I.               Introduction

SHETRAN is a physically based hydrological modelling software derived from SystemeHydrauliqueEuropeen (SHE). It is based on an understanding of the physics of the processes controlling catchment response to precipitations, and provides spatially-distributed models that allow spatial variation of catchment characteristics.

This report shows how Team 1 learned how to build a hydrological model using SHETRAN for flood event simulations. The flood event used for this study occurred in 1994 on the Var region. Section 2 discusses how the model was calibrated with a hydrograph recorded during the flood event. Section 3 discusses the impact of land use and rainfall data on the simulation output.


II.            Calibration of the SHETRAN model

Calibration is “the procedure of adjustment of parameter values of a model to reproduce the response of reality within the range of accuracy specified in the performance criteria” (Newcastle University, 2016). In order to build a model where the output discharge are close to the observed discharge the model needs to be calibrated. Section II.A discusses the manipulation of input parameters (saturated hydraulic conductivity, soil layer depth, and surface roughness coefficient) and the impact on the model. Section II.B discusses model optimisation to obtain a simulated hydrograph close to the observed discharge of the 1994 flood event.


A.             Manipulation of input parameters

1.              Saturated hydraulic conductivity (K)


The first input parameter to be manipulated is saturated hydraulic conductivity, which represents the capacity of water to infiltrate the soil when it is saturated. The simulation was run with three different values for K: 0.5 m/d, 4 m/d and 15 m/d. The resulting hydrographs are shown in Figure 1. It is observed that as K increases the runoff decreases, as expected, and that peak discharge is smoothed.

Figure 1 : Hydrograph showing observed and modelled flood discharge (m3/s) for different K values. 


2.              Soil Layer Depth


The soil layer depth represents the storage capacity of the soil before saturation. Simulations were run with three different values: 0.5 m, 10 m and 20 m. The resulting hydrographs are shown in Figure 2. It is observed that as soil depth increases runoff and peak flow decreases. This is due to the spare water capacity of the soil which is greater when the soil level depth is greatest. There is a direct link between soil layer depth and saturated hydraulic conductivity, as the soil layer depth can be considered as a porous tank and the saturated hydraulic conductivity as permeability within this tanks that allows water to flow to the river as groundwater flow.

Figure 2 : Hydrograph showing observed and modelled flood discharge (m3/s) for different K values. 


3.              Surface roughness coefficient


Strickler coefficient values are used in SHETRAN to represent the roughness of the surface. Strickler coefficient is the inverse of Manning’s n value for surface roughness. It is common practice to use Manning’s n values calculated by Engman (1986) for different surface types. For this study the surface roughness was changed only for forests, as this is the dominant land cover in the catchment. Simulations were run with three different surface roughness coefficients for forest: 1 m1/3.s-1, 5 m1/3.s-1 and 8 m1/3.s-1. The resulting hydrographs are shown in Figure 3. It is observed that as the coefficient increases the peak flow increases, as a higher coefficient represents a smoother surface therefore less obstructions to overland flow. 


Figure 3 : Hydrograph showing observed and modelled flood discharge (m3/s) for different surface roughness coeff values.


B.              Model optimisation


In order to construct a model which can be used to accurately simulate different scenarios calibration is required by manipulating all input parameters discussed in Section II.A. There are many sources in literature which contain common values for the parameters manipulated in this study (e.g. Engman, 1986, Clapp and Hornberger, 1978), and suggested values for saturated conductivity are presented in the SHETRAN Data Requirements documentation (Newcastle University, undated). Examples of saturated hydraulic conductivity for different soil types are shown in Table 1 and 2.

Table 1 : Strickler coefficient ranges takenfrom Mémoire d’hydraulique à surface libre G.Dégoutte (2002)

Table 2 : Saturated hydraulic conductivity for different soil types taken from Clapp and Hornberger (1978).

3 simulations were run using parameters from literature. It was understood that the Strickler coefficient of the forest land cover is key due to the the large proportion of catchment being covered by this land type. Therefore three Strickler coefficient values were used for the forest: 4 m1/3.s-1, 6 m1/3.s-1 and 7 m1/3.s-1, with the input for other parameters show in Table 3. The resulting hydrographs are shown in Figure 4. It was decided to keep the model with forest surface coefficient value equal to 7 as this has the best fit to the observed discharge of the 1994 flood event.

Table 3 : Input parameters for optimisation of model

Figure 4 : Hydrograph showing observed flood discharge (m3/s) and optimised modelled discharge (m3/s)


III.         Impact of land use and rainfall data on simulation output


Using the optimized calibrated model discussed in Section II.B, an analysis was made on the impact of land use and rainfall on modelled discharge. This section discusses how changes were made to the model, what impact this had on modelled discharge and whether the impacts are expected based on hydrological understanding.

Figure 5 : Hydrograph showing observed flood discharge (m3/s) and optimised modelled discharge (m3/s) 


A.             Land Use Change


To assess land use change 25% of the land use within the catchment was changed from Forest to Urban. The change was made by altering the land value of 25% of the catchment area from value “3” to “1” in the “LandCover-map-var2km.txt” file. A simulation was run using the optimized model, and the output hydrograph is shown by the dashed green line in Figure 4. Although not clear on the hydrograph, peak discharge is 3% higher in the urbanized model, with a slightly steeper rising limb than the calibrated model.

It is expected that urbanization leads to increased runoff and faster time to peak flow. Unfortunately this is not easily shown graphically in this study, despite multiple patterns used to urbanize 25% of the catchment.

B.              Rainfall Change


To assess rainfall change input rainfall was halved in the “Rainfall-6stations-hourly-1-5Nov-94.xls” file. A simulation was run using the optimized model, and the output is shown by the blue line in Figure 4. It is observed that halving the rainfall results in a 90% decrease in peak flow, and a delay of c. 12 hours in the start of discharge rise between the halved rainfall model and optimized model.

The results show that the catchment has storage capacity which acts as a buffer between peak rainfall and peak discharge, as expected especially given the dominance of forest land cover. When low is low, water is stored and slowly moves to the river causing a small increase in discharge. Conversely, when rainfall is high and storage is overwhelmed, discharge increases rapidly and causes flooding such as that observed in 1994.

C.              Land Use and Rainfall Change


The effect of both land use and rainfall change was simulated by running the optimized model with updated land use map and updated input rainfall data. The output of the simulation is shown by the orange line in Figure 4. It is observed that the change in land has much more of an effect on discharge when rainfall is halved, with discharge rising earlier and more rapidly. This fits better with the expectation that urbanization leads to faster run off and higher peak flows. Peak discharge in this case is 8% higher than in the model where rainfall is halved but urbanization does not occur.

The two-stage rise in discharge in this simulation may be caused by the methodology by which urbanization was modelled. In the land use text file, entire rows of input values were changed, and this may lead to a non-uniform rising leg as water intercepts multiple urban and forested areas on its modelled path to the river.


IV.          Conclusion


The purpose of this study was to understand how to calibrate, optimize, and analyze the impact of changing inputs on a physically based hydrological model built using SHETRAN software.

Section II discussed how input parameters saturated hydraulic conductivity, soil layer depth and surface roughness coefficient can be individually manipulated to calibrate the model by comparing observed and modelled hydrographs. These best values for each parameter were then combined in order to create an optimized model which could be used for analysis of different physical inputs. Section III discussed the use of the calibrated model to assess the impact that changes in land use and rainfall would have on the discharge of the catchment.

It is concluded by Team 1 that SHETRAN is a powerful hydrological modelling tool, and that once calibrated it can be used to model the impact of many different changes in input. However, it is noted that care must be taken at all stages, and that hydrological reasoning must be applied to both the input and outputs when using the SHETRAN software