A Machine Learning-Based Rain Rate Estimation from the OceanSat-3 Scatterometer Measurements

The OceanSat-3 scatterometer (OSCAT-3) is a Ku-band radar instrument designed specifically to measure wind vectors over the ocean surface. Utilizing a conical scanning design with dual-polarization pencil beams at incidence angles of ~49° and ~58°, OSCAT-3 scans the Earth's surface to measure the normalized radar cross section (sigma0) and brightness temperature, covering a swath width of 1800 km. In this paper, we introduce a supervised machine learning approach for rain rates estimation from OSCAT-3 measurements. Specifically, we tested ability of the Support Vector Machine (SVM) algorithm to estimate rain rate and flag suspect wind vector retrievals by utilizing regression and classification analysis respectively. For regression analysis, the machine learning model was trained using features derived from a combination of brightness temperatures, sea surface temperature (SST) and OSCAT-3 wind speed retrievals. The training rain rate targets were obtained from Global Precipitation Mission (GPM) Microwave Imager (GMI) measurements. In the case of rain flag classification analysis, training features were derived from brightness temperature and sigma0. The SVM rain rate model application for rain rate retrievals showed strong potential with the OSCAT-3 rain rates product exhibiting a bias of -0.2 mm/hr and a standard deviation of 1.0 mm/hr when compared to GMI rain rate measurements. However, the application of the SVM algorithm for data flagging purposes resulted in an over-flagging of the data under light rain condition. Consequently, for improved accuracy the final rain flag is determined based on OSCAT-3 rain rate estimation.