The Precursor Characteristics and Prediction of Convective Precipitation Based on Atmospheric Temperature and Humidity Profiles

ABSTRACT Based on multi‐source atmospheric temperature and humidity profiles over the Northern Hemisphere during summer (2020–2021), this study explores precursors of convective precipitation (CP) through comparative analysis of convective parameters under different precipitation conditions and machine learning classification results. The results indicate that for most profile sources, convective parameters such as Most Unstable Convective Inhibition (MUCIN), level of free convection (LFC), K index (Ki) and Showalter Index (Si) exhibit precursor signals for CP. Among all profile sources, the radiosonde profiles from the China Meteorological Administration (CMA) reveal the most pronounced precursor signals of CP across most convective parameters, while Constellation Observing System for Meteorology, Ionosphere and Climate‐2 (COSMIC2) radio occultation data show stronger precursor features among satellite‐based profiles, particularly for Ki and Si. Convective parameters under CP conditions are generally insensitive to the CP intensity, but several parameters vary significantly with geographic location, especially latitude and local standard time (LST). Based on the classification results of the random forest (RF) model, profiles from different sources all can predict CPs with reasonably high accuracy, ranging from 0.64 to 0.79. Satellite profiles achieve comparable accuracy to radiosondes, largely due to their substantially larger sample size. The important feature factors for the CP prediction vary across different data sources, but with LST is a consistently important factor. LFC and Ki are highly important factors for COSMIC2 and radiosonde profiles, while MUCIN and relative humidity emerge as key factors for satellite microwave data.