Day 3: mHM Calibration

[MuellerSeb]

All task will be done on the test domain 1 from the mHM repository. You can download it with:

mhm-download --verbose --domain 1

This will create a folder called test_domain and you need to modify the namelists inside: test_domain/mhm.nml.

You can run the domain with:

mhm test_domain/

All calibrations will be done with a relatively small budget for the purpose of the exercise and a relatively short modelling period. Set the simulation period as follows:

&time_periods
!> specification of number of warming days [d] and the simulation period.
warming_Days(1)    = 365
!> first year of wanted simulation period
eval_Per(1)%yStart = 1990
!> first month of wanted simulation period
eval_Per(1)%mStart = 01
!> first day   of wanted simulation period
eval_Per(1)%dStart = 01
!> last year   of wanted simulation period
eval_Per(1)%yEnd   = 1990
!> last month  of wanted simulation period
eval_Per(1)%mEnd   = 12
!> last day    of wanted simulation period
eval_Per(1)%dEnd   = 31
/

Make sure you have openMP disabled by adding in the terminal! The domain is too small for efficient parallelization.

export OMP_NUM_THREADS=1

Task 1: Calibration with DDS

Switch on calibration by setting

optimize = .True.

in the mainconfig_mhm_mrm namelist. mHM offers multiple objective functions and calibration algorithms. Choose KGE objective function by setting

opti_function = 9

and DDS optimization by

opti_method = 1

in the mainconfig_mhm_mrm namelist. Please note that DDS is run with relatively small number of iterations. This can be changed by setting nIterations in the Optimization namelist. Run the optimization with a budget of nIterations=100.

You can use mhm-tools to plot a hydrograph of the results. This can be achieved by executing this command in the output_b1 folder of the test-domain:

mhm-tools hydrograph -i ./ -o hydrograph.pdf

Investigate the difference between the KGE of the default run and the optimized parameter. Compare hydrographs of calibrated and default parameters.

Task 3: Choice of seed

The optimization algorithm is sensitive to the seed chosen for the optimizers. What is the difference between using a fixed seed and a seed derived from the system clock. This can be changed by:

seed = -9

in the Optimization namelist.

---

Task 2: Calibration with SCE and NSE as objective function

Similar to the DDS algorithm, mHM also offers the shuffled-complex evolution (SCE) algorithm. This can be chosen by

opti_method=3

To select the Nash-Sutcliffe efficiency (NSE), set

opti_function=1

Use the same number of iterations as for DDS.

nIterations = 100

Compare results.

---

Task 3: Split-sampling test

It is critical for parameter calibration to perform split-sampling test, which is to validate the parameter set derived from a calibration during a period that was not used for calibration. Use the optimized parameter set from Task 1 contained in the file FinalParam.nml and run the model for the period.

!******************************************************************************************
! Time periods (mHM and mRM-related)
!******************************************************************************************
&time_periods
!> specification of number of warming days [d] and the simulation period.
warming_Days(1)    = 365
!> first year of wanted simulation period
eval_Per(1)%yStart = 1993
!> first month of wanted simulation period
eval_Per(1)%mStart = 01
!> first day   of wanted simulation period
eval_Per(1)%dStart = 01
!> last year   of wanted simulation period
eval_Per(1)%yEnd   = 1993
!> last month  of wanted simulation period
eval_Per(1)%mEnd   = 12
!> last day    of wanted simulation period
eval_Per(1)%dEnd   = 31
/

You need to deactivate optimization in order to do a forward run.

Hint ...

optimize = .false.

Also, you need to specify the namelist with the parameters in the mhm call. You can get information on the command line flags with:

mhm --help

Hint ...

mhm -p FinalParam.nml

Take a look with:

mhm-tools hydrograph -i output_b1/ -o hydrograph-cross.pdf

---

Task 4: Multi-domain calibration

Now we want to do a multi-domain calibration and therefore we need to download the second test domain:

mhm-download --verbose --domain 2

This will create a folder called test_domain_2.

Also we need a common namelist to run both domains, so please download the default mhm.nml with this LINK
and store it next to test_domain and test_domain_2 folders.

We also need the parameter and output namelist, that we can copy from the test domain:

cp test_domain/mhm_parameter.nml .
cp test_domain/mhm_outputs.nml .
cp test_domain/mrm_outputs.nml .

Now you can execute mhm for both domains by simply running:

mhm

Please update the mhm.nml file again to do an optimization:

optimize = .True.
opti_function = 9
opti_method = 1

And also adjust the simulation period for both domains to only one year of warmup and one year of evaluation:

&time_periods
  warming_Days(1)    = 365
  warming_Days(2)    = 365
  !> first year of wanted simulation period
  eval_Per(1)%yStart = 1991
  eval_Per(2)%yStart = 1994
  !> first month of wanted simulation period
  eval_Per(1)%mStart = 01
  eval_Per(2)%mStart = 01
  !> first day   of wanted simulation period
  eval_Per(1)%dStart = 01
  eval_Per(2)%dStart = 01
  !> last year   of wanted simulation period
  eval_Per(1)%yEnd   = 1991
  eval_Per(2)%yEnd   = 1994
  !> last month  of wanted simulation period
  eval_Per(1)%mEnd   = 12
  eval_Per(2)%mEnd   = 12
  !> last day    of wanted simulation period
  eval_Per(1)%dEnd   = 31
  eval_Per(2)%dEnd   = 31
/