Bridging the Data Latency Gap for Real-Time Precipitation Monitoring: A U-Net CNN Approach Using GOES-16 in the Tropical Andes

Timely precipitation information is essential for resilient water resources management disaster risk reduction, and climate adaptation, particularly in mountainous and data-scarce regions. While Satellite Precipitation Products (SPPs) such as IMERG Early Run (IMERG-ER) offer valuable spatial and temporal coverage, their latency of more than 4 hours limits their use for real-time applications, including flash flood early warning and operational decision-making. This study presents a hydroinformatics-based solution to bridge this critical latency gap by combining deep learning with near-time geostationary satellite observations. We developed a U-Net convolutional neural network driven by GOES-16 infrared imagery to emulate IMERG-ER precipitation fields with a latency of only minutes. The framework is applied to the Jubones river basin (3,340 km²) in the tropical Andes of Ecuador, a region characterized by complex topography and limited ground observations. The model was trained using five years (2019–2023) of GOES-16 data and evaluated across 15 spectral input configurations. Results show that a combination of water vapor (6.2, 6.9, 7.3 µm) and longwave infrared bands (8.4, 11.2 µm) yielded the best performance, effectively capturing atmospheric moisture dynamics and cloud-top characteristics. The proposed approach successfully reduced precipitation data latency from 4 hours to approximately 11 minutes. Model evaluation yielded an RMSE of 0.46 mm/h, a Pearson correlation of 0.60, and a Critical Success Index of 0.53. While performance decreased for high-intensity precipitation due to data imbalance, the model performed robustly for low-intensity precipitation (<3 mm/h), which accounts for 97% of events in the study area and is critical for hydrological monitoring and water management. Overall, the results demonstrate how integrating deep learning with geostationary satellite data can enhance near-real-time precipitation monitoring, supporting climate resilience, early warning systems, and operational hydrology in vulnerable and data-limited regions.