5.13.2-dev0
 |
mHM
The mesoscale Hydrological Model
|
|
mHM
The mesoscale Hydrological Model
|
Objective Functions for Optimization of mHM/mRM against runoff. More...
Functions/Subroutines | |
| real(dp) function, public | single_objective_runoff (parameterset, eval, arg1, arg2, arg3) |
| Wrapper for objective functions optimizing agains runoff. | |
| real(dp) function, public | single_objective_runoff_master (parameterset, eval, arg1, arg2, arg3) |
| Wrapper for objective functions optimizing agains runoff. | |
| subroutine, public | single_objective_runoff_subprocess (eval, arg1, arg2, arg3) |
| Wrapper for objective functions optimizing agains runoff. | |
| subroutine, public | multi_objective_runoff (parameterset, eval, multi_objectives) |
| Wrapper for multi-objective functions where at least one is regarding runoff. | |
| real(dp) function | loglikelihood_stddev (parameterset, eval, stddev, stddev_new, likeli_new) |
| Logarithmic likelihood function with removed linear trend and Lag(1)-autocorrelation. | |
| real(dp) function | loglikelihood_evin2013_2 (parameterset, eval, regularize) |
| Logarithmised likelihood with linear error model and lag(1)-autocorrelation of the relative errors. | |
| real(dp) function | parameter_regularization (paraset, prior, bounds, mask) |
| TODO: add description. | |
| real(dp) function | loglikelihood_trend_no_autocorr (parameterset, eval, stddev_old, stddev_new, likeli_new) |
| Logarithmic likelihood function with linear trend removed. | |
| real(dp) function | objective_lnnse (parameterset, eval) |
| Objective function of logarithmic NSE. | |
| real(dp) function | objective_sse (parameterset, eval) |
| Objective function of SSE. | |
| real(dp) function | objective_nse (parameterset, eval) |
| Objective function of NSE. | |
| real(dp) function | objective_equal_nse_lnnse (parameterset, eval) |
| Objective function equally weighting NSE and lnNSE. | |
| real(dp) function, dimension(2) | multi_objective_nse_lnnse (parameterset, eval) |
| Multi-objective function with NSE and lnNSE. | |
| real(dp) function, dimension(2) | multi_objective_lnnse_highflow_lnnse_lowflow (parameterset, eval) |
| Multi-objective function with NSE and lnNSE. | |
| real(dp) function, dimension(2) | multi_objective_lnnse_highflow_lnnse_lowflow_2 (parameterset, eval) |
| Multi-objective function with NSE and lnNSE. | |
| real(dp) function, dimension(2) | multi_objective_ae_fdc_lsv_nse_djf (parameterset, eval) |
| Multi-objective function with absolute error of Flow Duration Curves low-segment volume and nse of DJF's discharge. | |
| real(dp) function | objective_power6_nse_lnnse (parameterset, eval) |
| Objective function of combined NSE and lnNSE with power of 5 i.e. | |
| real(dp) function | objective_kge (parameterset, eval) |
| Objective function of KGE. | |
| real(dp) function | objective_multiple_gauges_kge_power6 (parameterset, eval) |
| combined objective function based on KGE raised to the power 6 | |
| real(dp) function | objective_weighted_nse (parameterset, eval) |
| Objective function of weighted NSE. | |
| real(dp) function | objective_sse_boxcox (parameterset, eval) |
| Objective function of sum of squared errors of transformed streamflow. | |
| subroutine, public | extract_runoff (gaugeid, runoff, runoff_agg, runoff_obs, runoff_obs_mask) |
| extracts runoff data from global variables | |
Objective Functions for Optimization of mHM/mRM against runoff.
This module provides a wrapper for several objective functions used to optimize mRM/mHM against runoff.
If the objective contains besides runoff another variable like TWS move it to mHM/mo_objective_function.f90. If it is only regarding runoff implement it here.
All the objective functions are supposed to be minimized!
COPYING and COPYING.LESSER provided with this software. The complete GNU license text can also be found at http://www.gnu.org/licenses/. | subroutine, public mo_mrm_objective_function_runoff::extract_runoff | ( | integer(i4), intent(in) | gaugeid, |
| real(dp), dimension(:, :), intent(in) | runoff, | ||
| real(dp), dimension(:), intent(out), allocatable | runoff_agg, | ||
| real(dp), dimension(:), intent(out), allocatable | runoff_obs, | ||
| logical, dimension(:), intent(out), allocatable | runoff_obs_mask | ||
| ) |
extracts runoff data from global variables
extracts simulated and measured runoff from global variables, such that they overlay exactly. For measured runoff, only the runoff during the evaluation period are cut, not succeeding nodata values. For simulated runoff, warming days as well as succeeding nodata values are neglected and the simulated runoff is aggregated to the resolution of the observed runoff. see use in this module above
| [in] | integer(i4) :: gaugeId | current gauge Id |
| [in] | real(dp), dimension(:, :) :: runoff | simulated runoff |
| [out] | real(dp), dimension(:) :: runoff_agg | aggregated simulated |
| [out] | real(dp), dimension(:) :: runoff_obs | extracted measured |
| [out] | logical, dimension(:) :: runoff_obs_mask | mask of no data values |
Definition at line 2478 of file mo_mrm_objective_function_runoff.F90.
References mo_common_mhm_mrm_variables::evalper, mo_mrm_global_variables::gauge, mo_mrm_global_variables::nmeasperday, mo_common_mhm_mrm_variables::ntstepday, and mo_common_mhm_mrm_variables::warmingdays.
Referenced by loglikelihood_evin2013_2(), loglikelihood_stddev(), loglikelihood_trend_no_autocorr(), multi_objective_ae_fdc_lsv_nse_djf(), multi_objective_lnnse_highflow_lnnse_lowflow(), multi_objective_lnnse_highflow_lnnse_lowflow_2(), multi_objective_nse_lnnse(), objective_equal_nse_lnnse(), objective_kge(), mo_objective_function::objective_kge_q_bfi(), mo_objective_function::objective_kge_q_et(), mo_objective_function::objective_kge_q_rmse_et(), mo_objective_function::objective_kge_q_rmse_tws(), mo_objective_function::objective_kge_q_sm_corr(), objective_lnnse(), objective_multiple_gauges_kge_power6(), objective_nse(), objective_power6_nse_lnnse(), mo_objective_function::objective_q_et_tws_kge_catchment_avg(), objective_sse(), objective_sse_boxcox(), objective_weighted_nse(), and get::runoff_eval().
|
private |
Logarithmised likelihood with linear error model and lag(1)-autocorrelation of the relative errors.
This loglikelihood uses a linear error model and a lag(1)-autocorrelation on the relative errors. This is approach 2 of the paper Evin et al. (WRR, 2013). This is opti_function = 8. mHM then adds two extra (local) parameters for the error model in mhm_driver, which get optimised together with the other, global parameters. ADDITIONAL INFORMATION Evin et al., WRR 49, 4518-4524, 2013
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval | |
| [in] | logical, optional :: regularize |
Definition at line 726 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), mo_common_variables::global_parameters, loglikelihood_evin2013_2(), and parameter_regularization().
Referenced by loglikelihood_evin2013_2(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Logarithmic likelihood function with removed linear trend and Lag(1)-autocorrelation.
The logarithmis likelihood function is used when mHM runs in MCMC mode. It can also be used for optimization when selecting the likelihood in the namelist as opti_function.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval | |
| [in] | real(dp) :: stddev | standard deviation of data |
| [out] | real(dp), optional :: stddev_new | standard deviation of errors with removed trend and correlationbetween model run using parameter set and observation |
| [out] | real(dp), optional :: likeli_new | logarithmic likelihood determined with stddev_new instead of stddev |
Definition at line 584 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and loglikelihood_stddev().
Referenced by loglikelihood_stddev(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Logarithmic likelihood function with linear trend removed.
The logarithmis likelihood function is used when mHM runs in MCMC mode. It can also be used for optimization when selecting the likelihood in the namelist as opti_function.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval | |
| [in] | real(dp) :: stddev_old | standard deviation of data |
| [out] | real(dp), optional :: stddev_new | standard deviation of errors with removed trendbetween model run using parameter set and observation |
| [out] | real(dp), optional :: likeli_new | logarithmic likelihood determined with stddev_new instead of stddev |
Definition at line 952 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and loglikelihood_trend_no_autocorr().
Referenced by loglikelihood_trend_no_autocorr(), and single_objective_runoff().
|
private |
Multi-objective function with absolute error of Flow Duration Curves low-segment volume and nse of DJF's discharge.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. The first objective is using the routine "FlowDurationCurves" from "mo_signatures" to determine the low-segment volume of the FDC. The objective is the absolute difference between the observed volume and the simulated volume. For the second objective the discharge of the winter months December, January and February are extracted from the time series. The objective is then the Nash-Sutcliffe efficiency NSE of the observed winter discharge against the simulated winter discharge. The observed data \( Q_{obs} \) are global in this module. To calibrate this objective you need a multi-objective optimizer like PA-DDS.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1835 of file mo_mrm_objective_function_runoff.F90.
References mo_common_mhm_mrm_variables::evalper, extract_runoff(), mo_mrm_signatures::flowdurationcurve(), mo_mrm_global_variables::gauge, multi_objective_ae_fdc_lsv_nse_djf(), mo_mrm_global_variables::ngaugestotal, and mo_mrm_global_variables::nmeasperday.
Referenced by multi_objective_ae_fdc_lsv_nse_djf(), and multi_objective_runoff().
|
private |
Multi-objective function with NSE and lnNSE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. A timepoint \(t\) of the observed data is marked as a lowflow timepoint \(t_{low}\) if
\[ Q_{obs}(t) < min(Q_{obs}) + 0.05 * ( max(Q_{obs}) - min(Q_{obs}) )\]
and a timepoint \(t\) of the observed data is marked as a highflow timepoint \(t_{high}\) if
\[ t_{high} if Q_{obs}(i) > percentile(Q_{obs},95.)\]
This timepoint identification is only performed for the observed data. The first objective is the logarithmic Nash-Sutcliffe model efficiency coefficient \( lnNSE_{high} \) of discharge values at high-flow timepoints
\[ lnNSE_{high} = 1 - \frac{\sum_{i=1}^{N_{high}} (\ln Q_{obs}(i) - \ln Q_{model}(i))^2} {\sum_{i=1}^N (\ln Q_{obs}(i) - \bar{\ln Q_{obs}})^2} \]
. The second objective is the logarithmic Nash-Sutcliffe model efficiency coefficient \( lnNSE_{low} \) of discharge values at low-flow timepoints
\[ lnNSE_{low} = 1 - \frac{\sum_{i=1}^{N_{low}} (\ln Q_{obs}(i) - \ln Q_{model}(i))^2} {\sum_{i=1}^N (\ln Q_{obs}(i) - \bar{\ln Q_{obs}})^2} \]
. Both objectives are returned. The observed data \( Q_{obs} \) are global in this module. To calibrate this objective you need a multi-objective optimizer like PA-DDS.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1559 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and multi_objective_lnnse_highflow_lnnse_lowflow().
Referenced by multi_objective_lnnse_highflow_lnnse_lowflow(), and multi_objective_runoff().
|
private |
Multi-objective function with NSE and lnNSE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. A timepoint \(t\) of the observed data is marked as a lowflow timepoint \(t_{low}\) if
\[ Q_{obs}(t) < min(Q_{obs}) + 0.05 * ( max(Q_{obs}) - min(Q_{obs}) )\]
and all other timepoints are marked as a highflow timepoints \(t_{high}\). This timepoint identification is only performed for the observed data. The first objective is the logarithmic Nash-Sutcliffe model efficiency coefficient \( lnNSE_{high} \) of discharge values at high-flow timepoints
\[ lnNSE_{high} = 1 - \frac{\sum_{i=1}^{N_{high}} (\ln Q_{obs}(i) - \ln Q_{model}(i))^2} {\sum_{i=1}^N (\ln Q_{obs}(i) - \bar{\ln Q_{obs}})^2} \]
. The second objective is the logarithmic Nash-Sutcliffe model efficiency coefficient \( lnNSE_{low} \) of discharge values at low-flow timepoints
\[ lnNSE_{low} = 1 - \frac{\sum_{i=1}^{N_{low}} (\ln Q_{obs}(i) - \ln Q_{model}(i))^2} {\sum_{i=1}^N (\ln Q_{obs}(i) - \bar{\ln Q_{obs}})^2} \]
. Both objectives are returned. The observed data \( Q_{obs} \) are global in this module. To calibrate this objective you need a multi-objective optimizer like PA-DDS.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1703 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and multi_objective_lnnse_highflow_lnnse_lowflow_2().
Referenced by multi_objective_lnnse_highflow_lnnse_lowflow_2(), and multi_objective_runoff().
|
private |
Multi-objective function with NSE and lnNSE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the Nash-Sutcliffe model efficiency coefficient \( NSE \)
\[ NSE = 1 - \frac{\sum_{i=1}^N (Q_{obs}(i) - Q_{model}(i))^2} {\sum_{i=1}^N (Q_{obs}(i) - \bar{Q_{obs}})^2} \]
and the logarithmic Nash-Sutcliffe model efficiency coefficient \( lnNSE \)
\[ lnNSE = 1 - \frac{\sum_{i=1}^N (\ln Q_{obs}(i) - \ln Q_{model}(i))^2} {\sum_{i=1}^N (\ln Q_{obs}(i) - \bar{\ln Q_{obs}})^2} \]
are calculated and both returned. The observed data \( Q_{obs} \) are global in this module. To calibrate this objective you need a multi-objective optimizer like PA-DDS.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1454 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and multi_objective_nse_lnnse().
Referenced by multi_objective_nse_lnnse(), and multi_objective_runoff().
| subroutine, public mo_mrm_objective_function_runoff::multi_objective_runoff | ( | real(dp), dimension(:), intent(in) | parameterset, |
| procedure(eval_interface), intent(in), pointer | eval, | ||
| real(dp), dimension(:), intent(out), allocatable | multi_objectives | ||
| ) |
Wrapper for multi-objective functions where at least one is regarding runoff.
The functions selects the objective function case defined in a namelist, i.e. the global variable opti_function. It return the multiple objective function values for a specific parameter set.
| [in] | REAL(dp), DIMENSION(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval | |
| [out] | REAL(dp), DIMENSION(:) :: multi_objectives |
Definition at line 510 of file mo_mrm_objective_function_runoff.F90.
References multi_objective_ae_fdc_lsv_nse_djf(), multi_objective_lnnse_highflow_lnnse_lowflow(), multi_objective_lnnse_highflow_lnnse_lowflow_2(), multi_objective_nse_lnnse(), and mo_common_mhm_mrm_variables::opti_function.
|
private |
Objective function equally weighting NSE and lnNSE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the Nash-Sutcliffe model efficiency coefficient \( NSE \)
\[ NSE = 1 - \frac{\sum_{i=1}^N (Q_{obs}(i) - Q_{model}(i))^2} {\sum_{i=1}^N (Q_{obs}(i) - \bar{Q_{obs}})^2} \]
and the logarithmic Nash-Sutcliffe model efficiency coefficient \( lnNSE \)
\[ lnNSE = 1 - \frac{\sum_{i=1}^N (\ln Q_{obs}(i) - \ln Q_{model}(i))^2} {\sum_{i=1}^N (\ln Q_{obs}(i) - \bar{\ln Q_{obs}})^2} \]
are calculated and added up equally weighted:
\[ obj\_value = \frac{1}{2} (NSE + lnNSE) \]
The observed data \( Q_{obs} \) are global in this module.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1357 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_equal_nse_lnnse().
Referenced by objective_equal_nse_lnnse(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Objective function of KGE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the Kling-Gupta model efficiency coefficient \( KGE \)
\[ KGE = 1.0 - \sqrt{( (1-r)^2 + (1-\alpha)^2 + (1-\beta)^2 )} \]
where \( r \) = Pearson product-moment correlation coefficient \( \alpha \) = ratio of similated mean to observed mean \( \beta \) = ratio of similated standard deviation to observed standard deviation is calculated and the objective function is
\[ obj\_value = 1.0 - KGE \]
\((1-KGE)\) is the objective since we always apply minimization methods. The minimal value of \((1-KGE)\) is 0 for the optimal KGE of 1.0. The observed data \( Q_{obs} \) are global in this module.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 2097 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_kge().
Referenced by objective_kge(), mo_objective_function::objective_kge_q_sm_corr(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Objective function of logarithmic NSE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the logarithmic Nash-Sutcliffe model efficiency coefficient \( lnNSE \)
\[ lnNSE = 1 - \frac{\sum_{i=1}^N (\ln Q_{obs}(i) - \ln Q_{model}(i))^2} {\sum_{i=1}^N (\ln Q_{obs}(i) - \bar{\ln Q_{obs}})^2} \]
is calculated.
\[ obj\_value = lnNSE \]
The observed data \( Q_{obs} \) are global in this module.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1085 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_lnnse().
Referenced by objective_lnnse(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
combined objective function based on KGE raised to the power 6
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the Kling-Gupta model efficiency coefficient \( KGE \) for a given gauging station
\[ KGE = 1.0 - \sqrt{( (1-r)^2 + (1-\alpha)^2 + (1-\beta)^2 )} \]
where \( r \) = Pearson product-moment correlation coefficient \( \alpha \) = ratio of similated mean to observed mean \( \beta \) = ratio of similated standard deviation to observed standard deviation is calculated and the objective function for a given gauging station ( \( i \)) is
\[ \phi_{i} = 1.0 - KGE_{i} \]
\( \phi_{i} \) is the objective since we always apply minimization methods. The minimal value of \( \phi_{i} \) is 0 for the optimal KGE of 1.0. Finally, the overall \( OF \) is estimated based on the power-6 norm to combine the \( \phi_{i} \) from all gauging stations ( \( N \)).
\[ OF = \sqrt[6]{\sum((1.0 - KGE_{i})/N)^6 } \]
. The observed data \( Q_{obs} \) are global in this module.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 2199 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_multiple_gauges_kge_power6().
Referenced by objective_multiple_gauges_kge_power6(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Objective function of NSE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the Nash-Sutcliffe model efficiency coefficient \( NSE \)
\[ NSE = 1 - \frac{\sum_{i=1}^N (Q_{obs}(i) - Q_{model}(i))^2} {\sum_{i=1}^N (Q_{obs}(i) - \bar{Q_{obs}})^2} \]
is calculated and the objective function is
\[ obj\_value = 1-NSE \]
The observed data \( Q_{obs} \) are global in this module.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1264 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_nse().
Referenced by objective_nse(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Objective function of combined NSE and lnNSE with power of 5 i.e.
the p-norm with p=5.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the Nash-Sutcliffe model efficiency coefficient \( NSE \)
\[ NSE = 1 - \frac{\sum_{i=1}^N (Q_{obs}(i) - Q_{model}(i))^2} {\sum_{i=1}^N (Q_{obs}(i) - \bar{Q_{obs}})^2} \]
and the logarithmic Nash-Sutcliffe model efficiency coefficient \( lnNSE \)
\[ lnNSE = 1 - \frac{\sum_{i=1}^N (\ln Q_{obs}(i) - \ln Q_{model}(i))^2} {\sum_{i=1}^N (\ln Q_{obs}(i) - \bar{\ln Q_{obs}})^2} \]
are calculated and added up equally weighted:
\[ obj\_value = \sqrt[6]{(1-NSE)^6 + (1-lnNSE)^6} \]
The observed data \( Q_{obs} \) are global in this module.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1997 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_power6_nse_lnnse().
Referenced by objective_power6_nse_lnnse(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Objective function of SSE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the sum squared errors
\[ SSE = \sum_{i=1}^N (Q_{obs}(i) - Q_{model}(i))^2 \]
is calculated and the objective function is
\[ obj\_value = SSE \]
The observed data \( Q_{obs} \) are global in this module.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 1174 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_sse().
Referenced by objective_sse(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Objective function of sum of squared errors of transformed streamflow.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the sum of squared error \( tSSE \)
\[ tSSE = \sum_{i=1}^N (z(Q_{obs}(i), \lambda) - z(Q_{model}(i), \lambda))^2 \]
is calculated where \( z \) is the transform and given by
\[ z(x, \lambda) = \frac{x^\lambda -1}{\lambda} \]
for \( \lambda \) unequal to zero and
\[ z(x, \lambda) = log x \]
for \( \lambda \) equal to zero. The objective function is
\[ obj\_value = tSSE \]
The observed data \( Q_{obs} \) are global in this module.
The boxcox transformation uses a parameter of 0.2, suggested by Woldemeskel et al. Hydrol Earth Syst Sci, 2018 vol. 22 (12) pp. 6257-6278. "Evaluating post-processing approaches for monthly and seasonal streamflow forecasts." https://www.hydrol-earth-syst-sci.net/22/6257/2018/
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 2380 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_sse_boxcox().
Referenced by objective_sse_boxcox(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
Objective function of weighted NSE.
The objective function only depends on a parameter vector. The model will be called with that parameter vector and the model output is subsequently compared to observed data. Therefore, the weighted Nash-Sutcliffe model efficiency coefficient \( NSE \)
\[ wNSE = 1 - \frac{\sum_{i=1}^N Q_{obs}(i) * (Q_{obs}(i) - Q_{model}(i))^2} {\sum_{i=1}^N Q_{obs}(i) * (Q_{obs}(i) - \bar{Q_{obs}})^2} \]
is calculated and the objective function is
\[ obj\_value = 1- wNSE \]
The observed data \( Q_{obs} \) are global in this module.
| [in] | real(dp), dimension(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval |
Definition at line 2287 of file mo_mrm_objective_function_runoff.F90.
References extract_runoff(), and objective_weighted_nse().
Referenced by objective_weighted_nse(), single_objective_runoff(), and single_objective_runoff_subprocess().
|
private |
TODO: add description.
TODO: add description
| [in] | real(dp), dimension(:) :: paraset | |
| [in] | real(dp), dimension(size(paraset)) :: prior | |
| [in] | real(dp), dimension(size(paraset), 2) :: bounds | (min, max) |
| [in] | logical, dimension(size(paraset)) :: mask |
Definition at line 868 of file mo_mrm_objective_function_runoff.F90.
References parameter_regularization().
Referenced by loglikelihood_evin2013_2(), and parameter_regularization().
| real(dp) function, public mo_mrm_objective_function_runoff::single_objective_runoff | ( | real(dp), dimension(:), intent(in) | parameterset, |
| procedure(eval_interface), intent(in), pointer | eval, | ||
| real(dp), intent(in), optional | arg1, | ||
| real(dp), intent(out), optional | arg2, | ||
| real(dp), intent(out), optional | arg3 | ||
| ) |
Wrapper for objective functions optimizing agains runoff.
The functions selects the objective function case defined in a namelist, i.e. the global variable opti_function. It return the objective function value for a specific parameter set.
| [in] | REAL(dp), DIMENSION(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval | |
| [in] | real(dp), optional :: arg1 | |
| [out] | real(dp), optional :: arg2 | |
| [out] | real(dp), optional :: arg3 |
Definition at line 121 of file mo_mrm_objective_function_runoff.F90.
References loglikelihood_evin2013_2(), loglikelihood_stddev(), loglikelihood_trend_no_autocorr(), objective_equal_nse_lnnse(), objective_kge(), objective_lnnse(), objective_multiple_gauges_kge_power6(), objective_nse(), objective_power6_nse_lnnse(), objective_sse(), objective_sse_boxcox(), objective_weighted_nse(), mo_common_mhm_mrm_variables::opti_function, mo_common_mhm_mrm_variables::opti_method, and single_objective_runoff().
Referenced by mo_mhm_interface::mhm_interface_run_optimization(), and single_objective_runoff().
| real(dp) function, public mo_mrm_objective_function_runoff::single_objective_runoff_master | ( | real(dp), dimension(:), intent(in) | parameterset, |
| procedure(eval_interface), intent(in), pointer | eval, | ||
| real(dp), intent(in), optional | arg1, | ||
| real(dp), intent(out), optional | arg2, | ||
| real(dp), intent(out), optional | arg3 | ||
| ) |
Wrapper for objective functions optimizing agains runoff.
The functions selects the objective function case defined in a namelist, i.e. the global variable opti_function. It return the objective function value for a specific parameter set.
| [in] | REAL(dp), DIMENSION(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval | |
| [in] | real(dp), optional :: arg1 | |
| [out] | real(dp), optional :: arg2 | |
| [out] | real(dp), optional :: arg3 |
Definition at line 236 of file mo_mrm_objective_function_runoff.F90.
References mo_common_variables::domainmeta, mo_mrm_global_variables::ngaugestotal, mo_common_mhm_mrm_variables::opti_function, mo_common_mhm_mrm_variables::opti_method, and single_objective_runoff_master().
Referenced by mo_mhm_interface::mhm_interface_run_optimization(), and single_objective_runoff_master().
| subroutine, public mo_mrm_objective_function_runoff::single_objective_runoff_subprocess | ( | procedure(eval_interface), intent(in), pointer | eval, |
| real(dp), intent(in), optional | arg1, | ||
| real(dp), intent(out), optional | arg2, | ||
| real(dp), intent(out), optional | arg3 | ||
| ) |
Wrapper for objective functions optimizing agains runoff.
The functions selects the objective function case defined in a namelist, i.e. the global variable opti_function. It return the objective function value for a specific parameter set.
| [in] | REAL(dp), DIMENSION(:) :: parameterset | |
| [in] | procedure(eval_interface) :: eval | |
| [in] | real(dp), optional :: arg1 | |
| [out] | real(dp), optional :: arg2 | |
| [out] | real(dp), optional :: arg3 |
Definition at line 378 of file mo_mrm_objective_function_runoff.F90.
References mo_common_variables::domainmeta, loglikelihood_evin2013_2(), loglikelihood_stddev(), objective_equal_nse_lnnse(), objective_kge(), objective_lnnse(), objective_multiple_gauges_kge_power6(), objective_nse(), objective_power6_nse_lnnse(), objective_sse(), objective_sse_boxcox(), objective_weighted_nse(), mo_common_mhm_mrm_variables::opti_function, and mo_common_mhm_mrm_variables::opti_method.
Referenced by mo_mhm_interface::mhm_interface_run_optimization().
1.9.7