Open pag13 opened 3 months ago
Next, a function to compute DGSM (Derivative Based Sobol indices) ... cf. Sobol and Kuruchenko (2010) doi:10.1016/j.matcom.2009.01.023 ; Lamboni et al. (http://dx.doi.org/10.1016/j.matcom.2013.02.002) ; used with adjoint and another prior sampling in Chelil et al. https://doi.org/10.1016/j.jhydrol.2022.128006
A python function that enables to compute maps of spatial gradients of a cost function (global in time, no time series of gradient maps which would require to prostproc numerical adjoint model and add a new fortran variable to store backward grad trajectory...)
Based on Truyen's scripts used for WRR paper review ?
given a structure, setup incl. time period and cost func definition,
=> call of adjoint model, with user defined parameter values (either constant ones directly imposed or params maps, for ex. from an optim...)
return maps of dJ/dtheta_i, i = 1.. Nparam_type