Bridging Time-Delayed Microwave Radiometric Observations and Deep Convection Characteristics: A Machine Learning Approach for the C²OMODO Mission

Deep convective cloud systems are central to the global water and energy cycle, and yet their representation in climate models remains challenging. This study explores the potential of machine learning to classify and characterize cloud structures inside cloud systems using radiometric measurements from the planned C²OMODO (Convective Core Observation through MicrOwave Derivative in the trOpics) mission. The relationships between cloud structure (anvil, stratiform, convective, and deep convective) and geophysical variables such as ice water path as well as integrated vertical ice momentum are investigated using a gradient boosting algorithm both for classification and regression purposes. The gradient boosting classification method achieves an overall accuracy (True Positive) above 70%. Retrievals of the geophysical variables yield R² ranging from 0.60 to 0.99. Furthermore, it is shown that applying a prior classification of the scenes improves the performances of the retrieval. This study highlights the potential of the forthcoming C²OMODO mission in advancing our understanding of convective systems and paves the way for in-depth studies on alternative or refined classification schemes and inputs, which could deliver even better results.