Automatic detection and classification of low-level orographic precipitation processes from space-borne radars using machine learning

Ground-clutter is a significant cause of missed-detection and underestimation of precipitation in complex terrain from space-based radars such as the Global Precipitation Measurement Mission (GPM) Dual-frequency Precipitation Radar (DPR). This research proposes an Artificial Intelligence (AI) framework consisting of a precipitation detection model (PDM) and a precipitation regime classification model (PCM) to improve orographic precipitation retrievals from GPM-DPR using machine learning. The PDM is a Random Forest Classifier using GPM Microwave Imager (GMI) calibrated brightness temperatures (Tbs) and low-level precipitation mixing ratios from the High-Resolution Rapid Refresh (HRRR) analysis as inputs. The PCM is a Convolutional Neural Network that predicts the precipitation regime class, defined independently based on quantitative features of ground-based radar reflectivity profiles, using GPM DPR Ku-band (Ku-PR) reflectivity profiles and GMI Tbs. The AI framework is demonstrated for warm-season precipitation in the Southern Appalachian Mountains over three years (2016–2019), achieving large reductions in false alarms (77%) and missed detections (82%) relative to GPM Ku-PR precipitation products. The spatial distribution of predicted precipitation classes within the GPM overpass reflects the complex interactions between storms and topography that determine orographic precipitation regimes. For each GPM pixel, the local precipitation class informs on the vertical structure of rainfall microphysics aiming to capture low-level processes missed in GPM DPR reflectivity profiles contaminated by ground-clutter (i.e., the radar blind-zone).