A Fast Microwave Ocean Surface Emissivity and BRDF Model Based on Successive Machine Learning in Subphysical Spaces

Surface emissivity modeling remains one of the challenges for the radiance data assimilation of surface sensitive and window channels, as well as the retrieval of many key surface and atmospheric parameters (e.g. precipitation, soil moisture). At NOAA/NESDIS/STAR, the Community Surface Emissivity Model (CSEM) system has been developed and will be implemented in the CRTM 3.0 release with a number of features in model physics improvement and the model software structure. We will present our latest work on the microwave ocean emissivity and BRDF model development, which supports the multi-stream radiation transfer modeling in NOAA all-sky data assimilation applications. The new microwave ocean emissivity and BRDF model is based on the application of the machine learning techniques to the numerical reconstruction of a two-scale physical ocean surface model. The reconstruction is successively performed at sub-physical spaces, which retains the model physics while significantly reducing the requirement of the computing resources in the original physical space. Due to the limitation of the ground-based observations, the calibration and validation of the model is performed with satellite brightness temperature observations in the CRTM-CSEM coupled modeling. The model demonstrates better performance than the existing CRTM FASTEM in low frequency bands and larger zenith angles. The implementation of this new BRDF model in CSEM together with the case studies at SMAP, SMOS, AMSR2 and ATMS channels will be covered in the presentation.