Deep Learning for Multisatellite Precipitation Retrievals: Impact of Tomorrow.io’s Microwave Sounders

Forest Cannon, Simon Pfreundschuh, Brandon Taylor, S. Joseph Munchak, Randy J. Chase, Ethan Nelson, John L’Heureux, Camille Owens, Luke Conibear, Stylianos Flampouris, Arun Chawla

Journal of Hydrometeorology · 23 February 2026 · Colorado State University (incl. CIRA), Tomorrow.io

Paper

In one sentence. CNN fuses geostationary imagery and microwave sounder data for 10-minute, 4 km precipitation retrievals, gaining 30 to 50% skill over IMERG-Early against radar and gauges.

Abstract

Abstract Understanding precipitation intensity and distribution is critical for applications such as real-time flood monitoring and water resource management. However, accurate monitoring often depends on costly, ground-based radar networks that are not globally available. This study introduces a satellite-based precipitation retrieval that combines publicly available geostationary imagers and polar-orbiting microwave sensors with two commercially launched Tomorrow.io microwave sounders. The retrieval produces near-surface precipitation rate estimates every 10 min at a spatial resolution of 0.04° (∼4 km), covering latitudes between 60°S and 72°N. Precipitation estimates are derived from a convolutional neural network that ingests sequences of multichannel geostationary imagery and polar-orbiting passive microwave observations within the preceding hour. The network is designed to accommodate sparse data from individual microwave swaths, delivering a spatially and temporally complete trajectory of precipitation rate across the hour-long input window. Results demonstrate improvements of 5%–10% in categorical and continuous precipitation metrics over previous algorithm versions, with 30%–50% skill gains relative to publicly available near-real-time products such as IMERG-Early. Verification was conducted against terrestrial radar networks (United States, Europe, Japan), U.S.-based gauge observations spanning all seasons, and several case studies outside the training domains. Data-denial experiments from impactful 2024 hurricanes further illustrate the benefit of incorporating microwave observations into precipitation retrievals using geostationary and polar-orbiting satellites.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.