You are on page 1of 10

Agricultural Water Management 97 (2010) 91100

Contents lists available at ScienceDirect

Agricultural Water Management


jo u rn a l h omep age : w ww .e ls e v ie r . co m / lo ca te /a g w a t

Simulation of automatic control of an irrigation canal


D. Lozano a, C. Arranja b, M. Rijo b, L. Mateos a,*
a
b

Instituto de Agricultura Sostenible, CSIC, Cordoba, Spain


Departamento de Engenharia Rural, Universidade de E vora, E vora, Portugal

A R T I C L E

I N F O

Article history:
Received 16 October 2008
Accepted 20 August 2009
Available online 19 September 2009
Keywords:
Flexibility of water delivery
On-demand operation
Local upstream control
Distant downstream control
Proportional-Integral controller

AB S T R A C T

Improved water management and efcient investment in the modernization of irrigation schemes are
essential measures in many countries to satisfy the increasing demand for water. Automatic control of
the main canals is one method for increasing the efciency and exibility of irrigation systems. In 2005,
one canal in the irrigation scheme Sector B-XII del Bajo Guadalquivir was monitored. This canal is
representative of irrigation schemes in Southern Spain; it is divided into four pools and supplies an area
of 5154 ha. Ultrasonic sensors and pressure transducers were used to record the gate opening and water
levels at the upstream and downstream ends of each canal pool. Using the recorded data and the SIC
(Simulation of Irrigation Canals) hydraulic model, two canal control options (local upstream control and
distant downstream control) were evaluated using a PI (Proportional-Integral) control algorithm. First,
the SIC model was calibrated and validated under steady-state conditions. Then the proportional and
integral gains of the PI algorithm were calibrated. The controllers were tested using theoretical demand
changes (constant outow followed by a sudden demand increase or decrease) and real demand changes
generated on the basis of a spatially distributed crop water balance that included a number of sources of
variability (random and not random) in the determination of eld irrigation timing and depth. The
results obtained show that only the distant downstream controller was able to adjust quickly and
automatically the canal dynamics to the varying water demands; it achieved this efciently and with few
spills at the canal tail, even when there were sudden and signicant ow variations.
2009 Elsevier B.V. All rights reserved.

1. Introduction
Competition for water between the irrigation sector and the
industrial, urban, recreational, and environmental sectors, and the
need for increasing agricultural water productivity (Comprehensive Assessment of Water Management in Agriculture, 2007) are
challenging irrigation engineers to save water and to provide more
exible water delivery services (Merriam et al., 2007).
In Spain, investment in the modernization of irrigation systems
is signicant (Anonymous, 1998). However, interventions tend to
be focused mainly on the farm irrigation systems and on the
transformation of open channel distribution systems into ondemand pressurized-pipe networks. In most cases, the modernization of conveyance canals has been neglected or received little
technical attention, so these canals remain unchanged since they
were constructed decades ago. Therefore, the bottleneck for a
exible, on-demand service is often at the level of the conveyance
or primary distribution system.

* Corresponding author. Tel.: +34 957 499228; fax: +34 957 499252.
E-mail addresses: ag1mainl@uco.es (L. Mateos).
0378-3774/$ see front matter 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.agwat.2009.08.016

Irrigation canal automation may contribute to introduce


exible water delivery and to save water. Early canal automation
(before the 1950s) was characterized by the use of self-controlled
hydraulic gates. In the 1960s and 1970s, electromechanical
controllers were developed and installed in the US. Thereafter,
local control with programmable logic controllers was implemented (Burt and Piao, 2004). With the advent of personal computers,
unsteady open channel ow simulation models were applied in
combination with control algorithms (Burt and Piao, 2004;
Clemmens et al., 2005). This approach has allowed signicant
advances in the engineering of canal control and automation.
Canal control algorithms can be heuristic, classical, predictive,
or optimal (Malaterre et al., 1998; Ruiz-Carmona et al., 1998). The
most recent studies have returned to classical algorithms of the
Proportional-Integral (PI) type, using new techniques for tuning
the gains of the algorithm (Clemmens and Whalin, 2004; Overloop
et al., 2005; Piao and Burt, 2005; Litrico and Fromion, 2006; Litrico
et al., 2007). Their robustness, accuracy and ease of implementation in the eld have favoured this new trend (Bautista et al., 2006).
However, there is no single solution or recipe applicable to all
problems (Burt and Piao, 2004; Rijo and Arranja, 2005).
The goals of this study were: (1) to calibrate and validate a
hydraulic model that allows the simulation of the actual operation

92

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100

Fig. 1. Canal control logics.

and resulting water ow regime in a real canal; and (2) the


simulation and evaluation of alternative automatic control
methods that may help to shift the operation of irrigation canals
from supply-oriented to demand-oriented operation. For this
study, the hydraulic model Simulation Irrigation Canal (SIC)
(Malaterre and Baume, 1997) was selected, and a study case canal
that is representative of the irrigation schemes in Southern Spain
was selected.
2. Control logics
There are two canal control logics (Burt, 1987; Buyalski et al.,
1991): upstream control (Fig. 1a) and downstream control (Fig. 1b
and c), each referring to the location from which information is
needed by the control logic in relation to the check structure.
Under upstream control, the check structure adjustments are
based on information from upstream (Fig. 1a); thus, the upstream
control is appropriate for canal systems that are supply-oriented.
Under downstream control, the check structure adjustments
are based on information from downstream (Fig. 1b and c). This
control transfers the offtake demands to the upstream water
supply source, thus, it is appropriate for demand-oriented delivery
systems.

Under downstream control, the measured/controlled water


depth may be located at different locations along the pool. If it is
located at the downstream end of the canal pool (Fig. 1c), it will be
called distant downstream control, using the terminology adopted
in Litrico and Fromion (2006). Analogously, if the measured/
controlled water depth is located at the head of the pool, i.e., close
to the controller/check structure (Fig. 1b), it will be called local
downstream control.
Usually, for the three control methods presented above, the
target water depth is the normal depth for the design ow of the
canal pool.
Under upstream control, canals can be sized to convey the
maximum steady ow because the water depth in steady ow
conditions never exceeds the depth for the design ow. The free
surface proles (for varying steady ows) pivot around the
prescribed constant water depth just upstream of the check
structure. A storage wedge between consecutive steady-state ow
proles is created (Fig. 1a represents the free surface proles for
maximum and null steady ows, and therefore the maximum
storage wedge). When ow changes, the water storage volume
must also change in the same sense (increasing or decreasing).
That is why upstream control is particularly effective when
associated with supply-oriented delivery schedules, like rotations
(Clemmens, 1987). However, this method presents disadvantages
when combined with demand-driven-operation because pool
water storage must change opposite to the natural tendency
(Buyalski et al., 1991) and, for this reason, operational water losses
may be signicant.
Local downstream control was the rst control method
developed for demand-oriented-operation. Under this control
method, ow changes originated at the downstream end of the
pool make the storage volume within the pool change in the
opposite sense. The storage wedge responds to the outows
variations rapidly and efciently (Buyalski et al., 1991; Goussard,
1993). However, the canal bench has to be horizontal to
accommodate the null ow surface prole, and canal building
becomes much more expensive and difcult.
Under distant downstream control (as under local upstream
control), when there is a change in pool outow, the tendency for
pool water storage is to change in the opposite sense. To pivot the
water surface on the downstream end, however, pool storage
should change in the same sense (Buyalski et al., 1991). Therefore,
to achieve the required volume changes with distant downstream
control (as with local downstream control), inow must be
changed by a greater amount than outow, until the new steadystate prole is reached.
If changes in water demand can be predicted, the inow can be
changed in advance and the operation becomes more effective and
efcient. As changes in water depth at the end of the pool can be
detected, distant downstream control allows anticipating the
response. For this reason, distant downstream control often is an
option for upgrading traditional upstream control in modernization projects (Rijo and Arranja, 2005).
3. Materials and methods
3.1. Description of the study canal
The canal selected for this study was canal B in the irrigation
scheme Sector B-XII del Bajo Guadalquivir, Lebrija, Spain. It
consists of four pools separate by check sluice gates (named G1, G2,
G3 and G4) (Fig. 2). Pumping stations located just upstream of the
check gates (labelled PS I to PS III in Fig. 2) and at the canal tail (PS
IV in Fig. 2) deliver the water to the farms through pressurizedpipe networks. The canal, entirely concrete lined, is 7.8 km long.
The lengths of the four pools are 1.320, 2.155, 2.170, and 2.144 km,

93

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100


Table 1
Elevation of the weir crests and target water levels.

Fig. 2. Longitudinal sketch of canal B of the Sector B-XII del Bajo Guadalquivir
indicating the lines of the canal bed and bench.

respectively, and their design ows are 5.4, 4.5, 3.35, and
1.98 m3 s1, respectively. The trapezoidal cross-section of the
canal reduces in area downstream of each pumping station. The
slope of the canal walls is 2 horizontal to 1 vertical, the width of the
canal bottom is 1 m and the canal height is 2, 1.85, 1.6, and 1.35 m
at pools one to four (from upstream to downstream), respectively.
In the transition between the pools, the trapezoidal cross-sections
become rectangular (width: 2.5 m; height: 2.4 m). In addition, the
rst 89 m of the rst pool has a rectangular cross-section (also
2.5 m 2.4 m), with a bottom slope of 0.00058. The bottom slope
of the rest of the canal is 0.0002. An inverted siphon is located in
the fourth pool to cross a drainage ditch. Side weirs acting as
spillways are placed just upstream of the check gates. The
elevations of the weir crests are presented in Table 1.
The study canal was constructed during the 1970s. The canal is
operated manually using local upstream control. Two operators
control both the pumps and the canal check gates. The operation
aims to satisfy changes in on-farm water demand. This operation is
cumbersome and inefcient. However, the irrigation scheme is
currently being modernized, so this is an opportunity to consider
canal control alternatives. The possibility of a level canal bank has
been dismissed because the construction cost would be prohibitive. Thus, an evaluation of local upstream control versus distant
downstream control for canal automation is pertinent.
The canal geometry was surveyed in detail using an electronic
total station (GTS-210, TOPCON). Water levels upstream and
downstream of each check gate and downstream of the inverted
siphon were measured using ultrasonic The Probe sensors
manufactured by Milltronics, Siemens Milltronics Process Instruments Inc., Ontario, Canada. Water level upstream of the inverted
siphon was measured with a pressure transducer LMK 309
manufactured by SENSOTEC Instruments S.A., Barcelona, Spain.
The vertical gate opening was measured using UC2000-30GMIUR2-V15 ultrasonic sensors manufactured by Pepperl+Flush
GmbH in Mannheim, Germany. All the water level and gate
opening information was recorded using electronic data loggers.
3.2. Description of the hydraulic model SIC
The Simulation Irrigation Canal model, SIC (Malaterre and
Baume, 1997), is based on Saint-Venant equations. These
equations are nonlinear hyperbolic partial differential equations
dealing with mass and momentum conservation:

@A @Q

0
@t @ x
@Q @Q 2 =A
@h

@t
gA
gAJ 0
@x
@x

(1)

(2)

Pool

Elevation of the weir crest (m)

Target water level (m)

I
II
III
IV

101.55
101.25
100.82
100.15

101.50
101.00
100.64
100.15

where x (m) and t (s) are the distance and time dimensions,
respectively, A (m2) is the area of the ow cross-section, Q (m3 s1)
is the discharge, h (m) is the elevation of the water surface, g
(m s2) is the acceleration due to gravity, and J (m m1) is the
friction slope. In SIC, J is calculated on the basis of Mannings
formula.
Two boundary conditions are necessary for solving this system
of differential equations. Typically, Q(0,t) = Q0(t) is the upstream
boundary condition, where Q0(t) is a known inow hydrograph and
Q(L,t) = QL(t) is the downstream boundary condition, with L being
the length of the canal and QL(t) the discharge hydrograph at the
canal tail (usually determined using a discharge equation function
of the water level at x = L). The initial condition is given by the
water level prole at t = 0: h(x,0).
Eqs. (1) and (2) are not valid for modelling water ow through
hydraulic singularities like cross structures. Therefore, when gates
are present, SIC replaces the two equations for the continuity
equation and a discharge equation of the form Q = Q(hus, hds, W),
with hus (m) and hds (m) being, respectively, the water surface
elevations upstream and downstream of the gates, and W (m2)
being the area of the ow cross-section below the gates. In the case
of a weir, the general form of the discharge equation is Q = Q(hus),
i.e., it assumes free ow, with hus in this case referring to the weir
crest (Baume et al., 2005).
In the SIC model, Eqs. (1) and (2) and the discharge equations
are linearized and discretized in time (Dt, time step) and space (Dx,
space step) using the well known, Preissmann implicit scheme
(Cunge et al., 1980).
The version of SIC used in this study was SIC 3.9.
3.3. Tuning and validation of SIC in the study canal
First, the following gate discharge equation was chosen:
q
Q Cd W

2ghus hds

(3)

where Cd (dimensionless) is the discharge coefcient. Cd values


were taken from Lozano et al. (2009), who calibrated the gates of
the study canal using independent ow measurements. According
to their study, the calibrated Cd values were rather constant at
gates G2, G3 and G4, whereas they varied at gate G1 mainly as a
function of gate opening. However, for the purpose of the
simulation with the SIC model, it was considered that a mean
discharge coefcient was appropriate. With gate openings greater
than 0.2 m and differential heads greater than 0.06 m, the relative
uncertainty of the discharge determinations was less than 15% for
gate G1 and less than 8% for gates G2, G3 and G4 (Lozano et al.,
2009). Thus, the selected Cd values were: 0.87 for gate G1, 0.70 for
G2, 0.66 for G3 and 0.64 for G4.
The calibration and validation of the SIC model under unsteady
ow conditions would require knowledge of the outow hydrographs at the offtakes. Unfortunately, this information was not
available. However, it is considered that if the relevant parameter
(Mannings roughness coefcient, n) is calibrated adequately
under steady ow, the model will also perform well under
unsteady ow conditions (Litrico et al., 2005). Therefore, to obtain
n, 35 days from the 2005 irrigation season were selected along with

94

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100

corresponding periods of steady-state conditions (during which


the outows in the canal at the pumping stations could be
calculated by difference between discharges under consecutive
gates). The calibration was performed for each pool. Then,
by inverse modelling, combined with the calculated ow and
the measured water levels and gate openings, SIC determined n
for each pool. Variations in n throughout the irrigation season,
caused by the growth of algae, were observed (Lozano, 2008).
The accumulation of algae was at the downstream pools more
notable than at the upstream pools. However, for the purpose of
this study it was considered appropriate to use the average n
values determined from multiple calibrations for the periods out of
the algae peak growth. The average n values were 0.016, 0.019,
0.019, and 0.022 for canal pools IIV, respectively. The sensitivity
analysis of the controller to n variations will be the matter of a
separate study.
Flow over the spillways was estimated using the general weir
discharge equation:
2=3
(4)
Q C d lhus
p
2 g Cd is the spillway discharge coefcient, l (m) the width of the
where

weir (9.28 m for the spillway in pool IV, the only pool where spill
ow occurred in the simulations) and hus (m) the elevation of the
water surface in relation to the weir crest.
Next, the SIC model was validated. The data set used for this
purpose was independent of that used for the model calibration. As
in the case of the calibration, the validation had to be restricted to
steady-state conditions, but it was done with two different SIC
modules.
First, the SIC module that simulates steady-state regimes was
validated. Seven steady-state regimes corresponding to seven days
during the 2005 irrigation season were selected. The canal inows
during these periods were between 1.82 and 3.2 m3 s1. The
outows at the pumping stations were calculated by the difference
between discharges under consecutive gates. The water depths
upstream of the gates and at the canal end were input into the
steady-state computations. Therefore, the model was validated by
comparing measured and simulated water depths downstream of
the gates and gate openings.
Second, the SIC module that simulates unsteady-state regimes
was tested. A transitory ow from a known initial steady-state
regime to a known nal steady-state regime was selected. The
actual inow hydrograph was entered into the model, along with
the actual gate movements, and approximated outow hydrographs at the offtakes (estimated by the difference between ows
under consecutive gates). After running the ow variation, the
resulting water depths were compared with those observed in the
canal under the nal steady-state regime.
3.4. Application of the SIC model to the study canal
Once the model was validated, two automatic control methods
were evaluated: distant downstream and local upstream. These
two control methods were selected in view of the actual canal
characteristics and the potential changes that might result from
modernization.
The scenarios used to evaluate the control methods were of two
types. First, the response of the controllers to two theoretical
outow changes was simulated: (1) a sudden increase in outow at
PS IV from 50 to 60% of its capacity; and (2) a sudden decrease in
outow at PS IV from 50 to 40% of its capacity. Secondly, the
response of the controllers to two real demand hydrographs, which
were produced in the canal during two days in May, were
simulated: (1) one week day when demand increased throughout
the day, with a plateau in the early afternoon and another plateau

in the late evening; and (2) one Sunday, when demand typically
decreases throughout the day. Thus, both the theoretical and the
real scenarios included increasing and decreasing demands.
The controller response is totally automatic under distant
downstream control. The control scenarios for distant downstream
control were called DS-In and DS-De, where DS refers to
downstream and In and De refer to the increase and decrease of
demand, respectively. For upstream control, three operation
hypotheses were considered (that are referred to using nomenclature equivalent to that used for distant downstream control):
- Skilful operation: this represents the activities of a very
experienced and skilful operator, who adjusts the gates optimally
and immediately after the demand changes occur. The upstream
control skilful operations are referred to as US-In 1 and US-De 1.
- Delayed-decit ow operation: the operator does not have the
ability to respond immediately, thus the gates are adjusted half
an hour after the changes in demand occur, and the primary
operational criterion is to reduce spill ows, irrespective of
whether the controller is able to maintain the target water levels.
The delayed-decit ow operations under upstream control are
referred to as US-In 2 and US-De 2.
- Delayed-surplus ow operation: as in the previous hypothesis, the
operator does not have the ability to respond immediately, thus
the gates are adjusted half an hour after the changes in demand
occur. However, the primary criterion under delayed-surplus
ow operation is maintaining the target water levels even at the
cost of signicant spill ows. The delayed-surplus ow
operations under upstream control are referred to as US-In 3
and US-De 3. This hypothesis is closest to the actual operation of
the study canal.
In the theoretical scenarios, the initial canal state was dened
by a constant outow at the offtakes, set at 50% of the capacity of
each pumping stations; a constant inow, resulting from adding
the outows at the offtakes; and predetermined gate openings.
To generate the real scenarios, the response of the canal
controller to the demand variations at the pumping stations that
mirrored real water demands was simulated. The model used to
generate the demand hydrographs at the pumping stations was
described in Lozano and Mateos (2008). This model is based on a
soil water balance applicable to each eld. The spatially distributed
crop irrigation requirements are aggregated rst from eld to farm
and then from downstream to upstream in the collective
distribution network. In order to produce realistic simulations,
the model simulates farmers irrigation tactics, strategies and
habits, and their variability. Therefore, irrigation time, duration
and depth vary from eld to eld and from farm to farm. The
hydrant capacity at the farms was set to 16.5 L s1. With this
hydrant capacity, the pipe system does not limit the water
distribution (Lozano and Mateos, 2008). Therefore, this model
resulted in demand hydrographs for each pumping station using an
on-demand delivery schedule.
The three operation hypotheses for local upstream control were
realized by assuming that the operator adjusts the head gate, if
necessary, at time intervals in multiples of one hour and that the
adjustments result in discrete discharge changes of 0.1 m3 s1.
3.5. Control algorithm
In this study, the Proportional-Integral (PI) control algorithm
was used. The PI algorithm is a simplication of the ProportionalIntegral-Derivative (PID) algorithm and is better adapted to canal
stro m and Hagglund, 1995), leading to its application in
control (A
recent canal control investigations (Clemmens and Schuurmans,
2004; Piao and Burt, 2005; Litrico et al., 2005).

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100

The PI algorithm can be expressed in the form:

95

3.6. Canal control performance indicators

DU j U j U j 1 K p e j e j 1] Kie j Dt

(5)

where U(j) and U(j 1) (m) are the gate openings at times j and
j 1, respectively, with time expressed in s, Dt is the regulation
time step, e(j) and e(j 1) (m) are the deviation of the controlled
variable (water level in this case) from its target value at times j
and j 1, respectively, and Kp (dimensionless) and Ki (s1) are the
proportional and integral gains, respectively.
Considering the dynamics of the canal pools (Fig. 1ac), the
target water levels at the downstream end of the pools were set to
the normal water depths for the design ow.
The method for optimizing the gains followed three steps
(Baume et al., 1999):

To assess canal control, three performance indicators were


selected from the list proposed by Clemmens et al. (1998): the
maximum absolute error (MAE), the integral of absolute error (IAE),
and the integrated absolute discharge change (IAQ):
maxjh ht j
j
MAE
(7)
ht
.
.
P
Dt=T N .h j ht .
j0
IAE
(8)
ht
IAQ

X
..
N ..
Q j Q j 1 jQ 0 Q N j
.
.

(9)

j0

(1) Denition of the optimization function. In irrigation canals,


oscillations and large deviations from the target values of the
water levels, and frequent variations in gate opening are
undesirable. Thus, the optimization function for applying the PI
algorithm is based on the integral of the water level deviations
and the integral of the gate opening variations. Therefore,
optimal values of Kp and Ki were found by minimizing the
function:
. .
X
.m
M .Z T ...
m
j
hit ht . dwi dt
(6)
.

where N is the number of time intervals for the test; hj (m) is the
simulated water level at time interval j; ht (m) is the target water
level; Dt (s) is the regulation time step, held constant over T (s), the
time period for the test (24 h); Qj (m3 s1) is the discharge under
the check gate at time j; and Q0 and QN (m3 s1) are, respectively,
the initial and nal discharges under the check gate.
The MAE (dimensionless) quanties the maximum deviation
from the target water depth of the actual water level. The IAE
(dimensionless) indicates how long (and how far) 3the1water level
stays away from the target water level. The IAQ (m s ) relates to

i1

where T (s) is the duration of the optimization period, hi (m)


and hti (m) are the simulated and target water levels (Table 1),
respectively, at pool i, dwi (m) is the gate opening variation, and
the exponent m takes value 1 or 2 if the selected cost function is
linear or quadratic, respectively. Although both values were
tested, the calibration presented in this paper were carried out
for m = 1. Note that the optimization is performed for the canal
as a whole, considering the four pools simultaneously (M = 4).
(2) Selection of a set of perturbations at the offtakes. The perturbations
suggested by the authors of SIC, to obtain robust controllers (P.O.
Malaterre, personal communication, 2003), consist of variations
of T25% of the maximum discharge (i.e., 5075502550% of
Qmax) at 10-h intervals, with the optimization process lasting the
equivalent of 50 h. This was the perturbation scenario used for local
upstream control. However, this scenario was unviable with
distant downstream control because it resulted in dry canal
situations. Thus, the variations of T25% of Qmax were reduced to
variations of T10% of Qmax (i.e., 5060504050% of Qmax).
(3) Multiple simulations of unsteady ow searching for the set of the PI
gains that minimize Eq. (6). The optimization algorithm
implemented in SIC (Baume et al., 1999) is the simplex method
(Nelder and Mead, 1965). One problem is the occurrence of
local minima, thus the solution found may not be an absolute
minimum. This problem is overcome by repeating the
optimization process using as initial values of the gains those
obtained in the previous run. The resulting values of Kp and Ki
are presented in Table 2.

Table 2
Optimized control gains for the PI algorithm.
Gate

G1
G2
G3
G4

Local upstream control

Distant
control

downstream

Kp ()

Ki (s1)

Kp ()

Ki (s1)

2.42
2.60
4.34

785.8
128.7
776.7

3.60
3.64
3.91
1.13

24.63
27.98
12.47
2.11

changes in discharge at the check gates. Sudden discharge


uctuations may cause excessive oscillations in water levels,
which are undesirable. The second term in Eq. (9), jQ 0 Q N j, is
simply the difference between the initial and nal ow rates, and
allows the minimum value of IAQ to be zero.
4. Results
4.1. Model validation
Fig. 3a compares measured and SIC-simulated water depths
downstream of the check gates and upstream and downstream of
the inverted siphon. The simulation results were obtained with the
steady-state regime simulator in SIC, for seven steady-state
periods. The agreement is excellent. As expected in view of the
simulations of the water depths, the observed and simulated gate
openings (Fig. 3b) were also in good agreement.
For the tests of the SIC simulator under unsteady-state
regimes, Fig. 4 shows the observed and simulated water depths
during the nal steady-state regime after the transitory regime
selected. In this case the measured and simulated water levels also
agreed well.
In view of these results and despite the impossibility of
validating the model under unsteady-state conditions, it was
considered that the model had been sufciently tested in the study
canal. Therefore, alternative control methods were investigated
using SIC for unsteady-state conditions.
4.2. Control simulation results under ideal water demand scenarios
Figs. 5 and 6 show the results of the simulation for the local
upstream and the distant downstream controls in response to
the sudden, unexpected outow increase (Fig.
5a)
and
decrease (Fig. 6a) at the pumping station PS IV; these are
referred to as theoretical scenarios. For local upstream control,
Figs. 5b and 6b simply reproduce the operation hypotheses
described in Section 3.4.
In the case of distant downstream control, Figs. 5b and 6b show
the inow hydrograph resulting from the automatic movements
of gate G1. In Fig. 5b, the inow increased for about one hour.

96

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100

This inow increase rapidly overcompensated for the demand


increase, thus increasing the water surface gradient and storage
volume. The inow then decreased continuously until it reached
the inow required for the new steady regime.
The distant downstream control, in response to the decrease in
demand (Fig. 6b), rapidly reduced the initial storage increment and
then established a new water regime adjusted to the new outow.
The control of gate G1 was excellent in both scenarios, DS-In and
DS-De, as indicated by the low value for IAQ (0.56 and 0.32 m3 s1,
respectively) (Table 3).
With respect to returning to the target water depth at the canal
tail (the water level most difcult to maintain), the distant

Fig. 3. Measured and simulated variables for seven independent steady-state ow


periods: (a) water depth downstream of gates G1, G2, G3 and G4, and upstream and
downstream of the siphon; (b) gate opening (gates G1, G2, G3 and G4).

Fig. 4. Measured and simulated water depths during a steady-state regime


simulated using the SIC unsteady-state module.

Fig. 5. Local upstream and distant downstream control responses to a sudden


outow increase at pumping station PS IV for four different operation hypotheses:
(a) pumping station outow, (b) canal inow, (c) water level deviation at the tail of
the canal (end of pool IV), (d) tail weir outow. The operation hypotheses are
dened in Section 3.4.

97

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100

downstream control responded more smoothly than the local


upstream control in both cases (increase and decrease of outow)
for the operation hypotheses (Figs. 5c and 6c). Nevertheless, the
return to the target water levels was similar for US-In 1 and US-In 3
as for DS-In (Fig. 5c). US-De 1 and US-De 3 returned to the target
water levels in a similar way to DS-De (Fig. 6c). The target water
level for delayed-surplus ow operation under upstream control
(US-In 3 and US-De 3) initially deviated from the target by more
than that achieved with skilful operation (Figs. 5c and 6c), but after
about four hours it had returned to the target level. However, the
delayed-decit ow operation hypotheses (US-In 2 and US-De 2)
did not result in a return to the target water depth at the canal tail
(Figs. 5c and 6c, respectively), resulting in MAEs for this control
point equal to 19.5 and 10.6% for US-In 2 and US-De 2, respectively,
and IAEs equal to 7.8 and 4.3%, respectively. These values were
notably higher than those obtained for the same control point
under the other scenarios and operation hypotheses (Table 3).
The difculties described above in maintaining the target water
depth at the tail end when applying delayed-decit ow operation
were not apparent for the rest of the canal pools (I, II, and III) (see
MAE and IAE in Table 3). The distant downstream control also
performed excellently in pools I, II and III, with MAE and IAE values
of less than 1.7 and 0.1%, respectively (Table 3).
However, the inability of the control methods to re-establish
immediately the target water depth caused operational losses
over the spillway in all cases except the US-In 2 operation
hypothesis. The water spills in response to the demand increase
or decrease were highest for the delayed-surplus ow operation
(US-In 3 and US-De 3), followed by the automatic operation
using distant downstream control (DS-In and
DS-De),
then by skilful operation with local upstream control (US-In
1and US-De 1), and lowest for the delayed-decit ow operation
(US-In 2 and US-De 2) (Figs. 5d and 6d).
The water spills were insignicant in the scenario involving
demand increases and relatively small (less than 0.5%) in the
scenario involving demand decreases. However, the hypotheses
considered address a single disturbance during the period of
analysis. In real canal operation, multiple disturbances may
occur in a similar time period, thus the conveyance efciency
could be smaller than indicated by the simple hypotheses as
discussed below.
4.3. Control simulation results under real water demand scenarios
The control response to the real scenarios did not contradict the
results of the theoretical scenarios; rather it revealed some new
and interesting aspects.
Figs. 7a and 8a show the data for real demand, simulated using
the spatially distributed water balance model for a week day and a
Sunday, respectively. On the week day, the demand increased after
6 am, it stabilised around midday and early afternoon, and it

Fig. 6. Local upstream and distant downstream control responses to a sudden


outow reduction at pumping station PS IV for four different operation hypotheses:
(a) pumping station outow, (b) canal inow, (c) water level deviation at the tail of
the canal (end of pool IV), (d) tail weir outow. The operation hypotheses are
dened in Section 3.4.

Table 3
Maximum absolute error (MAE), integral of absolute error (IAE) and integrated absolute discharge change (IAQ) for local upstream and distant downstream controls in
response to an outow increase and an outow decrease at pumping station PS IV. The operation hypotheses are dened in Section 3.4 in the text.
Operation hypothesis

US-In 1
US-In 2
US-In 3
DS-In
US-De 1
US-De 2
US-De 3
DS-De

MAE (%)

IAQ (m3 s1)

IAE (%)

Pool I

Pool II

Pool III

Pool IV

Pool I

Pool II

Pool III

Pool IV

0.2
0.1
0.3
0.9
0.2
0.2
0.2
0.9

0.4
0.2
0.6
0.9
0.7
0.9
0.5
0.6

0.1
0.1
0.2
1.7
0.2
0.2
0.1
0.8

8.2
19.5
13.6
5.4
2.3
10.6
2.6
1.8

0.1
0.1
0.1
0.1
0.0
0.0
0.0
0.1

0.2
0.2
0.2
0.1
0.1
0.1
0.1
0.1

0.0
0.0
0.0
0.1
0.0
0.0
0.0
0.0

0.3
7.9
0.5
0.2
0.1
4.3
0.1
0.1

G1

0.6

0.3

G2

G3

G4

0.5
0.1
0.8
0.5
0.6
1.0
0.2
0.3

0.4
0.1
0.6
0.3
0.5
0.9
0.2
0.1

1.1
0.1
1.3
0.2
0.5
0.9
0.2
0.0

98

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100

Fig. 7. Local upstream and distant downstream control responses to a real scenario
of demand on a week day in May: (a) pumping station outow, (b) canal inow, (c)
water level deviation at the tail of the canal (end of pool IV), (d) tail weir outow.
The operation hypotheses are dened in Section 3.4.

Fig. 8. Local upstream and distant downstream control responses to a real scenario
of demand on a Sunday in May: (a) pumping station outow, (b) canal inow, (c)
water level deviation at the tail of the canal (end of pool IV), (d) tail weir outow.
The operation hypotheses are dened in Section 3.4.

increased again during the late afternoon and early evening. On a


Sunday, demand typically decreases throughout the day. Figs. 7b
and 8b show the canal inow in response to the different operation
hypotheses. The inow hydrograph for the distant downstream
control was generated automatically, being one output of the
control method. The inow hydrographs under local upstream
control simply represent the operation hypotheses dened in
Section 3.4.
Distant downstream control maintained the tail water level
deviation with respect to the tail target level closest to zero (Figs.
7c and 8c). The skilful and the delayed-surplus ow operations also
maintained the tail water level deviation close to zero, but the
uctuations had more amplitude than those resulting from distant
downstream control. This occurred both during the week (Fig. 7c)
and on the Sunday (Fig. 8c). The delayed-decit ow operation
prevented maintenance of the target water level at the canal tail

end, both during the week (Fig. 7c) and on Sunday (Fig. 8c). This
may not be critical in a canal like the one being studied, where the
turnouts are pumping stations. However, in other canals, where
reliable delivery depends on a constant water level, the inability of
US-In 2 and US-De 2 to maintain the target water level would be a
problem. On the other hand, hypotheses involving delayed-decit
ow operation that are slightly more restrictive than the one
dened herein (for instance an operation slightly more delayed on
the week day or an operation with an inow slightly lower than
that used for US-De 2) would empty the canal unless the pumping
was reduced. In this situation, the canal would have failed to
deliver water on-demand.
With respect to the water spills, Figs. 7d and 8d show the
overow hydrograph at the tail weir. As expected, the delayedsurplus ow operation resulted in more water spills than the other
operation hypotheses. The operation hypotheses that generated

99

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100

Table 4
Maximum absolute error (MAE), integral of absolute error (IAE) and integrated absolute discharge change (IAQ) for local upstream and distant downstream controls in
response to a real outow increase (week day in May) and a real outow reduction (Sunday in May) at the canal pumping stations. The operation hypotheses are dened in
Section 3.4 in the text.
Operation hypothesis

US-In 1
US-In 2
US-In 3
DS-In
US-De 1
US-De 2
US-De 3
DS-De

MAE (%)

IAQ (m3 s1)

IAE (%)

Pool I

Pool II

Pool III

Pool IV

Pool I

Pool II

Pool III

Pool IV

0.2
0.2
0.3
0.7
0.3
0.4
0.3
0.4

0.5
0.6
0.6
0.6
0.8
1.1
1.0
0.3

0.2
0.1
0.2
0.2
0.2
0.2
0.2
1.7

6.6
19.8
3.5
2.0
2.1
16.8
2.4
1.5

0.1
0.1
0.1
0.2
0.2
0.1
0.2
0.1

0.2
0.2
0.2
0.1
0.3
0.3
0.3
0.1

0.0
0.0
0.0
0.2
0.1
0.1
0.1
0.3

0.8
3.9
1.2
0.3
0.6
4.7
0.8
0.3

lower spill losses were US-In 2 (no losses) and US-De 2 (0.4%). The
losses under distant downstream control were the lowest among
the operation hypotheses that allowed maintenance of the tail
water near the target level (0.1 and 0.8% for DS-In and DS-De,
respectively). The tail water losses during the simulation period
exceeded 1% of the water pumped in US-In 3, US-De 3 and US-De 1
(with losses of 6, 2.5 and 2%, respectively). Taking into account the
fact that delayed-surplus ow operation is the hypothesis
considered closest to the current operation of the study canal,
the advantage of distant downstream control is also evident when
viewed in terms of conveyance efciency.
The control performances illustrated in Figs. 7 and 8 were
quantied based on the same performance indicators used for the
simulations of theoretical scenarios. The values of these indicators
for the control options compared under real water demand
scenarios can be found in Table 4.
5. Conclusions
The SIC hydraulic model proved to be an extremely useful tool
for assessing canal control alternatives. Considering a demandoriented canal operation, local upstream control and distant
downstream control were the two options to be considered for an
existing, sloping canal divided into four pools. Of the two options,
the distant downstream control was able to adjust the entire canal
dynamics quickly and automatically in response to unexpected,
sudden demand variations and to variations that mirrored real
demands. The spills simulated at the canal tail, unavoidable for
conservative operation under upstream control, were smallest
when applying an operation hypothesis that did not maintain the
water at the canal tail near the target level. The operation losses
under distant downstream control were less than 1% in all
simulations. Estimates of the spill losses under the current
operational regime of the canal are about 6% on week days and
about 2.5% on Sundays, during periods in the irrigation campaign
when water demand is increasing (as it is in May).
Therefore, the tested digital local upstream control would be
an improvement on the current manual upstream control,
but automation could not be extended to the head canal gate,
thus operation would still rely on the skill of the operator. Only
the distant downstream control would guarantee a totally
automatic canal operation with minimal spill losses. In the case
study, the current complement of seven canal operators for the
irrigation scheme could be reduced to only one, and the delivery
service, improved.
Acknowledgements
This study was supported by the research project Control y
Automatizacio n de Canales de Riego, funded by Instituto Andaluz
de Reforma Agraria (Junta de Andaluc a, Spain), and the Integrated

G1

3.1

0.6

G2

G3

G4

3.0
3.0
2.4
3.1
1.4
1.5
1.4
0.5

2.2
2.3
1.7
1.3
0.8
1.0
1.1
0.4

2.7
2.8
2.2
0.6
1.3
1.5
1.5
0.3

Action Automatizacio n y Control de Canales de Riego (HP20040110), funded by the Ministries of Education and Science of Spain
and Portugal. The authors acknowledge the technical assistance of
Mr. Daniel Lozano.
References
Anonymous, 1998. Plan Nacional de Regados, horizonte 2008. Madrid, Spain,
Ministerio de Agricultura, Pesca y Alimentacio n.
A stro m, K.J., Hagglund, T., 1995. PID Controllers: Theory, Design and Tuning, 2nd ed.
Instrument Society of America, Research Triangle Park, North Carolina, US.
Baume, J.P., Malaterre, P.O., Belaud, G., Le Guennec, B., 2005. SIC: a 1D hydrodynamic model for river and irrigation canal modeling and regulation. In:
Vieira da Silva, R.C. (Ed.), Me todos Nume ricos em Recursos Hidricos.Associacao Brasileira de Recursos Hidricos, Coppetec Fundacao, Brasil, pp. 181,
http://www.canari.free.fr/papers/mnhr7_cap1.pdf.
Baume, J.P., Malaterre, P.O., Sau, J., 1999. Tuning of PI controllers for an irrigation
canal using optimization tools. In: Clemmens, A.J., Anderson, S.S. (Eds.), Proceedings of the USCID Workshop, Phoenix, US, pp. 483500.
Bautista, E., Clemmens, A.J., Strand, R.J., 2006. Salt river project canal automation
pilot project: simulation tests. Journal of Irrigation and Drainage Engineering
132 (2), 143152.
Burt, C., 1987. Overview of canal control concepts. In: Zimbelman, D.D. (Ed.),
Planning, Operation, Rehabilitation and Automation of Irrigation Water Delivery Systems. American Society of Civil Engineers, New York, US, pp. 81109.
Burt, C.M., Piao, X., 2004. Advances in PLC-based irrigation canal automation.
Irrigation and Drainage 53, 2937.
Buyalski, C.P., Ehler, D.G., Falvey, H.T., Rogers, D.C., Serfozo, E.A., 1991. Canal
systems automation manual, vol. I. U.S. Department of the Interior, Bureau
of Reclamation, Denver, US.
Clemmens, A.J., 1987. Delivery system schedules and required capacities. In:
Planning, Operation, Rehabilitation and Automation of Irrigation Water
Delivery Systems. American Society of Civil Engineers, New York, US, pp.
1834.
Clemmens, A.J., Bautista, E., Wahlin, B.T., Strand, R.J., 2005. Simulation of automatic
canal control systems. Journal of Irrigation and Drainage Engineering 131 (4),
324335.
Clemmens, A.J., Kacerek, T.F., Grawitz, B., Schuurmans, J., 1998. Test cases for canal
control algorithms. Journal of Irrigation and Drainage Engineering 124 (1), 23
30.
Clemmens, A.J., Schuurmans, J., 2004. Simple optimal downstream feedback canal
controllers: ASCE Test Case Results. Journal of Irrigation and Drainage Engineering 130 (1), 3546.
Clemmens, A.J., Whalin, B.T., 2004. Simple optimal downstream feedback canal
controllers. Theory. Journal of Irrigation and Drainage Engineering 130 (1), 26
34.
Comprehensive Assessment of Water Management in Agriculture, 2007. Water for
Food, Water for Life: A Comprehensive Assessment of Water Management in
Agriculture. International Water Management Institute, London/Earthscan/
Colombo.
Cunge, J.A., Holly, J.F.M., Verwey, A., 1980. Practical Aspects of Computational River
Hydraulics. Pitman Publishing Limited, London, UK, pp. 420.
Goussard, J., 1993. LAutomatisation des Re seaux dIrrigation en Canaux. Commission Internationale des Irrigations et du Drainage (CIID), New Delhi, India.
Litrico, X., Fromion, V., 2006. Tuning of robust distant downstream PI controllers for
an irrigation canal pool. I. Theory. Journal of Irrigation and Drainage Engineering
132 (4), 359368.
Litrico, X., Fromion, V., Baume, J.P., Arranja, C., Rijo, M., 2005. Experimental validation of a methodology to control irrigation canals based on Saint-Venant
equations. Control Engineering Practice 13 (11), 14251437.
Litrico, X., Malaterre, P.O., Baume, J.P., Vion, P.Y., Ribot-Bruno, J., 2007. Automatic
tuning of PI controllers for an irrigation canal pool. Journal of Irrigation and
Drainage Engineering 133 (1), 2737.
Lozano, D., 2008. Modelo para el ana lisis y gestio n de redes de riego. Ph.D. Thesis.
University of Cordoba, Spain, 112 pp.

100

D. Lozano et al. / Agricultural Water Management 97 (2010) 91100

Lozano, D., Mateos, L., 2008. Usefulness and limitations of decision support systems
for improving irrigation scheme management. Agricultural Water Management
95 (4), 409418.
Lozano, D., Mateos, L., Merkley, G., Clemmens, A.J., 2009. Field calibration of
submerged sluice gates in irrigation canals. Journal of Irrigation and Drainage
Engineering, in press.
Malaterre, P.O., Baume, J.P., 1997. SIC 3.0, a simulation model for canal automation
design.In: International Workshop on the Regulation of Irrigation Canals: State
of the Art of Research and Applications, RIC97, Marrakech, Morocco.
Malaterre, P.O., Rogers, D.C., Schuurmans, J., 1998. Classication of canal control
algorithms. Journal of Irrigation and Drainage Engineering 124 (1), 310.
Merriam, J.L., Styles, S.W., Freeman, B.J., 2007. Flexible irrigation systems:
concept, design, and application. Journal of Irrigation and Drainage Engineering
133 (1), 111.

Nelder, J.A., Mead, R., 1965. A simplex method for function minimization. Computer
Journal 7, 308313.
Overloop, van P.J., Schuurmans, J., Brouwer, R., Burt, C.M., 2005. Multiple-model
optimization of proportional integral controllers on canals. Journal of Irrigation
and Drainage Engineering 131 (2), 190196.
Piao, X., Burt, C.M., 2005. Tuning algorithms for automated canal control. ITRC
Report No. R05-005. California Polytechnic State University, San Luis Obispo,
CA, US, pp. 94.
Rijo, M., Arranja, C., 2005. Hydraulic performance of a downstream controlled
irrigation canal equipped with different offtakes types. Agricultural Engineering
International: The CIGR Journal of Scientic Research and Development II
(March).
Ruiz-Carmona, V.M., Clemmens, A.J., Schuurmans, J., 1998. Canal control algorithm
formulations. Journal of Irrigation and Drainage Engineering 124 (1), 3139.

You might also like