Towards constraining soil and vegetation dynamics in land surface models: Modeling ASCAT backscatter incidence-angle dependence with a Deep Neural Network

A Deep Neural Network (DNN) is used to estimate the Advanced Scatterometer (ASCAT) C-band microwave normalized backscatter (σ40o), slope (σ′) and curvature (σ″) over France. The Interactions between Soil, Biosphere Atmosphere (ISBA) land surface model (LSM) produce variables (LSVs) that are input DNN. DNN trained simulate σ40o, σ′ σ″ from 2007 2016. predictive skill of evaluated during an independent validation period 2017 2019. Normalized sensitivity coefficients (NSCs) computed study ASCAT observables changes in LSVs as a function time space. Model performance yields near-zeros bias σ40o σ′. domain-averaged values ρ 0.84 0.85 for σ′, compared 0.58 σ″. unbiased RMSE 8.6% dynamic range 13% with cover having some impact on performance. NSC results show DNN-based could reproduce physical response LSVs. Results indicated sensitive soil moisture LAI these sensitivities vary time, highly dependent type. was shown be LAI, but also root zone due dependence vegetation water content moisture. potentially serve observation operator data assimilation constrain dynamics LSMs.