Warm-cloud Organization Enables Earlier Satellite Prediction of Convection Initiation

Abstract Warm-cloud edges can sharpen and organize before cloud tops cool, providing an earlier satellite signal of convection initiation (CI). We analyze 8,607 isolated CI events and matched non-CI samples over South China during 2018–2023. The 10.7-µm brightness-temperature gradient ( G 10.7 ) shows sustained CI–non-CI separation 30 min before initiation, 20 min earlier than the cloud-top cooling signal. Adding edge organization and its persistence to the five cloud-growth signals advances CI prediction by 18.6 min. Averaged over 10 deep-learning models and seven lead times (0–60 min before CI), the area under the receiver operating characteristic curve (AUC) rises from 0.751 to 0.804, and the Brier score falls from 0.204 to 0.176. Replacing G 10.7 history with an equally long cloud-growth history yields a much smaller gain, showing that edge persistence adds information beyond cloud growth alone. Warm-cloud edge organization is therefore a distinct and underused source of early CI predictability.