Rainfall Retrieval from Challenging 7 GHz Commercial Microwave Links in Burkina Faso Using a Deep Learning Wet-Dry Classification Framework

Commercial Microwave Links (CML) have demonstrated their potential for opportunistic rainfall estimation. They offer a promising solution for improving the spatial and temporal coverage of precipitation observations in poorly instrumented regions. However, the exploitation of CML operating in the 6 - 7 GHz frequency range remains a major challenge. At these frequencies, rain-induced attenuation is relatively weak and can be strongly affected by non-rain-related fluctuations, making the extraction of rainfall information particularly difficult. This study proposes a dedicated processing framework for 7 GHz CML in Burkina Faso. The approach relies on a deep learning-based wet-dry classification model developed using MSG-SEVIRI satellite observations and rain gauge measurements. The model outputs are projected along the microwave link paths using intersection weights and are subsequently used to identify rainy periods, estimate rainfall rates, and reconstruct spatial rainfall fields. The results demonstrate that the proposed framework can extract meaningful rainfall information despite the strong non-rain-related attenuation affecting this frequency band. The wet-dry classification framework achieved consistent performance across the analyzed links, with a median Matthews Correlation Coefficient (MCC) of approximately 0.30. Comparisons with rain gauge observations show encouraging temporal agreement for the main rainfall events, although quantitative uncertainties remain due to the point-scale nature of rain gauge measurements and the path-integrated nature of CML observations. In addition, daily and cumulative rainfall maps demonstrate the capability of 7 GHz CML networks to capture both the spatial variability and temporal evolution of precipitation. These findings represent an important step toward the use of low-frequency commercial microwave links as opportunistic rainfall sensors in regions where weather radar coverage is unavailable and conventional observation networks remain sparse.