You are on page 1of 10

METEOROLOGICAL APPLICATIONS

Meteorol. Appl. 16: 513–522 (2009)


Published online 4 June 2009 in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/met.149

A method for deriving a future temporal spectrum of heavy


precipitation on the basis of weather patterns in low
mountain ranges
Johannes Franke* and Christian Bernhofer
Department of Meteorology, Technical University of Dresden, Institute of Hydrology and Meteorology, Germany

ABSTRACT: A weather-pattern-based multiple regression model to derive future possible changes in the level of the
higher temporal resolution spectrum of heavy precipitation has been developed. The temporal spectrum was described
using statistical precipitation amounts as a function of the event’s duration (1–24 h) and return period (once in 5 yr to
once in 100 yr). The principle of the method consists in projecting a statistical relationship between the parameters of a
transformed Gumbel distribution (theoretical extreme value distribution) and the distribution of classes of objective weather
patterns to time slices in the near future of climate. Changes in distribution parameters were constructed in the model from
changes in the distribution of weather patterns. Possible change signals were calculated for the catchment of the Weißeritz
River (Ore Mountains, Germany) for the time slices centred around 2025 (2011–2040) and 2050 (2036–2065) as changes
versus the reanalyses of the reference period 1961–2000 (May–September). For the climate conditions to be expected in
the near future (IPCC A1B scenario), increases in the amounts of heavy precipitation, i.e. decreases in the return periods
of equal amounts of heavy precipitation from the reference period, were obtained for the entire temporal spectrum covered
by this paper. Overall, the change signals derived on the basis of a concept of weather patterns seem plausible because
they represent a possible continuation of the already observed increase in frequency and intensification of events of heavy
precipitation in the extended study area. Copyright  2009 Royal Meteorological Society
KEY WORDS heavy precipitation; extreme value statistics; statistical downscaling; objective weather pattern classification;
climate projection; low mountain range; precipitation-related risk management
Received 7 December 2008; Revised 24 March 2009; Accepted 25 March 2009

1. Introduction management (for agriculture, as well as design issues,


e.g. in the construction of reservoirs) emphasize the need
Studies of recent regional effects of global climate
for updated information on changes of extreme precip-
change for Saxony (Germany) have shown that the gen-
itation. It is necessary to derive statistical probabilities
eral decrease in precipitation amounts during the entire
of rare events of heavy precipitation, both under recent
growing season is combined with an increase in dry
climate conditions and under the climate conditions to be
periods (frequency, duration) and heavy precipitation
expected up to the mid-twenty first century.
(≥20 mm per day: five-fold increase for 1971–2000
versus 1961–1990) (Franke et al., 2004; SMUL, 2008). The objective of this paper is to introduce a method
In addition to being of great importance to agriculture of deriving changes in the time spectrum of local heavy
(drought risk and potential of soil erosion), this rep- precipitation events. This had to be done without a sim-
resents a potential for conflict between flood preven- ulated time series for precipitation in the near climate
tion and drinking-water supply with regard to reser- future: Due to its spatial and temporal heterogeneity, pre-
voir management in the Ore Mountain region and their cipitation is marked by being generally difficult to model
foothills in Saxony (Franke et al., 2006). Climate pro- (Sillmann, 2003). Therefore, the assumption was made
jections for Saxony have shown that this conflict is that the time series of precipitation generated using down-
likely to intensify (Enke et al., 2005a, 2006). The recent scaling methods may involve uncertainties, particularly
frequency of extreme events during summer (e.g. the in mountain ranges. According to Enke (2004) statisti-
flood of the Oder River in 1997 and of the Elbe cal downscaling methods, such as WETTREG (Spekat
River (Becker and Grünwald, 2003) and its tributaries et al., 2007) and STAR (Orlowsky et al., 2007), have
in the eastern Ore Mountains in 2002 (LfUG, 2004)) a damping effect on the simulation of heavy precipi-
and the general great importance for issues of water tation. In contrast, dynamic methods, such as REMO
(Jacob, 2001; Jacob et al., 2008) and CLM (Rockel et al.,
2008), involve particular location-related inaccuracies in
* Correspondence to: Johannes Franke, Department of Meteorology,
Technical University of Dresden, Institute of Hydrology and Meteorol- the mapping of simulated precipitation. An evaluation
ogy, Germany. E-mail: johannes.franke@forst.tu-dresden.de study conducted by Bachner et al. (2008) of the CLM

Copyright  2009 Royal Meteorological Society


514 J. FRANKE AND C. BERNHOFER

regional climate model with regard to precipitation in


summer shows that the model is unable to reflect ade-
quately the spatial patterns of precipitation observed for
Germany and that the model’s quality is generally low in
regions with complex terrain.
Here, an attempt was made to establish a link between
atmospheric circulation patterns and local heavy pre-
cipitation behaviour for recent climate using classes of
weather pattern and theoretical distributions of extreme
values, with the aim to project that link to time slices
in the twenty first century under different climate condi-
tions. Time series with classes of regionally-valid weather
patterns were used as the master variable. These were
derived using ERA40 reanalyses (Uppala et al., 2005)
and ECHAM5/MPI-OM T63L31 simulations (Roeckner
et al., 2003) of meteorological fields using an objective
method (Enke et al., 2005b). The local heavy precipi-
tation behaviour was described via the statistical heavy
precipitation, ρs , being a function of the duration δ and
the return period τ .
The model was conceived to obtain a more robust
derivation of possible shifts within the temporal spectrum
of local heavy precipitation. Furthermore, attention was
paid to ensure that the method may be applied to other
areas. Figure 1. Altitude distribution of the Weißeritz catchment area with
For the catchment area of the Weißeritz River (east- gauging stations, DEM: 100 × 100 m.
ern Ore Mountains), first, the temporal spectrum of
1 ≤ δ ≤ 24 h and 5 ≤ τ ≤ 100 yr, was calculated and parts of the catchment area of the Weißeritz (Mannsfeld
mapped for ρs (δ, τ ) for the reference period 1961–2000 and Richter, 1995). Although originally built to guar-
in the growing season (May–September). Second, this antee minimum runoff, today the purpose of any reser-
was projected to the time slices of 2025 (2011–2040) voirs/dams in the catchment area is flood protection and
and 2050 (2036–2065), using weather pattern changes drinking-water supply. The land cover in the catchment
constructed, for example, assuming the emission scenario comprises 45% forest, 40% agricultural land, and 15%
A1B (IPCC, 2000). For example, the value ρs (24 h, sealed area (e.g. urban areas and traffic routes), with the
100 yr) is the amount of rain over a duration of 24 h that sealed areas giving way to forest as the terrain elevation
occurs or is exceeded once every 100 yr from a statistical increases (LfUG, 2004).
point of view.
2.2. Data
2. Case study region and data 2.2.1. Precipitation data
2.1. Catchment area of the Weißeritz river
The measured daily amounts of precipitation ρ(t) were
The catchment area of the Weißeritz River covers approx- obtained from the gauging network records of the Ger-
imately 390 km2 . It extends along the northern slope of man Weather Service. These time series were subject to
the Ore Mountains (Saxony, Germany). There is approx- a complex checking algorithm for homogeneity in the
imately 800 m height difference between the maximum climate database of Saxony (see Franke et al., 2004).
headwater elevation level of the Rote Weißeritz and Typical homogeneity checks were based on a reference
Wilde Weißeritz rivers at about 900 m above sea level series (e.g. Alexandersson, 1986), with the advantage that
(ASL) and the confluence of the united Weißeritz with potential outliers in the test series were not removed from
the Elbe near Dresden at 100 m ASL (Figure 1). The the data sample if they were simultaneously registered in
morphology of the north-west-directed catchment area is the reference series. With regard to availability and homo-
characterized by relatively flat elevated areas and deep, geneity, in this work precipitation data from 13 gauging
narrow and steep valleys providing little natural reten- stations (Figure 1) for the period, t, from 1961–2000
tion of precipitation runoff (Liedtke and Marcinek, 2002; (May–September) were used. Equation (1) (in Dyck and
LfUG, 2004). Because the Ore Mountains extend from Peschke, 1995) was applied to the consistent description
southwest to northeast, troughs located in the east of of the temporally higher-resolution precipitation spectrum
Germany are particularly conducive to heavy precipita- to derive the amounts of precipitation ρ(δ, t) for lower
tion (LfUG, 2004). In addition, the frequency of thun- duration levels (1–24 h). The reason for this was the lack
derstorms is already known to have increased for large of higher-resolution precipitation data. Equation (1) uses

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met
FUTURE TEMPORAL SPECTRUM OF HEAVY PRECIPITATION 515

a 13% offset for ρ(t) to correct the systematic underesti- dependent correction fields were taken into considera-
mation of ρ(δ = 24 h, t) by ρ(t, recorded at 0730–0730 tion for the predictors (the difference between ERA40
CET (= UTC + 2 h)) (DVWK, 1991; Dyck and Peschke, reanalysis and control run ECHAM5/MPI-OM T63L31
1995): for 1961–2000) during screening. A recognition rate of
 0.25 80% of the given weather (precipitation) classes was
δ reached with the screening process (Enke et al., 2005b).
ρ(δ, t) = 1.13 ρ(t) × (mm) (1)
1440 The construction of the daily weather pattern time series
for the projection period 2001–2065 was based on a
with, δ, duration (in minutes) screening discriminant analysis (cluster method). Using
a measure of distance (the root mean square distance,
2.2.2. Weather pattern data RMSD), this analysis was used to look among the pre-
The time series of daily weather patterns available for dictor fields simulated for each model day for each grid
the period 1961–2065 was derived using an objective point, for the combination with the highest similarity (i.e.
classification of weather patterns in accordance with min(RMSD)) to the seasonally selected composites for
Enke et al. (2005b). The principle of the method consists the past. The weather pattern time series used for the
in approximating empirically given weather classes by projection period was based on the output of the global
a combination of meteorological predictor fields from circulation model ECHAM5/MPI-OM T63L31 (Roeck-
the ERA40 reanalyses (1961–2000) and recognizing the ner et al., 2003), assuming the emission scenario A1B.
field combinations found in the predictor fields simulated
using the global circulation model ECHAM5/MPI-OM
T63L31 (2011–2065). 3. Method
For seasons, the daily weather classes are derived from
the range of spatially averaged measuring data for the In principle, the assumption was made that local (heavy)
temperature or the precipitation (optimization elements). precipitation conditions were determined by atmospheric
In this paper the weather pattern time series optimized circulation patterns (weather patterns) and their interac-
with regard to areal precipitation for Saxony (Germany), tion with local conditions (e.g. topography), which varies
divided into eight classes, was used. By clearly assigning depending on the season and is due to physical rea-
a weather (precipitation) class to each day of the reanal- sons. In this context, Kotlarski et al. (2004) and Werner
ysis period (1961–2000) it was possible to calculate the et al. (2008) proved the relationship between weather pat-
mean meteorological field types (predictors, e.g. relative terns and precipitation events in the extended study area
humidity at the 850 hPa level) from the ERA40 reanal- for the recent past climate. To increase the robustness
yses (Uppala et al., 2005) for individual seasons at each of a possible change signal, the temporal spectrum of
point of an equidistant, geo-referenced grid. The combi- local heavy precipitation behaviour was used (via station-
nation (or the composite) of mean predictor fields that related statistics of extreme values) as statistical heavy
best approximate the seasonal precipitation conditions precipitation ρs in (mm) depending on the duration δ in
for each given weather class was selected by screen- (h) and the return period τ in (yr). The developed statis-
ing. In order to minimize systematic errors, seasonally tical model is represented schematically in Figure 2.

Figure 2. Scheme of the model of statistical heavy precipitation.

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met
516 J. FRANKE AND C. BERNHOFER

3.1. Calculation of the past temporal heavy 3.2. Derivation of future level shifts in the temporal
precipitation spectrum heavy precipitation spectrum
The calculation of statistically extreme values of tem- The possible changes in the temporal spectrum of
poral precipitation variability is generally achieved by heavy precipitation were derived on the basis of a
adapting a theoretical extreme value distribution to series weather-pattern-based projection of a statistical relation-
separated as a function of duration levels (Bartels et al., ship between the distribution parameters ω(δ, t), υ(δ,
2005). Using Equation (1), consistent series ρ(δ, t) were t) and the frequency distribution γ (t) of the classes of
generated for 1–24 h for t from the station-related time weather patterns j  (reanalysis) to future time slices t˜.
series of daily precipitation amounts ρ(t). According to The distribution parameters ω and υ were substituted
Bartels et al. (1997), the calculation of ρs (δ, τ , t) for by for reasons of analogy.
5 yr ≤ τ ≤ 3η yr (according to DVWK (1991): 3η: per- A weather-pattern-based multiple regression model
missible extrapolation range) for a length of measured (Equation (4)) according to von Stroch and Zwiers
time series of η ≥ 20 yr is based on an adapted two- (1999) was built to identify the statistical relationship
parametric Gumbel distribution (Gumbel, 1967). For this  (δ, γ (t)). The purpose of this was to isolate the
purpose, by filtering the annual maximum values i, the weather-pattern-related and the local-scale-related param-
series ρ(δ, t) of each station was reduced to annual eter parts. Equation (4) shows that the regression con-
series ρi (δ, t, κ) and arranged according to size using the stant and the partial distribution parameter are only
counter variable κ(κ = 1, . . . , η − 1, η). The estimated a function of δ (i.e. indirectly dependent on ρ(δ, t).
return periods τ̂i (δ, t, κ), referred to as plotting posi- In an adapted state (Section 3.3), normalization of the
tions (Fuchs, 1983; Grieser et al., 2007), were calculated composite distribution parameters (Equations (4) and (5))
for the related amounts of rain ρi (δ, t, κ) as a function to local (heavy) precipitation characteristics may be
of κ in accordance with Equation (2): achieved. These local normalization variables materially
η + 0.2 affect the areal distribution structure of ρs in the study
τ̂i (δ, t, κ) = (yr) (2) area (see e.g. Figure 5(a)). Based on the assumption of
κ − 0.4
a similarity between the spatial patterns of the recent
with, η, length of measured time series (yr) past climate and a near future climate, the normaliza-
tion variables were assumed to be stationary in order to
According to DVWK (1991), the regression method ensure model-related stability. In this context, possible
is the most suitable for adapting the Gumbel distribu- changes in the level of the local temporal (heavy) precip-
tion to the semi-logarithmic scatter plots [− ln ln (τ̂i (δ, itation spectrum originate from changes in the frequency
t, κ)); ρi (δ, t, κ)]. This leads to the use of the Gum- of occurrence of macro-scale-related circulation patterns
bel distribution transformed into the linear Equation (3) ˜
(weather patterns). Thus, the distribution parameter (δ,
(Bartels et al., 1997). Dyck and Peschke (1995) provide γ (t˜)) projected to t˜ may be constructed from a changed
detailed information on the transformation of the theoret- distribution of weather patterns γ (t˜) using the regression
ical extreme value distribution: model. Equation (5) was derived from Equation (4) for
this purpose:
ρs (δ, τ, t) = υ(δ, t) + ω(δ, t)
 
τ 
m
× − ln ln (mm) (3)
τ −1  (δ, γ (t)) = A (δ, t) + j  (δ, t) × γj  (t) (4)
j  =1
where ρs is statistical heavy precipitation, υ and ω

m
full distribution parameters, δ duration of the heavy ˜
(δ, γ (t˜)) = A (δ, t) + j  (δ, t) × γj˜ (t˜) (5)
precipitation event, τ the return period (yr) and t the
j  =1,j˜=1
observed time slice.
The distribution parameters ω(δ, t) (slope) and υ(δ, t)
(offset) were estimated consistently across all δ levels where is substituted distribution parameters ω and υ,
using the Least Squares Method (see Storm, 1995). In Aφ regression constant, j class of weather pattern (j = 1,
order to avoid discontinuities in the rain amount-duration- . . . , m = 8), j partial distribution parameter, γj partial
frequency lines (see e.g. Figure 5(b)), the distribution relative frequency and t time slice (reanalysed observa-
parameters estimated across all δ values were homog- tion) as well as’ is weather-pattern-based reanalysis and
enized by linear regression based on semi-logarithmic weather-pattern-based projection.
scatter plots [ln(δ); ω(δ, t)] and [ln(δ); υ(δ, t)]. Bar- In this paper, the model’s output for the local scale
tels et al. (1997, 2005) provide detailed information for (Figure 2), was the reanalysis ρs  of the rain amount-
this parameter adjustment. In the model (Figure 2), cur- duration-frequency lines derived from Equation (3) for
rent statistics of extreme values were calculated for the the temporal precipitation spectrum of 1–24 h of return
temporal precipitation spectrum of 1–24 h of the return periods of 5–100 yr and calculated using Equation (6)
periods 5–100 yr on the local scale, using Equation (3) (based on the re-substituted Equation (4)). Based on
based on the measured time series of precipitation. this, the possible changes ρ̃s versus the reanalysis

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met
FUTURE TEMPORAL SPECTRUM OF HEAVY PRECIPITATION 517

were derived for time slices in the near future of cli- part for the gauging station at Zinnwald-Georgenfeld
mate using Equation (7) (based on re-substituted Equa- may be calculated using the following function (re-
tion (5)). Changed return periods τ̃ for equal amounts of substituted): Aυ (δ, t) = 3.6615 × δ 0.2527 (coefficient of
heavy precipitation from the reanalysis may be derived determination r 2 = 0.9993) and Aω (δ, t) = 0.6084 ×
using Equation (8): δ 0.2452 (r 2 = 0.9664). For a consistent description of the
local- and weather-pattern-related parameter parts, they
ρs  (δ, τ, γ (t)) = υ  (δ, γ (t)) + ω (δ, γ (t)) are shown versus the altitude H in Figure 3 as an exam-
  ple for ρs  (24 h, 100 yr, γ (t)). A function in the form
τ
× − ln ln (mm) (6) of y(H ) = a − b × cH was adapted to each of the scat-
τ −1
ter plots (r 2 = 0.82206). Figure 3 shows the general
ρ̃s (δ, τ, γ (t˜)) = υ̃(δ, γ (t˜)) + ω̃(δ, γ (t˜)) dominance of the weather-pattern-related (i.e. circulation-
  caused) effect on the construction of the composite dis-
τ
× − ln ln (mm) (7) tribution parameters versus the local-scale-related effect,
τ −1
with a distinct dependence on altitude. The circulation-
τ̃ (δ, γ (t˜)) caused effect changes proportionally, whereas the local-
  −1 −1 scale-related effect changes inversely proportional to the
υ̃(δ, γ (t˜))− height above sea level. From about 500 m ASL in the
   
  ρ  (δ, τ, γ (t))   study area, the weather-pattern-related effect reaches the
= 1 − exp exp s   (yr)(8)
  ω̃(δ, γ (t˜))   70% asymptote and the local-scale-related effect reaches
the 30% asymptote (Figure 3).

3.3. Adaptation and uncertainty of the projection 3.3.2. Uncertainty


model A model uncertainty was derived by comparing results
obtained directly from precipitation measurements (Equa-
3.3.1. Adaptation tion (3)) with model results (Equation (6), control run).
The temporal and spatial randomness of the occurrence The control run is commonly calculated for a period
of heavy precipitation events requires adaptation of the of the past outside the adaptation period. As a com-
normalization variables (Section 3.2) over the longest promise between the limited availability of reanalysis
possible period of data availability. The adaptation made data and the need for long time series, the adaptation
in this paper over the period 1961–2000 was based on period was divided in this paper into overlapping periods
the availability of re-analyses for the classes of weather of 3 × 20 yr, i.e. 1961–1980, 1971–1990, 1981–2000.
patterns j  and the areal ensemble of precipitation mea- Figure 4 shows uncertainty-duration-frequency lines in
surements (Section 2.2). (%) as areal means for the study area over the three sub-
The distribution parameters υ and ω were used to periods. It reveals that the model underestimated results
substitute the distribution parameter, (below) along the from Equation (3) by 2–7%, with the amount of under-
lines of Section 3.2. estimation depending on δ and τ . As a general rule, the
The weather-pattern-related parameter part is separated model error for individual duration levels is proportional
in turn into (δ, t) by calculating the partial distribution to the return period. The error level within the return peri-
parameter j  (δ, t) (similar to (δ, t) in Section 3.1) and ods is three times larger. The uncertainty for the duration
weighting it with the relative frequency of occurrence
γj  (t) of the classes of weather patterns j  . This is based
on time series of precipitation ρ(t) filtered according to
the temporal occurrence of j  in ρj  (t). The following
must apply to the adaptation period in Equation (4):
 (δ, γ (t)) = (δ, t). Therefore, the local-scale-related
parameter part, A (δ, t), is calculated as a remainder
using Equation (9):


m
A (δ, t) = (δ, t) − j  (δ, t) × γj  (t) (9)
j  =1

substitutes the full distribution parameters υ, ω (from


Equation (3))

In order to describe consistently the local-scale-related


Figure 3. Local-scale-related (dots, dashed curve) and weather-pattern-
parameter part, a power function was adapted to the related (triangles, solid curve) parameter parts (%) as a function of
remainders calculated over the spectrum 1–24 h: y(δ) = altitude in the Weißeritz catchment area, example of ρs  (24 h, 100 yr),
c × δ b . For example, the local-scale-related parameter 1961–2000 (May–September).

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met
518 J. FRANKE AND C. BERNHOFER

4. Results and discussion


Figure 5(a) shows, as an example, the areal distribution
pattern in the study area of ρs  (24 h, 100 yr) (reanalysis),
i.e. a heavy precipitation event of 24 h duration based
on the data for 1961–2000 (May–September), which,
from a statistical point of view occurs or is exceeded
once every 100 yr. In accordance with the topographical
structure (Figure 1) Equation (10) (r 2 = 0.7105) explains
that 71% of the spatial variance (approximately. 100 ≤
ρs  ≤ 125 mm) is due to proportional dependence on the
altitude. In this regard, the effect of the Ore Mountains as
a barrier is associated with a continuous 5.6 mm increase
in amounts of 24 h heavy precipitation per 100 m only
up to an altitude of 550 m ASL (Equation (10)). Based
on the regionalization method applied (Section 3.4),
Figure 4. Model uncertainty-duration-frequency lines (%) of the rain amounts above 125 mm were found for objectively
Weißeritz catchment area, spectrum: 1–24 h and 5–100 yr, area mean. calculated windward areas. By comparison, the return
period for such amounts of precipitation in the Black
Forest (a low mountain range in south-western Germany)
range of 1 ≤ δ < 9 h was proportional to the duration δ, is 10 yr (Kunz, 2003).
whereas for 11 < δ ≤ 24 h it was inversely proportional. For the temporal spectrum of statistically heavy pre-
A reversal of error development takes place within the cipitation (1–24 h for 5–100 yr) covered by this paper,
range of 9 ≤ δ ≤ 11 h. A reason for this behaviour is reanalyses of rain amount-duration-frequency lines are
that by reanalysing middle duration events, tendentious shown as areal means in Figure 5(b) and their areal stan-
uniform distributions (on average) over the eight weather dard deviations are shown in Figure 5(c). The spatial
classes were calculated by the model, in contrast to the variance of statistical heavy precipitation in the study area
events of smaller and larger duration, respectively. This was different, depending on an event’s duration and its
can also be attributed to the spatial generalization degree return period. In this regard, it appears that the orographic
of the objective weather pattern classification. It can be intensification of larger (i.e. events of longer duration)
assumed that a use of fuzzy time series of weather pattern and more rarely occurring events decreases for the benefit
entails a minimization of the model uncertainties (com- of the convective influence if the events become smaller
ments in Section 5). (i.e. of shorter duration) and occur more frequently
(Figure 5(b) and (c)). The reanalyses for the period of
1961–2000 (May–September) varied within the range
3.4. Calculation of areal precipitation in the study area of 23.5 ≤ ρs  ≤ 125 mm (derived from Figure 5(b) and
(c)) and were used in this paper as a reference for pos-
To demonstrate the areal distribution structure of statis- sible changes in the time slices of 2025 and 2050. The
tical heavy precipitation ρs in the study area, an areal time slices are 30 yr averaging periods of 2011–2040
distribution was calculated, as an example, for the reanal- and 2036–2065 for statistical safety, despite the transient
ysis of ρs (24 h, 100 yr) in accordance with Franke et al. character of climate in the twenty first century.
(2008) (Figure 5(a)). The method takes into consideration With regard to future possible changes in the higher
the orographic effects on precipitation, i.e. the formation temporal resolution spectrum of statistical heavy pre-
of a vertical gradient (Equation (10) in the example), as cipitation, positive and non-linear level shifts were cal-
well as the windward and leeward effects, in a manner culated assuming an A1B scenario. In this regard, it
true to scale. The objective calculation of the local-scale seems that the change in signal is generally inversely
proportional to the return period for every event duration
windward and leeward effects, which was achieved by
(Figures 6(a) and 7(a)). Events of heavy precipitation that
combining wind climatology with the geometry of the
are smaller and occur more frequently tend to be subject
terrain, has modified the spatially superposed orographic
to the biggest relative changes versus the reanalysis (ref-
effect (vertical gradient) on a local scale (Franke et al.,
erence) with increases by between c. 1.5% (2025) and c.
2008):
4.5% (2050), whereas comparable changes that are less
pronounced are possible for events of longer duration
   (Figures 6(a) and 7(a)). Lower increases were calculated

 91.4 mm + 5.6 mm × H for events of medium duration. The consideration of the

 100 m
 H :≤ 550 m ASL uncertainty range (2–7%, Figure 4) would lead to a gen-
ρs  (H ) = (mm) eral intensification of the relative change signal.



 The change signal obtained may also be interpreted
 122.2 mm as a changed return period of the rain amount-duration-
H :> 550 m ASL
(10) frequency lines for the reference period (Figure 5(b)).

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met
FUTURE TEMPORAL SPECTRUM OF HEAVY PRECIPITATION 519

(a)

(b) (c)

Figure 5. (a) Areal distribution of ρs  (24 h, 100 yr) (mm) of the Weißeritz catchment area, 1961–2000 (May–September), DEM: 500 × 500 m.
(b) Rain amount-duration-frequency lines (mm) of the Weißeritz catchment area, spectrum: 1–24 h and 5–100 yr, 1961–2000 (May–September),
reanalysed area mean. (c) Standard deviation-duration-frequency lines (mm) of the Weißeritz catchment area, spectrum: 1–24 h and 5–100 yr,
1961–2000 (May–September), reanalysed area mean.

Accordingly, the resulting return periods for equal et al., 2004; Enke et al., 2005a, 2006; SMUL, 2008)
amounts of rain in the near future are shorter in the study is probably accompanied by intensification of primarily
area (Figures 6(b) and 7(b)). For example, it is possible thermally-induced convective heavy precipitation. Thus,
that for the mean amount of rain of a ρs  (24 h, 100 yr) more frequent events of shorter duration will occur and
event of c. 117 mm (Figure 5(b)), the return period may events of longer duration will be associated with addi-
be reduced to c. 91 (2025) and 78 (2050) yr, respec- tional orographic intensification. The latter may become
tively. Under current climate conditions, which reflect an increasingly important if the already observed increases
already ongoing change, such trends in the probabilities in frequency of occurrence of weather patterns con-
of exceeding threshold values have already been identi- ducive to heavy precipitation events of longer duration
fied for Germany (Schönwiese et al., 2005; Trömel and in the study area (e.g. troughs over Central Europe with
Schönwiese, 2007). Vb–paths (see Blüthgen and Weischet, 1980)) continue
The change signal obtained for the temporal spec- in the future.
trum of statistical heavy precipitation covered by this Overall, the change signals obtained indicate a possible
paper may be used to conclude that the continuing gen- non-linear continuation of the already observed increase
eral warming trend in the extended study area (Franke in frequency and intensification of heavy precipitation

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met
520 J. FRANKE AND C. BERNHOFER

(a) (b)

Figure 6. (a) Changed statistical heavy precipitation (%) for 2025 (2011–2040, scenario A1B) versus 1961–2000 (reanalysis) of the Weißeritz
catchment area, spectrum: 1–24 h and 5–100 yr, May–September, projected area mean. (b) Changed return period (yr) for 2025 (2011–2040,
scenario A1B) versus 1961–2000 (reanalysis) of the Weißeritz catchment area, spectrum: 1–24 h and 5–100 yr, May–September, projected
area mean.

(a) (b)

Figure 7. (a) Changed statistical heavy precipitation (%) for 2050 (2036–2065, scenario A1B) versus 1961–2000 (reanalysis) of the Weißeritz
catchment area, spectrum: 1–24 h and 5–100 yr, May–September, projected area mean. (b) Changed return period (yr) for 2050 (2036–2065,
scenario A1B) versus 1961–2000 (reanalysis) of the Weißeritz catchment area, spectrum: 1–24 h and 5–100 yr, May–September, projected
area mean.

events in the extended study area (Grieser and Beck, macro-scale-related) parts in the parameters were sepa-
2002; Franke et al., 2004; Trömel and Schönwiese, 2007; rated for this purpose, and thus the changes in distribution
SMUL, 2008). parameters were constructed from changes in the distribu-
The method was successfully tested for its applicability tion of weather patterns. Because the local-scale-related
to other areas for the catchment of the Zwickauer Mulde parameter parts were assumed to be stationary, it was
(western Ore Mountains, Saxony, Germany) (see Görner ensured that the areal distribution patterns for statistical
et al., 2008). amounts of heavy precipitation in the study area were
nearly time-invariant. As the method conserves local-
scale variability in precipitation patterns, it produces
5. Summary and conclusion results that appear realistic and applicable. The basic
The study yields plausible results deriving future possi- objective of applying the concept of weather patterns
ble changes of heavy precipitation with higher tempo- was to avoid the inaccuracies of simulated time series
ral resolution by a multiple regression model based on for precipitation, particularly in low mountain ranges. In
weather patterns. This was achieved by projecting statisti- this context, it has been shown that the local behaviour
cal relationships between the parameters of a transformed of heavy precipitation may be well described within a
Gumbel distribution of extreme values and distributions 2–7% uncertainty range (for areal means) using region-
of classes of objective weather patterns to time slices ally valid, objective weather patterns. Direct comparison
in the near future derived from global climate mod- of daily precipitation measurements with the reanalysis
els. Local-scale-related and weather-pattern-related (i.e. of classes of weather patterns enabled the conclusion to

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met
FUTURE TEMPORAL SPECTRUM OF HEAVY PRECIPITATION 521

be drawn that the model error for the study area was also Bartels H, Malitz G, Asmus S, Albrecht FM, Dietzer B, Günther T,
Ertel H. 1997. Starkniederschlagshöhen für Deutschland –
caused by a spatial degree of generalization involved in KOSTRA, Department Hydrometeorology of the German Weather
objectively deriving classes of weather patterns. Service: Offenbach; 52.
Future optimization of the regression model based on Becker A, Grünwald U. 2003. Flood risk in central Europe. Science
300: 1099.
weather patterns might deal with the statistical relation- Blüthgen J, Weischet W. 1980. Allgemeine Klimageographie. Gruyter:
ship between the parameters of distributions of extreme Berlin; 887.
values and the distributions of classes of objective DVWK (Deutscher Verband für Wasserwirtschaft und Kulturbau e.V.)
(eds). 1991. Starkniederschläge in der Bundesrepublik Deutschland.
weather patterns as a fuzzy problem. The background for In DVWK-Schriften 97. Verlag Paul Parey: Hamburg; 190.
this is that, firstly, daily weather conditions for a spatial Dyck S, Peschke G. 1995. Grundlagen der Hydrologie. Verlag für
section may be influenced by several classes of weather Bauwesen: Berlin, 536.
Enke W. 2004. WEREX III – Anwendung Eines Verfahrens zur
patterns (Sillmann, 2003) and, secondly, the objective Wetterlagenkonsistenten Projektion von Zeitreihen und Deren
classification of weather patterns according to Enke et al. Extreme mit Hilfe Globaler Klimasimulationen. Final report of the
(2005b) may yield fuzzy allocations to individual weather Meteo-Research Company: Stahnsdorf; 44.
Enke W, Deutschländer T, Schneider F, Küchler W. 2005a. Results
classes per time step (day). of five regional climate studies applying a weather pattern
As far as risk management is concerned, combining based downscaling method to ECHAM4 climate simulations.
current and projected statistics of extreme values allows Meteorologische Zeitschrift 14: 247–257.
Enke W, Schneider F, Deutschländer T. 2005b. A novel scheme to
the option of adapting management regulations in a derive optimized circulation pattern classifications for downscaling
manner that is both promptly and spatially relevant. and forecast purposes. Theoretical and Applied Climatology 82:
Possible applications exist in agriculture, forestry and 51–62.
Enke W, Kreienkamp F, Spekat A. 2006. WEREX IV – Neuentwicklung
water management (e.g. of reservoirs) of catchments in von regional hoch aufgelösten Wetterlagen für den Freistaat Sach-
low mountain ranges. sen und Bereitstellung regionaler Klimaszenarios auf der Basis von
globalen Klimasimulationen mit ECHAM5/MPI-OM T63L31 2010
bis 2100 für die Szenarios B1, A1B und A2 . Final report of the Cli-
mate and Environment Consulting Potsdam GmbH, Potsdam; 79.
Acronyms Franke J, Goldberg V, Eichelmann U, Freydank E, Bernhofer C. 2004.
Statistical analysis of regional climate trends in Saxony, Germany.
h Hour(s) Climate Research 27: 145–150.
yr Year(s) Franke J, Goldberg V, Mellentin U, Bernhofer C. 2006. Risiken
des regionalen Klimawandels in Sachsen, Sachsen-Anhalt und
CET Central European Time Thüringen. Wissenschaftliche Zeitschrift der TU Dresden 55:
CLM Climate Local Model 97–104.
DEM Digital Elevation Model Franke J, Häntzschel J, Goldberg V, Bernhofer C. 2008. Application of
a trigonometric approach to the regionalization of precipitation for
IPCC Intergovernmental Panel on a complex small-scale terrain in a GIS environment. Meteorological
Climate Change Applications 15: 483–490.
REMO Regional Atmosphere Model Fuchs L. 1983. Plotting positions für die extremal-i-verteilung, monte-
carlo-untersuchungen von plotting formeln. Deutsche Gewässer-
STAR Statistical Regional Model kundliche Mitteilungen 27: 21–25.
WETTREG Weather pattern based Görner C, Franke J, Bernhofer C, Hellmuth O. 2008. Climate change
in extreme precipitation events in the Elbe catchment of Saxony.
Regional model In Flood Risk Management – Research and Practice, Samuels P,
ECHAM5/MPI-OM Atmosphere-Ocean coupled Huntington S, Allsop W, Harrop J (eds). Taylor & Francis Group:
T63L31 General Circulation Model London; 309.
Grieser J, Beck C. 2002. Extremniederschläge in Deutschland –
(AOGCM) Zufall oder Zeichen? In: Klimastatusbericht 2002, Deutscher
Wetterdienst: Offenbach; 209.
Grieser J, Staeger T, Schönwiese CD. 2007. Estimates and uncertain-
Acknowledgements ties of return periods of extreme daily precipitation in Germany.
Meteorologische Zeitschrift 16: 553–564.
The authors wish to thank the German Weather Service Gumbel E. 1967. Statistics of Extreme, 4th edn. Columbia University
for the provision of daily precipitation data and the Press: New York; 375.
Climate & Environment Consulting Potsdam GmbH for IPCC (Intergovernmental Panel on Climate Change). 2000. Emissions
Scenarios. Cambridge University Press: Cambridge; 570.
the provision of daily weather pattern data. In the context Jacob D. 2001. A note to the simulation of the annual and inter-annual
of the projects ‘EMTAL’ (grant number: 02 WT 0337) variability of the water budget over the Baltic Sea drainage basin.
Meteorology and Atmospheric Physics 77: 61–73.
and ‘LandCaRe 2020’ (grant number: 01 LS 05104) this Jacob D, Göttel H, Kotlarski S, Lorenz P, Sieck K. 2008. Kli-
study was funded by the German Federal Ministry of maauswirkungen und Anpassung in Deutschland – Phase 1: Erstel-
Education and Research. lung regionaler Klimaszenarien für Deutschland. German Federal
Ministry for the Environment, Nature Conservation and Nuclear
Safety: Dessau; 154.
References Kotlarski S, Demuth S, Uhlenbrook S, Reich T. 2004. Der Zusam-
menhang zwischen atmosphärischer Zirkulation und Niederschlag
Alexandersson H. 1986. A homogeneity test applied to precipitation in Baden-Württemberg. Hydrologie und Wasserbewirtschaftung 6:
data. Journal of Climatology 6: 661–675. 214–225.
Bachner S, Kapala A, Simmer C. 2008. Evaluation of daily Kunz M. 2003. Simulation von Starkniederschlägen mit Langer
precipitation characteristics in the CLM and their sensitivity to Andauer über Mittelgebirgen, Wissenschaftliche Berichte des
parameterizations. Meteorologische Zeitschrift 17: 407–419. Institutes für Meteorologie und Klimaforschung der Universität
Bartels H, Dietzer B, Malitz G, Albrecht FM, Guttenberger J. 2005. Karlsruhe: 31, Karlsruhe; 170.
KOSTRA-DWD – Starkniederschlagshöhen für Deutschland LfUG (Sächsisches Landesamt für Umwelt und Geologie) (eds).
(1951–2000) – KOSTRA. Department Hydrometeorology of the Ger- 2004. Ereignisanalyse – Hochwasser August 2002 in den Osterzge-
man Weather Service: Offenbach; 85. birgsflüssen. Saxoprint: Dresden; 188.

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met
522 J. FRANKE AND C. BERNHOFER

Liedtke H, Marcinek J. 2002. Physische Geographie Deutschlands. Bereitstellung regionaler Klimaszenarios auf der Basis von globalen
Klett-Perthes: Stuttgart; 786. Klimasimulationen mit dem Regionalisierungsmodell WETTREG auf
Mannsfeld K, Richter H. 1995. Naturräume in Sachsen. Forschungen der Basis von globalen Klimasimulationen mit ECHAM5/MPI-OM
zur Deutschen Landeskunde, 238, Trier; 228. T63L31 2010 bis 2100 für die SRES-Szenarios B1, A1B und A2.
Orlowsky B, Gerstengarbe FW, Werner PC. 2007. A resampling German Federal Ministry for the Environment, Nature Conservation
scheme for regional climate simulations and its performance and Nuclear Safety: Dessau; 112.
compared to a dynamical RCM. Theoretical and Applied Climatology Storm R. 1995. Wahrscheinlichkeitsrechnung, Mathematische Statistik
92: 209–223. und Statistische Qualitätskontrolle. Fachbuchverlag Leipzig-Köln:
Rockel B, Will A, Hense A. 2008. The Regional Climate Model Leipzig; 423.
COSMO-CLM (CCLM). Meteorologische Zeitschrift 17: 347–348. Trömel S, Schönwiese CD. 2007. Probability change of extreme
Roeckner E, Baeuml G, Bonaventura L, Brokopf R, Esch M, Giorgetta precipitation observed from 1901 to 2000 in Germany. Theoretical
M, Hagemann S, Kirchner I, Kornblueh L, Manzini E, Rhodin A, and Applied Climatology 87: 29–39.
Schlese U, Schulzweida U, Tompkins A. 2003. The atmospheric Uppala SM, Kallberg PW, Simmons AJ, Andrae U, Bechtold VD,
general circulation model ECHAM5. PART I: Model description. Fiorino M, Gibson JK, Haseler J, Hernandez A, Kelly GA, Li
Report of the Max Planck Institute of Meteorology 349, Hamburg, X, Onogi K, Saarinen S, Sokka N, Allan RP, Andersson E, Arpe
127. K, Balmaseda MA, Beljaars ACM, Berg LVD, Bidlot J, Bormann
Schönwiese CD, Staeger T, Trömel S. 2005. Klimawandel und N, Caires S, Chevallier F, Dethof A, Dragosavac M, Fisher M,
extremereignisse in Deutschland. In: Klimastatusbericht 2005, Fuentes M, Hagemann S, Holm E, Hoskins BJ, Isaksen L, Janssen
Deutscher Wetterdienst (eds). Deutscher Wetterdienst: Offenbach; PAEM, Jenne R, McNally AP, Mahfouf JF, Morcrette JJ, Rayner NA,
191. Saunders RW, Simon P, Sterl A, Trenberth KE, Untch A, Vasiljevic
Sillmann J. 2003. Regionale Klimamodellierung – Wetterlagenklassi- D, Viterbo P, Woollen J. 2005. The ERA-40 re-analysis. Quarterly
fikation auf Basis des globalen Atmosphärenmodells ECHAM . Journal of the Royal Meteorological Society 612: 2961–3012.
Research Report of the Technische Universität Bergakademie von Stroch H, Zwiers FW. 1999. Statistical Analysis in Climate
Freiberg, Freiberg; 78. Research. Cambridge University Press: Cambridge, 484.
SMUL (Sächsisches Staatsministerium für Umwelt und Land- Werner PC, Gerstengarbe FW, Wechsung F. 2008. Großwetterlagen
wirtschaft) (eds). 2008. Sachsen im Klimawandel – Eine Analyse. and precipitation trends in the Elbe river catchment. Meteorologische
Eigenverlag: Dresden; 211. Zeitschrift 17: 61–66.
Spekat A, Enke W, Kreienkamp F. 2007. Neuentwicklung von
regional hoch aufgelösten Wetterlagen für Deutschland und

Copyright  2009 Royal Meteorological Society Meteorol. Appl. 16: 513–522 (2009)
DOI: 10.1002/met

You might also like