5. Results‎ > ‎

C/ Hydrological Modelling

U
MIKE SHE


MIKE SHE is an integrated hydrological modelling system. It models hydrological process. It transforms rainfall data into runoff with specific coefficients. It considers rivers and lakes.The majority of hydrological processes are represented: rainfall, evapotranspiration, infiltration, runoff, overland flow and groundwater flow. So, this model can simulate the behavior of the hydrologic cycle at the land level. Ameliorations can be added at the calibration phase.

The main purpose of our exercise is to build up a 2D hydrological model using two types of data: 300m resolution data and 75m resolution data. The goal being: comparing the results to the measured discharge at the outlet of the catchment after modifying the values of friction coefficient.

To do this study, the bathymetry data with a resolution of 300 m and 75 m are available. The two models will be simulated to compare the impact of the accurate resolution. The precipitations on the Var catchment are register with gaging stations. Last, the network file of the catchment is available.

There are six steps to obtain the hydrograph. First is to specify the model domain, second is to insert the topography of the catchment, third is to specify the land uses and so, the friction coefficient, fourth is to import precipitation data, fifth is to associate with MIKE 11 and sixth is to calibrate the model.

I. FIRST STEP: MODEL DOMAIN AND TOPOGRAPHY

For this exercise, we are planning to build up a hydrological model for flood event simulation, which means that he simulation is carried out for short period, and the loss due to the evapotranspiration can be ignored comparing to the huge rainfall depth. Therefore, the flood event starts from November 4th 1994 at 6pm (18H00) to November 6th 1994 at midnight (00H00).

For the model domain, we used a prepared dfs2 file named « topography_300 » as the topography for the first part of the exercise. We did the same for the second part of the exercise, by using « topography_75 » as the topography.

II. SECOND STEP: RAINFALL

Using the sections climate configuration and precipitation rate configuration, we used the data of six meteorological stations and assigned it thanks to Thiessen polygon.


After choosing station based as spatial distribution, we selected Grid code (dfs2) as data type, then we used the Thiessen polygon file leading us towards these results:


III. THIRD STEP: MANNING’S VALUE

We had two parameters to modify in this section. The first one is the net rainfall fraction (0.9 uniformly on the area ) which represents the ratio between the net rainfall that causes runoff and the total rainfall.

The second parameter is K, Strickler number and named Manning number on MIKE SHE. It influences the flow and is assigned according to the land use. We used the same Manning number for both exercises: a file of fully distributed map of Stickler values.


IV. FOURTH STEP: CALIBRATION AND COMPARISON OF RESULTS

Here are the results of the variations of discharges during time simulation for the two types of topographies 300m and 75m, according to different net rainfall fraction number (NRF= 0.5, NRF= 0.7, NRF=0.9 and NRF=1).


Here is a table to resume the simulations and the results:

With these graphs and this table, we show that NRF number modifies the results of the hydrographs. First, with a low NRF number, the infiltration is important. So, the quantity of water at the outlet is lower than with a high NRF number.

Second, the peak discharge with a low NRF number is lower than a high NRF number.

Third, the time to reach the peak discharge is later with low NRF number than a high NRF number.

Fourth, the grid resolution has an impact of water quantity, peak discharge and peak of discharge. In average, with the 75 m X 75 m grid, the quantity of water is lower than the 300 m X 300 m grid (as the peak discharge). And more the runoff is important, more the amount of water is important, more the peak is high and less the time to peak of discharge gaps is high between the two grids resolutions.


CONCLUSION

Before the model is set up hydrological assumptions must be made to fully understand and analyze the parameters. For this flood event the soil was already totally saturated and evapotranspiration has been neglected, also there is no interaction between the river and the aquifer. Therefore; the only parameter to which can be varied is the infiltration.



Y
SHETRAN


INTRODUCTION

SHETRAN is a physically-based, spatially-distributed, catchment modelling system. It is closely related to another software using “Système Hydraulique Européen” (SHE) which is MIKE SHE. SHETRAN can calculate different phenomenon as physics of snowmelt, movement of water over lands, transpiration of plants and movement of water within unsaturated and saturated zones. So SHETRAN can be used to predict hydrological impacts of urbanization for example, predict impacts of land changes.

We are given a data set on discharges for the Var River Basin in November of 1994. The purpose of these exercises was to optimize the relationship between the measured data and the discharge output from the SHETRAN model.


I. EXERCISE ONE: CALIBRATION

In order to calibrate, parameters within the SHETRAN model needed to be modified. The model was then run with various parameter changes and the hydrographs for the measured data and the model output were compared. The Nash Sutcliffe-Efficiency (NSE) and Bias were also calculated for each simulation run to have a more quantitative method for analyzing the calibration. A simulation log was kept for each trial which can be found in the Appendix.


I.1. DESCRIPTION OF PARAMETER CHANGES

The parameters that were changed for the simulation runs were: Strickler coefficient, saturated conductivity, and soil depth. For the November 1994 event in the Var catchment, there was a high flow event, and so we wanted the model to be calibrated so that it would represent the high flows well. This gave some insight in to how the parameters should be changed. Higher Strickler coefficient means a smoother surface, and therefore enables more runoff. The lower the saturated conductivity, the less good the soil is at absorbing water, and therefore, the higher the runoff. If soil depth increases, runoff might also increase due to less absorption. For these reasons, we decided that the simulations should be done by (a) Increasing Strickler coefficient, (b) Reducing saturated conductivity, and (c) Reducing soil depth. First, the SHETRAN model was run with no parameters changed to create a comparison for how parameter changes were affecting the calibration. These parameters were first tested individually and then combined appropriately to get the best calibration.


I.2. CALIBRATED MODEL

By increasing the Strickler coefficient, it was found that the model calibration improved significantly and the high flows were much better modelled than when no parameters were changed. Decreasing the saturated conductivity resulted in a small improvement in the calibration, but the effect was not as strong as the change in Strickler coefficient. Increasing the soil depth actually resulted in lower values for the NSE and a higher bias and so the calibration did not improve by changing this parameter. For this reason, we did choose to include any parameter change for soil depth.

The accepted calibrated model had parameter changes of doubling the Strickler’s coefficient and reducing the conductivity by a value of 1. This simulation (2nredK) is shown below in Figure 1. Many other trials of parameters were done, but this proved to better model high flows and also have the best NSE and bias values. Values for these objective functions are shown in Table 1. Figure 1 also shows an additional model, 4n, run where we further increased the Strickler’s coefficient to be multiplied by a factor of 4 from it’s original value. It can be seen that this increases how well the peak is represented, however, this does not well model what is happening over the entire time frame. If we were to use this model, it would show a much longer flood wave, and there would be increased concerns about the duration of flood. While we want to model the peak flow well, it is difficult to do so without overestimating the discharge throughout the entire time frame. Therefore, the simulation chosen to represent the calibrated model was 2nredK.


Figure 1: Select simulations compared against measured flow data in the Var River Basin



II. EXERCISE TWO: LAND USE AND RAINFALL CHANGE

It is important to be able to adapt the SHETRAN model for any changes in the catchment, for example, changes in land use, vegetation, soil, etc. It is important to test what effects these changes may have on flow. A common implication of a change to more urban land, is a result of more runoff and increased flood risk. For this exercise, we used the calibrated SHETRAN model (2nredK) to run simulations of both land use change and precipitation changes in the catchment.


II.1. LAND USE CHANGE

The Var Catchment is mostly forest land cover. We wanted to see what the change in flows would be if we changed 20% of this forested land to urban. We expected to see an increase in flows overall, due to increased runoff. Figure 2 shows the result of the land use change simulation. The flows were actually much less affected than was expected with peak flow only increasing 2.96% (Table 2).


Figure 2: Discharge in Var River Basin for land use and rainfall change scenarios


II.2. RAINFALL CHANGE

The rainfall change simulation ran the calibrated SHETRAN model with all rainfall magnitudes from the six stations in the Var catchment halved. The impact that this has on the flow peak I much larger than the impact the land use simulation had. The flow peak reduces by 84.16%.


II.3. LAND USE CHANGE AND RAINFALL CHANGE

The final simulation ran the 2nredK model with both the reduced rainfall and land use change scenario. The peak flow reduction is 78.47%. We can see from this that changing the rainfall input in to the model has a huge effect on peak flows whereas, in this case of the Var River Basin, changing 20% of the land to urban, has very little effect on increasing the flow peak.


III. APPENDIX: SIMULATION LOG






T
HEC-HMS


INTRODUCTION

The aim of this report is to produce a discharge hydrograph of the 1994 Var flood. First, there will be a work with excel and second, a work with HEC-HMS. HC-HMS is a precipitation-runoff modelling. It models a quantity of water in an outlet (Napoleon’s bridge).


I. FIRST STEP : EXCEL

Input GIS result:

To realise this study, we use data from ArcGis. We have these three tables:



Thiessen Polygon:

In this step, we do hyetograph of each sub-catchment. First, we calculate proportion of importance of each gage by sub-catchment. We obtain this table:


Then, we calculate rainfall with Thiessen parameter in order to find hyetograph for each subcatchment. We represented one hyetograph in Lower Var:


All hyetographs are similar. There are two rainfall but intensity is different. Rainfall on these subcatchments is varied. So, the response of sub-catchments will be separate in the time. But if we compare the peak of precipitation of sub-catchments, we see that the interval hour is the same for all. So the maximum quantity of water will arrive as the same time to the outlet. Consequently, there will be a flood.

Concentration Lag time:

We use four empirical equation in order to compare results. There are “Kirpich” which is appropriate for catchment with high slope, “Pasini” which is appropriate for rural basins in Italy, “SCS lag” which is for SCS method and “Bransby-Williams” which is appropriate for rural basins. First, we calculate concentration time (time for a drop of water to realise the longest way from catchment to outlet).


To calculate “SCS lag”, we must give value of CN (Curve Number). Curve Number is an empirical parameter used for predicting infiltration and runoff. CN number is based on type of land: we use this table to define CN number.


Soil types:

    A _ sand, loess, silts
    B_ shallow loess, sandy loam
    C_ clay loam, soil low in organic content
    D_ saline soils, heavy clay content

We use the column “C” because the soil is constituted of clay (so, not A and B) and there isn’t many salt in soil (so, not D). I fill CN column of next table.


For imperviousness column, we use these values:

            -5 for forest because there is a lot infiltration in this area;
            -20 for agricultural because there is infiltration but less than forest because of action of human;
            -30 for semi-natural because we consider that there are some infrastructures and small village, so there is less                      infiltration;
            -85 for artificial because there are mainly infrastructures;
            -100 for water bodies and wetlands because I consider there are always water in these areas.

Then, we calculate table of proportion of each land use to total sub-catchment area. We divide area
of land use of one sub-catchment by total area of this sub-catchment. We get this table:


We can see that results of four equation are very various. The method’s choice is depending of type of catchment. With HEC HMS, we will use “Kirpich” because our catchment has high slope. Consequently, the concentration time is low. So, the results of Bransby-Williams and SCS lag are too high. These equations are not interested for our catchment. And we don’t use Passini because application is not accurate. To have a good value, we think we should have more equation and we would do an average. In our case, we prefer only use Kirpich method because other results are very different.

SCS method:

It’s an empirical model which use rainfall and give us flow.

A limp ratio and a peaking factor must be defined. We choose “typical” land because there are forest,
agricultural, artificial area and high several altitude.

First, we define time to peak, the time to get peak of discharge with Tp=60.


After, we calculate peak discharge, the higher discharge.


For Esteron, Vesubie, Lower Var, values of discharge are similar whereas for Tinee and Upper Var, they are different (more than 100 m3/s between higher and lower value). Kirpich is central value, it’s why we think it’s the best method for this catchment.

Results figures:

We realise in this step graph for all methods but we only present tables for Kirpich, method that we choose. 

First table is a summary of our results that we already have.


When we realise the graph, we obtain this:


We can see that peak of discharge for observed values is approximately height hours before values calculated. If we compare time interval, we see interval is different. Time begins six hours more lately for observed values than values for calculated values.

We change interval for observed values in add six discharges before. We put same value that first value. We have. In our case, it’s 300 m3/s.



In this case, with the same beginning of interval, peak discharge of measured values is nearly than other peaks. Five other graphs are representation of SCS triangular hydrograph of each subcatchment. If we sum value of peak discharge, we get next graph.



Peak of discharge is equal to 1 561.22 m3/s when we sum all sub-catchment whereas peak discharge of measurement values is 3 680 m3/s. Lake of water can be explicate with not use baseflow. For each sub-catchment, if we add baseflow which is approximately equal to 300 m3/s before flood, we obtain the baseflow of 1 500 m3/s. So, 1 561.22 + 1 500 = 3 061.22 m3/s. Finally, our simulation is pessimistic (-500 m3/s) than observation.

We think, to have a better result for peak value, it would do an average between simulated values and observed values. I get 3 370 m3/s for average peak ((3 061 + 3680)/2).



II. SECOND STEP: HEC-HMS

HEC-HMS is a software for simulate hydrological event.

First step is to fill geographical data of catchment and we put values from excel sheet in HEC-HMS (area of sub-catchment, canopy method is none, surface method is none, loss method is SCS Curve number, transform method is Unit hydrograph, baseflow method is none, we fill curve number, impervious and lag time).

Second step is to add two meteorological models. One is for precipitation: data source is Manual entry, unit is incremental millimetres, time interval is one hour, and time window is duration of scale precipitation. Other is for discharge: Data source is manual entry, unit is cubic meters per second, time interval is one hour and time window is duration of scale discharge. Then, we use meteorological model for rainfall distribution. Shortware is none, precipitation is gage weights,
evapotranspiration is none, snowmelt is none, unit system us metric, replace missing is set to default, include sub-basins is yes. 

After, we fill contribution of gauges by sub-catchment. We put values from Thiessen for depth weight and we put 1 for time weight for all gauges. It means that all gauges are the same importance. Then I create a control specification.

Third step is to run the model.

We get this graph. In blue, it’s simulate discharge and in red, it’s observation of discharge. Difference on the left and right of graph is baseflow. With this simulation, baseflow is not used. It’s why we haven’t got the same discharge.

For the peak of discharge, there are approximately 500 m3/s of difference. Our model over dimensioned (it is optimistic) the discharge and it time of peak is one hour before that observation flow (but it’s acceptable).

It’s difficult to conclude about the good graph because it’s very difficult to give good value of type of flood. Measurement are not effective for these discharges. Mean discharge of Var is 60-70 m3/s. So machine to measure are not calibrated for discharge of 3 000 m3/s.

Simulation is based on precipitation registered on each gauges. It’s the same, we don’t be sure that values are good. So there is incertitude on my results. 

We think, to have a better result for peak value, it would do an average between simulated values and observed values. We get 3 863 m3/s for average peak ((4 047 + 3680)/2).

Furthermore, during the flood, two weirs broke down. So the water storage (just before the weirs) was added to the real rainfall. Observed values didn’t represent net rainfall runoff. In consequence, observed values were over estimated. Finally, peak flood simulated might be overestimated too.


CONCLUSION

This study give us two different hydrographs for flood in Nice in 1994. Compare with observed values (which are not really right because of huge flood and measurement measuring device not calibrated for these type of discharge), hydrograph from Excel is pessimistic and hydrograph from Hec-Ras is optimistic. If we do an average between peak discharges (mean peak) values from the two methods, we find a peak discharge of 3 615 m3/s ((3 370.5+3 863)/2). This value is sensibly the same that observed peak.

Furthermore, to have a better result, we can adjust parameters. For example, we can choose other CN numbers. It’s depend on soil type, it’s difficult to be accurate on this parameter. It’s the same for imperviousness, each land use has its personal value and it’s an approximation to give this kind of value. Moreover, equation to determine lag time is tricky. Choice of equation depends on land use and topography of soil. We think the best solution to have appropriate value is to do an average between many methods. In our case, four methods are not enough to do a good value. Maybe we can use ten methods and realise average. It would be more representative. Likewise, in our model, we have not used baseflow. With baseflow, we have a better representation of knowledge of flood. Then, for the limp ratio and the peaking factor, I used typical value. These data are an important eight in our results because we use them to find recession time and peak discharge.

To finish, modelling is complicate. To have best model, we should use average of the most possible of models in order to test maximum of possibilities and to get closer of reality.
Ċ
jonathan PLISSON,
17 Feb 2017, 10:40
Ċ
jonathan PLISSON,
17 Feb 2017, 10:40
Ċ
jonathan PLISSON,
17 Feb 2017, 10:40
Comments