Stepwise sensitivity analysis from qualitative to quantitative: Application to the terrestrial hydrological modeling of a Conjunctive Surface‐Subsurface Process (CSSP) land surface model

Abstract

AbstractAn uncertainty quantification framework was employed to examine the sensitivities of 24 model parameters from a newly developed Conjunctive Surface‐Subsurface Process (CSSP) land surface model (LSM). The sensitivity analysis (SA) was performed over 18 representative watersheds in the contiguous United States to examine the influence of model parameters in the simulation of terrestrial hydrological processes. Two normalized metrics, relative bias (RB) and Nash‐Sutcliffe efficiency (NSE), were adopted to assess the fit between simulated and observed streamflow discharge (SD) and evapotranspiration (ET) for a 14 year period. SA was conducted using a multiobjective two‐stage approach, in which the first stage was a qualitative SA using the Latin Hypercube‐based One‐At‐a‐Time (LH‐OAT) screening, and the second stage was a quantitative SA using the Multivariate Adaptive Regression Splines (MARS)‐based Sobol’ sensitivity indices. This approach combines the merits of qualitative and quantitative global SA methods, and is effective and efficient for understanding and simplifying large, complex system models. Ten of the 24 parameters were identified as important across different watersheds. The contribution of each parameter to the total response variance was then quantified by Sobol’ sensitivity indices. Generally, parameter interactions contribute the most to the response variance of the CSSP, and only 5 out of 24 parameters dominate model behavior. Four photosynthetic and respiratory parameters are shown to be influential to ET, whereas reference depth for saturated hydraulic conductivity is the most influential parameter for SD in most watersheds. Parameter sensitivity patterns mainly depend on hydroclimatic regime, as well as vegetation type and soil texture.