Boundary‐Aware Physical Constraints Refine Satellite Precipitation Occurrence Over the Qinghai‐Tibet Plateau

Abstract

Over complex terrain, hourly precipitation occurrence estimates are often distorted by weak, persistent, and mistimed wet hours that are poorly constrained by independent time‐step classification. Using 1,232 hourly gauge records over the Qinghai‐Tibet Plateau (QTP), we develop a boundary‐aware physical‐constraint framework that learns rainfall onset and cessation probabilities from short‐window environmental contrasts, extracts interpretable physical rules, and fuses them into transferable boundary evidence for downstream correction of hourly satellite precipitation occurrence. The learned boundary probabilities are discriminative and well calibrated. The extracted rules reveal boundary asymmetry: onset emphasizes cloud‐liquid‐water, midlevel‐humidity and hydrometeor contrasts, whereas cessation shifts to humidity–cloud‐fraction loss and land–wind persistence controls. Applied conservatively, the framework increases station‐median CSI by 15.5% for GSMaP and 17.1% for IMERG, while reducing median FAR by 4.5% and 13.7%, respectively. These results show that boundary‐aware physical constraints can refine the structural realism of hourly satellite precipitation occurrence over complex terrain.