Climate change poses significant threats to global life and biodiversity. In addition to rising global mean temperatures, precipitation patterns are projected to shift in unpredictable ways. For example, damages from climate change and flooding in Europe alone are expected to reach €45 billion annually by 2050, while in Mexico, gross domestic product may decline by approximately 1.97% due to erosion, sea-level rise, and severe weather.

Recent high-resolution climate models have enhanced the ability to resolve convective events, which frequently precede extreme precipitation. However, these models continue to struggle with accurately reproducing the observed aggregation of convective cells identified in satellite imagery. Furthermore, the availability of higher-resolution observations has not reduced uncertainties in precipitation estimates. Precipitation prediction remains particularly challenging due to the complexity of the integrated Earth system.

One approach to address these challenges is to conduct targeted observational campaigns to capture critical processes that remain poorly understood. This strategy is central to the Global Precipitation EXperiment (GPEX), a new initiative under the World Climate Research Program (WCRP) aimed at improving precipitation forecasts (Zeng et al., 2025). Additionally, deep learning methods offer innovative techniques for analyzing complex multisensor and multiplatform observations, extracting information from extensive satellite datasets, and developing new frameworks for model evaluation. In this talk, we will explore what can be learned from high-resolution observations of clouds and precipitation processes conducted over challenging regions of the trades and Alpine complex orography. We will then show how deep learning self-supervised methods can be used to extract information from observations and characterize the evolution of precipitation systems for nowcasting, model evaluation, and process understanding, and discuss the future perspectives of these methods.