5. Model calibration
To calibrate a specific model, CREST or SAC-SMA, you need to set up the calibration parameters range in the control file. The calibration process will optimize the parameters to minimize the difference between observed and simulated streamflow. EF5 uses the DREAM method for optimization. You can specify the objective function -CC, NSCE, SSE-, which are statistic metrics to evaluate the model performance. The possible parameter ranges for calibration are also defined in the configuration file. The following is an example of how to set up the calibration parameters in the configuration file:
Lumped (aggregated) vs distributed (grided) parameters.
Take a look at the lumped and distributed description in the Coltrolfile section for more details about the difference between lumped and distributed parameters.
Table: Kinematic wave lumped parameters definition and possible ranges for calibration.
Parameter |
Definition |
Ref min val |
Ref max val |
|---|---|---|---|
alpha0 |
Used for overland, not channel, routing |
0.01 |
5 |
alpha |
Used in the equation Streamflow = alpha*(cross-sectional channel area)^beta |
0.01 |
3 |
beta |
Used in the equation Streamflow = alpha*(cross-sectional channel area)^beta |
0.0001 |
1 |
under |
Interflow flow speed multiplier |
0.0001 |
3 |
leaki |
Amount of water leaked from interflow reservoir in each time step |
0.0001 |
1 |
th |
Number of grid cells needed to flow into a cell for it to be part of a channel |
30 |
300 |
isu |
Initial value of the interflow reservoir |
0 |
0.0001 |
Table: CREST lumped parameters definition and possible ranges for calibration.
Parameter |
Definition |
Ref min val |
Ref max val |
|---|---|---|---|
wm |
Maximum soil water capacity |
5 |
600 |
b |
Exponent of the variable infiltration curve (VIC) |
0.1 |
20 |
im |
Impervious area ratio |
0.01 |
0.5 |
ke |
Multiplier to convert between input PET and local PET |
0.001 |
1 |
fc |
Soil saturated hydraulic conductivity (Ksat) in mm/hr |
0 |
150 |
iwu |
Initial value of soil water |
2.5 |
250 |
5.1. Calibration output
The calibration process will generate an output file, “cali_dream.gauge_name.crest.csv”, that contains the optimized parameters, at the end of the .csv file, and the objective function value. The output will be saved in the specified directory.
The optimized parameters will be saved in the following format:
wm,b,im,ke,fc,iwu,under,leaki,th,isu,alpha,beta,alpha0,nsce,nsce/2
1.243439,1.027347,0.028567,2.212074,0.690118,232.437332,0.002938,0.339752,12.009190,13.275298,1.238962,0.246814,2.164778,-8814420.000000,-4407210.000000
...
...
...
[WaterBalance]
wm=2.861236
b=1.986400
im=0.016121
ke=2.344525
fc=1.774454
iwu=281.206879
[Routing]
under=0.002960
leaki=0.707585
th=12.006120
isu=19.988890
alpha=2.894948
beta=2.375221
alpha0=2.960580
Warning
Common EF5 error message in this step:
This error message could be related to the gauge basin area verification, but it is also caused by a memory overload. Adding more RAM to your job would solve this problem.
5.2. Parameters’ sensitivity
The following image shows the parameters’ sensitivity in the simulated time series. Each parameter is varied while the others are kept constant. The sensitivity analysis helps to understand how each parameter influences the model output.
Parameters sensitivity analysis. Blue line represents the observed streamflow, the read lines goes from the lowest to the highest value of each parameter, where thickness indicates the increasing value of the parameter.
5.3. Control file example for Calibration
The following is an example of a control file for calibrating the CREST model with Kinematic wave routing. The calibration parameters ranges are set up according to the tables above. The task section specifies the calibration process using the DREAM method.
[Basic]
...
[PETForcing CLIMO]
...
[PrecipForcing MRMS]
...
[Gauge 03404900]
lon=-84.093599999999995
lat=36.951400000000000
obs=data/observations/usgs/Q_03404900.csv
[Basin 0]
gauge=03404900
# The following code is used for the simulation process.
# It is kept here to replace the optimized parameters later and run the simulation.
[CrestParamSet EF5KY]
gauge=03404900
wm=9.883508
...
# The following code is used for the simulation process.
# It is kept here to replace the optimized parameters later and run the simulation.
[KWParamSet EF5KY]
gauge=03404900
under=0.000100
leaki=5.144720
...
[CrestCaliParams 0CRESTCALI]
# To set up the following ranges, please refer to the above table
gauge=03404900
objective=CC # Possible options: CC, NSCE, SSE
dream_ndraw=20000
wm=0.05,10
b=0.05,10
im=0.005,1
ke=0.001,1
fc=0.0,150.0
iwu=0.0,300.0
[kwcaliparams 0KWCALI]
# To set up the following ranges, please refer to the above table
gauge=03404900
under=0.0001,0.00010001
leaki=0.02,10.0
th=10,10.00001
isu=0.0,0.000001
alpha=0.05,10
beta=0.05,10
alpha0=0.05,10
[Task TaskCalibration]
STYLE=CALI_DREAM
MODEL=crest
ROUTING=KW
BASIN=0
PRECIP=MRMS
PET=CLIMO
OUTPUT=outputs
STATES=data/states
defaultparamsgauge=03404900
PARAM_SET=EF5KY
ROUTING_PARAM_Set=EF5KY
CALI_PARAM=0CRESTCALI
ROUTING_CALI_PARAM=0KWCALI
TIMESTEP=2u
TIME_BEGIN=20220727120000
TIME_END=20220730120000
# The following code is used for the simulation process.
# It is kept here to replace the optimized parameters later and run the simulation.
[Task CREST_Simulation]
...
[Execute]
task=TaskCalibration
#task=CREST_Simulation # Comment this line, and then, after the calibration, update the parameters
# with the optimized values and un-comment it to run the simulation
Warning
🚧 This documentation is currently under active development. Content may change frequently and some sections may be incomplete.