Modeling the Influence of Precipitation on L-Band SMAP Observations of Ocean Surfaces Through Machine Learning Approach

A new forward model (FM) was developed to characterize the influence of precipitation on L-band passive ocean surface measurements. The FM, which relates rain-induced brightness temperature (TB) variations to the rain rate and wind speed (WS), was established through a machine learning approach (referred to as the ML-FM). The soil moisture active passive (SMAP) data matched with integrated multisatellite retrievals for global precipitation measurement (IMERG) rain rate data and cross-calibrated multiplatform (CCMP) wind data were binned as a function of the rain rate, WS, and wind direction. The ML-FM was validated by comparing the simulated top-of-atmosphere (TOA) TB values with SMAP measurements. The results showed favorable agreement between the ML-FM outputs and SMAP data, with a root mean square error (RMSE) smaller than 0.55 K for both the horizontal and vertical polarizations. The validation results for ensuring more reasonable rainfall intensity distributions showed that the ML-FM returned stable results with a slightly reduced RMSE of ∼0.75 K for both the horizontal and vertical polarizations. Based on the ML-FM, we found that sea surface emission exhibited significant dependence on the rain rate for both polarizations. In addition, the ML-FM demonstrated signal saturation when the rain rate exceeded 45 mm/h, while precipitation slightly affected the directional characteristics of sea surface emission. These effects accounted for ∼0.3 K at a rain rate of 50 mm/h. Overall, our analyses demonstrated that the proposed ML-FM achieved superior performance in retrieving the TOA TB for both the vertical and horizontal polarizations with a higher accuracy than existing models.