Sensitivity Analysis and Identification of Noninformative Parameters of the Land Surface Model HTESSEL

Abstract:

Within the context of computational hydrology, land surface models couple physical processes that describe the flux of energy and water on the surface and sub-surface of the modeled domain. Given that the predicted quantities are used as boundary conditions for general circulation models, it becomes important to understand to which extent each parameter affects the outcome of the underlying model. The current study is concerned with the sensitivity analysis for the HTESSEL land-surface model using the Multiscale Parameter Regionalization tool, MPR. The latter is used to calculate spatially distributed model parameters from high-resolution predictors (e.g., soil parameters) for HTESSEL simulations. The HTESSEL code has gone through a substantial refactorization, resulting in extraction of more than 280 hard-coded parameters and the introduction of spatially distributed soil and vegetation parameters. The sensitivity analysis of these parameters is performed using the Efficient Elementary Effects (EEE) package [1] that uses the Morris method to separate informative from noninformative parameters. The screening is conducted at basins across the entire globe to ensure the coverage of all the parameters. The results of this analysis will be used in the calibration of the refactored model, by giving priority in optimization to those parameters that are found to affect the outcome the most.

References:

[1] https://doi.org/10.1002/2015WR016907 , https://github.com/julemai/EEE