100 simulator. The sandstone block was discretized into 8668 three-dimensional elements for the porous matrix and 435 twodimensional elements for the plane fracture. They used
a symmetric streamline stabilization technique and an
implicit time stepping with 2000 constant time steps.
The spatial discretization is largely similar to that of
Wendland and Himmelsbachs mesh in that, considering the expected concentration profile in the matrix,
logarithmic grid spacing is used. The first nodal row is
5
located at a distance of 2 10 m from and parallel to
the fracture interface. The subsequent nodes are at dis5
4
3
tances of 5 10 m, 1.4 10 m, and 4 10 m. All
cbcilt==OSR

OQK= t~= ~= e~= bW= p= q~= = ~= qJJ


~=c~Jj~=p
further nodes parallel to the vertical fracture are located
at a constant horizontal distance of 1 cm. The resulting
finite-element mesh of the entire sandstone block is
shown in Figure 24.3. Note that in the FEFLOW simulations only the symmetric half of the domain is considered. This leads to a half-mesh consisting of 20160
a)

three-dimensional hexahedral elements for the porous


matrix and 1904 two-dimensional bi-linear fracture
elements. In order to account for the sealed areas in the
fracture plane (Figure 24.2), the corresponding twodimensional elements of the fracture were deleted from
the mesh.
b)

fracture

fracture

Figure 24.3 Finite-element mesh of the sandstone block with a vertical fracture: a) view of the entire block, b)
zoomed mesh at the fracture.

In the FEFLOW simulation a variable time stepping


technique is applied which is based on the FE/BE predictor corrector scheme with an initial time step of
5
4
10 d and a RMS error criterion of 10 . The simulations are performed for a period of 660 min, which
required 157 variable time steps. Similar to Wendland
and Himmelsbach3 a Petrov-Galerkin least-square
(PGLS) upwind technique2 is used to stabilize the
numerical solution.
OSS==o=j~~=J=m~=ff

OQKQ o
Figure 24.4 illustrates the resulting flow field and
head distribution in the fracture-matrix system. The
computed distribution of solutes within the fracture at
the final time of 660 min is shown in Figure 24.5.
Wendland and Himmelsbachs3 results are displayed in
Figure 24.6.

OQKQ=o

Figure 24.4 Computed pathlines in the fracture and head


distributions in the contacted sandstone (half-space view).

Figure 24.5 FEFLOW results of solute distribution within


the fracture at final simulation time t = 660 min.

Figure 24.6 Wendland and Himmelsbachs3 results of solute distribution within the
fracture at final simulation time t = 660 min.

cbcilt==OST

OQK= t~= ~= e~= bW= p= q~= = ~= qJJ


~=c~Jj~=p
1.75

1.50

FEFLOW
measured

1.25

concentration [mg/l]

A comparison of the FEFLOW results with solution


given by Wendland and Himmelsbach3 reveals differences in the solute concentration at the sealed areas and
near the outlet of the fracture. We note, however, that
the magnitudes of solute concentrations are in good
agreement (the same concentration levels are used both
in Figure 24.5 and Figure 24.6). Perhaps more significant are the results of the breakthrough behavior at the
outlet shown in Figures 24.7 and 24.8. The agreement
with Wendland and Himmelsbachs3 measurements is
quite well. Wendland and Himmelsbach3 obtained a
higher peak concentration in their simulations compared to the measurements (Figure 24.8) and the
FEFLOW simulation (Figure 24.7). Obviously, the solute diffusion into the matrix and its accurate numerical
representation in the three-dimensional fracture-matrix
system is of high importance. The better agreement of
the FEFLOW results can result from the more refined
spatial resolution.

1.00

0.75

0.50

0.25

0.00

60

120

180

240

300

360

420

480

540

600

660

time [min]

Figure 24.7 Breakthrough curve at the outlet: FEFLOW


results compared to the measurements3.

Figure 24.8 Measured and simulated breakthrough curves


at the outlet obtained by Wendland and Himmelsbach3.

OSU==o=j~~=J=m~=ff

OQKR=k~

OQKR k~
b
C
D
Dd
g
h
I
K
K
Q
Qc
q
v
L, T

L
3
ML
2 1
L T
2 1
L T
2
LT
L
1
1
LT
1
LT
1
T
3 1
ML T
1
LT
1
LT
L

1
1
3
ML
1 1
ML T

aperture of fracture.
solute concentration.
tensor of hydrodynamic dispersion.
molecular diffusion.
gravitational acceleration.
hydraulic head.
unit (identity) tensor.
tensor of hydraulic conductivity.
generalized conductivity.
fluid mass sink(-)/source(+) term.
solute mass sink(-)/source(+) term.
specific flux.
fluid velocity vector.
longitudinal and transverse dispersivity, respectively.
porosity of porous matrix.
generalized void space.
fluid density.
dynamic viscosity of fluid.

OQKS o
1. Diersch, H.-J.G. (2002) Discrete feature modeling of flow, mass
and heat transport processes by using FEFLOW. White Papers Vol. I, Chapter 9, WASY Ltd., 147-190.
2. Diersch, H.-J.G. (2002) The Petrov-Galerkin least square method
(PGLS). White Papers - Vol. I, Chapter 13, WASY Ltd., 219-262.
3. Wendland, E. and Himmelsbach, T., Transport simulation with
stochastic aperture for a single fracture - comparison with a laboratory experiment. Advances in Water Resources 25 (2002) 1, 1932.

cbcilt==OSV

OQK= t~= ~= e~= bW= p= q~= = ~= qJJ


~=c~Jj~=p

OTM==o=j~~=J=m~=ff

qJ~= ^= ~= m~=
d~=t

ORKN p~==m
Three-dimensional anisotropy in layered aquifers
can lead to somewhat unusual flow patterns, even in
steady-state flow situations. This was firstly discovered
and reported by Hemker et al.4, who termed this type of
flow pattern as groundwater whirls. The cause and
appearance of groundwater whirls has been thoroughly
discussed and studied using both numerical and analytical solution techniques1-7. Hemker et al.4 and recently
Hemker and Bakker5,6 introduced prototypical model
problems for groundwater whirls, which can also be
used as benchmark test for 3D anisotropic aquifer systems. We chose the following two examples from
Hemkers studies.

the center of the box is 150 m long, 20 m wide, and 20


m thick. Inside the anisotropic block, the major principal directions of the horizontal hydraulic conductivity
tensor are orthogonal. The general flow direction is
lengthwise (straight north) and makes an angle of 45
with either of these directions (see Figure 25.1).
25 m
25 m

ORKNKN b~= NW= p= j=


qJi~=`=^

20 m

25 m

25 m

20 m

70 m

Hemker et al.4 considered a box-shaped, two-layer,


homogeneous aquifer. The model box is 200 m long,
70 m wide, and 20 m thick, and consists of only two
layers, each 10 m thick. An anisotropic block located at

150 m

qJ~=^=~=m~==d~=t

200 m

OR

OR

20 m
general flow direction
anisotropy in top layer
anisotropy in bottom layer

Figure 25.1 The two-layer anisotropic model problem, example 1 (modified from4).

cbcilt==OTN

ORK=qJ~=^=~=m~==d~=t
The hydraulic conductivity of the isotropic outer
5
area is 1 m/d ( 1.15741 10 m/s). Within the anisotropic block, the major principal value of the horizontal
conductivity tensor is Kmax = 1 m/d , and the minor
principal value is K min = 0.1 m/d , so that the anisotropy ratio is K max K min = 10 . The major principal
directions of the horizontal hydraulic conductivity in
the two layers are perpendicular to each other - southeast-northwest in the upper layer, and southwest-northeast in the lower layer, which is referred as a crosswise
anisotropy. The vertical hydraulic conductivity in the
isotropic and anisotropic parts is Kv = 1 m/d .
The flow in the aquifer is fully confined. The western and eastern sides are no-flow boundaries, while the
short southern and northern sides are open boundaries
with fixed potentiometric heads that differ by 1 m. It is
noted that the specific value of the gradient has an

effect on the values of the potentiometric contours and


velocity only, and does not affect the pattern of potentiometric contours and flow lines4.

ORKNKO b~= OW= j= = kJ


i~=o~=a=^
Hemker and Bakker5 studied a 18 m thick confined
prototypical aquifer at steady state, consisting of nine
equally thick layers. The horizontal hydraulic conductivity is heterogeneous in a 100 m wide section of the
model; each layer in this section is divided in 10 strips
of equal width. A cross-section perpendicular to these
strips shows a regular pattern of 9-by-10 cells, where
each cell is 10 m wide and 2 m high (Figure 25.2). On
each side of this central zone a 100 m wide homogeneous block serves to reduce boundary effects.

300 m

K2
layer 1

K1

Kz

18 m

layer 9

100 m

100 m

100 m

Figure 25.2 A stratified confined aquifer with a laterally heterogeneous anisotropic central zone (taken from5).

OTO==o=j~~=J=m~=ff

ORKO=k~=^~
The major and minor principal values of the horizontal hydraulic conductivity tensor are 10 m/d and 5
m/d in the entire model. The vertical hydraulic conductivity is 1 m/d in all layers. The general flow direction
is in the direction of the strips (straight north). In the
presented model the major principal direction of the
horizontal hydraulic conductivity tensor is also chosen
straight north in the two large side blocks, while it varies between -45 (N45W, northwest) and 45 (N45E,
northeast) in the 90 cells of the central zone. To obtain
a two-dimensional spatial distribution, ten uniformly

distributed anisotropy directions (-angle, see


Figure 25.2) were chosen (-45, -35, -25, , 45)
and for each layer these ten values were randomly
assigned to the cells. The resulting distribution is given
in Table 25.1.
The model size is 300 by 300 m by 18 m. The east
and west sides are no-flow boundaries, while the south
and north sides are open boundaries with fixed potentiometric heads that differ by 0.3 m in all layers5.

Table 25.1 Principal directions of horizontal anisotropy in all 9 by 12 cells of the model (angle in )
0
0
0
0
0
0
0
0
0

-35
25
-35
-35
15
-45
25
35
45

5
35
25
35
-5
-35
-45
5
-25

15
-35
-15
5
25
25
-5
-15
-35

35
15
-25
-25
-45
-25
35
-35
-5

-15
-45
-5
15
35
5
-25
25
-15

ORKO k~=^~
ORKOKN b~=N
Hemker et al.4 analyzed the steady-state problem by
using the finite-element code MicroFEM3. They
applied a 3D finite element mesh consisting of 2470
nodes and 4800 triangular elements in each layer. The
nodal distance decreases from 4 m at the model boundaries to 2 m inside the anisotropic block.

-25
-5
15
-5
45
-15
5
-45
15

-5
-25
5
45
5
15
15
-25
5

25
5
-45
25
-35
35
-15
15
-45

-45
-15
45
-15
-15
-5
45
45
35

45
45
35
-45
-25
45
-35
-5
25

0
0
0
0
0
0
0
0
0

For the present FEFLOW simulations we prefer a


refined 3D mesh of hexahedral brick elements, where
the vertical cross-section of the two-layer aquifer problem coincides with the x 1 x 2 -plane, while the x 3 coordinate coincides with the horizontal north direction. For such an coordinate orientation the Eulerian
anisotropic angles have to be defined for the general
rotation matrix of anisotropy according to Eq. (5-11),
viz.,
cbcilt==OTP

ORK=qJ~=^=~=m~==d~=t
cos 0 sin
a =
0 1 0
sin 0 cos

at

= 90
= 90

(25-1)

where the rotation is performed about the x 2 -axis by


the angle as the only anisotropic angle to be input for
the present problem at each layer of the inner area of

the model domain. The mesh used in FEFLOW is


shown in Figure 25.3. It consists in total of 664,000
hexahedral elements with 667,521 nodes. In the crosssectional x 1 x 2 -plane the nodal distances are 1.25 m
in the x 1 -direction and 0.5 m in the x 2 -direction for the
isotropic outer area and 0.5 m in both directions for the
anisotropic inner area. The nodal distance along the
horizontal x 3 -direction is chosen always 1 m.

a)

x2

x3

x1

b)

x2
x1
Figure 25.3 FEFLOW mesh used for the example 1 consisting of 664,000 hexahedral elements with 667,521
nodes: a) 3D view, b) cross-sectional view.

OTQ==o=j~~=J=m~=ff

ORKO=k~=^~
The geometric and boundary conditions used for the
FEFLOW simulation of example 1 are summarized in
Table 25.2. Material parameters are listed in Table
25.3.
Table 25.2 Geometric and boundary conditions of
example 1
Quantity

Value

geometry:
length of outer area

200 m

vertical plane is oriented in the x 1 x 2 -coordinate directions (x - y plane)


horizontal plane is oriented in the x 1 x 3 -coordinate
directions (x-z plane)
length of inner area

150 m

width of outer area

70 m

width of inner area

20 m

thickness of each layer

10 m

steady-state flow:
potential difference in north
direction

h = 1 m

Table 25.3 Material parameters for example 1

isotropic conductivity of outer


m
m
m
area, K 1 = K 2 = K 3

Value
1.15741 10

major principal conductivity


of inner area,
m
m
K 1 = K 2 = K max

1.15741 10

ms

minor principal conductivity


m
of inner area, K 3 = K min

1.15741 10

ms

Eulerian angles for the isotropic outer area, = =


for all two layers

Eulerian angle for the


anisotropic inner area for all
two layers

90

Eulerian angle for the


anisotropic inner area for all
two layers

90

Eulerian angle for the


anisotropic inner area of the
upper layer

45

Eulerian angle for the


anisotropic inner area of the
lower layer

45

ORKOKO b~=O

impervious boundaries, except at vertical front and back


faces

Quantity

Table 25.3 Material parameters for example 1

ms

The steady-state problem 2 was computed by


Hemker and Bakker5 with the MicroFEM code3, where
a 3D finite-element mesh of 18 layers was built up, i.e.,
each 2 m thick layer was subdivided by two computational layers. They used a nodal spacing along the horizontal direction of exactly 2 m within the central zone,
which increases to nearly 10 m at the east and west
model boundaries to attain 11,668 nodes per layer
(210,024 nodes in total).
cbcilt==OTR

ORK=qJ~=^=~=m~==d~=t
For the FEFLOW simulations we choose a similar
mesh resolution comparable to Hemker and Bakker5.
Again, it is convenient in FEFLOW to use the vertical
cross-section as the x 1 x 2 -plane so that the horizontal
extent of the aquifer system is oriented to the x 3 -coordinate (north) direction. According to Table 25.1 the
anisotropic angle is only defined around the vertical
axis. In using our coordinate alignments it means - in
the same manner as for example 1 - that the rotation is
done explicitly via the Eulerian angle for the rotation
around the vertical x 2 -axis (see Eq. (25-1)).
The FEFLOW mesh for example 2 is shown in
Figure 25.3. It consists of 259,200 hexahedral elements
with 273,097 nodes. In the cross-sectional x 1 x 2 plane the nodal distances are 2.5 m in the x 1 -direction
and 0.5 m in the x 2 -direction. The nodal distance along
the horizontal x 3 -direction is constantly 5 m.
The geometric and boundary conditions used for the
FEFLOW simulation of example 2 are summarized in
Table 25.4. Material parameters are listed in Table
25.5.

Table 25.4 Geometric and boundary conditions of


example 2 (continued)
width of inner zone

100 m

width of outer zone (left and


right)

100 m

9 layers with a thickness of


each layer

2m

steady-state flow:
potential difference in north
direction

h = 0.3 m

impervious boundaries, except at vertical front and back


faces

Table 25.5 Material parameters for example 2


Quantity
m
K1

Value

for all ele-

1.15741 10

ms

conductivity K 2 for all elements

1.15741 10

ms

conductivity K 3 for all elements

5.78704 10

ms

Eulerian angle for all elements

90

Eulerian angle for all elements

90

vertical plane is oriented in the x 1 x 2 -coordinate directions (x - y plane)

Eulerian angle for elements


of the isotropic outer zone

horizontal plane is oriented in the x 1 x 3 -coordinate


directions (x-z plane)

Eulerian angle for elements


of the anisotropic inner zone

-angle pattern according to Table 25.1

Table 25.4 Geometric and boundary conditions of


example 2
Quantity

Value

geometry:
domain measure

OTS==o=j~~=J=m~=ff

300 m x 300 m x 18 m

conductivity
ments

ORKP=o

a)

b)

x2
x1
Figure 25.4 FEFLOW mesh used for the example 2 consisting of 259,200 hexahedral elements with 273,097
nodes: a) 3D view with drawn -angle pattern of anisotropy, b) cross-sectional view; (exaggeration 3 : 1).

ORKP o
ORKPKN b~=N
We compare the FEFLOW results with the outcome
from Hemker et al.4. Five flow pathlines are shown in
Figure 25.5, all starting in the upper layer at the center

of the southern model boundary, at depths of 1, 3, 5, 7,


and 9 m from the confined top; the pathlines are numbered 1 through 5 from top to bottom. Isochrone markers on the pathlines indicate a residence time interval of
10 years, based on 30% porosity. The computed total
travel times for the five pathlines are compared in
Table 25.6 with the results of Hemker et al.4.
Figure 25.6 displays the pathlines in a 3D view, which
clearly indicate a typical whirl pattern of the flow field.
cbcilt==OTT

ORK=qJ~=^=~=m~==d~=t

a)

b)
1

5 4 3 2

1 3 5
2 4

Figure 25.5 Plan, side, and front view of the two-layer model (example 1) with hydraulic head contours of both layers and five pathlines (1, 2, 3, 4, 5) starting at different levels in the upper layer on the
southern boundary: a) results by Hemker et al.4, b) FEFLOW results.

OTU==o=j~~=J=m~=ff

ORKP=o
times (Figure 25.6), which are obviously caused by the
different mesh resolutions used (note that the FEFLOW
mesh is much more refined than the MicroMFEM
mesh). Particularly, pathlines 3, 4, and 5 increasingly
depart from Hemker et al.s results with progressing
travel time. This has to be expected because these pathlines start at decreasing distance from the interface of
the two layers, where the influence of the strong crosswise anisotropy becomes more dominant and consequently a higher numerical accuracy should be required
there.

Figure 25.6 Whirling flow pattern appearing in the 3D


view of the computed flow pathlines (FEFLOW results).

Table 25.6 Comparison of the computed travel times


for the five pathlines of example 1
Pathline
No.

Hemker et al. reported a total discharge of 6.3 m3/d.


The FEFLOW simulation result a discharge of 6.18
m3/d. Water balance computations for each layer show
that vertical flow components between both layers are
induced. We find an exchanging rate of 3.2 m3/d in
both directions at the interface of the two layers, which
is identical to Hemker et al.s result.

Travel times [years]


Hemker et al.4

FEFLOW

35.2

35.4

40.2

40.6

46.7

50.4

56.6

55.6

57.7

53.2

There is a reasonable quantitative agreement


between Hemker et al.s and the present FEFLOW
results; however, some differences exist in the pathline
characteristics (Figure 25.5) and the computed travel

ORKPKO b~=O
The results reveal a complex whirling flow pattern
due to the anisotropy of the layered aquifer system.
This can be clearly seen in Figure 25.7. Large and
small whirls exist next to each other, rotating in opposite directions. The detailed whirl pattern shows that
several smaller whirls may occur within larger whirls.
The slight extension of the whirls into the homogeneous side blocks is on the order of 20 m. The vertical
exchange of water between the adjacent layers as a
result of the whirling flow can be clearly detected.
A comparison of the FEFLOW results with the
numerical and analytical results presented by Hemker
cbcilt==OTV

ORK=qJ~=^=~=m~==d~=t
and Bakker5 reveal a rather good agreement.
Figure 25.8 displays the computed whirl pattern at a
vertical cross-section. Evidently, MicroFEM and

FEFLOW compute the same results, which also agree


very well with the analytical predictions.

a)

b)

Figure 25.7 Groundwater whirls occurring for example 2 simulated by FEFLOW: a) 3D view of computed flow
pathlines, b) pathlines projected on the vertical cross-section (exaggeration 3 : 1).

OUM==o=j~~=J=m~=ff

ORKP=o

a)

b)

20

15

10

c)

50

100

150

200

250

d)

-2

-4

-6

-8

-1 0

-1 2

-1 4

-1 6

-1 8
50

100

150

200

250

Figure 25.8 Cross-section showing the pattern of groundwater whirls: a) MicroFEM results by Hemker and Bakker5.
Projection of 20 pathlines in each of the 9 layers, pathlines start in the centre of each cell and at equal distances to the
left and right in the homogeneous side blocks. All starting points are located at 100 m north of the southern boundary
and all pathlines run for 100 m due north. Different colors are used for different starting depths. b) FEFLOW results
using the same starting points at 100 m north of the southern boundary, however, pathlines run for 200 m due north. c)
Analytical results of the streamfunction contours in the cross-section. d) FEFLOW results for a refined pathline pattern at the cross-section, pathlines run over the full horizontal distance (300 m) due north.

The same pathlines projected on a side view are


depicted in Figure 25.9 comparing the FEFLOW
results with the MicroFEM results. Slight differences
in the divergent characteristics of the pathlines can be
recognized that are obviously caused by the different
meshes, velocity computations and pathline tracking
techniques used in the different codes. Comparing
FEFLOWs hydraulic head distribution along a cross-

section with the analytical (exact) results presented by


Hemker and Bakker5 we found in Figure 25.10 that
while the profiles agree rather well, the peaks in the
zigzag-profiles of the hydraulic head appear generally
somewhat smaller in the numerical predictions. More
refined 3D meshes seem to be needed to improve the
agreement with the analytical results.
cbcilt==OUN

ORK=qJ~=^=~=m~==d~=t

a)

b)

-5

-10

-15

10

20

30

40

50

60

70

80

90

Figure 25.9 The 9 * 20 pathlines projected on a side view: a) MicroFEM results5 b) FEFLOW results.

OUO==o=j~~=J=m~=ff

100

ORKQ=`

a)

b)

-3

x 10-3

Heads in all layers

x 10

3
layer 1
layer 2
layer 3
layer 4
layer 5
layer 6
layer 7
layer 8
layer 9

Head

-1

-1

-2

-2

-3
50

100

150

200

layer 1
layer 2
layer 3
layer 4
layer 5
layer 6
layer 7
layer 8
layer 9

250

-3

50

100

150

200

250

Figure 25.10 Hydraulic heads in an east-west cross-section of the layered aquifer: a) analytical profiles5, b) present
FEFLOW results.

ORKQ `
Groundwater whirls are linked to an anisotropy in a
fully 3D flow regime. This was firstly discovered and
reported by Hemker et al.4. The present FEFLOW simulations fully confirm their findings4,5. We found good
agreement both with the MicroFEM results and the
analytical solutions.

^
Dr. Kick Hemker generously provided data and

information on the subject matter. We are grateful for


his suggestions and critical remarks.

ORKR k~
a

h
h
K
m
Ki

L
L
1
LT
1
LT

K max, K min LT

rotation matrix (for more see


Chapter 5: Anisotropy, p. 79 and
following.)
hydraulic head.
difference of hydraulic head.
hydraulic conductivity.
hydraulic conductivity along the
principal directions (i = 1,2,3).
major and minor principal value of

cbcilt==OUP

ORK=qJ~=^=~=m~==d~=t
Kv
x 1 , x 2, x 3

, ,

LT
L

hydraulic conductivity, respectively.


vertical hydraulic conductivity.
Cartesian coordinates.
horizontal anisotropy angle.
Eulerian
rotation
angles
of
anisotropy.

ORKS o
1. Bakker, M. and Hemker, K. A Dupuit formulation for flow in layered, anisotropic aquifers. Advances in Water Resources 25
(2002) 7, 747-754.
2. Bakker, M. and Hemker, K. Analytical solutions for groundwater
whirls in box-shaped, layered anisotropic aquifers. Advances in
Water Resources 27 (2004) 11, 1075-1086.
3. Hemker, C.J. and Nijsten, G.J. Groundwater flow modeling using
MicroFem: Version 3. Amsterdam, The Netherlands; Hemker
Geohydroloog Amsterdam, 1996.
4. Hemker, K., van den Berg, E. and Bakker, M. Ground water
whirls. Groundwater 42 (2004) 2, 234-242.
5. Hemker, K. and Bakker, M., Groundwater whirls in heterogeneous and anisotropic layered aquifers. In: FEM_MODFLOW
Intern. Conf. on Finite Element Models, MODFLOW, and More:
Solving Groundwater Problems, ed. K. Kovar et al., Proceedings,
Karlovy Vary (Czech Republic), 2004, 27-30; long text version
on the proceedings CD.
6. Hemker, K. and Bakker, M., Analytical solutions for whirling
groundwater flow in two-dimensional heterogeneous anisotropic
aquifers. Water Resour. Res. 42 (2006), W12419, doi:10.1029/
2006WR004901.
7. Meesters, A.G.C.A., Hemker, C.J. and can den Berg, E.H. An
approximate analytical solution for well flow in anisotropic layered aquifer systems. Journal of Hydrology 296 (2004) 241-253.

OUQ==o=j~~=J=m~=ff

p=f

p=f
^
~==ON
~==OOI=RP
~=TVUSI=OTN
~J~~=URI=UT
J~=UNUR
J~=UMUN
~
=RQ
~=UU
~=NOV
~=QP
obs=E^obsF=OR
=QS
=PM
=PNI=QPI=QPQQI=RR
^=~=NV
~==PPI=OQV

~~=~
~J~~=OQOS
~=NS
~=NR
~=NRNU
=NQ
_^pa==QSQUI=SMI=UQ
~==RTSV
`~=RUI=SQI=SU
a=RUI=SPI=ST
~~=~J=OPR
=~=RV
~=`~=SMI=SRI=SV
~=k~=SMI=SRI=SV
~~=SNI=SRI=SV
k~=RUI=SPI=ST

==RVI=SQI=SU
_=~~=PUPVI=NRVNSMI=OMP
b=PQI=PVI=NSMNSN
==RO
~=OSU
=PVI=NUPI=OMPI=ONTI=ONU
=OON
_=~=p==NOV
~
~=NVR
=PR

~~=~=OQV
`~==~K==OPN
~==PV
~
~=NUI=OM
=NV
=NU
~==~=NV
=NUNV
=NU
~~=NU
~=NU
~==NR
~=SN
~==TM
=PQ
=OS
~=RO
=NT
~~=NR
~=QU
~=OT
~=OM
=POI=UTI=NMN
=~=ON

~=TV

cbcilt==OUR

p=f
=UR
~=OOU
~=RP
~
~~=NP
=NPI=OQ
=~=NP
=NP
~==~=~
~~=~=NS
=NS
=~=NS
=NS
~=~~
~~=OR
=OR
~=OR
=OR
~==NPNR
~=TMTU
~~=TO
~~==NR
=~=PQ

~=PV
=PV
=PV
~~=PVI=NRV
~=NVP
==PT
~=
~=PPI=RN
`~~=OMI=TV
b~=NP

a~
~=NUI=PR
=NSO

OUS==o=j~~

~=OQ
J~=OR
~=QSI=RV
~=NSN
=RO
~
=OM
~=OPI=RO

~~=OT
=RP
=~=OS
=OMI=OSPM
=PS
==OM
=ROI=RP
=OMQOMR
~=~~==OT
~
~~=E~~F=NP

~=RP
~==UQ
=NT
~=RO
~~=NU
J=NT
~=RO
==RP
=OQP
==PTI=SR
~~==OQP
~=TOI=NOM
==ORT
a=~=OQ

b=~==NQ
b==NRV
~=QSI=UU

p=f

~=PM
~=E~F=NT
=~=NS
~==~=OSPP

~=NU
~=OQ
=
~=NOR
b~=~=OTS
b~=~=UM
~~=OPR
~
~=OT
~~=OTI=OV
~==OS
~=~J=OT
~=
=~==PQ
==NP
~==NR

cbcilt
~==RN
c~=~==NT

c~=~=NT
c~=~=NT
c~=~==NT
~=NPVI=OSP
~=~=OSP
==PV
=~=QRQSI=TN
=QS
=QP
c
~=OO
=OOI=RP

d
d~==OOT
~=~=NV
dJe=~=OMP
~=OM
~==NT
~=
=~~=ONP
~=~=QRI=QS

~J=OPI=RO
e~~==OQQ
~=~~=OQI=PM
=RO
=RO
=UT
e
~=ON
=OOI=RP
e==OMP
e=~=e~~==NQV
~==OMI=PRI=PS
~=~=OMI=RO
evaol`lfk=ONP
~==RO
=ORT
==ORT

~=OPRI=OQV
~=QRI=QS
~==RT
==NP
=~=~~=ON

c=OOI=RP
e=ONI=RP
i~=ONI=RP

cbcilt==OUT

p=f
g
g~~=~=UP
=QT

h==NU

i~
=OOI=RP
i~=~=ON
~==~=~=Eij^F=NU
~=QT
~==NOV
~=~~=SS
~~==SN
i==NSP
~==RP
=~=~~==TS

~~=NR
~=~=NS
~=~=RP

~=QT
jcbj=OTP
==PV
=~~=OOU
=~=NS
==QPI=QSQU

~~==PV

l~~=~=_~==NOP

m==PVI=NSPI=ONV
~=PS
=NT
~

OUU==o=j~~

~=OM
=NV
=NVI=OM
=~=NV
~=~=QP
=~=OM
=PMI=RO
~~=PR
=PRI=QS
~=PR
=NTI=PRI=RP
~~=OOU
~=QR
=PR
~==TV
~J=UP
=
J=OP

o~==PVI=NSPI=OON
~=QM
~=QM
=QM
~
=NU
~=OO
=NU
=~=OM
=OP
~=~=NT
~=~==EobsF=NR
~~=OOI=RP
~~=~=RO
o=~==NQ
~=~=TV

~===ONP
~~=OMP
J~=PN

p=f
=TT
~~==NRVI=NVTI=OMR
~~
=OPOI=OPS
=OPO
==PM
=QS

J~=OQP
=OPN
=OPR
=~=ON
==OOU
=RP

=NU

=NU
=NU
~=
=PQ
~~=QT
~==NSQ
p~==NSP
=TO

~=NT
~=NT
=~==OOM

q~==OUI=OMR
~=NUI=OMI=OQI=PMI=RO
~=RP
=OTI=PNI=UTI=NMN
=OQ
q==NNT
=NTI=RP
~
=~==NP

~==NVP
~==RUI=RVI=SOI=SQI=SSI=SUI=SV
~=~=RUI=SQI=SUI=UVI=VMI=VNI=VOI=VPI=VQI=VRI=VSI
VTI=VUI=VVI=NMM
~=ORI=UN
~=~
==PSPU
==PSPU
~=
~J=PV
~==RP

~~===OOT

~=d=J=j~==OPOI=OPS
~=d==OPR
s~~==~K==OQP

~=NQ
=ONR
a~=OM
=OM
~=OM
~=~=NT

~=NTI=PNPP
~J=PO
~J=PO
=OMI=PNI=PS
~==PN
=~==RO
=~=PN
=~==~=~=NT
=NSPI=NSQ

~=~=QPI=QUI=SMI=TMI=TOI=TTI=NNTI=OOTI=OQP
==ORT
=OTN

cbcilt==OUV

p=f

OVM==o=j~~

^=f

^=f
^
^=QMI=NRV
^~~=NVP
^=QM

_~~=NRI=PM
_~=OTN
_~=NOR
_~=OT
_~=NQI=NRI=NTI=NUI=OQI=PMI=NSM
_=QM
_~=PUI=QM
_=PU
_=QM
_=PUI=QM
_=NPMI=NPOI=NPS

`~~=QM
`~~=NNT
`~=OPNI=OPPI=OPQ
`=NVP
`=QM
`~=ORTI=OSM
`~=PUI=QM

a=p=QMI=NRV
a=NRI=NSI=ONI=OOI=ORI=PVI=QMI=QUI=RNI=SRI=NOVI=NPMI
NPOI=NPQI=NPRI=NPSI=NRVI=NVPI=ONRI=OPN

b=QMI=NRVI=NSQI=NSRI=NSSI=NSTI=NTR
b~=NSMI=NSQI=NVP

c=NNUI=NNVI=NOM
c=QMI=NRV

d
d~~=OMUI=OMV
d~=OOTI=OOU
d~=OR
d~=NPVI=NQQI=NQS
d~=OQPI=OQRI=OQS

e~~=NQVI=NRMI=NRN
e~~~=NRI=PN
e=OTN
e=OMPI=OMQ
e=ONPI=ONQI=ONRI=ONTI=ONUI=ONVI=OOM
e~=OSP
e=QM
e=OT
e=QM
e=NQVI=NRMI=NRN
e=QM
e=QM
e=QM
e~=NROI=NRQ

g~=PN
g~=NNT

h~=QM
h~=NOP
h=QR
h=NSI=QMI=NRVI=NVPI=ONR
h=OMQ
h~=NNT

i~=ORT
i=PN

jh=QM
j=PN
j=QU

cbcilt==OVN

^=f
j~=NVP

k=OQP
k=PU
k=NVP
k~=OQP

lp~=QM
l~~=NOR
l=ONTI=ONUI=ONVI=OOMI=OONI=OQV

m~=QM
m=OTI=OUI=OPN
m=OOU
m=NPVI=NQQI=NQS
m=PNI=NNUI=NNVI=NOM
m=ONTI=ONUI=ONVI=OOMI=OONI=OQV

o~=NSMI=NSQ
o=NNT
o=QM
o=QM
o=OQV
o=QM
o=QM
o~~=QM

p~=NOP
p=NT
p=QM
p=QRI=NNUI=NQVI=OMP
p=NVP
p=QM
p=QM
p~=NRVI=NSSI=NSTI=NSUI=NTRI=NVPI=OMUI=OMVI=ONR
p=UP
p~=QM
p~~=NVP

OVO==o=j~~

p~=QM
p=NPMI=NPOI=NPS

q~=NPVI=NQNI=NQQI=NQRI=NQS
q=NNT
q=NOVI=NPMI=NPOI=NPQI=NPRI=NPS
q~=QM
~=d=OPRI=OPSI=OPVI=OQMI=OQN
s~~=OQPI=OQS
s=NRVI=NSSI=NSTI=NSUI=NTRI=NVPI=OMUI=OMVI=ONR
t~~=OT
t=OQVI=ORP
t~=OSP
t=QM

v=QR
v=QMI=ONQ

w=QM
w=NS

You might also like