Abstract This study presents a proof of concept for a storm Detection, Delineation, and Tracking (DDT) method based on the vertical integral of condensed water and ice content—namely the Ice Water Path (IWP). Traditional DDT methods rely solely on regularly sampled variables, such as brightness temperatures, which do not directly capture storm evolution. In contrast, IWP is a direct indicator of storm presence and intensity, being zero when absent and high in convective cores. A machine learning approach is proposed to forecast IWP during gaps in passive microwave (PMW) observations. It integrates an initial IWP retrieval with the temporal evolution of geostationary IR data. The method was first evaluated using numerical weather model (NWP) simulations of convective episodes. Then, it was applied to 26 cases of near‐coincident (20 min apart) PMW and IR observations of real storms. The prediction of IWP is addressed at three levels: (a) a cloud/no‐cloud classification achieving probabilities of detection (POD) consistently above 0.60 and often exceeding 0.80; (b) a regression model with some challenges on the spatial structure forecast but normally achieving correlations above 0.60 (c) an intensity‐based, three‐class classification that outperformed simple variable persistence and demonstrated strong spatial coherence.
Leganés et al. (Sun,) studied this question.