You are on page 1of 58

Technical Report No.

26


SIMULATION OF LOW FLOWS AND DROUGHT EVENTS
IN WATCH TEST BASINS:
IMPACT OF DIFFERENT CLIMATE FORCING DATASETS




























Marjolein H.J. van Huijgevoort, Anne F. van Loon, Martin Hanel, Ingjerd Haddeland, Oliver Horvt,
Aristeidis Koutroulis, Andrej Machlica, Graham Weedon, Miriam Fendekov, Ioannis Tsanis,
Henny A.J. van Lanen


18 March 2011

Technical Report No. 26
ii


Technical Report No. 26
ii


WATCH is an Integrated Project Funded by the European Commission under the Sixth Framework
Programme, Global Change and Ecosystems Thematic Priority Area (contract number: 036946).
The WACH project started 01/02/2007 and will continue for 4 years.



Title: Simulation of low flows and drought events in WATCH test
basins: impact of climate forcing datasets
Authors: Marjolein H.J. van Huijgevoort, Anne F. van Loon, Martin Hanel,
Ingjerd Haddeland, Oliver Horvt, Aristeidis Koutroulis,
Andrej Machlica, Graham Weedon, Miriam Fendekov, Ioannis
Tsanis, Henny A.J. van Lanen

Organisations: - Wageningen University - Hydrology and Quantitative Water
Management Group (WUR)
- Comenius University in Bratislava, Faculty of Natural Sciences,
Department of Hydrogeology (UC)
- T.G. Masaryk Water Research Institute, v.v.i. (TGM-WRI)
- Technical University of Crete - Water Resources Management &
Coastal Engineering Laboratory (TUC)
- Norwegian Water Resources and Energy Directorate (NVE)
- Met Office (JCHMR)

Submission date: 18 March 2011

Function: This report is an output from Work Block 4 Extremes: frequency,
severity and scale, and contributes to: (i) Task 4.1.1 Investigate
processes controlling the propagation of drought, and (ii) Task 1.3.4.
Evaluating uncertainty of means and extremes.

Deliverable WATCH deliverables D 4.1.4 Report on the increased understanding
of the propagation of drought in different hydro-climatological regions,
physical catchment structures and different scales, D 4.1.5 Generic
method to quantify the propagation of a drought, and D 1.3.4 Report
on the uncertainty of the global water cycle of the 20th Century. The
technical report contributes to: (i) M4.1.6a Overview of major
historical events; part: drought, and M4.1-7 Analysis of major
historical extreme events with metamodel.



Photos cover:
upper left: Narsj catchment Norway (Van Lanen, 2007),
upper right: Upper-Metuje catchment Czech Republic (Van Loon, 2006),
lower left: Upper-Szava catchment Czech Republic (Van Loon, 2008),
lower right: Nedoery catchment Slovakia (Oosterwijk, 2009).



Technical Report No. 26
iii



Technical Report No. 26
iv
Table of Contents

Table of Contents .................................................................... iv
1. Introduction ......................................................................... 1
Outline ............................................................................................................. 1
2. Catchment descriptions ..................................................... 3
2.1. Narsj ........................................................................................................... 5
2.2. Upper-Metuje ............................................................................................... 5
2.3. Upper-Szava ............................................................................................... 5
2.4. Nedoery ...................................................................................................... 6
2.5. Platis ............................................................................................................. 6
3. Methodology ........................................................................ 7
3.1. Model descriptions ...................................................................................... 7
3.1.1. BILAN ..................................................................................................... 7
3.1.2. FRIER ..................................................................................................... 8
3.1.3. HBV ........................................................................................................ 8
HBV-WUR ........................................................................................................ 9
HBV-NVE ......................................................................................................... 9
HBV-TUC ......................................................................................................... 9
3.2. Drought analysis ....................................................................................... 10
3.3. Large-scale forcing data ........................................................................... 10
3.4. Local forcing data ...................................................................................... 11
4. Differences between forcing datasets ............................ 13
5. Results ............................................................................... 17
5.1. Narsj ......................................................................................................... 17
5.1.1. Modelling .............................................................................................. 17
HBV-WUR model ........................................................................................... 17
HBV- NVE model ........................................................................................... 18
5.1.2. Drought analysis ................................................................................... 18
5.2. Upper-Metuje ............................................................................................. 21
5.2.1. Modelling .............................................................................................. 21
HBV-WUR model ........................................................................................... 21
BILAN model .................................................................................................. 22
5.2.2. Drought analysis ................................................................................... 22
5.3. Upper-Szava ............................................................................................. 25
5.3.1. Modelling .............................................................................................. 25
HBV-WUR model ........................................................................................... 25
BILAN model .................................................................................................. 26
5.3.2. Drought analysis ................................................................................... 26
5.4. Nedoery .................................................................................................... 29
5.4.2. Modelling .............................................................................................. 29
HBV-WUR model ........................................................................................... 29
FRIER model ................................................................................................. 30
BILAN model .................................................................................................. 30
5.4.3. Drought analysis ................................................................................... 31
5.5. Platis ........................................................................................................... 35
5.5.2. Modelling .............................................................................................. 35

Technical Report No. 26
v
HBV-TUC model ............................................................................................ 35
5.5.3. Drought analysis ................................................................................... 36
5.6. Discussion ................................................................................................. 37
6. Conclusions....................................................................... 39
Narsj catchment (Norway) ........................................................................... 39
Upper-Metuje (Czech Republic) ..................................................................... 39
Upper-Szava (Czech Republic) ................................................................... 39
Nedoery (Slovakia) ....................................................................................... 40
Platis (Crete) .................................................................................................. 40
References .............................................................................. 41
List of abbreviations .............................................................. 44

Annex 1 Total monthly precipitation for four test basins;
original and elevation-corrected by the HBV-WUR model .... i

Annex 2 Nash-Sutcliffe values for all models with local
and WFD forcing for all catchments ...................................... iii

Annex 3 Influence of (re)calibration on drought
characteristics .......................................................................... v







Technical Report No. 26
1
1. Introduction
Drought is a natural hazard that occurs all over the world that can have large economic, social and
environmental impacts (Wilhite, 2000). Drought is caused by below-average natural water availability
due to low precipitation and/or high evaporation rates. It is characterized as a deviation from normal
conditions of the physical system (climate and hydrology), which is reflected in variables such as
precipitation, soil moisture, groundwater, and discharge (Tallaksen and van Lanen, 2004).

For drought analysis, time series of hydrometeorological variables are required. These time series
should be long enough to sufficiently capture climate variability. In many catchments around the world,
no or insufficient hydrological and meteorological observations are available. As an alternative, time
series of hydrological data can be simulated, e.g. with rainfallrunoff models. However, for this type of
model, time series of hydrometeorological data are required for forcing and calibration. To overcome the
problem of lack of local forcing data, global gridded meteorological datasets might be suitable for this
type of hydrological modelling. Over the last decade, global gridded re-analysis meteorological datasets
have been developed based on observations and modelling, e.g. the ERA-40 re-analysis (Uppala et al.,
2005), the Climate Research Unit (CRU) dataset (Mitchell and Jones, 2005). Gridded, large-scale
(0.5 x 0.5) meteorological datasets have already been used for soil moisture drought analyses in de
USA and globally (Andreadis et al., 2005; Sheffield and Wood, 2007) and for discharge drought at the
continental scale (Shukla and Wood, 2008). These drought analyses at large scale gave reasonable
results when compared with broad characteristics derived from observations. However, the suitability of
large-scale meteorological datasets to force models for drought analysis at catchment scale still needs
to be investigated.

The objective of this study is to assess the suitability of large-scale meteorological datasets for drought
analysis at the catchment scale. To reach this objective, the potential of one of these large-scale forcing
datasets, the WATCH Forcing Data (Weedon et al., 2010) was investigated by comparing drought
characteristics based on simulations using this large-scale forcing dataset with those derived from
simulations using local, more detailed, forcing data.

Several WATCH test basins were used in this study to test the large-scale forcing dataset. The test
basins are Narsj (Norway), Upper-Metuje (Czech Republic), Upper-Szava (Czech Republic),
Nedoery (Slovakia), and Platis (Crete, Greece). In each of the test basins, discharge was simulated
with one or more rainfall-runoff models using both local forcing data and the WATCH Forcing Data
(WFD). The same drought analysis was done on these simulations and drought characteristics were
compared.

Outline
First, a short description of the WATCH test basins used in this study is given (Chapter 2). In Chapter 3,
all rainfall-runoff models that were applied are described and drought analysis with the variable
threshold is explained. Also, a very short description of the large-scale forcing dataset, the WFD, is
provided. The comparison between the two forcing datasets for the test basins can be found in
Chapter 4. The results of the hydrological modelling and drought analysis with both forcing datasets are
given in Chapter 5. Conclusions are presented in Chapter 6. Abbreviations used in this report are
explained on page 44. Finally, three Annexes are included, containing 1) figures showing the
comparison between WFD and local corrected precipitation data, 2) an overview table of model results,
and 3) the effects of (re)calibration of models using large-scale data.

Technical Report No. 26
2

Technical Report No. 26
3
2. Catchment descriptions
In this chapter, a short description is given of the WATCH test basins used in this study. For more
detailed information about the catchments, the reader is referred to Van Lanen et al. (2008), Van
Huijgevoort et al. (2010), and Van Loon et al. (2010).
Figure 1 shows the location of the test basins in Europe and Table 1 gives an overview of the most
important catchment characteristics of the studied test basins.



Figure 1 a) Location of the studied catchments in Europe; and gauging station and meteorological stations in b)
Upper-Metuje and c) Upper-Szava (Czech Republic), d) Narsj (Norway), e) Nedoery (Slovakia), and f) Platis
(Crete).

Table 1

Technical Report No. 26
4

T
a
b
l
e

1

C
a
t
c
h
m
e
n
t

c
h
a
r
a
c
t
e
r
i
s
t
i
c
s

o
f

t
h
e

s
t
u
d
i
e
d

c
a
t
c
h
m
e
n
t
s

N
a
r
s
j


(
N
o
r
w
a
y
)
,

U
p
p
e
r
-
M
e
t
u
j
e

a
n
d

U
p
p
e
r
-
S

z
a
v
a

(
C
z
e
c
h

R
e
p
u
b
l
i
c
)
,

N
e
d
o

e
r
y

(
S
l
o
v
a
k
i
a
)
,

a
n
d

P
l
a
t
i
s

(
C
r
e
t
e
)
.




N
a
r
s
j


U
p
p
e
r
-
M
e
t
u
j
e

U
p
p
e
r
-
S

z
a
v
a

N
e
d
o

e
r
y

P
l
a
t
i
s








a
r
e
a

(
k
m
2
)


1
1
9

7
3
.
6

1
3
1
.
3

1
8
1

2
1
0








a
l
t
i
t
u
d
e

(
m

a
.
m
.
s
.
l
.
)


m
e
a
n


(
m
i
n

-

m
a
x
)

9
4
5

(
7
3
7

1
5
9
5
)

5
9
1

(
4
5
9
-
7
8
0
)

6
2
8

(
4
8
7
-
8
0
5
)

5
7
3

(
2
8
8

1
1
7
2
)

6
9
8

(
5
-
2
4
5
4
)








c
l
i
m
a
t
e

t
y
p
e

(

)


D
f
c

C
f
b

C
f
b

D
f
b

C
s
a








o
b
s
e
r
v
a
t
i
o
n

p
e
r
i
o
d
*


1
9
5
8
-
2
0
0
7

1
9
8
2
-
2
0
0
5

1
9
6
3
-
1
9
9
9

1
9
7
4
-
2
0
0
6

1
9
7
4
-
1
9
9
8








t
e
m
p
e
r
a
t
u
r
e

(

C
)
*
*

m
e
a
n

a
n
n
u
a
l


(
m
i
n
;

m
a
x

m
o
n
t
h
l
y
)

0
.
7

(
J
a
n
:

1
0
.
1
;

J
u
l
:

1
1
.
9
)

5
.
9

(
J
a
n
:

3
.
9
;

J
u
l
:

1
5
.
5
)

6
.
8

(
J
a
n
:

3
.
2
;

J
u
l
:

1
6
.
3
)

7
.
6

(
J
a
n
:

2
.
8
;

J
u
l
:

1
7
.
5
)

1
5

(
J
a
n
:

7
.
4
;

J
u
l
:

2
3
.
7
)








p
r
e
c
i
p
i
t
a
t
i
o
n

(
m
m
)
*
*

m
e
a
n

a
n
n
u
a
l


(
m
i
n
;

m
a
x

m
o
n
t
h
l
y
)

5
9
4

(
M
a
r
:

2
7
;

J
u
l
:

8
1
)

7
4
6

(
A
p
r
:

4
2
;

J
u
l
:

9
2
)

7
1
7

(
F
e
b
:

3
6
;

J
u
n
:

9
2
)

8
7
3

(
F
e
b
:

5
2
;

J
u
n
:

9
6
)

9
3
0

(
A
u
g
:

1
;

D
e
c
:

2
0
1
)








d
i
s
c
h
a
r
g
e

(
m
m
/
d
)
*
*

m
e
a
n

a
n
n
u
a
l


(
m
i
n
;

m
a
x

m
o
n
t
h
l
y
)

2
.
2

(
M
a
r
:

0
.
2
9
;

M
a
y
:

8
.
0
)

0
.
9
9

(
O
c
t
:

0
.
6
6
;

M
a
r
:

1
.
9
)

0
.
8
2

(
A
u
g
:

0
.
4
8
;

M
a
r
:

1
.
7
)

0
.
9
6

(
A
u
g
:

0
.
4
2
;

M
a
r
:

2
.
1
)

1
.
6

(
J
u
l
-
S
e
p
:

0
;

J
a
n
:

3
.
9
)








*

u
s
e
d

f
o
r

c
a
l
c
u
l
a
t
i
o
n

o
f

c
a
t
c
h
m
e
n
t

c
h
a
r
a
c
t
e
r
i
s
t
i
c
s

t
e
m
p
e
r
a
t
u
r
e
,

p
r
e
c
i
p
i
t
a
t
i
o
n
,

a
n
d

d
i
s
c
h
a
r
g
e

*
*

t
e
m
p
e
r
a
t
u
r
e
,

p
r
e
c
i
p
i
t
a
t
i
o
n
,

a
n
d

d
i
s
c
h
a
r
g
e

d
a
t
a

t
a
k
e
n

f
r
o
m

v
a
r
i
o
u
s

h
y
d
r
o
-
m
e
t
e
o
r
o
l
o
g
i
c
a
l

s
t
a
t
i
o
n
s

a
t

d
i
f
f
e
r
e
n
t

l
o
c
a
t
i
o
n
s

(
s
e
e

F
i
g
u
r
e

1

a
n
d

C
h
a
p
t
e
r

4
&
5
)

Technical Report No. 26
5
2.1. Narsj
The Narsj catchment is located in an open mountainous area in Eastern Norway, between Oslo and
Trondheim (Figure 1). It is a sub-basin of the Upper-Glomma, which is the headwater catchment of the
largest river in Norway, the Glomma. The Narsj catchment covers an area of 119 km
2
and its mean
elevation is 945 m a.m.s.l., with a minimum of 737 and a maximum of 1595 m a.m.s.l. (Engeland, 2002).
Land use types in the Narsj catchment are open area (60.9%), forest (24%), bogs (11.7%), and
agriculture (0.4%). Lakes cover 3.1% of the catchment (Hohenrainer, 2008). The Narsj catchment has
a Nordic continental climate with cold winters and relatively warm summers (Kppen-Geiger climate
Dfc). Mean annual temperature is 0.7C (Table 1). Every winter, snow covers the catchment
continuously for on average 7 months, from approximately the middle of October until the end of May,
depending on altitude (Engeland, 2002). Mean annual precipitation in Narsj is about 594 mm. The flow
regime in the catchment is dominated by the snowmelt flood, which on average has its peak in May.
The mean annual discharge of Narsj is 2.2 mm day
-1
. The catchment is dominated by hard rock, which
is covered by glacial deposits and a weathering layer with a variable thickness. During winter, when
precipitation accumulates as snow, long low-flow periods occur. The presence of many bogs and lakes
in the catchment delays the discharge during such dry periods and results in long recessions. The
minimum discharge is usually reached by late winter, just before the snow melt (Hohenrainer, 2008). In
summer, discharge is mainly determined by the catchments fast response to rainfall, resulting in a
flashy hydrograph.
2.2. Upper-Metuje
The Upper-Metuje catchment is situated in northeast Czech Republic and partly in Poland
(approximately 10% of the catchment area) (Figure 1). It is the headwater catchment of the river Metuje,
which discharges into the river Elbe. The area of the Upper-Metuje catchment is 73.6 km
2
and its mean
altitude is 591 m a.m.s.l., with a minimum of 459 and a maximum of 780 m a.m.s.l. (Rakovec et al.,
2009). Deep valleys, gentle and steep slopes and plateaus are the characteristic elements of the
landscape. Land use of the catchment consists mainly of cropland and grass fields (51%), and forest
(46%) (Rakovec et al., 2009). The Upper-Metuje catchment has a Central European continental climate
(Kppen-Geiger climate Cfb) with a mean annual temperature of 5.9C and a mean annual precipitation
of 746 mm (Table 1). The mean annual discharge is 0.99 mm day
-1
. Discharge peaks occur in spring
due to melting of snow accumulated in winter, whereas low discharges are mostly observed in summer
(Rakovec et al., 2009). The subsurface consists of thick permeable Cretaceous deposits overlying
rather impermeable Permian-Carboniferous rocks. Groundwater in the Upper-Metuje catchment is
characterised by deep circulation and high storage. This makes the catchment slowly responding to
precipitation.
2.3. Upper-Szava
The Upper-Szava catchment is the headwater of the river Szava, which eventually drains into the
river Vltava and subsequently into the river Elbe. The focal area is the Upper-Szava catchment
upstream from Zdar nad Sazavou (Figure 1). It only drains Czech territory and has an area of 131 km
2
,
which is about 3% of the total Szava catchment (Rakovec et al., 2009; Van Lanen et al., 2008). The
catchment is hilly with gentle slopes and flat wide valleys. The catchment altitude varies from 487 to
805 m a.m.s.l., with a mean altitude of 628 m a.m.s.l. (Table 1). The dominant land use types in the
catchment are forest (50%) and cropland and grassland (40%). The bedrock of Upper-Szava
catchment consists predominantly of Proterozoic impermeable metamorphic rocks, which consists of
black mica migmatite, gneiss and mica schist. There is no extensive groundwater storage (Rakovec et
al., 2009). Mean annual precipitation sum in the catchment is 717 mm (Table 1). The lowest monthly
precipitation is observed in February, the largest in June. The long term mean daily temperature is
6.8C. The warmest month is July and the coldest is January. Mean annual discharge is 0.82 mm day
-1
.

Technical Report No. 26
6
Floods occur regularly in spring because of snowmelt and low discharges predominantly occur in
summer and beginning of autumn (Table 1). In an average year there is snow from November until April,
with largest amounts in JanuaryMarch (2530 mm). In the Upper-Szava catchment withdrawal of
surface water and discharge of waste water into the streams takes place.
2.4. Nedoery
The Nedoery catchment is located in the upper part of the Nitra catchment (Figure 1). The Nitra
discharges into the Vah and, finally into the Danube. The catchment is located in the Prievidza district in
central Slovakia. It has an area of 181 km
2
and an average altitude of 573 m a.m.s.l. (Table 1). The
Nedoery catchment is an asymmetric valley, with the lowest parts in the east and most of the highest
parts in the west (Oosterwijk et al., 2009). Two-thirds of the catchment is covered by forest. Other land
cover types are agriculture (23%), natural meadow (6%), and urban area (5%) (Oosterwijk et al., 2009).
The catchment has a moderately warm, humid continental climate (Kppen-Geiger climate Dfb), with a
mean annual precipitation of 873 mm and a mean annual air temperature of 7.6C. Annual average
discharge is 0.96 mm day
-1
(Table 1). Maximum discharges occur in spring, minimum discharges occur
in summer and autumn (Machlica and Stojkovova, 2008). The largest part of the catchment consists of
Mesozoic rocks of the Inner Carpathians. These rocks are located in the northern, eastern and western
parts of the catchment. Since the catchment is dominated by hard rock, Nedoery shows a quick
response to rainfall.
2.5. Platis
The Platis catchment is located in the south-central part of the island of Crete in Greece and covers an
area of 210 km
2
(Figure 1). The mean annual precipitation is estimated to be 930 mm and its mean
elevation is 698 m a.m.s.l. which varies from 5 to 2454 m (Figure 1). The climate ranges between sub-
humid Mediterranean and semi-arid with long hot and dry summer and relatively humid and cold winter
with a mean annual temperature of 15C (Pavlakis, 2004). The mean annual discharge is 1.6 mm day
-1
.
It is estimated that from the total precipitation onto the catchment about 46% evapotranspirates, 19%
flows to the sea, and 35% recharges the groundwater. The land cover consists predominantly of forest
and semi-natural areas (53.5%) and agricultural areas (46.5%), and a small part of the catchment is
covered with artificial surfaces (0.1%). The hydrogeological bedrock of the area consists of impermeable
quartzites and phyllites, as well as permeable carboniferous, limestone formations, neogene, and
quaternary deposits (Pavlakis, 2004).


Technical Report No. 26
7
3. Methodology
As mentioned in the introduction, we used rainfall-runoff models to simulate discharge with local and
large-scale forcing, with the objective to test the suitability of large-scale forcing data in smaller
catchments. To analyse drought in these simulated time series, the variable threshold method is used.
In this chapter, the rainfall-runoff models, the approach for the drought analysis, and the forcing data are
described.
3.1. Model descriptions
In each test basin, different rainfall-runoff models were used (Table 2). For each model, a short
description is given below. All models were calibrated separately for both forcing datasets. That is
because the objective of this research was to investigate whether large-scale forcing datasets are
suitable in catchments where local meteorological observations are not available. So, to compare
drought characteristics in simulations with local and large-scale forcing, both datasets should be used
similarly. The effect of not (re)calibrating the models for the large-scale forcing dataset is described in
Section 5.6 and Annex 3.

Table 2 Overview of models used in each catchment and organisations performing the simulations (for
abbreviations see page 44)
Test basin BILAN FRIER HBV
Narsj
X
(NVE & WUR)
Upper-Metuje
X
(TGM-WRI)

X
(WUR)
Upper-Szava
X
(TGM-WRI)

X
(WUR)
Nedoery
X
(UC)
X
(UC)
X
(WUR)
Platis
X
(TUC)


3.1.1. BILAN
The structure of the BILAN model is formed by a system of relationships describing basic principles of
water balance on the land surface, in the zone of aeration (including the effect of vegetation cover), and
in groundwater. Air temperature is used as an indicator of energy conditions, which affect significantly
the water balance components. The input data of the model are daily series of catchment precipitation,
air temperature, and relative air humidity. For the calibration of the model parameters, a daily discharge
series at the outlet of the catchment is used.

The potential evapotranspiration is estimated from saturation deficit by using functions (in form of tables)
that have been derived for individual days (by interpolation between monthly values) and for different
bioclimatic zones from empirical graphs given by Gidrometeoizdat (1976). The saturation deficit is
calculated from data on the air temperature and relative air humidity. It is possible to use potential
evapotranspiration calculated externally, then the abovementioned estimation is bypassed.

The model generates daily series of catchment average potential evapotranspiration, actual evaporation,
water storage components in the snow cover, zone of aeration (soil), direct runoff storage, and
groundwater. The total runoff consists of two components, which are direct runoff and base flow. The
model has six free parameters and uses an optimisation algorithm for calibration in gauged catchments.

Technical Report No. 26
8
The standard calibration procedure of these parameters consists of two steps. In the first step, the
standard error or the mean absolute error of the simulated runoff is minimized to estimate the
parameters significantly affecting the mean runoff. The remaining three parameters affecting the runoff
distribution into its individual components (direct and subsurface runoff and base flow) are then
calibrated using the mean of absolute values of relative deviations. It has been demonstrated by
experimental calculations that in most cases this calibration procedure ensures an acceptable fit in
terms of both mean runoff and low flow runoff, which is formed predominantly by base flow. In addition,
different objective functions can be used in both calibration steps.

3.1.2. FRIER
FRIER is a physically-oriented rainfall-runoff model with distributed parameters (Horvt, 2007). The
model divides a catchment into uniform spatial units on a grid scale, in which the water balance is
simulated and discharge at the catchments outlet is generated. Transformation of the surface runoff in
the catchment is simulated by approximating a diffusive wave model using the geometric and hydraulic
characteristics of hillslopes and the stream network. The subsurface flow and percolation of each cell is
calculated using Darcys law and a method of approximating the kinematic wave model.

The necessary input files are time series of discharge at the catchment outlet, total precipitation, and air
temperature in any time step (min. 1 hour), and spatial layers of digital elevation model, soil texture, and
land use of the catchment. From these maps, other physio-graphical characteristics are derived as
digital maps: e.g. maps of the soil and land use parameters, flow direction, stream network, slope. The
filling in of missing data is possible but not necessary.

Data from the time series in the model can be spatially distributed by the arithmetic mean of closest
stations, nearest neighbours, lapse rate, or kriging. Potential global radiation can be computed with or
without the slope orientation of each cell and the shading of its neighbouring cells. The difference
between short-wave and long-wave solar radiation is expressed by the net radiation balance. In the
surface energy balance, they are required for the determination of potential evapotranspiration. It is
possible to choose among many methods for the calculation of potential evapotranspiration, which were
selected on the basis of a detailed study (Horvt, 2007).

The routing parameters are generated in a developed extension of the ESRI ArcView GIS program in a
GIS interface. Ten global parameters serve to simplify some processes and for the best setting of the
initial values. They are constants for all the cells in the catchment and they can be calibrated. Several
methods for the calibration of the global parameters and also several objective functions for assessing
the models efficiency (e.g. BIAS, NashSutcliffe) are incorporated in the model.

Output of the FRIER model are time series of the water balance components or spatial maps. The time
series contain simulated discharge and its three components (overland flow, interflow and base flow) for
any time step. The mean quantities for the whole catchment are calculated for each time step, e.g. air
temperature, potential and actual evapotranspiration, rainfall, snowmelt. The output maps can be e.g.
layers of total overland flow, interflow, base flow.

3.1.3. HBV
The HBV model (originally developed at SMHI (Bergstrm, 1976; Bergstrm, 1992; Bergstrm and
Forsman, 1973)) is a rainfall-runoff model, which includes conceptual numerical descriptions of
hydrological processes at the catchment scale using various model routines. HBV can be used as a
semi-distributed model by dividing the catchment into subbasins. Each subbasin is then divided into
zones according to altitude, lake area, and vegetation. It can also be run as a lumped catchment model,
using similar elevation and vegetation zones. The model is normally run on daily values of rainfall and

Technical Report No. 26
9
air temperature (that are corrected during calibration according to an altitude gradient and a snowfall
correction factor), and daily or monthly potential evaporation regimes. The model is used for flood
forecasting in the Nordic countries, and many other purposes. There are many different versions of HBV
Model software besides the original SMHI version.

HBV-WUR
The model version HBV light (Seibert, 2005), used by WUR (and called HBV-WUR in this report), is a
lumped version of the HBV model. It does not divide the catchment into subbasins, but makes use of
elevation zones. The HBV-WUR model was forced with observed daily temperature and precipitation,
and calculated daily potential evaporation for all catchments (Table 2). Temperature and precipitation
were corrected for elevation according to predefined elevation zones. The model consists of four
routines, i.e. a distributed snow routine and soil moisture routine, a lumped response routine
representing groundwater, and a routing routine. Snow accumulation and melt are calculated by the
degree-day method for a number of elevation (max. 10) and vegetation (max. 3) zones separately. In
each of these zones, groundwater recharge and actual evapotranspiration are simulated as a function of
actual water storage in the soil. Subsequently, the lumped response function, consisting of two linear
reservoirs, transforms recharge into discharge. Finally, channel routing is computed by a triangular
weighting function. A more comprehensive description of the model can be found in Seibert (2000;
2005) and Oosterwijk et al. (2009). Calibration of the HBV-WUR model was done on time series of
observed discharge using the genetic calibration algorithm described by Seibert (2000). To give more
weight to low flows, the logarithm of the Nash-Sutcliffe efficiency (lnReff) (Nash and Sutcliffe, 1970;
Seibert, 2005) was used to evaluate the agreement between simulated and observed discharge. The
first year of data was used as starting up year to initialize the model state.

HBV-NVE
The HBV model version used by NVE is the "Nordic" HBV model (Killingveit and Slthun, 1995;
Slthun, 1996). This version (called HBV-NVE in this report) has the same setup as the HBV-WUR
model. It is a lumped catchment model in which the spatial structure of the catchment is not explicitly
modelled. Ten equal area height zones from the hypsometric curve for the catchments are defined, and
land cover data is distributed by height zone. The model consists of the same four routines as the HBV-
WUR model, i.e. a distributed snow routine and soil moisture routine, a lumped response routine
representing groundwater, and a routing routine. All processes contribute directly to discharge at the
outlet without internal routing between elevation zones. Processes are represented as linear or simple
non-linear relationships, and all are controlled by parameters determined during calibration.

The model is driven by daily time series of air temperature and precipitation, and model parameters are
adjusted to achieve a best fit relative to discharge observed at the catchment outlet. In the applications
reported here, evapotranspiration was estimated by HBV-NVE using the temperature index method,
rather than using monthly values as model input. In the HBV-NVE model calibrations, PEST parameter
estimation routines (Doherty, 2004) based on PEST v. 11.2 were used to calibrate parameters for HBV-
NVE model. The HBV-NVE model was calibrated on the Nash-Sutcliffe efficiency (Reff) (Nash and
Sutcliffe, 1970) and volume bias, see also Lawrence et al. (2009).

HBV-TUC
The version of HBV used in the Platis catchment by TUC (called HBV-TUC in this report) is the
Integrated Hydrological Modelling System (IHMS 5.10.1) HBV 7.1, developed and provided by Swedish
Meteorological and Hydrological Institute (SMHI) (Integrated Hydrological Modeling System, 2006). This
model has the same setup and routines as the HBV-WUR and HBV-NVE models. Input data are
observations of precipitation, air temperature, vapour pressure, wind speed, and estimates of potential
evaporation. The evapotranspiration values used are long-term monthly averages. Air temperature,

Technical Report No. 26
10
vapour pressure, and wind speed are used for calculations of snow accumulation and melt. Discharge
observations are used to calibrate the model, and to verify and correct the model before a runoff
forecast. The first year of the simulation period is used as starting up year to initialize the model state.
The model was calibrated manually, based on both efficiency criteria lnReff and Reff (Nash and Sutcliffe,
1970). More weight was given to lnReff, to focus on a better performance of low flows, according to the
aims of the present study.
3.2. Drought analysis
To determine drought events from the time series of simulated discharge, the threshold level method
(Hisdal et al., 2004; Yevjevich, 1967) was applied. With this method, a drought occurs when the variable
of interest (e.g. discharge, precipitation, recharge) is below a predefined threshold (Figure 2). The start
of a drought event is indicated by the point in time when the variable falls below the threshold and the
event continues until the threshold is exceeded again. Drought characteristics commonly derived with
this method are beginning, end, duration, deficit volume, and minimum flow during an event (Fleig et al.,
2006; Hisdal et al., 2004). The characteristics taken into account in this study are the number, mean
duration, and mean deficit of drought events. Both a fixed and variable (seasonal, monthly, or daily)
threshold can be used. In this study, a monthly threshold derived from the 80-percentile of the flow
duration curve was applied. The discrete monthly threshold values were smoothed by applying a
centred moving average of 30 days (Van Loon et al., 2010). To eliminate minor drought events, a
minimum duration of 3 days was used. The thresholds were calculated for each time series of discharge
(observed and simulated) separately, so each time series had a different threshold (see for discussion
Section 5.6).



Figure 2 Threshold level method with a variable (monthly) threshold (data from Tallaksen and Van Lanen (2004)).

3.3. Large-scale forcing data
The large-scale forcing dataset used in this study is the WATCH Forcing Data (WFD (Weedon et al.,
2010)). It consists of gridded time series of meteorological variables (e.g. rainfall, snowfall, temperature,
wind speed), both on a sub-daily and daily basis for 19582001. In this study, the daily data were used.
For hydrological modelling with these data, the same periods were used as were available for the local
forcing data in each catchment. The data have a resolution of 0.5 x 0.5. The WFD originate from
modification (bias-correction) of the ECMWF ERA-40 re-analysis data, which are sub-daily data on a
one-degree spatial resolution. The different weather variables have been interpolated and corrected for
the elevation differences between the ERA-40 one-degree grid and the CRU half-degree grid. For
precipitation, the ERA-40 data were firstly adjusted to have the same number of wet (i.e. rain- or snow-)
days as the CRU wet day data. Next the data were bias corrected using monthly GPCC precipitation
totals (Schneider et al., 2008) and finally gauge-catch corrections were applied separately for rainfall
and snowfall. Additionally, the interpolated ERA-40 near-surface temperatures were elevation corrected

Technical Report No. 26
11
and bias-corrected using both CRU monthly average temperatures and CRU monthly average diurnal
temperature ranges. For more information the reader is referred to Weedon et al. (2010).

3.4. Local forcing data
The local forcing data is different for each catchment and sometimes even for the different models used
in one catchment. The description of the local forcing data that each model uses in a certain catchment,
is given in Chapter 5 before the description of the modelling results.

Technical Report No. 26
12


Technical Report No. 26
13
4. Differences between forcing datasets
For each catchment, the WFD are compared with measured local values (same as used as local forcing
in HBV-WUR and HBV-TUC simulations, see Chapter 5) to check the credibility of the large-scale
forcing data. For the comparison of the two forcing datasets, only WFD grid cells that cover the
catchments are used. Time series of catchment average forcing data were computed by calculating the
weighted average according to the relative area of the catchment in each grid cell in case of two grid
cells covering the area. The area of a WFD grid cell is much larger than the areas of the studied
catchment (~2500 km
2
vs. 73-210 km
2
) and also altitudes are different (Table 3). The grid cell averages
might not be representative for the catchments, especially in regions with complex orography. Some
form of altitude correction can be applied to correct for this difference. In some models used in this
research (e.g. HBV-WUR and HBV-NVE), both the local and large-scale forcing datasets are corrected
based on an altitude gradient and a snowfall correction factor. The corrected precipitation values are
also compared (Annex 1), but show the same results as presented in this paragraph.

The differences in mean annual temperature (T) and precipitation (P) between the two datasets are
given in Table 3. The differences in mean annual temperature vary from -0.2 C to 2 C (WFD compared
to local forcing data). In most catchments WFD temperatures are higher than temperatures measured
locally, but this is not consistent for all catchments. The WFD either overestimate or underestimate the
mean annual precipitation by maximally about 10%. Again there is no systematic bias for all catchments.
There are some differences between the catchments, for example, in the Narsj catchment the
differences between the datasets are smaller than in the Platis catchment.

In Figure 3 to Figure 7 monthly temperature and precipitation of both WFD and local forcing data are
shown for each catchment. Differences between catchments are visible: in the Narsj and Nedoery
catchments temperature differences are minimal, but in the other catchments temperature is
overestimated by the WFD throughout the year. The WFD precipitation is not consistently higher or
lower than local precipitation throughout the year: in some months the WFD overestimate the
precipitation, in other months precipitation is underestimated. Overall, the differences between the
forcing datasets seem to be acceptable for further hydrological model applications.

Table 3 Long-term average mean annual temperature and precipitation for both forcing datasets and comparison
of WFD with local forcing data (see Chapter 5 HBV-WUR and HBV-TUC models for the origin of the local forcing
data)
Catchment
(overlapping
time period)
Forcing
dataset
No of
grid
cells
Elevation (m
a.m.s.l.)
Mean
annual T
(C)
Difference
T (C)
Mean
annual P
(mm)
Difference
P (mm)

Narsj Local T: 628, P: 713 0.5 586
(19582001) WFD 2 785 0.3 -0.2 632 46 (7.8%)

Upper-Metuje Local 490 5.8 754
(19812001) WFD 1 446 7.6 1.8 824 70 (9.3%)

Upper-Szava Local T: 530, P: 628 6.7 717
(1962-1999) WFD 2 461 7.6 0.9 782 65 (9.1%)

Nedoery Local 573 7.6 849
(1973-2001) WFD 1 580 7.1 -0.5 795 -54 (-6.4%)

Platis Local 698 15.1 931
(1975-1998) WFD 1 469 17.1 2 836 -95 (-10.2%)



Technical Report No. 26
14


Figure 3 Long-term average mean monthly temperature and total monthly precipitation for both forcing datasets
in Narsj.


Figure 4 Long-term average mean monthly temperature and total monthly precipitation for both forcing datasets
in Upper-Metuje.



Figure 5 Long-term average mean monthly temperature and total monthly precipitation for both forcing datasets
in Upper-Szava.



Figure 6 Long-term average mean monthly temperature and total monthly precipitation for both forcing datasets
in Nedoery.



Technical Report No. 26
15


Figure 7 Long-term average mean monthly temperature and total monthly precipitation for both forcing datasets
in Platis.




Technical Report No. 26
16



Technical Report No. 26
17
5. Results
In each catchment, discharge is simulated with locally measured meteorological data and with the WFD.
This chapter gives the results of the modelling and the drought analysis for each test basin. The Nash-
Sutcliffe values of all model results for all catchments are presented in Annex 2.

5.1. Narsj

5.1.1. Modelling

HBV-WUR model
The local meteorological data used for the HBV-WUR model of the Narsj catchment are measured at
three stations. Daily precipitation was obtained from two stations (Figure 1), i.e. Ellefsplass and
Tufsingdal (moved to Tufsingdal-Midtdal in 1991), located on either side of the catchment. Catchment
precipitation was calculated by computing the arithmetic mean. Daily mean temperature was taken from
the meteorological station Rros (about 25 km from Narsj). Daily discharge was recorded at the outlet
of the catchment (gauging station Narsj, Figure 1). For modelling with HBV-WUR, data from all stations
were used for the period 19582001. This entire period was used as calibration period of the model and
calibration was done on the logarithm of the Nash-Sutcliffe efficiency (lnReff). Potential
evapotranspiration was calculated with the Penman-Monteith method (Allen et al., 1998) for both local
forcing data and WFD. In case of missing or incorrect meteorological data, assumptions and
recommendations of Doorenbos and Pruitt (1975) and Allen et al. (1998) were followed.

The model performs well for both forcing datasets (upper part of Table 4), especially if lnReff is
considered. The winter low-flow conditions in Narsj can be simulated quite well with the linear reservoir
in HBV. In a winter situation in Narsj, all precipitation is stored as snow and therefore there is no
recharge. This means that the forcing datasets have little influence on the recession, which could
explain the good performance of the model for both datasets. Hardly any difference was found between
simulated discharge using the two different forcing datasets (Figure 8). In both cases, peaks are often
underestimated (due to calibration on lnReff), but low flows are modelled well.

Table 4 Nash-Sutcliffe values for HBV-WUR and HBV-NVE models with local and WFD forcing for the Narsj
catchment (grey columns depict which calibration criterion was used)
Reff lnReff
HBV-WUR local
0.7691 0.9046
WFD 0.7814 0.8902
HBV-NVE local 0.7214 0.5165

WFD 0.7067 0.7884

a)


Technical Report No. 26
18
b)

Figure 8 Discharge (HBV-WUR model): (a) observed and simulated discharge for Narsj, (b) detail of the
discharge (low-flow range) for the period 19931994 (part of calibration period).

HBV- NVE model
The local forcing data used for the HBV-NVE model is based on a 1 x 1 km gridded daily temperature
and precipitation dataset of the Norwegian Meteorological Institute (met.no). For each day, the
catchment mean temperature and precipitation are calculated (i.e. the mean of the grid cell within the
Narsj catchment) and used as input to the model. PET is calculated based on a temperature index
method. The overlapping period between this source of local forcing data and WFD at the time of model
simulations was 01-09-1961 until 31-12-2001. The calibration period of the HBV-NVE model is 1980-
1989, the validation period is 1990-1999. The calibration criteria used are the Nash-Sutcliffe efficiency
(Reff) and volume bias, and the best parameter set is used in the model simulations for the validation
period.

As this model was calibrated on the Nash-Sutcliffe efficiency (Reff), it gives a lower performance on low
flows (see the lower lnReff values for HBV-NVE in Table 4). However, the model still shows reasonable
results (Figure 9), especially when timing of low flows is considered. The results of the simulation with
local forcing data seem to be better than the ones of the simulation with WFD.

a)

b)

Figure 9 Discharge (HBV-NVE model): (a) observed and simulated discharge for Narsj, (b) detail of the
discharge (low-flow range) for the period 19931994 (part of validation period).

5.1.2. Drought analysis
From the observed and simulated discharge time series, hydrological droughts were identified with the
threshold level method (Section 3.2). Several drought characteristics were determined, i.e. number,
mean duration, and mean deficit of droughts (Table 5). The relative difference between results from
simulated and observed discharge for each characteristic is also included.

These differences show that for the HBV-WUR model drought characteristics derived from simulations
with WFD are more or less similar to those obtained with local forcing data. When comparing drought

Technical Report No. 26
19
events in both simulations with drought events in observed discharge, all characteristics correspond well,
with deviations of 220%.

The HBV-NVE model gives larger differences between the characteristics derived from the simulations
and those from observed discharge than the HBV-WUR model. This can be caused by the different
calibration criterion used by HBV-NVE (Reff instead of lnReff) or the short calibration period used (10 yr
instead of 43 yr). For the HBV-NVE model, the drought characteristics derived from the simulations with
WFD are closer to the ones of the observations than those from simulations with local forcing data, in
particular the number of droughts and duration. This was not expected on the basis of the hydrographs
(Figure 9). The model run using local forcing better reproduced the shape of the hydrograph during low-
flow periods than the model run using WFD. However, the threshold level is different for all runs (see
Section 3.2 and Figure 11), which influences average drought characteristics (see Section 5.6). Another
reason for this difference can be the gridded local forcing data used in HBV-NVE.

Drought events found in the period 1979-1983 in both local and simulated discharge are shown in
Figure 10 for the HBV-WUR model and in Figure 11 for the HBV-NVE model. Overall, drought events
(indicated in red) in the observed discharge are reproduced in the simulations of both models and with
both datasets, especially the most severe events (e.g. summer 1982). However, there are also
differences in duration, deficit, and start of the drought events between simulated and observed
discharge (e.g. winter 1979-1980, and summer 1983).

Table 5 Summary of discharge drought characteristics for Narsj (number, mean duration, and mean deficit of
droughts) and differences between the two forcing datasets
Period % difference

Number of droughts
(-)
Observed discharge 1959-2001 140
HBV-WUR with local forcing 128 -8.6
HBV-WUR with WFD 128 -8.6
Observed discharge 1961-2001 136
HBV-NVE with local forcing 221 62.5
HBV-NVE with WFD 157 15.4

Mean duration drought
(days)
Observed discharge 1959-2001 23.29
HBV-WUR with local forcing 25.83 10.9
HBV-WUR with WFD 23.78 2.1
Observed discharge 1961-2001 22.24
HBV-NVE with local forcing 13.42 -39.7
HBV-NVE with WFD 19.66 -11.6

Mean deficit
(mm)
Observed discharge 1959-2001 6.59
HBV-WUR with local forcing 5.34 -19.0
HBV-WUR with WFD 5.52 -16.2
Observed discharge 1961-2001 6.11
HBV-NVE with local forcing 3.19 -47.8
HBV-NVE with WFD 3.84 -37.2


Technical Report No. 26
20
a)







b)







c)

Figure 10 Drought events in Narsj (HBV-WUR model): (a) observed discharge and threshold, (b) simulated
discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

a)







b)







c)

Figure 11 Drought events in Narsj (HBV-NVE model): (a) observed discharge and threshold, (b) simulated
discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

Technical Report No. 26
21
5.2. Upper-Metuje

5.2.1. Modelling

HBV-WUR model
The local meteorological data used for the HBV-WUR model of the Upper-Metuje catchment are
measured at a representative meteorological station, i.e. the Bunice station (Figure 1). Daily mean
temperature and precipitation measurements were obtained for the period 19812001. Daily discharge
was measured during the same period at the outlet of the catchment (gauging station MXII, Figure 1).
The entire period 1981-2001 was used as calibration period and calibration was done on the logarithm
of the Nash-Sutcliffe efficiency (lnReff). Potential evapotranspiration was calculated with the Penman-
Monteith method (Allen et al., 1998) for both local forcing data and WFD. In case of missing or incorrect
meteorological data, assumptions and recommendations of Doorenbos & Pruitt (1975) and Allen et al.
(1998) were followed.

The Nash-Sutcliffe efficiency (Reff) for the Upper-Metuje catchment is low (upper part of Table 6), mainly
due to a poor simulation of peak flows (Figure 12). However, focus is on low flows, for which the model
performance is better reflected by lnReff. This lnReff (0.71 and 0.65) is substantially higher than the Reff
(0.48 and 0.35), but still lower than lnReff of the Narsj catchment (0.90 and 0.89). In Upper-Metuje, low
flows show a recession curve typical to a large, multiple aquifer system, which is more difficult to
represent with the linear reservoir in HBV. Low flows in Upper-Metuje usually occur in summer and are
caused by low precipitation and/or high evaporation rates. The forcing data have a larger influence on
simulated discharge than in the Narsj catchment, especially for the winter season. The differences
between the forcing datasets of the Upper-Metuje catchment (Table 3) could cause the difference
between the lnReff values of the simulations using both datasets.

Table 6 Nash-Sutcliffe values for HBV-WUR and BILAN models with local and WFD forcing for the Upper-Metuje
catchment (grey columns depict which calibration criterion was used)
Reff lnReff
HBV-WUR local
0.4783 0.7111
WFD 0.3489 0.6509
BILAN
local
0.4016 0.5059

WFD 0.3973 0.4892

a)

b)

Figure 12 Discharge (HBV-WUR model): (a) observed and simulated discharge for Upper-Metuje, (b) detail of the
discharge (low-flow range) for the period 19931994.

Technical Report No. 26
22

BILAN model
For the BILAN model of the Upper-Metuje catchment the same local meteorological data were used as
for the HBV-WUR model. That means daily mean temperature and precipitation from Bunice
meteorological station and daily discharge from gauging station MXII (Figure 1) for the period 1981-
2001. Potential evapotranspiration was calculated with the standard method of BILAN (Section 3.1.1).
The entire period was used as calibration period. The standard two-step calibration procedure
(Section 3.1.1) emphasizing the mean and low flows was used. The lnReff was also tested as the
calibration criterion in both steps of the calibration. This has led, however, to unacceptable deviations in
medium and high flows.

A 2-yr time slice of the observed and simulated discharge series is shown in Figure 13. In general, the
simulated discharge is consistent with observed discharge. Low flows are slightly overestimated. The
difference between the simulation with local forcing and WFD is minimal. The Nash-Sutcliffe efficiencies
for the BILAN model are comparable to those of the HBV-WUR model (Table 6), only lnReff values of
BILAN are lower, which might be due to the different objective functions used in calibrating the two
models.

a)

b)

Figure 13 Discharge (BILAN model): (a) observed and simulated discharge for Upper-Metuje, (b) detail of the
discharge (low-flow range) for the period 19931994.

5.2.2. Drought analysis
From the observed and simulated discharge time series, several drought characteristics were
determined with the threshold level method (Section 3.2), i.e. number, mean duration, and mean deficit
of droughts (Table 7). The relative difference between results from simulated and observed discharge
for each characteristic is also included.

For the HBV-WUR model, drought characteristics derived from simulations with WFD are a bit closer to
those derived from observed discharge (6.5 23.3%), than results from simulations with local forcing
data (25.8 45.1%). This was not expected on the basis of the Nash-Sutcliffe values (upper part of
Table 6). Differences in the shape of the threshold of each time series could be the reason for this
difference (Figure 14).

For the BILAN model, the differences between the drought characteristics from the simulated
discharges (both local and WFD) and those from the observed discharges are larger (min. 52%) than for
the HBV-WUR model and similar for both datasets. The larger difference could partly be caused by the
different calibration criteria (lnReff for HBV-WUR and mean absolute and relative error for BILAN) and

Technical Report No. 26
23
partly by the fact that the BILAN simulations have very smooth recessions leading to longer droughts
than in the observations. This is due to the soil module in the daily version of the BILAN model which
does not allow the infiltrated water to runoff to subsurface flow until the soil water storage reaches its
maximal capacity. Therefore, if the cumulative infiltration minus evapotranspiration in relatively dry
periods with low intensities of precipitation does not exceed the maximal soil water storage (parameter
of the model) the runoff is formed exclusively by the outflow from the groundwater reservoir. This will be
modified in the coming version of the BILAN model. Since the smooth recessions are a consequence of
the structure of the model, simulations with WFD lead to similar drought characteristics as simulations
with local forcing data.

In Figure 14 and Figure 15, drought events in the time series of observed and simulated discharges are
indicated in red for the period November 1991 to November 1996. These figures illustrate that both
models and both datasets have problems with exactly mimicking the droughts in observed discharge:
some drought events are missed (e.g. summer 1992 and summer 1994) or extra drought events are
added (winter 1993-1994). Again, drought deficit and duration are different in the simulations. However,
the timing of most severe drought events in observed discharge is reproduced by the simulations,
especially the drought in spring 1996. Furthermore, it is clear that the small interruptions of droughts in
observed discharge are not reproduced by the smooth hydrographs of the BILAN model, leading to
longer droughts with higher deficits.

Table 7 Summary of discharge drought characteristics for Upper-Metuje (number, mean duration, and mean
deficit of droughts) and differences between the two forcing datasets
Period % difference

Number of droughts
(-)
Observed discharge 1981-2001* 98
HBV-WUR with local forcing 69 -29.6
HBV-WUR with WFD 82 -16.3
Observed discharge 1980-2001* 100
BILAN with local forcing 46 -54.0
BILAN with WFD 48 -52.0

Mean duration drought
(days)
Observed discharge 1981-2001* 14.59
HBV-WUR with local forcing 21.17 45.1
HBV-WUR with WFD 17.99 23.3
Observed discharge 1980-2001* 14.91
BILAN with local forcing 32.26 116.4
BILAN with WFD 31.46 111.0

Mean deficit
(mm)
Observed discharge 1981-2001* 0.93
HBV-WUR with local forcing 1.17 25.8
HBV-WUR with WFD 0.87 -6.5
Observed discharge 1980-2001* 0.98
BILAN with local forcing 2.95 201.0
BILAN with WFD 2.93 199.0
* = hydrological year from 1 November to 31 October

Technical Report No. 26
24
a)







b)







c)

Figure 14 Drought events in Upper-Metuje (HBV-WUR model): (a) observed discharge and threshold, (b)
simulated discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

a)







b)







c)

Figure 15 Drought events in Upper-Metuje (BILAN model): (a) observed discharge and threshold, (b) simulated
discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

Technical Report No. 26
25
5.3. Upper-Szava

5.3.1. Modelling

HBV-WUR model
The local meteorological data used for the HBV-WUR model of the Upper-Szava catchment are
measured at several meteorological stations in and around the catchment. Data were obtained for the
period 1-11-1961 until 30-10-2000. Records of temperature are available for two stations, i.e. daily
temperature from Pibyslav and minimum and maximum daily temperature from Svratouch. Daily data of
precipitation are available from Pibyslav, Krucemburk, r nad Szavou-Stranov, Kinky, and
Kadov meteorological stations. Some climatological data (minimum and maximum temperature, wind
speed, and solar radiation) are available from Svratouch and they are used to calculate potential
evapotranspiration with the Penman-Monteith method (Allen et al., 1998). In case of missing or incorrect
meteorological data, assumptions and recommendations of Doorenbos & Pruitt (1975) and Allen et al.
(1998) were followed. Although Vatn meteorological station has a suitable location close to the
catchment border, we did not make an effort to obtain these data because of the short observation
period, which started in the beginning of the 1990s. The location of mentioned stations is shown in
Figure 1. Daily discharge was measured at station 1550 (Szava u ru nad Szavou). The entire
period was used as calibration period, with focus on low flows (lnReff as calibration criterion).

The Nash-Sutcliffe values (upper part of Table 8) are reasonable and especially lnReff values are
comparable for both datasets (0.63 and 0.60). A 2-yr time slice of the observed and simulated discharge
series is shown in Figure 16. In the complete hydrograph (Figure 16a), peaks are not simulated correctly
due to the focus on low flows during calibration. Low flows, however, are represented quite well (Figure
16b), and both datasets seem to give a similar result.

Table 8 Nash-Sutcliffe values for HBV-WUR and BILAN models with local and WFD forcing for the Upper-Szava
catchment (grey columns depict which calibration criterion was used)
R
eff
lnR
eff

HBV-WUR local
0.6115 0.6313

WFD
0.5276 0.6048
BILAN local 0.5583 0.5363

WFD 0.5499 0.5408

a)

b)

Figure 16 Discharge (HBV-WUR model): (a) observed and simulated discharge for Upper-Szava, (b) detail of
the discharge (low-flow range) for the period 19931994.

Technical Report No. 26
26

BILAN model
For the BILAN model of the Upper-Szava catchment the same local meteorological data were used as
for the HBV-WUR model. That means daily temperature from Pibyslav, precipitation from Pibyslav,
Krucemburk, r nad Szavou-Stranov, Kinky, and Kadov meteorological stations and daily
discharge from gauging station 1550 (Szava u ru nad Szavou) for the period 1961-2000
(Figure 1). Potential evapotranspiration was calculated with the standard method of BILAN
(Section 3.1.1). The entire period was used as calibration period. The standard two-step calibration
procedure (Section 3.1.1) emphasizing the mean and low flows was used. The lnReff was also tested as
the calibration criterion in both steps of the calibration. This has led, however, to unacceptable
deviations in medium and high flows.

Also for the Upper-Szava catchment, the Nash-Sutcliffe efficiencies for the BILAN model are
comparable to, but slightly lower than those of the HBV-WUR model (Table 8). A 2-yr time slice of the
observed and simulated discharge series is shown in Figure 17. Both peaks and low flows are generally
simulated well, although the peaky behaviour of the observations is not reproduced by the model.
Simulated discharge, both with local forcing and WFD, is much more smoothed than observed
discharge (for explanation see Section 5.2.2. on page 22-23).

a)

b)

Figure 17 Discharge (BILAN model): (a) observed and simulated discharge for Upper-Szava, (b) detail of the
discharge (low-flow range) for the period 19931994.

5.3.2. Drought analysis
From the observed and simulated discharge time series, several drought characteristics were
determined with the threshold level method (Section 3.2), i.e. number, mean duration, and mean deficit
of droughts (Table 9). The relative difference between results from simulated and observed discharge
for each characteristic is also included.

For the HBV-WUR model, drought characteristics derived from simulations with local forcing data (2-
72%) are a bit closer to those derived from observed discharge than results from simulations with WFD
(47-100%). Especially the mean deficit of droughts in the simulation using local forcing (1.19 mm) is
almost similar to the mean deficit of droughts in observed discharge (1.17 mm).

For the BILAN model, the difference between drought characteristics of simulations and observations
(69-253%) is larger than for the HBV-WUR model, but results of BILAN simulations with both datasets
are similar. The larger difference could again partly be caused by the different calibration criteria, but
clearly the smooth simulated discharges by the BILAN model play a large role as well. In Figure 18 and

Technical Report No. 26
27
Figure 19 drought events in simulated and observed discharges are indicated in red. Both HBV-WUR
and BILAN models do not reproduce the small peaks during low-flow periods that interrupt droughts in
observed discharge. Therefore, in both models, the number of drought events is underestimated and
drought duration and deficit are overestimated. In BILAN this effect is even stronger than in HBV-WUR.

In the period displayed in Figure 18 and Figure 19, there was a series of severe drought events in
observed discharge starting in spring 1990 and lasting until autumn 1991. All simulated discharge time
series represent these droughts quite well looking at timing and total duration. The smaller drought
events in the observed discharge in 1988 and 1989 are sometimes completely missed or not well
identified in the simulations. The HBV-WUR model simulated an extra drought in 1992 that is not visible
in the observations. However, droughts determined in simulated discharge with both datasets are very
similar in both models, especially the severe droughts.

Table 9 Summary of discharge drought characteristics for Upper-Szava (number, mean duration, and mean
deficit of droughts) and differences between the two forcing datasets
Period % difference

Number of droughts
(-)
Observed discharge 1962-2000* 210
HBV-WUR with local forcing 124 -41.0
HBV-WUR with WFD 111 -47.1
Observed discharge 1961-2000* 215
BILAN with local forcing 67 -68.8
BILAN with WFD 66 -69.3

Mean duration drought
(days)
Observed discharge 1962-2000* 12.58
HBV-WUR with local forcing 21.6 71.7
HBV-WUR with WFD 25.14 99.8
Observed discharge 1961-2000* 12.64
BILAN with local forcing 42.49 236.2
BILAN with WFD 44.67 253.4

Mean deficit
(mm)
Observed discharge 1962-2000* 1.17
HBV-WUR with local forcing 1.19 1.7
HBV-WUR with WFD 1.84 57.3
Observed discharge 1961-2000* 1.24
BILAN with local forcing 3.44 177.4
BILAN with WFD 3.37 171.8
* = hydrological year from 1 November to 31 October


Technical Report No. 26
28
a)







b)







c)

Figure 18 Drought events in Upper-Szava (HBV-WUR model): (a) observed discharge and threshold, (b)
simulated discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

a)







b)







c)

Figure 19 Drought events in Upper-Szava (BILAN model): (a) observed discharge and threshold, (b) simulated
discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

Technical Report No. 26
29
5.4. Nedoery

5.4.2. Modelling

HBV-WUR model
The local meteorological data used for the HBV-WUR model of the Nedoery catchment are measured
at a number of stations in and around the catchment (Figure 1). Daily temperature data were derived
from two meteorological stations: Prievidza and Turcianske Teplice, and daily precipitation
measurements from five stations: Nitrianske Pravno, Chvojnica, Vricko, Slovensk Pravno, and Valask
Bel Gapel. Catchment average temperature and precipitation were calculated using Thiessen
polygons (Oosterwijk et al., 2009). Daily discharge is measured at gauging station Nedoery (Figure 1).
The modelling period for Nedoery was 19742001. The entire period was used as calibration period
and calibration was done on the logarithm of the Nash-Sutcliffe efficiency (lnReff). Potential evapotrans-
piration was calculated with the Penman-Monteith method (Allen et al., 1998) for both local forcing data
and WFD. In case of missing or incorrect meteorological data, assumptions and recommendations of
Doorenbos & Pruitt (1975) and Allen et al. (1998) were followed.

For Nedoery the lnReff (upper part of Table 10) is similar to that of the Upper-Metuje catchment (Table
6) and Upper-Szava catchment (Table 8), but lower than that of the Narsj catchment (Table 4). The
hydrological regime of Nedoery is less regular and more determined by a fast response to rainfall than
that of Narsj. Model results in Nedoery are therefore more dependent on the quality and
representativeness of precipitation measurements. The difference in precipitation between the local and
large-scale forcing data is quite small (Table 3) and this leads to quite similar simulations of the
discharge with both datasets (Figure 20).

Table 10 Nash-Sutcliffe values for HBV-WUR, BILAN, and FRIER models with local and WFD forcing for the
Nedoery catchment (grey columns depict which calibration criterion was used) for the period 1981-2001
Reff lnReff
HBV-WUR local 0.6639 0.671

WFD
0.6723 0.7338
FRIER local 0.6065 0.6332

WFD 0.516 0.5634
BILAN
local
0.4108 0.3071
WFD 0.2231 -0.1243

a)




Technical Report No. 26
30
b)

Figure 20 Discharge (HBV-WUR model): (a) observed and simulated discharge for Nedoery, (b) detail of the
discharge (low-flow range) for the period 19911992.

FRIER model
Data from 6 rain-gauge (R) stations and 3 meteorological (M) stations (Figure 1) were used for
modelling: Nitrianske Pravno (R, inside the catchment), Chvojnica (R, inside), Valask Bel - Gapel (R,
outside the catchment boundary), Slovensk Pravno (R, outside), Vricko (R, outside), Zliechov (R,
outside), Prievidza (M, outside), Turianske Teplice (M, outside) and Krna (M, outside). Missing data
were filled in from the nearest station which measured at that time. Four fictive stations were created in
the mountains and near the catchment outlet due to missing meteorological stations in the catchment
and missing measurements at all altitude levels, especially at the highest points of the catchment. The
highest rain-gauge station is located in the Vricko village at 603 m a. s. l., the highest point in the
catchment is the Rev hill at 1204 m a. s.l.. Kriging method was used for spatial distribution of point
measurements. Daily data of observed discharge was taken from Nedoery gauging station. The
simulation period was 27 years, from 1981 to 2007. The entire period was used as calibration period
and manual calibration was based on the Nash-Sutcliffe efficiency. Potential evapotranspiration was
calculated by the Schendel method (Hoelting, 1980) for both local forcing data and WFD. Actual
evapotranspiration was determined by the Krickij-Menkel-Rosinskij method (Hoelting, 1980).

The Nash-Sutcliffe efficiency (Reff) was 0.61 for the model with local forcing data (Horvt and Machlica,
2009) and 0.52 for the model with WFD (middle part of Table 10). Figure 21 shows that the simulation
with WFD is much more peaky that the one with local forcing data. Overall, the FRIER model shows a
good agreement with observations for the low flows.

a)

b)

Figure 21 Discharge (FRIER model): (a) observed and simulated discharge for Nedoery, (b) detail of the
discharge (low-flow range) for the period 19911992.

BILAN model
The local meteorological data used for the BILAN model of the Nedoery catchment are measured at a
number of stations in and around the catchment (Figure 1). Daily precipitations were taken from 6

Technical Report No. 26
31
stations: Nitrianske Pravno, Chvojnica, Valaska Bela, Slovenske Pravno, Vricko, and Prievidza.
Thiessen polygons were used for calculation of mean precipitation. Daily mean temperature from
meteorological station Prievidza was modified according to the mean altitude of the catchment. The
daily mean air humidity was used from the same station. Discharges were taken from the Nedoery
gauging station and they were converted to the runoff depth. The modelling time period was 1981 -
2007. Calibration of the BILAN model was done on time series of observed data (mean daily
discharges) using the standard method of BILAN (Section 3.1.1). For the simulation with local forcing
data the calibration period was 1981 1986, and for the simulation using WFD the calibration period
was 1958 - 1960.

The Nash-Sutcliffe values of the BILAN model are lower than those of HBV-WUR and FRIER (Table 10),
especially for low flows (lnReff). Figure 22 shows that peaks are highly underestimated, especially in the
run with WFD. Low flows are occasionally simulated reasonably well, but most are underestimated and
small peaks during a recession are not reproduced by the model. Overall, the response of the BILAN
model is much too smooth (for explanation see Section 5.2.2. on page 22-23). The simulation with WFD
gives a lower agreement than the simulation with local forcing; some peaks are completely missed.
Differences in the forcing data for the Nedoery catchment (Figure 6) can be part of the cause, but the
short calibration period used is probably the main reason for the low performance of BILAN.

a)

b)

Figure 22 Discharge (BILAN model): (a) observed and simulated discharge for Nedoery, (b) detail of the
discharge (low-flow range) for the period 19911992.

5.4.3. Drought analysis
From the observed and simulated discharge time series, several drought characteristics were
determined with the threshold level method (Section 3.2), i.e. number, mean duration, and mean deficit
of droughts (Table 11). The relative difference between results from simulated and observed discharge
for each characteristic is also included.

For the HBV-WUR model, drought characteristics derived from simulations with WFD (14-49%) are
about equal or a bit closer to those derived from observed discharge than results from simulations with
local forcing data (36-55%). However, the differences between droughts in simulations and observations
are much larger than the differences between droughts in the two simulations with different forcing data.


Technical Report No. 26
32
For the FRIER model, the difference between simulations and observations (22-39%) is smaller than for
the HBV-WUR model
1
. The absolute differences between drought characteristics in observed and
simulated discharge are quite similar for both datasets, but the local forcing data lead to an
underestimation of the number of droughts (and hence an overestimation of duration), while the WFD
simulations overestimate the number of droughts (and underestimate duration). So, the FRIER model
using local forcing data simulates less but more severe drought events than the model run using WFD.

For the BILAN model, difference between drought characteristics derived from simulations and
observations (59-224%) is larger than for the HBV-WUR and FRIER models. The number of droughts is
underestimated and the mean duration and mean deficit are highly overestimated. This is due to the
smooth hydrographs of the BILAN model, in which drought events are not interrupted by small peaks.
The model using WFD performs slightly better than the one using local forcing data.

Drought events in observed and simulated discharge for all models for the period January 1989 to
January 1993 (Figure 23, Figure 24, and Figure 25) show the same pattern as the drought
characteristics. The FRIER model is peaky and shows short drought events and the BILAN model is
smooth and shows long drought events. Both the HBV-WUR and FRIER models give drought events of
similar magnitude in the same period as the observations, whereas the BILAN model does not
reproduce observed drought events or gives droughts when no drought is observed. The differences
between drought events in simulations with both datasets are relatively small for all models, but the
influence of the forcing datasets is clearly visible, for example in 1990 and 1993. This is due to the fast
response of the Nedoery catchment to rainfall.

Table 11 Summary of discharge drought characteristics for Nedoery (number, mean duration, and mean deficit
of droughts) and differences between the two forcing datasets
Period % difference

Number of droughts
(-)
Observed discharge 1974-2001 161
HBV-WUR with local forcing 103 -36.0
HBV-WUR with WFD 102 -36.6
Observed discharge 1981-2006 163
FRIER with local forcing 127 -22.1
Observed discharge 1958-2001 232
FRIER with WFD 300 29.3
Observed discharge 1981-2001* 117
BILAN with local forcing 39 -66.7
BILAN with WFD 48 -59

Mean duration drought
(days)
Observed discharge 1974-2001 12.4
HBV-WUR with local forcing 19.17 54.6
HBV-WUR with WFD 18.44 48.7
Observed discharge 1981-2006 11.07
FRIER with local forcing 13.87 25.3
Observed discharge 1958-2001 13.75
FRIER with WFD 9.86 -28.3
Observed discharge 1981-2001* 11.48
BILAN with local forcing 37.23 224.3
BILAN with WFD 30.17 162.8

1
Note that the FRIER model runs with local forcing and WFD use a different simulation period. Therefore, the drought
characteristics of the simulations are compared with droughts in observed discharge over two different time periods. This can
partly explain the different results for both forcing datasets.


Technical Report No. 26
33

Mean deficit
(mm)
Observed discharge 1974-2001 1.22
HBV-WUR with local forcing 1.77 45.1
HBV-WUR with WFD 1.39 13.9
Observed discharge 1981-2006 0.95
FRIER with local forcing 1.32 38.9
Observed discharge 1958-2001 1.35
FRIER with WFD 1.07 -20.7
Observed discharge 1981-2001* 0.99
BILAN with local forcing 2.97 200
BILAN with WFD 2.45 147.5
* = hydrological year from 1 November to 31 October


a)







b)







c)

Figure 23 Drought events in Nedoery (HBV-WUR model): (a) observed discharge and threshold, (b) simulated
discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.


Technical Report No. 26
34
a)







b)







c)

Figure 24 Drought events in Nedoery (FRIER model): (a) observed discharge and threshold, (b) simulated
discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

a)







b)







c)

Figure 25 Drought events in Nedoery (BILAN model): (a) observed discharge and threshold, (b) simulated
discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

Technical Report No. 26
35
5.5. Platis

5.5.2. Modelling

HBV-TUC model
The local meteorological data used for the HBV-TUC model of the Platis catchment were obtained from
meteorological stations located in and around the catchment. Daily precipitation is measured at three
stations located within the basin and eight surrounding stations. Average daily temperature data of the
nearby meteorological stations were provided by the Hellenic National Meteorological Service. Time-
series of measured potential evaporation for Platis test basin were available from one nearby
meteorological station, at monthly time-step. Mean daily precipitation was derived from IDW
interpolation and daily average temperature was calculated from multiple linear regression for the period
1974-1999. Daily potential evaporation (PET) values were estimated through a combination of a locally
calibrated Blaney-Cridlle equation (Allen and Pruitt, 1986) and adjustment indices from generic daily
WFD PET data. Daily discharge time-series at the outlet of the catchment were used for model
calibration. The first year of the period was used as a warm-up period to initialize model states. The
model was calibrated manually, based on both efficiency criteria lnReff and Reff. More weight was given
to lnReff, to focus on a better performance of low flows (Integrated Hydrological Modeling System, 2006).
No data were preserved for validation purposes, in order to maximize the calibration period and, likely,
calibration reliability.

The HBV-TUC model delivered acceptable calibration results, as Reff for the whole period was 0.76 and
lnReff was 0.80 (Table 12). The simulation with local forcing data better reproduces observed discharge
than the simulation with WFD (Reff = 0.50 and lnReff = 0.72; also visible in Figure 26). Erroneous rainfall
peaks in the WFD cause discharge responses that are not observed, e.g. in the autumn of 1986. The
blocky observations in the low-flow range reflect measurement errors, which are quite common in
catchments with zero-flow situations.

Table 12 Nash-Sutcliffe values for HBV-TUC model with local and WFD forcing for the Platis catchment (grey
columns depict which calibration criterion was used)
Reff lnReff
HBV-TUC local
0.7602 0.8001
WFD 0.5019 0.7201

a)




Technical Report No. 26
36
b)

Figure 26 Discharge (HBV-TUC model): (a) observed and simulated discharge for Platis, (b) detail of the
discharge (low-flow range) for the period 19851986.

5.5.3. Drought analysis
From the observed and simulated discharge time series, several drought characteristics were
determined with the threshold level method (Section 3.2), i.e. number, mean duration, and mean deficit
of droughts (Table 13). The relative difference between results from simulated and observed discharge
for each characteristic is also included.

The number of droughts in the simulated time series correspond well to those in the observations (1
to -13% difference), mean drought duration is overestimated by the simulations (53-87%), and the
simulations of mean deficit are again quite close to observations (-12 to 22%). The simulation with WFD
seems to perform slightly better. This was not according to the expectations based on the Nash-Sutcliffe
values (Table 12) and the hydrographs (Figure 26). This difference is also not spotted in the period
plotted in Figure 27, in which drought events are indicated in red. In the period 1981-1986, drought
events in observed discharge are reproduced by both model runs, except for the one in spring 1983.
Furthermore, the deficit of the large drought in winter 1985-1986 is lower in the simulation with WFD
than in the observations. In the Platis catchment, the forcing data have an influence on the simulation of
discharge peaks and therefore on the drought characteristics.

Table 13 Summary of discharge drought characteristics for Platis (number, mean duration, and mean deficit of
droughts) and differences between the two forcing datasets
Period % difference

Number of droughts
(-)
Observed discharge 1974-1999* 80
HBV-TUC with local forcing 69 -13.8
HBV-TUC with WFD 81 1.3

Mean duration drought
(days)
Observed discharge 1974-1999* 16.26
HBV-TUC with local forcing 30.39 86.9
HBV-TUC with WFD 24.9 53.1

Mean deficit
(mm)
Observed discharge 1974-1999* 3.2
HBV-TUC with local forcing 3.89 21.6
HBV-TUC with WFD 2.83 -11.6
* = 31-8-1974 to 30-1-1999


Technical Report No. 26
37
a)







b)







c)

Figure 27 Drought events in Platis (HBV-TUC model): (a) observed discharge and threshold, (b) simulated
discharge with local forcing data and threshold (c) simulated discharge with WFD and threshold.

5.6. Discussion
The model simulations of the various partners in the WATCH project had a different focus, which
governed the choice of what calibration method to use. If the focus was not (only) on low flows and
drought during the calibration process, but f.e. on floods or hydropower production, like in the NVE-HBV
model, model results for low flows and drought characteristics are less similar to observed low flows and
droughts. For these cases, it is not straight-forward to draw conclusions about the suitability of the WFD
for drought characteristics. However, we considered the differences in calibration procedure in our
performance assessment.

As indicated before (in Section 3.1 and Annex 3), the model parameters of the runs using WFD are
(re)calibrated. This is required as the objective is to test whether large-scale forcing datasets can be
used at catchment scale instead of local forcing data. In catchments where local forcing data is not
available, calibration would also be done as parameter sets can not be taken from a model run using
local forcing. The influence of calibrating or not calibrating on (low) discharges and drought
characteristics has shown to be small (Annex 3). However, difficulties can arise in simulated soil
moisture and groundwater storage, so, preferably, these variables should be used carefully.

For the same reason, the threshold level used for drought analysis (Section 3.2) is separately calculated
for the discharge simulated using local forcing data and discharge simulated using WFD. In catchments
where local forcing data is not available, threshold levels would also need to be calculated and can not
be taken from the simulation with local forcing data. A problem could arise if discharges simulated using
WFD have the same dynamics as the ones simulated using local data, but are shifted up or down. Such
an offset could yield similar values for the drought characteristics (i.e. number of droughts, drought
duration, and deficit), as the threshold level would also be shifted. Therefore, we did not only study the
summary tables of drought characteristics, but also a number of specific drought events. It depends on

Technical Report No. 26
38
the purpose of a drought analysis (e.g. focus on duration or deficits only or also on absolute discharge
values) whether the approach with different thresholds for each time series is valid. In this research, the
use of different threshold levels for model runs using local forcing and model runs using large-scale
forcing is necessary. However, one should bear in mind that if, for example, a pre-defined Ecological
Minimum Flow would be used as threshold for all simulations, results might be different.

Average drought characteristics may give the an incomplete impression of simulated drought events
when simulations are less peaky than observations. More smooth simulated hydrographs result in a low
number of droughts and large duration and deficit values, whereas, for a certain severe drought event,
timing, total duration, and total volume of the event might be similar to those of the observed drought
event. Therefore, it is essential to use a pooling method (Fleig et al., 2006; Tallaksen et al., 1997), or, as
we did in this research, also investigate time series of drought events and compare the most severe
droughts by visual inspection.

Large-scale forcing datasets (like the WFD used in this research) are based on observed meteorological
data of many meteorological stations around the world. These stations are not equally spread, but data-
rich (e.g. North-America and Europe) and data-poor (e.g. Africa and Asia) regions exist. In data-rich
areas, where many meteorological stations are located and much observed data is available, large part
of the observations is incorporated in the large-scale datasets and, therefore, these datasets show a
high reliability. In data-poor areas, the large-scale datasets might be less capable of reproducing local
climate. As little data for ground truthing is available in these regions, the reliability is unknown. In this
research, we studied catchments in a data-rich region. It is unclear, however, which meteorological
stations are used in the WFD for the selected cells. For further research, it is interesting to investigate if
the results of the comparison between drought characteristics simulated with both local and large-scale
forcing would change if the local forcing data used was not incorporated in the large-scale dataset. This
is especially important in data-poor regions, as the large-scale data will be based on less local
meteorological data.


Technical Report No. 26
39
6. Conclusions
In this report, the results of modelling using both local and large-scale forcing data in the WATCH test
basins, i.e. five small, contrasting catchments in Europe, are given. The objective of this study was to
assess the suitability of large-scale forcing data in smaller catchments.

The overall conclusions of this research are:
1. The differences between the WFD and local forcing seem to be acceptable as input for
hydrological model applications (Chapter 4).
2. In all studied catchments and for all models, the difference between simulations and
observations is much larger than difference between simulations with different forcing data
(Chapter 5).
3. In all studied catchments, the difference between simulations with different models is much
larger than difference between simulations with different forcing data (Chapter 5).
4. All models seem to be able to reproduce the most severe events in the observed discharge with
both forcing datasets in all catchments (Chapter 5).

Of course, there are differences between catchments and models:

Narsj catchment (Norway)
- HBV-WUR model: simulations with local forcing and WFD give the same low flow and drought
results, due to:
o High similarity of both datasets
o Calibration focussed on low flows, which mainly occur in winter when the forcing data
have little influence
- HBV-NVE model: drought characteristics of simulations with WFD are more similar to drought
characteristics of observations, than simulations with local forcing data, due to:
o Gridded local forcing data
o Calibration on short period and not focussed on low flows
- (Severe) drought events are mostly captured by both models using both datasets.

Upper-Metuje (Czech Republic)
- HBV-WUR model: simulations with WFD give lower Nash-Sutcliffe values for low flows, but
more similar drought characteristics than simulations with local forcing, due to:
o Differences in the datasets
- BILAN model: simulations with local forcing and WFD give the same low flow and drought
results, but large differences with observations, due to:
o Smooth BILAN hydrographs do not reproduce small peaks in observed discharge which
is caused by the structure of the BILAN model
o Different calibration criterion used
- (Severe) drought events are not always captured by the models using both datasets.

Upper-Szava (Czech Republic)
- HBV-WUR model: simulations with local forcing and WFD give the same low flow and drought
results
- BILAN model: simulations with local forcing and WFD give the same low flow and drought
results, but large differences with observations, due to:
o Smooth BILAN hydrographs do not reproduce small peaks in observed discharge which
is caused by the structure of the BILAN model
o Different calibration criterion used

Technical Report No. 26
40
- (Severe) drought events are captured by both models using both datasets.

Nedoery (Slovakia)
- Simulations with local forcing and WFD give similar low flow and drought results, due to the
high similarity of both datasets
o FRIER model: simulated discharge more peaky than observed discharge
o BILAN model: simulated discharge smoother than observed discharge
o HBV-WUR model: simulated discharge similar to (or slightly smoother than) observed
discharge
- Drought characteristics are best reproduced by FRIER using both datasets and severe drought
events are captured best by HBV-WUR using both datasets.

Platis (Crete)
- Simulations with WFD give lower Nash-Sutcliffe values for low flows, but more similar drought
characteristics (as compared with observations) than simulations with local forcing, due to:
o Differences in the datasets
- (Severe) drought events are captured by the model using both datasets.

In conclusion, this study demonstrates that the large-scale forcing dataset used here (WFD) is suitable
for drought analysis in these small, contrasting catchments in Europe.



Technical Report No. 26
41
References
Allen, R.G., Pereira, L.S., Raes, D., 1998. Crop evapotranspiration : guidelines for computing crop water
requirements. FAO irrigation and drainage papers;56. FAO, Rome.
Allen, R.G., Pruitt, W.O., 1986. Rational use of the FAO Blaney-Criddle formula. Journal of Irrigation &
Drainage Engineering - ASCE, 112(2): 139-155.
Andreadis, K.M., Clark, E.A., Wood, A.W., Hamlet, A.F., Lettenmaier, D.P., 2005. Twentieth-century
drought in the conterminous United States. J. Hydrometeorol, 6(6): 985-1001.
Bergstrm, S., 1976. Development and application of a conceptual runoff model for Scandinavian
catchments; PhD thesis; SMHI Reports RHO No. 7, Norrkping, Sweden.
Bergstrm, S., 1992. The HBV model-its structure and applications. Report RH No. 4. In: Swedish
Meteorological and Hydrological Institute (SMHI) (Ed.), Norrkping, Sweden, pp. 35.
Bergstrm, S., Forsman, A., 1973. DEVELOPMENT OF A CONCEPTUAL DETERMINISTIC
RAINFALL-RUNOFF MODEL. NORDIC HYDROL., 4(3): 1973.
Doherty, J., 2004. PEST: Model-independent parameter estimation, User Manual 5th ed. Watermark
Numerical Computing, Brisbane, Australia.
Doorenbos, J., Pruitt, W.O., 1975. Guidelines for predicting crop water requirements. Irrigation and
drainage paper. FAO;no. 24. FAO, Rome.
Engeland, K., 2002. ECOMAG - Application to the Upper Glomma catchment. In: Department of
Geosciences (Ed.). University of Oslo, Norway.
Fleig, A.K., Tallaksen, L.M., Hisdal, H., Demuth, S., 2006. A global evaluation of streamflow drought
characteristics. Hydrol. Earth System Sci., 10(4): 535-552.
Gidrometeoizdat, 1976. Rekomendatsii po roschotu ispareniia s poverhnosti suchi. In: Gidrometeoizdat
(Ed.), St. Peterburg.
Hisdal, H., Tallaksen, L.M., Clausen, B., Peters, E., Gustard, A., 2004. Hydrological Drought
Characteristics. In: Tallaksen, L.M., van Lanen, H.A.J. (Eds.), Hydrological Drought Processes
and Estimation Methods for Streamflow and Groundwater. Elsevier Science B.V, Developments
in Water Science, 48, pp. pp. 139-198.
Hoelting, B., 1980. Hydrogeologie, Einfuhrunfg in the Allgemeine und Angewandte Hydrogeologie.
Ferdinand Enke Verlag, Stuttgart, 340 pp.
Hohenrainer, J., 2008. Propagation of drought through the hydrological cycle in two different climatic
regions, der Albert-Ludwigs-Universitt Freiburg i. Br. .
Horvt, O., 2007. Parameterization of Hydrologic Processes in Runoff Modelling, Slovak University of
Technology Bratislava, 129 pp pp.
Horvt, O., Machlica, A., 2009. Inter-comparison of modelling results of Bilan and FRIER models in the
upper Nitra catchment, 21st Conference of young scientists, Slovak Hydrometeorological
Institute, Slovak Committee for Hydrology of the IHP, Bratislava, Slovakia, pp. 1-5.
Integrated Hydrological Modeling System, 2006. IHMS v5.10, Swedish Meteorological and Hydrological
Institute (SMHI).
Killingveit, ., Slthun, N.R., 1995. Hydrological models. Hydropower Development 7: Hydrology.
Trondheim: Norwegian Institute of Technology: 99-128.
Lawrence, D., I. Haddeland, Langsholt, E., 2009. Calibration of HBV hydrological models using PEST
parameter estimation.
Machlica, A., Stojkovova, M., 2008. Groundwater drought in different geological conditions, IOP Journal
Conf. Series: Earth and Environmental Science 4, pp. 9.
Mitchell, T.D., Jones, P.D., 2005. An improved method of constructing a database of monthly climate
observations and associated high-resolution grids. International Journal of Climatology, 25(6):
693-712.
Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models part I - A discussion
of principles. J. Hydrol., 10(3): 282-290.

Technical Report No. 26
42
Oosterwijk, J., van Loon, A.F., A. Machlica, Horvt, O., van Lanen, H.A.J., Fendekov, M., 2009.
Hydrological drought characteristics of the Nedoery sub catchment, Upper Nitra, Slovakia,
based on HBV modelling. In: WATCH Technical report No. 20 (Ed.). http://www.eu-
watch.org/nl/25222760-Technical_Reports.html.
Pavlakis, P., 2004. Hydrological study of Platis basin.
Rakovec, O., van Loon, A.F., Horek, S., Kaprek, L., van Lanen, H.A.J., Novick, O., 2009. Drought
analysis for the Upper Metuje and Upper Szava catchments (Czech Republic) using the
hydrological model HBV. In: WATCH Technical report No. 19 (Ed.). Wageningen University,
Wageningen, The Netherlands, T. G. Masaryk Water Research Institute, Prague, Czech
Republic http://www.eu-watch.org/nl/25222760-Technical_Reports.html.
Slthun, N.S., 1996. The Nordic HBV Model.
Schneider, U., Fuchs, T., Meyer-Christoffer, A., Rudolf, B., 2008. Global Precipitation Analysis Products
of the GPCC. In: Global Precipitation Climatology Centre (GPCC) (Ed.).
http://www.dwd.de/bvbw/generator/Sites/DWDWWW/Content/Oeffentlichkeit/KU/KU4/KU42/en/
Reports__Publications/GPCC__intro__products__2008,templateId=raw,property=publicationFil
e.pdf/GPCC_intro_products_2008.pdf, Deutscher Wetterdienst, Offenbach a. M., Germany.
Seibert, J., 2000. Multi-criteria calibration of a conceptual runoff model using a genetic algorithm. Hydrol.
Earth System Sci., 4(2): 215-224.
Seibert, J., 2005. HBV light version 2, User's manual. In: Department of Physical Geography and
Quaternary Geology (Ed.). Stockholm University,
http://people.su.se/~jseib/HBV/HBV_manual_2005.pdf.
Sheffield, J., Wood, E.F., 2007. Characteristics of global and regional drought, 1950-2000: Analysis of
soil moisture data from off-line simulation of the terrestrial hydrologic cycle. Journal of
Geophysical Research-Atmospheres, 112: D17.
Shukla, S., Wood, A.W., 2008. Use of a standardized runoff index for characterizing hydrologic drought.
Geophysical Research Letters, 35(2).
Tallaksen, L.M., Madsen, H., Clausen, B., 1997. On the definition and modelling of streamflow drought
duration and deficit volume. Hydrol. Sci. J.-J. Sci. Hydrol., 42(1): 15-33.
Tallaksen, L.M., van Lanen, H.A.J. (Eds.), 2004. Hydrological drought : processes and estimation
methods for streamflow and groundwater. Elsevier Science BV, Developments in water science;
48, The Netherlands.
Uppala, S.M., Kallberg, P.W., Simmons, A.J., Andrae, U., Bechtold, V.D., Fiorino, M., Gibson, J.K.,
Haseler, J., Hernandez, A., Kelly, G.A., Li, X., Onogi, K., Saarinen, S., Sokka, N., Allan, R.P.,
Andersson, E., Arpe, K., Balmaseda, M.A., Beljaars, A.C.M., Van De Berg, L., Bidlot, J.,
Bormann, N., Caires, S., Chevallier, F., Dethof, A., Dragosavac, M., Fisher, M., Fuentes, M.,
Hagemann, S., Holm, E., Hoskins, B.J., Isaksen, L., Janssen, P., Jenne, R., McNally, A.P.,
Mahfouf, J.F., Morcrette, J.J., Rayner, N.A., Saunders, R.W., Simon, P., Sterl, A., Trenberth,
K.E., Untch, A., Vasiljevic, D., Viterbo, P., Woollen, J., 2005. The ERA-40 re-analysis. Q. J. R.
Meteorol. Soc., 131(612): 2961-3012.
Van Huijgevoort, M.H.J., Anne F. van Loon, O. Rakovec, I. Haddeland, S. Hocek, H.A.J. van Lanen,
2010. Drought assessment using local and large-scale forcing data in small catchments. In:
Servat, E., S. Demuth, A. Dezetter, T. Daniell (Ed.), Global Change: Facing Risks and Threats
to Water Resources. IAHS Publ. No. 340, Fez (Maroc).
Van Lanen, H.A.J., Tallaksen, L.M., Candel, M., Carrera, J., Crooks, S., Engeland, K., Fendekov, M.,
Haddeland, I., Hisdal, H., Horacek, S., Bermdez, J.J., van Loon, A.F., Machlica, A., Navarro,
V., Novick, O., Prudhomme, C., 2008. Database with hydrometeorological variables for
selected river basins: Metadata Catalogue. In: Technical report No. 4 (Ed.). WATCH.
Van Loon, A.F., Fendekov, M., Hisdal, H., Horvt, O., Van Lanen, H.A.J., Machlica, A., Oosterwijk, J.,
Tallaksen, L.M., 2010. Understanding hydrological winter drought in Europe. In: Servat, E., S.

Technical Report No. 26
43
Demuth, A. Dezetter, T. Daniell (Ed.), Global Change: Facing Risks and Threats to Water
Resources. IAHS Publ. No. 340, Fez (Maroc).
Weedon, G.P., Gomes, S., Viterbo, P., sterle, H., Adam, J.C., Bellouin, N., Boucher, O., Best, M.,
2010. The WATCH Forcing Data 1958-2001: a meteorological forcing dataset for land surface-
and hydrological-models. In: WATCH Technical report No. 22 (Ed.). http://www.eu-
watch.org/nl/25222760-Technical_Reports.html.
Wilhite, D.A. (Ed.), 2000. DROUGHT A Global Assesment, Vol I &II, Routledge Hazards and Disasters
Series, Routledge, London.
Yevjevich, V., 1967. An objective approach to definition and investigations of continental hydrologic
droughts. In: Hydrology Papers 23 (Ed.). Colorado State University, Fort Collins, USA.





Technical Report No. 26
44
List of abbreviations

BILAN = "Balance" in Czech
FRIER = Water distribution (Flow, Routing, IUH) model with accent to Evapotranspiration and
Radiation methods
HBV = Hydrologiska Byrns Vattenbalansavdelning model
NVE = Norwegian Water Resources and Energy Directorate, Oslo, Norway
TGM-WRI = TGM Water Research Institute, Prague, Czech Republic
TUC = Technical University of Crete, Greece
UC = Comenius University, Bratislava, Slovakia
WATCH = EU-FP6 Integrated Project Water and Global Change
WFD = WATCH Forcing Data
WUR = Wageningen University and Research Centre, Wageningen, the Netherlands


Technical Report No. 26
i
Annex 1 Total monthly precipitation for four test basins;
original and elevation-corrected by the HBV-WUR model


Figure A1.1 Total monthly precipitation for both forcing datasets in Narsj; original and corrected by the HBV-
WUR model.



Figure A1.2 Total monthly precipitation for both forcing datasets in Upper-Metuje; original and corrected by the
HBV-WUR model.


Technical Report No. 26
ii

Figure A1.3 Total monthly precipitation for both forcing datasets in Upper-Szava; original and corrected by the
HBV-WUR model.



Figure A1.4 Total monthly precipitation for both forcing datasets in Nedoery; original and corrected by the HBV-
WUR model.


Technical Report No. 26
iii
Annex 2 Nash-Sutcliffe values for all models with local
and WFD forcing for all catchments



Narsj Upper-Metuje
Upper-
Szava Nedoery Platis
BILAN local Reff
- 0.4016 0.5583 0.4108 -

lnReff
- 0.5059 0.5363 0.3071 -

WFD Reff
- 0.3973 0.5499 0.2231 -
lnReff
- 0.4892 0.5408 -0.1243 -
FRIER local Reff
- - - 0.6065 -

lnReff
- - - 0.6332 -

WFD Reff
- - - 0.516 -
lnReff
- - - 0.5634 -
HBV-WUR local Reff
0.7691 0.4783 0.6115 0.6639 -

lnReff
0.9046 0.7111 0.6313 0.671 -

WFD Reff
0.7814 0.3489 0.5276 0.6723 -
lnReff
0.8902 0.6509 0.6048 0.7338 -
HBV-NVE local Reff
0.7214 - - - -

lnReff
0.5165 - - - -

WFD
Reff
0.7067 - - - -

lnReff
0.7884 - - - -
HBV-TUC local Reff

- - - 0.7602

lnReff

- - - 0.8001

WFD
Reff

- - - 0.5019

lnReff

- - - 0.7201
(grey columns depict which calibration criterion was used)

Technical Report No. 26
iv

Technical Report No. 26
v
Annex 3 Influence of (re)calibration on drought
characteristics
The objective of this research is to assess the suitability of large-scale meteorological datasets for
drought analysis at catchment scale. To test whether these large-scale forcing datasets can be used
instead of local forcing data (e.g. in catchments where local forcing data are not available), these
datasets should be used in exactly the same way. That means that if a model run using local forcing
data is calibrated, a model run using large-scale forcing data should also be calibrated (recalibration). In
the catchments used in this study, both local and large-scale forcing data are available (which enables
comparison). Consequently, the parameters obtained with calibration using local forcing data, can also
be applied in the model run with large-scale forcing data. This is done only to explore the influence of
(re)calibration on the model results. Of course, this can not be done in real-life situations where
parameters from a calibration with local forcing data are not available.

The HBV-WUR model was developed for four out of five test basins, i.e. Narsj (Norway), Upper-Metuje
(Czech Republic), Upper-Szava (Czech Republic), and Nedoery (Slovakia). For these catchments
three runs were done: 1) using local forcing data, parameters calibrated, 2) using WFD, parameters
(re)calibrated, and 3) using WFD, parameters not (re)calibrated, but taken from run with local forcing
data.

In this Annex the results of this test are presented. The Nash-Sutcliffe values of the runs with WFD that
are not calibrated, are lower than the ones of the calibrated runs (Table A3.1). This applies to all
catchments and for both Reff and lnReff. The hydrographs of Upper-Metuje and Upper-Szava (Figure
A3.1 and A3.2) reveal that low flows deviate slightly between calibrating and not calibrating. For Upper-
Metuje the differences are much larger than for Upper-Szava.

Table A3.1 Nash-Sutcliffe values for the HBV-WUR model with local and WFD forcing (calibrated and not
calibrated) for all catchments except Platis (grey columns depict which calibration criterion was used)


Narsj
Upper-
Metuje
Upper-
Szava Nedoery
HBV-WUR local_calibrated Reff
0.7691 0.4783 0.6115 0.6639

lnReff
0.9046 0.7111 0.6313 0.671

WFD_calibrated Reff
0.7814 0.3489 0.5276 0.6723
lnReff
0.8902 0.6509 0.6048 0.7338

WFD_not calibrated
Reff
0.5583 0.2812 0.4569 0.5897
lnReff
0.8121 0.564 0.5222 0.6761

a)


Technical Report No. 26
vi
b)

Figure A3.1 Discharge of the Upper-Metuje simulated with HBV-WUR: (a) observed and simulated discharge,
(b) detail of the discharge (low-flow range) for the period 19931994.

a)

b)

Figure A3.2 Discharge of the Upper-Szava simulated with HBV-WUR: (a) observed and simulated discharge,
(b) detail of the discharge (low-flow range) for the period 19931994.

The effect of calibration on the discharge drought characteristics (Table A3.2) is small, especially for the
number of droughts and mean duration. Deficit values are slightly more different when the parameters of
the run using WFD are not calibrated, but taken from the run with local forcing. Again, differences exist
between the catchments, e.g. the Upper-Metuje catchment shows the largest differences.

Table A3.2 Summary of discharge drought characteristics for all catchments except Platis (number, mean
duration, and mean deficit of droughts)

Narsj
Upper-
Metuje
Upper-
Szava Nedoery

Number of droughts
(-)
Observed discharge 140 98 210 161
HBV-WUR with local forcing 128 69 124 103
HBV-WUR with WFD calibrated 128 82 111 102
HBV-WUR with WFD not calibrated 123 68 112 106

Mean duration drought
(days)
Observed discharge 23.29 14.59 12.58 12.4
HBV-WUR with local forcing 25.83 21.17 21.6 19.17
HBV-WUR with WFD calibrated 23.78 17.99 25.14 18.44
HBV-WUR with WFD not calibrated 25.74 21.25 24.2 18.95

Mean deficit
(mm)
Observed discharge 6.59 0.93 1.17 1.22
HBV-WUR with local forcing 5.34 1.17 1.19 1.77
HBV-WUR with WFD calibrated 5.52 0.87 1.84 1.39
HBV-WUR with WFD not calibrated 3.66 1.51 1.99 1.83


Technical Report No. 26
vii
Up to now, we only looked at discharge and not at other hydrological variables, like soil moisture and
groundwater storage. Figure shows that calibration can lead to incorrect simulation of the stores of the
model. In Upper-Metuje, soil moisture storage in the calibrated run using WFD (the green line in
Figure A3.3b) has the same dynamics as soil moisture storage in the run using local forcing (the green
line in Figure A3.3a), but much higher values. Groundwater storage in the calibrated run using WFD (the
blue line in Figure A3.3b) is not only much higher than groundwater storage in the run using local
forcing (the blue line in Figure A3.3b), but also has completely different dynamics. The soil moisture
storage and groundwater storage in the not-calibrated run using WFD (Figure A3.3c) are comparable to
the ones in the run using local forcing (Figure A3.3a). So, by calibrating the model, possible errors in the
large-scale forcing dataset are compensated in the stores. This is not the case in all catchments. In
Upper-Szava, for example, both soil moisture and groundwater storage show the same dynamics for
all three runs (Figure A3.4), although slight differences in the exact values are visible.

a)

b)

c)


Figure A3.3 Water balance (HBV-WUR model): P, ET, soil moisture, groundwater, observed and simulated
discharge for Upper-Metuje, for the period 19811985, a) local forcing data, b) WFD calibrated, c) WFD not
calibrated.

Technical Report No. 26
viii

a)

b)

c)


Figure A3.4 Water balance (HBV-WUR model): P, ET, soil moisture, groundwater, observed and simulated
discharge for Upper- Szava, for the period 19811985, a) local forcing data, b) WFD calibrated, c) WFD not
calibrated.

The conclusion of this exercise is that the differences between calibrating and not calibrating are
relatively small and that the soil moisture and groundwater storage of the calibrated model run using
WFD should not be carefully. This is not problematic in this research because our focus is on discharge
droughts, but it limits the application of large-scale datasets for other purposes.

You might also like