GRADE 2009

(1)

GRADE

(2)

(3)

GRADE

2009 1202382-005 © Deltares, 2010 Nienke Kramer Hessel Winsemius Otto de Keizer

(4)

(5)

Title GRADE Client Waterdienst Project 1202382-005 Reference 1202382-005-VEB-0002 Pages 84 Keywords

Generator of Rainfall And Discharge Extremes, Extreme value analysis, hydrological modelling

Summary

The project ‘GRADE’ is carried out for RWS Waterdienst in Lelystad. GRADE (Generator of Rainfall and Discharge Extremes) is a new methodology to provide a better physical basis for the estimation of the design discharge of the main Dutch rivers compared to the present method based on frequency analysis of extreme discharge values. GRADE can also be used for the determination of the shape of the design hydrograph, important for the duration of the stress on the river dikes, for the assessment of the impact of scenarios, such as climate change or major changes in the river geometry, as well as for the impact of measures on the design discharge hydrograph on the Rhine and Meuse river in the Netherlands This report describes investigations and developments to the GRADE instrument for the Meuse and Rhine rivers performed in 2009.

Rhine

The new HBV calibration from SMHI, which has been commissioned by BfG in cooperation with Rijkswaterstaat Waterdienst, has been compared to the version currently used in GRADE-Rhine.

The comparison shows that the new calibration results in a large over-estimation of peak flows, which does not make it suitable for application in GRADE. This can be explained by the fact that this calibration has been focused in particular on low flows, whereas the previous calibration was focused on peak flows.

Meuse

The development of GRADE-Meuse is further advanced than GRADE-Rhine and sufficient confidence is obtained now in the GRADE-Meuse instrument for application in the project ‘Wettelijk Toets instrumentarium’ (WTI) in pre-operational mode (Dutch: “schaduwdraaien”) for the determination of the Hydraulic Boundary conditions (HBC). This is evident from the results of various studies carried out in 2009 and which are summarized in this report. These include the study of the newly derived parameter sets with the GLUE methodology that result in a better performance given all criteria. Furthermore, the potential for future improvements in the hydrological component of the instrument is demonstrated. Finally, the effect of new, improved model structures (with respect to HBV) on the trade-off between several hydrograph behaviours (volume, hydrograph shape, and wave irregularity) has been investigated.

An investigation of the wave patterns produced with the new parameter set of HBV for the Meuse shows that with this new set much more realistic patterns can be simulated than with the original (Van Deursen) parameter set. This is a major improvement over the original results that were available before in the year 2008 and also opens the way to start using GRADE directly for the derivation of the wave patterns belonging to the design discharges instead of relying on an upscaling of historical wave patterns as has been the case until now. The main conclusion on Meuse is that the significant improvements of GRADE-Meuse, since introduction of the GLUE parameter sets, give enough confidence to include GRADE-Meuse pre-operationally in WTI.

(6)

(7)

1202382-005-VEB-0002, 18 March 2010, final

1 Introduction

1.1 Background

In the Netherlands, the flood protection situation must be evaluated every 5 years, which includes the evaluation of the Hydraulic Boundary Conditions (HBC) along the Meuse and Rhine branches. For the determination of the HBC, use is made of the 1250-year design discharge at Lobith and Borgharen. The next determination of the HBC in 2011 is carried out in the project “Wettelijk Toets Instrumentarium” (WTI).

The current estimation of the 1250-year design discharges from statistical analyses of the measured peak discharges faces various problems. The estimation of the 1250-year discharge event from statistical information in a discharge record of about 100 years involves a strong extrapolation, and is therefore hampered by a large uncertainty (Figure 1.1).

Q

100 years

Q

1250 years

Extrapolation

Q 1250 T

Q

100 years

Q

1250 years

Extrapolation

Q 1250 T

Figure 1.1 Present method for the determination of design discharges

In 1996 Rijkswaterstaat RIZA, KNMI and WL|Delft Hydraulics started to work together on a new methodology to provide a better physical basis for the estimation of the design discharge of the main Dutch rivers. The new methodology is a combination of various components (Figure 1.2). The first component of this new methodology is a stochastic multivariate weather generator, which generates long synthetic simultaneous records of daily rainfall (P) and temperature (T) records (up to 20,000 years). Evaporation (E) series are derived from the temperature series.

The second component consists of hydrological and hydraulic models, which transform the generated rainfall and temperature records into discharge series.

Altogether, this new methodology is indicated as GRADE: Generator of Rainfall And Discharge Extremes.

(10)

2 of 84 GRADE 1202382-005-VEB-0002, 18 March 2010, final

P,T,E

66 years

P,T,E

20.000 years

Q

20.000 years HBV rainfall generator

Q

20.000 years SOBEK 1250 T Q

P,T,E

66 years

P,T,E

20.000 years

Q

20.000 years HBV rainfall generator

Q

20.000 years SOBEK 1250 T Q

Figure 1.2 Overview of the GRADE instrument

Advantages of the GRADE methodology are that: long discharge records can be simulated,

meteorological conditions and basin characteristics can be taken into account, the shape and duration of the flood can be analysed,

the method can iteratively be improved and uncertainties can be reduced on the basis of new knowledge,

the method can potentially assess the effects of future development like climate change and upstream interventions such as retention basins and dike relocations.

As mentioned in this paragraph for the determination of the HBC’s there is a potential role for GRADE on the following three parts:

1. determination of the design discharge line; 2. determination of shape of the hydrograph; and 3. determination of the hydrograph of the waterlevels.

The GRADE instrument is now in the ‘experimental’ phase. The ambition is to use the instrument for the river Meuse ‘semi-operationally’ for the determination of the HBC 2011. This means that GRADE will be used parallel to the original method and the GRADE results will not be used yet for the final determination of the HBC’s. When the model gives sufficient and reliable results, the model is a serious candidate to be used in the ‘operational’ phase for the period after 2011. The criteria that will be applied to decide whether the instrument is reliable will need to be defined as part of the pre-operational application. This will need to be done in close cooperation with the ‘Expertise Netwerk Waterkering’ (ENW), which carries out the quality control of the WTI programme. For the determination of the proposed improvements to the WTI instruments the following basic principles are used:

• acceptation by managers of the dikes and policy makers; • reproducibility;

• reliability; • consistence;

• user friendliness (applicable, no too laborious, good instructions and helpdesk); • production of the results on time;

• continuity in the HBC’s; • uniformity of the models.

(11)

1202382-005-VEB-0002, 18 March 2010, final

For the GRADE model it is important to guaranty the continuity of the HBC. In the second place, it is important to have a good overview of the reliability of the model in order to approve the system by the managers of the dikes and policy makers.

1.2 GRADE 2007 & 2008 project

The current GRADE 2009 project is the continuation of the former GRADE project, which was carried out in the period 2007-2008 and completed was with a workshop in December 2008. The GRADE 2007 and 2008 project consisted of three parts:

1. configuration of GRADE-Rhine and GRADE-Meuse in Delft-FEWS; 2. analysis of performance of the GRADE-Meuse instrument by:

a. analysis of the shape and duration of the extreme flood hydrograph; b. qualification of the uncertainties in the GRADE instrument.

3. improvement of the GRADE-Meuse model by selection of new parameter sets for the HBV model using a GLUE analysis.

The main findings of the analysis of the performance of the GRADE-Meuse model were: 1. The original (Van Deursen) parameter set of the HBV model underestimates the peak

discharges for the river Meuse.

2. The volume of the wave pattern is significantly larger than the measured wave pattern 3. The observed wave pattern is much more irregular than the form of the simulated peaks. In Figure 1.2 the observed flood wave patterns are shown. The pattern is irregular and most waves have multiple peaks. Figure 1.3 shows the simulated flood wave patterns using the GRADE instrument. All flood waves are smooth and have one single peak.

(12)

Figure 1.4 Generated wave patterns using the GRADE instrument in 2008 with the Van Deursen parameterset. (Roelevink, 2008, HKV) 1

Possible explanations for the difference in volume were identified:

1. The parameter set (Van Deursen, 2004) of the hydrological model is not applicable for peak flows; the simulated peak discharges are too low and too smooth. Consequently the selection method of Roelevink (2008) causes too large volume of the wave, 2. It is difficult to simulate the discharge at the location Borgharen. This is caused by the

presence of the weir upstream of the gauging station and many extractions and inflow points in the surrounding of this location (Figure 1.5) (Weir operation and abstractions are not taken into account in the hydrological model).

Possible explanations for the smooth pattern of the simulated wave at Borgharen were also identified:

1. The geological situation at the location Borgharen (see point 2 in upper section) 2. The daily time step used in the hydrological model. It is known that in tributaries such

as the Ourthe and other small tributaries, the flow is generated by processes that occur at a smaller time scale than a day. This may compromise the ability of the hydrological models to capture the real flow processes during extremes,

3. Effects of the weirs at the Belgian part of the river Meuse,

4. Imperfection of the rainfall generator, mainly the spatial patterns of the discharge events.

1

For this figure, only peak waves having a discharge of 1750 - 2000 (m3/s) have been selected. The discharges are based on 10 synthetic meteorological series of each 3000 years. For the translation of rainfall to discharge the HBV model, including the Van Deursen parameter set, were used.

(13)

1202382-005-VEB-0002, 18 March 2010, final

Julianakanaal

Grensmaas

Zuid

Willemsvaart

stuw Borgharen

sluis

Measurement point Borgharen Maastricht

Julianakanaal

Grensmaas

Zuid

Willemsvaart

stuw Borgharen

sluis

Measurement

point Borgharen

Maastricht

Figure 1.5: location Borgharen

1.3 GRADE 2009 project

In the 2007-2008 GRADE project, the focus has been on the Meuse basin. In 2009 the research on GRADE for the Rhine has been resumed. The rainfall-runoff part of the modelling chain has been analysed, in particular with regard to the performance of the new HBV calibration for the Rhine published early 2009 (Berglöv et al., 2009). A general overview of the modelling chain for the Rhine is also given.

In GRADE 2009 fundamental research for the Meuse has primarily concentrated on the issues brought forward during the 2008 GRADE workshop and has therefore focussed on the evaluation of the performance of the status quo of GRADE compared to the presented GRADE-Meuse version in 2008 (using the Van Deursen parameter set). A second topic was the testing of certain sensitivities within the hydraulic part of the model such as the impact of weirs and their operation on flood hydrographs and the use of the SOBEK hydraulic model in general.

Furthermore, one of the advantages of GRADE is that (unlike statistics based extreme value distributions) it can be further developed by incorporating the knowledge of the hydrological and hydraulic characteristics of extreme events and reduce the uncertainty in the estimation of design discharges. Therefore, investigations into future developments of the hydrology within GRADE have been undertaken in 2009, which have resulted in a number of recommendations for future research and a framework, which can be used to test possible improvements. The results of these investigations are summarized in this Report.

(14)

1.4 Report outline

In detail, investigations have been performed on: 1. GRADE for the Rhine basin, which includes:

a. performance of the new HBV calibration for the Rhine published early 2009. b. general overview of the modelling chain for the Rhine.

This part is described in Chapter 2.

2. GRADE for the Meuse basin, which includes:

a) Hindcast using the HBV model. This hindcast describes the improvements in performance of the HBV model by introduction of the GLUE parameter set. A thorough comparison has been performed between the Van Deursen and the GLUE parameter sets, based on criteria that are deemed important for operational use of the GRADE instrument (i.e. flood peaks, flood volumes and flood wave shape). In this hindcast also the impact of the choice of the location and the use of Sobek is analysed. This assessment is presented in Chapter 3.

b) Analysis to investigate the sensitivity of the timestep of the HBV model. In this analysis the difference between the original simulated daily and simulated hourly flood wave patterns (using the same parameter set, calibrated on daily basis) is investigated. The findings are presented in Chapter 4.

c) Analysis of the sensitivity of the discharge to weir operations (Chapter 5).

3. Investigation of hydrological behaviour of the HBV model and possible future improvements of the hydrological model by enhanced process understanding.

a. Introduction of a new criterion for assessing the model’s capability to reproduce flood wave irregularity

b. Assessment of possible irregularities in the performance of the HBV model. c. Testing of process-based improvements of the hydrological model, based on

the aforementioned criteria. This part is described in Chapter 6.

The conclusions and recommendations formulated on basis of the results form this study are given in respectively Chapter 7 en 8. In these two Chapters the results of the external study by HKV (Barneveld and Van den Berg, 2010) towards the impact of the new HBV hydrological model calibration on the wave pattern produced by GRADE have been incorporated. For details on this study, reference is made to the original report.

(15)

1202382-005-VEB-0002, 18 March 2010, final

2 GRADE for the Rhine basin

Both the international Rhine and Meuse river basins are of major importance for the Netherlands. For the Rhine the design discharge at Lobith, near the German border, is taken as a reference for the Dutch hydraulic boundary conditions (e.g. Diermanse, 2004).

The development of GRADE has been initiated both for the Rhine and Meuse basins. Considering the much smaller size of the Meuse basin, GRADE-Meuse has been put on a faster track during last years, which has resulted in a new calibration parameter set for the HBV model and some improvements on the hydrological modelling side. Although GRADE-Rhine is being constructed based on existing modelling tools for the GRADE-Rhine basin, it has not been documented neither validated yet as a whole.

The objective of the GRADE 2009 Rhine study is to analyse the new calibration of the HBV model for the Rhine basin. Besides, as current knowledge on the hydrological toolbox used in GRADE-Rhine is rather fragmented, also an overview on the status of the whole hydrological modelling chain is provided

2.1 Introduction

GRADE, Generator of Rainfall And Discharge Extremes, has been developed to estimate low probability (high) flows. It is a combination of a stochastic weather generator, which creates synthetic time series based on historical precipitation and temperature, and a hydrological / hydraulic model to facilitate the estimation of low probability discharges (Figure 2.1).

Figure 2.1 Schema of GRADE

GRADE is a collaborative initiative between KNMI, Deltares and Rijkswaterstaat. KNMI is has developed the Rainfall Generator and Deltares the hydrological modelling phase. Here we will focus on the hydrological modelling toolbox that is being developed for the Rhine basin. This “hydrological modelling toolbox” is implemented in FEWS, which is an operational system particular useful for the coupling of different types of models and data management. Current applications include FEWS-NL which is used in operational flood forecasting for the Netherlands, as well as applications for flood and drought forecasting in other countries of Europe and the rest of the world.

Figure 2.2 presents the different modelling steps used in GRADE-Rhine for the estimation of low probability discharge extremes. The synthetic climate series, with a daily time step and in

(16)

general between 3.000 and 20.000 years long, are used as input for the hydrological model HBV that calculates a discharge series of the same length. The (relatively fast) Muskingum model is run to emulate the flow routing, the results being used to identify periods with maxima that require more precise flood routing. Then the selected flood waves are recalculated with the SYNHP routing module for the Rhine upstream from Maxau, and the hydrodynamic SOBEK model for the lower Rhine stretch starting off from Maxau. Subsequently statistical analyses and/or the calculation of low probability return flows (e.g. for a 1250 year return period) can be made for the location of Lobith.

Figure 2.2 GRADE modelling steps (Rhine basin)

The historical climate series for the Rhine that is used is the CHR-climate data set (Hydrological Commission of the Rhine). This data set contains areal precipitation and temperature of the 134 HBV subbasins.

In the following subchapters the hydrological and the hydraulic modelling steps respectively will be described in more detail. Information on the configuration of FEWS-GRADE can be found in Patzke (2007).

2.2 Hydrological modelling

The first step in the modelling chain within FEWS consists of the HBV-model. HBV is a conceptual rainfall runoff model that has been developed in the early 1970’s by SMHI, Sweden. The simple structure of the model makes it attractive for discharge calculations based on meteorological data at river basin level (see Figure 2.3). In 1996 an enhanced version of HBV has been published (Lindström et. al., 1997), which is currently used by Deltares and BfG (Bundesanstal für Gewässerkunde, Germany).

(17)

1202382-005-VEB-0002, 18 March 2010, final

Figure 2.3 Schematic presentation of the HBV-model.

Different phases in the development of the HBV-Rhine model have taken place leading to the version that is currently used (Eberle et al., 2005). This year a new calibration by SMHI has become available, see Table 2.1.

Table 2.1 HBV-Rhine versions

I Daily models for major tributaries Mülders et al. [1999] II Hourly model for the Rhine basin between Maxau and Lobith

(daily model between Basel and Maxau)

Eberle et al. [2001]

III Daily model for the entire Rhine basin upstream of Lobith Eberle et al. [2005]

IV Hourly model for the Rhine basin Berglöv et al. [2009]

The Rhine basin upstream from Lobith has been divided into 134 subbasins and some additional ones with surface area of zero that are only used for flow routing purposes. Each subbasin consists of an HBV model as shown in Figure 2.4 and the resulting flows are routed downstream between the sub basins.

(18)

Figure 2.4 Subbasins in the HBV Rhine model. Red dots mark the main gauging stations (Berglöv et al., 2009)

2.3 New HBV-calibration (2009)

In several test runs of the forecasting systems it was found that the previous HBV model has, in certain ranges, significant shortcomings, with the consequence that predictions of mean and low-flow conditions require manual corrections (Berglöv et al., 2009). Therefore, in 2009 a new HBV-calibration was published by SMHI that was commissioned by the German BfG. As a basis for this calibration Weerts et al. (2008) developed a new interpolation method of precipitation and temperature. Although the objective of the calibration was to improve model performance for mean and low-flow predictions, it was agreed that this should not affect the performance for simulation of peak discharges2.

2

(19)

1202382-005-VEB-0002, 18 March 2010, final

As it was hard to match flood peaks in summer and winter with the same parameter sets, a contributed area approach was applied. Normally, the recession curve of an observed discharge wave is less steep when the basin is wet (in winter) than when it is dry (in summer). This is because in dry circumstances a smaller part of the basin actively contributes to discharge generation. In order to include this effect in the modelling, the contributed area approach was applied in which the areal wetness in the soil routine (SM/fc) is used, letting it represent the contributing area. The percolation parameter PERC (between first and second HBV reservoir) is multiplied with this areal wetness, and the outflow Q0 is divided by this

factor (see Figure 2.5).

Figure 2.5 Principal behaviour of the response routine when the contributing area approach is used (Berglöv et al., 2009)

To evaluate the new calibration for its possible application in GRADE-Rhine, a comparison was made between the two HBV-versions. This has been done using the historical CHR data set. As the new calibration is based on hourly values and the current calibration on daily values, different approaches have been used to make them comparable; the first one being to transfer the new hourly calibration to a daily one using the same approach as in Eberle et al. (2005). In the following description the currently used HBV calibration is named “HBV-2005” and the new one “HBV-2009”.

Figure 2.6 shows a comparison between observed and modelled discharges at Lobith for the flood peak of 1988. HBV-2005 slightly overestimates the peak discharge at Lobith, but the HBV-2009 overestimates this value very strongly.

(20)

Figure 2.6 Flood wave at Lobith for March 1988, comparing both HBV-calibrations with the observed discharge values (new HBV calibration has been adapted to daily values).

Figure 2.7 Flood wave 1995 at Lobith, calculated with both HBV models and compared to measured values (for both calibrations hourly models have been used)

0 5 000 10 0 00 15 000 Date Q (m ³/ s) mrt 07 mrt 17 mrt 27 apr 06 apr 16 Q_m Q_HBV Q_HBV_new

(21)

1202382-005-VEB-0002, 18 March 2010, final

In order to make sure that this difference is not caused by the conversion from the hourly to the daily model of the new calibration set, also the hourly version of the current HBV-model has been compared to the new calibration (see Figure 2.7).

Finally Figure 2.8 shows the 2003 flood wave as taken from the extended calibration report of SMHI. Also here the overestimation of the flood waves can easily be confirmed visually.

Figure 2.8 Flood wave 2003 at Lobith according to Berglöv et al. (2009). Note that the discharge at Maxau has been excluded.

Table 2.2 presents the average discharge over the 1961-1995 period as calculated by each model calibration at different location in the Rhine basin. For Lobith also the average observed flow has been included. Here also an overestimation of the discharge is shown for the new calibration, in particular for the Neckar (Rockenau) and Main (Raunheim).

Table 2.2 Average discharge in m3/s of 1961-1995 period and systematic relative difference between calibrations

Lobith Cochem Rockenau Raunheim Maxau

HBV-2009 2766 396 188 296 1341 HBV-2005 2354 340 141 197 1271 Observed 2299 Relative difference between HBV-calibrations 17% 17% 33% 50% 5%

Now the question is what could cause this very significant overestimation of peak discharges while average and low discharges are calculated sufficiently well.

Different reasons can be thought of, each of which will be analysed in the following paragraphs:

1. Calibration and validation periods

(22)

3. Performance criteria used during calibration 4. Flow routing

Calibration and validation periods

The calibration periods used in the current HBV-2005 version and for different parts of the Rhine basin are presented in Table 2.3.

Table 2.3 Basics for the calibration of different model parts of HBV-2005 (Eberle et. al., 2005)

For HBV-2009, the period from November 2000 until October 2007 has been used for calibration and the period from November 1996 until October 2000 for validation (Berglöv et al., 2009). For some sub-catchments with incomplete time series of recorded discharge, the periods had to be shortened. Note that HBV-2009 has been calibrated on hourly values. Comparing the calibration periods of the HBV-2009 with those of HBV-2005 it becomes evident that these periods are different and do not overlap. During the (shorter) calibration period of HBV-2009 only one major flood wave was included in the calibration period (in 2003) and even this one was much smaller than the 1993 and 1995 peak discharges. This makes it particularly difficult to produce a new parameter set that reproduces faithfully the (extreme) flood waves that will need to be simulated with the GRADE-Rhine instrument.

Discharge data and interpolation technique used for calibration

We will focus here on the precipitation data, as for the lower part of the Rhine basin differences in peak flows are mainly defined by this input.

As explained by Eberle et al. (2005), for HBV-2005, precipitation data for the German part of the Rhine basin are calculated out of REGNIE grid data provided by the German Meteorological Service (DWD). From these data, daily areal precipitation values are calculated by computing the arithmetic mean of the grid values within a subbasin. For the River Moselle basin and the Swiss part of the basin, a different approach was chosen.

(23)

1202382-005-VEB-0002, 18 March 2010, final

The calibration data used for HBV-2009 are based on a new interpolation technique developed by Weerts et. al. (2008). As for operational purposes, a limited amount of operational weather stations is available, using the areal precipitation results in unsatisfactory results with HBV-2005. The new interpolation technique is based on the intention to approach the REGNIE data as much as possible within the operational systems FEWS-NL and –DE. One can conclude that although different calibration data have been used, the very large differences between both calibrations during peak flows cannot be explained by differences in the interpolation as the new interpolation intends to approach the REGNIE data as closely as possible.

Performance criteria

The choice of criteria for defining the quality of simulations depends on the questions the model should help answering. The criteria can be optimised applying predefined objective functions.

For HBV-2005 a calibration of all parts of the HBV model was done to obtain a good overall simulation of the discharge dynamics with some emphasis on the simulation of flood events (Eberle et. al., 2005). The simulation of low flows was not of special interest at the time of calibration.

In addition to comparing hydrographs of computed and observed runoff visually, three quality criteria that are implemented in the modelling software were used during calibration:

• the explained variance according to Nash-Sutcliffe

• the relative accumulated difference between observed and computed discharge

• the relative peak error

Results for gauging stations at the main river are presented in Table 2.4:

Table 2.4 Simulation results from HBV-2005 (Eberle et. al., 2005)

For HBV-2009 the calibration criterion was (Berglöv et al., 2009): crit = 0.5 R2 + 0.5 R2log + 0.1 Relaccdif

Where:

R2 efficiency criterion according to Nash-Sutcliffe

R2log as R2, but using the logarithmic discharge values (gives more weight to low flows)

(24)

Also, but separately, the peak err criterion was evaluated; it is based on one value per full year (i.e. 6 values for the calibration period and 4 for the verification period). The results for those criteria are shown in Table 2.5.

Table 2.5 Results for the calibration and validation for HBV-2009 (Berglöv et al., 2009)

When we only have a look at the peak error at each location, it is easy to see that these are much higher in the new HBV calibration. Note that also the other statistics for the main river are better for HBV-2005, although the R2log, an important (low flow) criterion in HBV-2009,

cannot be compared. Flow routing

In HBV-2005 flow routing has been adjusted as simulated flows during peak discharges were much too high. The option of simulating several branches has been used to model routing during flood. As soon as a certain threshold discharge is exceeded, water starts to run through another branch, usually flowing slower and a certain percentage even getting lost (branches flowing to nowhere) (Eberle et. al., 2005).

In a memo that came with the HBV-2009 calibration the following statement is found: “In the original HBV Rhine set-up the branch option was used to remove and delay water in some of the sub-basins along the Rhine. The reason was assumed to be a decrease in observed discharge values from upstream to downstream stations at high flows. This option was not used in the current set-up as the results seemed acceptable also without it.”

Apparently, flow routing has a critical influence on the discharge peaks. Normally SYNHP and SOBEK are used for the flow routing, skipping the routing within HBV. However, when only HBV is used, as is the case here, its impact on peak discharges is very relevant.

2.4 Hydraulic modelling

As SOBEK has a large calculation time for long time series, the periods with high discharges are selected from the discharge series from HBV. As explained before this is done using a certain threshold on the HBV discharges corrected with the Muskingum flow routing utility. In this way for each tributary a time series is created for each flood wave, which is used as input for SYNHP and SOBEK.

(25)

1202382-005-VEB-0002, 18 March 2010, final

For the area upstream of Maxau (see Figure 4) a SOBEK schematization does not exist yet. For this part of the basin the SYNHP model, administered by the German state Baden Würtenberg, is applied. This model describes the routing of the flood wave through a series of linear reservoirs. A number of configurations exist, being the situation of this reach before (1997) and after initiation of the major river training works in the last century. In GRADE-Rhine the model is run for the situation of 1997 (Patzke, 2007).

The area downstream from Maxau is modelled with SOBEK. Also different versions of the SOBEK model do exist. Details regarding the different versions have not been analysed yet. Using the HBV-output directly as input for the SYNHP and SOBEK models resulted in a significant over-estimation of the flood peaks. To correct for this, correction factors have been introduced (Buiteveld, 2004). Globally, discharge values have been diminished by 5% and the contribution of in-between-catchments has been set to zero. Two possible explanations for the use of these correction factors include (1) not taking into account the effect of groundwater and (2) backwater effects at the tributaries by the models. The last explanation might be important as SOBEK simulates only the main river stretch of the Rhine and does not include the tributaries.

Note that in the current development version of FEWS-GRADE, the factor for Maxau (earlier set at 0.9) has been left out for an unknown reason.

It is important to stress that the corrections factors are only valid during periods of high discharge, so the SOBEK-model within FEWS-GRADE can only be used for the calculation of flood waves. Considering the large influence of these factors on the final results of GRADE we suggest to re-evaluate their values regularly due to possible effects of differences in HBV, SYNHP or SOBEK versions.

(26)

(27)

1202382-005-VEB-0002, 18 March 2010, final

3 Hindcast with the HBV model for the Meuse river

In this chapter, a hindcast is made for the period 1968 to 1998 using the HBV model of the Meuse river. This hindcast describes the improvements in performance of the HBV model by introduction of the GLUE parameterset as defined in the GRADE 2008 project. In this hindcast the performance of the different parameter sets are analysed for different locations:

• Monsin (paragraph 3.2)

• Borgharen (paragraph 3.3)

• Maaseik (paragraph 3.4)

• Lith (paragraph 3.5)

For the simulations at Monsin and Borgharen only the HBV model is used. For Borgharen, Maaseik and Lith use is made of the Sobek model.

3.1 Parameter sets

During GRADE 2007-2008, Deltares has worked on the optimalisation of the parameter sets in the HBV model. A GLUE method was used and new (for the lateral inflows) and longer discharge series are used for the optimalisation of the parameter sets. The results are reported in Kramer et al (2008) and Kramer and Schroevers (2008).

In this chapter the HBV Meuse model with the Van Deursen parameter set (2004) is indicated as HBV old. The parameter sets as selected in Kramer en Schroevers (2008) is indicated as HBV new. The GLUE analysis produces a large number of possible parameter sets. Out of these sets five final sets were selected, which are indicated as HBV5%, HBV25%, HBV50%, HBV75%, HBV95%. For more details reference is made to Kramer and Schroevers (2008).

3.2 Comparison between HBV old and new for Monsin

The discharges simulated with the HBV model are compared with the measurements at the station Monsin instead of Borgharen, because this location is not influenced by upstream weir operations or abstractions. To define the discharge at Borgharen the simulated HBV results for Monsin are reduced with the discharge of the lateral inflow of the Jeker. The (synthetic) observed series of Monsin is calculated as the observations of Borgharen minus the extractions of the branches between Liège and Borgharen (Albertkanaal / Zuid-Willemsvaart / Julianakanaal).

3.2.1 Volume

In Figure 3.1 the discharge regimes for the observations and the simulations at Monsin are given. All models slightly overestimate the discharges in spring and underestimate the discharges in autumn. A possible explanation is that the under- and over estimation is caused by the crude monthly potential evaporation numbers used. It could be that in spring, the potential evapotranspiration is estimated too low, which leads to too wet soils and too high discharges in the model.

(28)

This absolute error is specified in more detail in Figure 3.2. The highest errors are found for the old model. The relative volume error is calculated by the following formula:

1

(

)

n m o i

AVE

Q

where Q discharge (m3/s) i the time step [d]

n the total number of time steps [d] o observed m modelled

Discharge regime

50 100 150 200 250 300 350 400 450 500 550

jan f eb mar apr may jun jul aug sep oct nov dec

dis c har ge (m 3/ s ) Measured HBV old HBV 5% HBV 25% HBV 50% HBV 75% HBV 95%

(29)

1202382-005-VEB-0002, 18 March 2010, final Monsin -40.0 -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0 40.0 50.0 60.0 jan feb m

ar apr may jun jul aug sep oct nov dec

v ol um e erro r (m 3/ s ) old 5% 25% 50% 75% 95%

Figure 3.2 Absolute volume error per month, the volume error is determined by comparing the simulated discharges at the Meuse-Monsin station of the period 1968-1998 with the measurements.

3.2.2 Peaks

In Table 3.1 the averaged yearly maxima for the years 1968 to 1998 are given. In this table two methods are use to calculate the yearly maxima. In the first column, the yearly simulated maxima are given at the day that the yearly maxima were observed. In the second column, the yearly maximum is the discharge of the real yearly maxima (timing of the peaks is not taken into account). For both comparisons, it can be concluded that the new parameter sets perform better in terms of peak simulations than the old HBV parameter set.

Table 3.1 Averaged yearly maxima for the period 1968-1998

Yearly day-maxima at t= tpeak measured Yearly maxima-day (m3/s) deviation (%) (m 3 /s) deviation (%) Measurements 1489 1462 HBV old 1235 -17.0 1313 -10.2 HBV 5% 1405 -5.7 1446 -1.1 HBV 25% 1402 -5.9 1466 0.3 HBV 50% 1464 -1.7 1520 4.0 HBV 75% 1519 2.0 1570 7.4 HBV 95% 1536 3.2 1602 9.6

In Figure 3.3 the yearly maxima are plotted for every year. The figure shows that only for the highest peaks 1993 and 1995 the simulated discharges are higher than the measured discharges. For less extreme situations, the measured discharge (square) is generally situated in between the peaks simulated by the new parameter sets. Furthermore, it can clearly be seen that the old HBV model underestimates the peak flows. Note that the variable performance of the models is also caused by the limited amount of rainfall data available for the simulations.

(30)

Monsin 0 500 1000 1500 2000 2500 3000 3500 4000 1965 1970 1975 1980 1985 1990 1995 2000 (year) di s c har ge (m 3/ s ) old 5% 25% 50% 75% 95% Measured 1993 1995

Figure 3.3 Yearly maximum discharge for the river Meuse at Monsin.

In Figure 3.4, Figure 3.5 and Figure 3.6 the measured and simulated discharges are given for three extreme situations. In both flood situations, the simulated discharges of the new HBV models overestimate the peak discharges. Especially for the simulation with the new HBV model, the peaks of the simulated discharges are earlier (mostly 1 day) than the measured peaks.

Based on a visual comparison the new parameters already give more irregularly shaped hydrographs. However, it cannot be concluded that the new HBV models simulate the extreme flood periods (1993 and 1995) and the extreme low flows (1976) better than the old HBV simulations. December 1993 1000 1500 2000 2500 3000 3500 15/12/93 20/12/93 25/12/93 30/12/93 04/01/94 09/01/94 m3/s Measured HBV old HBV 5% HBV 25% HBV 50% HBV 75% HBV 95%

(31)

1202382-005-VEB-0002, 18 March 2010, final January 1995 0 500 1000 1500 2000 2500 3000 3500 15/01/95 20/01/95 25/01/95 30/01/95 04/02/95 09/02/95 14/02/95 m3/s Measured HBV old HBV 5% HBV 25% HBV 50% HBV 75% HBV 95%

Figure 3.5 The second highest measured peak discharge in the river Meuse at Monsin occurred in December 1995. Low Flow 1976 0 50 100 150 200 250 300 05/03 /76 20/03 /76 04/04 /76 19/04 /76 04/05/ 76 19/05 /76 03/06/ 76 18/06 /76 03/07 /76 18/07 /76 02/08/ 76 17/08 /76 01/09/ 76 16/09 /76 01/1 0/76 16/10 /76 31/10/ 76 m3/s Measured HBV old HBV 5% HBV 25% HBV 50% HBV 75% HBV 95%

Figure 3.6 An extreme low flow period in the river Meuse was in the summer of 1976. In the figure the measured and simulated discharges for the river Meuse at Monsin are plotted.

(32)

3.2.3 Wave pattern

For the determination of the volume of the wave pattern, the 10-day volume of peak events has been calculated. The 10-days volume is important because the longer flood waves that produce several days of flood conditions along the dikes in the lower Meuse river reach in the Netherlands, put a severe stress on the dike systems. In this table, two methods are used to calculate the average of 10 contentious days around the yearly maxima over the period 1969 to 1998 (Yearly-maxima-10-days). In the first column, the simulated yearly-maxima-10-days are chosen at the same dates as observed days. In the second column the yearly-maximum-10 days is chosen at yearly-maximum-10 days around the date of the yearly maxima (timing of the peaks is not taken into account). When comparing the 10-daily volume the old HBV model performs much better than the new HBV models (Table 3.2). The new models produce a higher 10-days volume (4-12 %) than the measured 10-days volume.

Table 3.2 Averaged of 10-days discharge along the peaks of the yearly maxima for the period 1968-1998

Yearly-maxima 10-days at t= tpeak measured Yearly-maxima 10-days (m3/s) Deviation (%) (m 3_/s) Deviation (%) Measurements* 1021 1047 HBV old 1010 -1.0 1073 2.5 HBV 5% 1076 5.4 1107 5.8 HBV 25% 1073 5.1 1087 3.8 HBV 50% 1085 6.3 1114 6.3 HBV 75% 1142 11.9 1150 9.9 HBV 95% 1536 10.9 1135 8.4

*The comparison is made for the following years: 1986-1971, 1973, 1975-1978, 1982, 1984, 1986, 1987, 1990, 1992-1995, 1997, 19983.

3.3 Performance of the HBV parameter sets at Borgharen – with and without SOBEK

Borgharen is a difficult measurement location and it is not easy to simulate the discharge here. This is caused by the presence of the weir upstream and the river diversions and confluences upstream of Borgharen. HBV is strictly a hydrological model and is therefore not able to take into account hydraulic effects such as the river geometry and weirs. By coupling the SOBEK model to the HBV model it is possible to take into account these weirs and river geometry. In Appendix A.2 information is given about the construction of the SOBEK model. In this paragraph the difference between the performance of HBV with the different parameter sets with and without the SOBEK model for the location Borgharen is analysed.

3

The comparisons could not be made for the full period (1968-1998) because of the use of SOBEK. Because SOBEK has a long run time. Therefore, 10 days before and 10 days after the yearly maxima have been simulated. As there were not enough data available to carry out the comparison for all years, the analysis could only be carried out when:

1. the peaks of the yearly maximum are the same peak as the maximum of the hydrological year.

(33)

1202382-005-VEB-0002, 18 March 2010, final

3.3.1 Peaks

In Table 3.3 and Table 3.4 the yearly maxima and the 10-days-yearly maxima are given for the simulations with only the HBV model and those with the SOBEK model included for the hydraulic routing. In this table the method is used that the yearly simulated maxima are given at the day of the real yearly maxima (timing of the peaks is not taken into account). The following conclusions can be drawn:

The SOBEK model gives lower peak values than the HBV model. Due to the lower peak values also the volume of the wave decreases

The performance of the new parameter sets improves when using the SOBEK model in comparison with the single HBV model

The old parameter set performs better for the single HBV run (without SOBEK) than with SOBEK

For the new parameter sets HBV5% and HBV25%, the performance of the yearly maxima is lower when taking into account SOBEK

For the new parameter sets HBV50%, 75% and 95% the performance of the yearly maxima improves when taking into account SOBEK.

Table 3.3 Borgharen HBV (1968-1998).

Yearly maxima Yearly 10 days-maxima (m3/s) deviation (%) (m 3 /s) deviation (%) Measurements* 1541 1095 HBV old 1416 -8.1 1147 4.7 HBV 5% 1604 4.0 1213 10.8 HBV 25% 1629 5.7 1205 10.0 HBV 50% 1705 10.6 1231 12.4 HBV 75% 1752 13.7 1283 17.2 HBV 95% 1799 16.7 1279 16.8

Table 3.4 Borgharen SOBEK (1968-1998).

Yearly maxima Yearly 10 days-maxima (m3/s) deviation (%) (m 3 /s) deviation (%) Measurements* 1533 1090 HBV old 1250 -18.5 991 -9.1 HBV 5% 1412 -7.9 1054 -3.4 HBV 25% 1437 -6.3 1054 -3.3 HBV 50% 1493 -2.6 1057 -3.1 HBV 75% 1547 0.9 1122 2.9 HBV 95% 1590 3.7 1106 1.4

*The comparison is made for the following years: 1986-1971, 1973, 1975-1978, 1982, 1984, 1986, 1987, 1990, 1992-1995, 1997, 19984.

4

The comparisons could not be made for the full period (1968-1998) because of the use of SOBEK. Because SOBEK has a long run time. Therefore, 10 days before and 10 days after the yearly maxima have been simulated. As there were not enough data available to carry out the comparison for all years, the analysis could only be carried out when:

1 the peaks of the yearly maximum are the same peak as the maximum of the hydrological year. 2 the measured yearly maximum discharge falls inside the same peak period as all simulated yearly

(34)

3.3.2 Wave pattern

The wave pattern of the peak events is important to determine the standard wave pattern. The figures below give the simulated wave pattern for several flood periods.

For the simulation shown in Figure 3.8 the old parameter set has been used. Figure 3.9 gives the wave patterns using the GLUE 50% parameter set (HBV 50%). Based on visual comparison the wave pattern generated with the 50% parameter sets is smaller than the pattern based on the old parameter sets. Despite of the smaller wave pattern the volume of the 50% wave is still larger for the 50% set (see previous paragraphs), which is caused by the higher peak waves.

In Figure 3.10 the peaks of the simulations including the HBV50% and SOBEK are plotted. The shape of the peaks are comparable to the HBV50% peaks, however the height of the peaks is higher for the single HBV50% run than for the run which includes SOBEK.

0 500 1000 1500 2000 2500 3000 3500 -8 -6 -4 -2 0 2 4 6 8 time (days) d isch a rg e (m 3/ s) 1970 1980 1984 1985 1986 1988 1991 1994 1995 1999

Figure 3.7 Measured peak discharge river Meuse at Borgharen using daily timestep. In the figure the timescale is shifted to let all peaks coincide.

(35)

1202382-005-VEB-0002, 18 March 2010, final 0 500 1000 1500 2000 2500 3000 3500 -8 -6 -4 -2 0 2 4 6 8 time (days) d is ch ar g e (m3 /s) 1970 1980 1984 1985 1986 1988 1991 1993 1994 1995 1999

Figure 3.8 Old parameter set. Simulated discharge river Meuse at Borgharen using daily timestep. For the simulation the old parameter set has been used. In the figure the shifting of the peaks is equal to the shift of the measured data (Figure 3.8).

0 500 1000 1500 2000 2500 3000 3500 -8 -6 -4 -2 0 2 4 6 8 time (days) d is ch ar g e (m3/ s) 1970 1980 1984 1985 1986 1988 1991 1993 1994 1995 1999

Figure 3.9 HBV50% parameter set. Simulated peak discharge river Meuse at Borgharen using daily timestep. For the simulation the 50% parameter set has been used. In the figure the shifting of the peaks is equal to the shift of the measured data (Figure 3.8).

(36)

0 500 1000 1500 2000 2500 3000 3500 -8 -6 -4 -2 0 2 4 6 8 time (days) d is ch ar g e (m3 /s) 1970 1980 1984 1985 1986 1988 1991 1993 1994 1995 1999

Figure 3.10 HBV50% parameter set + SOBEK. Simulated peak discharge river Meuse at Borgharen using daily timestep. For the simulation in HBV the 50% parameter set has been used. In the figure the shifting of the peaks is equal to the shift of the measured data (Figure 3.8).

3.4 Performance of the parameter sets at Maaseik with SOBEK

In this paragraph the performance of HBV at downstream location Maaseik is analysed. As the HBV model only simulates discharges upstream of Borgharen, the SOBEK model is used to simulate the discharges at Maaseik.

In Table 3.5. the yearly maxima and the 10-days-yearly maxima are given. In this table the method is used that the yearly simulated maxima are given at the day of the real yearly maxima (timing of the peaks is not taken into account). It can be concluded that generally the new parameter sets perform better than the old parameter set. For the yearly maxima and the 10-days maxima the parameter sets between HBV5% and HBV50% perform best.

In Figure 3.11 the yearly maxima are plotted against the accompanying year. It is not possible to draw any conclusions regarding the performance of the parameter sets.

(37)

1202382-005-VEB-0002, 18 March 2010, final

Table 3.5 Maaseik SOBEK

Yearly maxima-day Yearly-maxima 10-days (m3/s) deviation (%) (m 3 /s) (m3/s) Measurements* 1701 1267 HBV old 1479 -13.1 1187 -6.3 HBV 5% 1672 -1.7 1257 -0.8 HBV 25% 1705 0.2 1260 -0.6 HBV 50% 1785 5.0 1270 0.2 HBV 75% 1847 8.6 1344 6.1 HBV 95% 1897 11.5 1332 5.2

*The Waterloopkundig Laboratorium of Vlaanderen kindly provided the discharge data of 1979 tot 1998. Because of limited SOBEK data the analysis is based on a limited number of years (1982, 1984, 1986, 1987, 1990, 1991, 1993-1995, 1997-1998). Maaseik 0 500 1000 1500 2000 2500 3000 3500 1975 1980 1985 1990 1995 2000 (year) d is c h arg e (m 3/ s ) old 5% 25% 50% 75% 95% Measured 1993 1995

Figure 3.11 Yearly maximum discharge for the river Meuse at Maaseik.

3.5 Performance of the parameter sets at Lith

In this paragraph the performance of HBV at downstream location Lith is analysed. As the HBV model does only simulates discharges upstream of Borgharen, the SOBEK model is used to simulate the discharges at Lith.

Dienst Limburg of RWS kindly provided the discharge data at Lith of 1968 tot 1998. However, the Q-H relation is considered unreliable for high discharges.

In Table 3.6 the yearly maxima and the 10-days-yearly maxima are given. In this table the method is used that the yearly simulated maxima are given at the day of the real yearly maxima (timing of the peaks is not taken into account). It can be concluded that generally the new parameter sets perform better than the old parameter set.

(38)

When comparing the yearly maxima and the 10 days maxima the parameter sets 75% and 95% performs best. It should be remarked that the unreliability of the measurements at this location does not allow for any well-founded conclusions and the results are only shown as an indication of the impact of the choice of the parameter sets.

In Figure 3.12 the yearly maxima are plotted against the accompanying year. Also for the location Lith there is a high variety in performance of the model.

Table 3.6 Lith SOBEK

Yearly maxima-day Yearly-maxima 10-days (m3/s) deviation (%) (m 3 /s) (m3/s) Measurements* 1378 1135 HBV old 1143 -17.0 938 -17.4 HBV 5% 1262 -8.4 989 -12.9 HBV 25% 1275 -7.5 989 -12.9 HBV 50% 1314 -4.6 999 -12.0 HBV 75% 1362 -1.1 1054 -7.2 HBV 95% 1389 0.8 1045 -7.9

* The analysis is based on the hydrological years: 1982, 1984, 1986, 1987, 1990, 1991, 1993-1995, 1997-1998

Lith 0 500 1000 1500 2000 2500 3000 3500 1965 1970 1975 1980 1985 1990 1995 2000 2005 (year) d is c ha rge ( m 3/ s ) old 5% 25% 50% 75% 95% Measured 1993 1995

Figure 3.12 Yearly maximum discharge for the river Meuse at Lith.

3.6 Conclusions

Based on the analysis in the foregoing paragraphs the following results are summarized Table 3.7 and Table 3.8). The location Lith is not taken into account because of the unreliability of the measured discharges. The results show a large variation.

(39)

1202382-005-VEB-0002, 18 March 2010, final

Table 3.7 Deviation of the average yearly maximum for the different parameter sets and for the different models.

Monsin HBV (%) Borgharen HBV (%) Borgharen SOBEK (%) Maaseik SOBEK (%) old -10.2 -8.1 -18.5 -13.1 5% -1.1 4.0 -7.9 -1.7 25% 0.3 5.7 -6.3 0.2 50% 4.0 10.6 -2.6 5.0 75% 7.4 13.7 0.9 8.6 95% 9.6 16.7 3.7 11.5

Table 3.8 Deviation of the 10-days yearly maximum of the different parameter sets and for the different models. Monsin HBV (%) Borgharen HBV (%) Borgharen SOBEK (%) Maaseik SOBEK (%) old 2.5 4.7 -9.1 -6.3 5% 5.8 10.8 -3.4 -0.8 25% 3.8 10.0 -3.3 -0.6 50% 6.3 12.4 -3.1 0.2 75% 9.9 17.2 2.9 6.1 95% 8.4 16.8 1.4 5.2

Based on these results, conclusions can be drawn about the influence of the new parameter sets and the use of the SOBEK model on the flood volume (10-daily), the irregular behaviour of flood waves and the design discharge:

1. Peaks:

a) The choice of the parameter sets significantly influences the peak and volume of

the flood waves. The new parameter sets (GLUE) give more satisfactory simulations of peaks than the old (Van Deursen, 2004) parameter set at locations where the impact of hydraulic effects is minimal (i.e. Monsin). The location Monsin gives better results than the location Borgharen because:

o the HBV model is calibrated for Monsin

o HBV is a hydrological model, which does not take into account the Borgharen abstractions/confluences (Juliana channel, Zuid Willemsvaart and Albertchannel) and weirs

o Discharge measurement at Borgharen are not always reliable. The weir and confluences influences the waterlevel.

b) The Sobek model gives lower peak values than the HBV model. The reason for

the additional loss of water in the hydraulic model in not trivial and is being investigated.

(40)

2. Flood volume

a. The volume of the 10-daily wave is dependent on the peak-height of the wave. 3. Irregular behaviour of the wave

a. The choice of the parameter sets have influence on the irregular behaviour of the wave. The old parameter sets has a smoother pattern than the new sets b. The use of SOBEK might have influence on the irregular behaviour of the

(41)

1202382-005-VEB-0002, 18 March 2010, final

4 Comparison hourly and daily timesteps

In an earlier stage of the GRADE project the decision was made to use daily time steps. The motive for this decision was that the rainfall generator in GRADE simulates daily timeseries of thousands of years. Secondly, it is too time-consuming to use an hourly time step both for the main river and its tributaries. In this chapter, the effects of the use of the daily HBV model instead of an hourly model is studied by comparing the simulated discharges using an hourly and daily time step. In the experiments performed in this study, it has been assumed that parameter sets, conditioned on daily rainfall and runoff records, are fully transferable to the hourly time domain, without recalibration.

4.1 Input

Parameter set

In this analysis the mean (50%) set is used, which is determined in the GLUE analysis in 2008. It should be mentioned that the used 50% parameter set was originally derived with a daily model. Weerts (2007) mentions that in case these parameter sets are used in the hourly HBV model the timing of the peaks is incorrect and an adjustment must be made to the MAXBAS parameter in the hourly model to come to better results. According to Weerts the MAXBAS parameter is the only parameter, which is time-dependent. The MAXBAS values of the hourly model are given in Table 4.1. Once these changes in the MAXBAS parameter are made, it was assumed that it is possible to use the hourly model with the other parameters equal to those used in the daily model and arrive at satisfactory simulation results. Note that this is in contradiction to the experience with similar conversions of hourly to daily models in the Rhine basin (see Chapter 5).

Table 4.1 HBV-96 calibration values for MAXBAS (day) for the daily and hourly model

model/basin 1 4 5 8 9 11 12 13 rest daily MAXBAS 3.52 1.2 2.5 2 2.2 1.4 1.1 1.1 1.0 hourly MAXBAS 2.52 0.2 1.5 1 1.2 0.4 0.1 0.1 0.0 Model

To make the hindcast based on hourly data the Hourly HBV-96 model (as used in the operational system FEWS-NL) was used.

For the daily data the Daily-HBV model, which is implemented in the GRADE instrument, was used.

Meteorological data

The input of the HBV model consists of historical precipitation- and temperature series for October 1998 to May 2004. MET-Sethy kindly provided this historical precipitation and discharge data for the Walloon part of the Meuse basin. Meteorological data has also been provided and/or collected by KNMI and the Deutscher Wetter Dienst (DWD). These data are similar to the data used in Weerts (2007). As no evaporation data were available, for the evaporation disaggregated monthly averages have been used. They are considered less important for the simulation of extreme events.

(42)

4.2 Analysis

To compare the daily and hourly discharges hourly meteorological data were used. These hourly data were used to simulate the hourly timeseries using the hourly HBV model. For the daily timeseries, first the hourly data were aggregated to daily timeseries before running the daily HBV model. In Figure 4.1 the simulated discharges using hourly and daily timesteps are plotted. In Appendix B plots are given for the tributaries. Secondly in Figure 4.2 and

Figure 4.3 the discharges of eight peaks in the period 1993 to 2004 are given. The peaks are shifted such that the maximum discharge of the wave is reached at t=0.

When comparing the daily and hourly results the following conclusions can be drawn:

• The timing of the peaks is correct for both the main river and the tributaries

• In the subcatchments Membre Pont, Salzinnes and Teignes the peak flows are

higher for the daily model than for the hourly model. This difference could be due to the conversion of the daily model to the hourly model. Apparently, the conversion can not be done by only adapting the MAXBAS value. The model using the hourly timesteps should be recalibrated

• The results of the daily model have a smoother pattern than the hourly model (see

Figure 4.2 and Figure 4.3). This is expected as many tributaries of the Meuse River are fast responding during floods and using the daily model these responses can possibly not be taken into account. However for both simulations (hour and daily), double peaks are found and it can be concluded that the use of the daily timestep does not cause the smooth- single peak-wave-pattern as found in 20.000 years GRADE simulations (Figure 1.4). The extreme differences in the shape of the hydrographs for the measured and simulated discharge series can not be explained by the choice for either a daily or hourly HBV model.

05/12/000 04/01/01 03/02/01 05/03/01 04/04/01 04/05/01 500 1000 1500 2000 2500 D is c har ge ( m 3 /s) Borgharen (H-MS-BORD)

(43)

1202382-005-VEB-0002, 18 March 2010, final 25/12/010 09/01/02 24/01/02 08/02/02 23/02/02 10/03/02 25/03/02 1000 2000 D is c harge (m 3 /s) Borgharen (H-MS-BORD) 05/12/020 20/12/02 04/01/03 19/01/03 03/02/03 18/02/03 1000 2000 D is c har ge (m 3 /s) Borgharen (H-MS-BORD) measured hour hour day

Figure 4.1 Daily and hourly discharge series for flood events in the river Meuse at Borgharen

-200 -150 -100 -50 0 50 100 150 500 1000 1500 2000 2500 3000 di s c harge(m 3/ s ) time (hours) 02-Nov-1998 27-Dec-1999 04-Mar-2000 07-Jan-2001 14-Feb-2002 04-Jan-2003 20-Jan-2004

Figure 4.2 Simulated discharge at Borgharen using hourly timestep. For the simulation the 50% parameter set has been used. All peaks are shifted such that the maximum discharge of the wave is reached at t=0.

(44)

-8 -6 -4 -2 0 2 4 6 8 0 500 1000 1500 2000 2500 3000 di s c harge(m3/ s ) time(days) 02-Nov-1998 27-Dec-1999 04-Mar-2000 07-Jan-2001 14-Feb-2002 04-Jan-2003 20-Jan-2004

Figure 4.3 Simulated discharge at Borgharen using daily timestep. For the simulation the 50% parameter set has been used.

-2000 -150 -100 -50 0 50 100 150 200 500 1000 1500 2000 2500 3000 d is c ha rge( m 3/ s ) time (hours) 02-Nov-1998 27-Dec-1999 04-Mar-2000 07-Jan-2001 14-Feb-2002 04-Jan-2003 20-Jan-2004 13-Feb-2005

(45)

1202382-005-VEB-0002, 18 March 2010, final

4.3 Conclusion

Results of the simulation of the peaks with the hourly and daily models, both for the main river and its tributaries, are comparable. The timing is good, but there is a difference in the height of the peaks, with the daily model producing higher peaks. The baseflow of the hourly model is also larger. This difference is likely caused by the conversion from a daily model to a hourly model and the conclusion must be made that converting models to another timestep cannot be done with only adapting the MAXBAS value. The calibration of the daily model must be renewed in order to improve the maximum discharges.

From the results of this chapter it can be concluded that the use of the daily model instead of hourly model will influence the height and irregular behaviour of the stimulated wave. In Chapter 6 a more fundamental study is presented towards the impact of the use of hourly and daily data on the behaviour of the stimulated waves.

(46)

(47)

1202382-005-VEB-0002, 18 March 2010, final

5 Influence of weirs upstream of Borgharen on the discharge

5.1 Introduction

The River Meuse is a typical rain-fed river with a strongly varying discharge regime over the year. In order to facilitate shipping more than 70 weirs have been built. In the GRADE instrument the HBV model is used to simulate the discharge in Borgharen. Because HBV is a semi-distributed conceptual rainfall-runoff model the influence of weirs is not taken into account. In this chapter, an analysis is made of the influence of the weirs upstream of Borgharen during a flood period.

First some general information is given about the weirs in the river Meuse. Secondly, the SOBEK application for the Meuse river is introduced. This model can simulate the effect of weirs as it is a one dimensional hydraulic model Finally, three calculations have been made, to analyse the sensitivity of simulated floods with respect to the weirs.

5.2 Weirs in the river Meuse

The mean yearly discharge of the river Meuse is approximately 240 m3/s, but high discharge in the winter periods can cause discharges of tenfold this amount. The French part, located in the south of the basin, is relatively flat with a permeable soil. It reacts relatively slow to precipitation. The Belgian part located in the hilly region of the Ardennes, is characterized by tributaries with steep slopes and rocky soils. This part reacts very quickly to precipitation. Depending on the sub-basin, the flood wave from the Belgian tributaries resulting from a rainfall period has a travel time of 7-14 h to Borgharen.

Water levels in the river Meuse are controlled by weirs, in order to facilitate shipping. During a flood the weirs are all open in order to quickly release the water to the sea. The weir at Borgharen is out of operation from about 1200 m3/s, while the weirs in the Liège reaches are still operating. From 1500 to 2000 m3/s at Borgharen, the weirs in the Liège areas are still in use and the water levels are lowered in the Belgian headwater reaches by about 0.6 meter to create storage capacity in anticipation of the arrival of the flood. Above 2500 m3/s it is difficult to maintain the backwater level there, and above about 2700 m3/s it is impossible to keep the weirs in operation. In the Liège region it is important to keep the weirs in operation as long as possible, in order (1) to generate hydropower for the town Liège and (2) to maintain the water level in the upstream reach of the Albert Canal, the shipping route to Antwerp (Gerretsen, 2009).

In Figure 5.1 the measured discharges of the period November 2001 to March 2002 are given. In the figure it can be seen that during normal flows the line is more irregular than during floods, when all weirs are open. The irregular behaviour during normal flow is also caused by extractions of branches (Albert Canal, Zuid-Willemsvaart and Juliana Canal) just upstream of Borgharen. Because of the shipping a minimum waterlevel in these branches is required during daytime and therefore the extractions fluctuate during the day. These extractions especially have impact on the low flow periods. During peaks the total extractions are only a small percentage of the total flow (30 m3/s of 3000 m3/s) and can therefore be disregarded.

(48)

Borgharen 0 500 1000 1500 2000 2500 3000 01-Nov -01 21-Nov -01 11-Dec -01 31-Dec -01 20-J an -02 09-F eb-02 01-M ar -02 21-M ar -02 Measurements

Figure 5.1 Measured discharge of the total period during a flood event from November 2001 to April 2002.

5.3 SOBEK

The influence of weirs is not taken into account in the HBV model. A possibility to take these weirs into account is to use the SOBEK-river model. The HBV discharges per catchments are used as input for SOBEK. More information about the SOBEK model can be found in Appendix A.1. In this paragraph the HBV results5 for Borgharen are compared with the SOBEK model.

For these calculations use was made of the FEWS-NL operational system. This system includes the SOBEK301 model, which simulates discharges from Chooz at the French-Belgium border to Keizersveer in the Netherlands.

The geography is given by several cross sections. Apart from that the model includes river constructions (e.g. weirs and locks), lateral inflows, riverbed roughness and groundwater interaction. The simulations are made based on hourly time steps6.

In Figure 5.2 measured and simulated discharges at Borgharen for a flood event in February 2001 are given. For this event the differences between HBV and SOBEK model are small; approximately 50 -100 m3/s.

5. For the HBV run the Van Deursen parameter sets are used.

GRADE 2009

GRADE

GRADE

Contents

1

Introduction

Q

Q

Extrapolation

Q

Q

Extrapolation

P,T,E

P,T,E

Q

Q

P,T,E

P,T,E

Q

Q

Julianakanaal

Grensmaas

stuw Borgharen

sluis

Julianakanaal

Grensmaas

stuw Borgharen

sluis

2 GRADE for the Rhine basin

Performance criteria

3 Hindcast with the HBV model for the Meuse river

(

)

AVE

Q

Q

Discharge regime

4 Comparison hourly and daily timesteps

5 Influence of weirs upstream of Borgharen on the discharge