Quantifying uncertainty and sensitivity of model parameterizations and parameters in WRF in simulating surface fluxes and Land-Atmosphere coupling over the Amazon region

Description

The objectives of this study are 1) to quantify the sensitivities and uncertainties of physical parameterizations in simulating surface fluxes and Land-Atmosphere (L-A) coupling, 2) to quantify the parametric sensitivity in YSU PBL parameterization for better understanding model behavior and improving the PBL parameterization, and 3) to explore the applications of Uncertainty Quantification (UQ) methods to identify the structural uncertainty and better understand the physical process associated with L-A interactions. We use two-legged coupling metrics, which include both terrestrial (soil moisture to surface fluxes) and atmospheric (surface fluxes to atmospheric state or precipitation) legs, to diagnose the land-atmosphere interaction and coupling strength. To quantify the uncertainty in physical parameterizations, we performed a 120-member ensemble of simulations with the WRF model using a stratified experimental design including 6 cloud microphysics, 3 convection, 6 PBL and surface layer, and 3 land surface schemes. A multiple-way analysis of variance (ANOVA) approach is used to quantitatively analyze the inter- and intra-group (scheme) means and variances. The results show that 1) The variance across parameterizations (model structural uncertainty) varies with interest variables, 2) Land surface and convection are two dominant processes in simulating surface fluxes and land-atmosphere coupling, and 3) First-order effects from individual parameterization or parameter dominate comparing to the interaction effects. To quantify parameter sensitivity in YSU PBL scheme, we conducted an additional 256 WRF simulations in which an efficient sampling algorithm is used to explore the multiple-dimensional parameter space. Three uncertainty quantification approaches are applied for sensitivity analysis (SA) of multiple variables of interest to 20 selected parameters in the YSU PBL and MM5 surface-layer schemes. Sensitivity analysis results show consistent parameter sensitivity across different UQ methods. Six out of twenty parameters contribute more than 90% total variance. Variability of surface fluxes induced by perturbation of parameterizations is 5-10 times larger than that by parameters. Results of this study help to quantify model uncertainties from various sources, better understand the roles of different physical processes in L-A interactions, and provide insights for improving the model physics parameterizations such as the PBL scheme.