Cannon et al. (2026) Deep Learning for Multi-Satellite Precipitation Retrievals: Impact of Tomorrow.io’s Microwave Sounders

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Journal of Hydrometeorology
• Year: 2026
• Date: 2026-02-23
• Authors: Forest Cannon, Simon Pfreundschuh, Brandon Taylor, S. Joseph Munchak, Randy J. Chase, Ethan Nelson, John L’Heureux, C. Owens, Luke Conibear, Stylianos Flampouris, Arun Chawla
• DOI: 10.1175/jhm-d-25-0007.1

Research Groups

Tomorrow.io (developer of commercially launched microwave sounders and the precipitation retrieval)

Short Summary

This study introduces a novel satellite-based precipitation retrieval system that integrates publicly available geostationary and polar-orbiting satellite data with commercial microwave sounders, utilizing a convolutional neural network to provide near-surface precipitation rates every 10 minutes at 4 km resolution, demonstrating significant accuracy improvements over existing products.

Objective

• To develop an accurate, globally available, satellite-based precipitation retrieval system that overcomes the limitations of ground-based radar networks by combining public geostationary/polar-orbiting data with commercial microwave sounders.

Study Configuration

• Spatial Scale: 0.04° (approximately 4 km) resolution, covering latitudes between 60°S and 72°N.
• Temporal Scale: Precipitation rate estimates every 10 minutes, with the convolutional neural network ingesting observations from the preceding 1 hour.

Methodology and Data

• Models used: Convolutional neural network (CNN).
• Data sources: Publicly available geostationary imagers, polar-orbiting microwave sensors, two commercially launched Tomorrow.io microwave sounders. Input data includes sequences of multi-channel geostationary imagery and polar-orbiting passive microwave observations. Verification data included terrestrial radar networks (USA, Europe, Japan) and U.S.-based gauge observations.

Main Results

• The new retrieval system achieved improvements of 5–10% in categorical and continuous precipitation metrics compared to previous algorithm versions.
• It demonstrated 30–50% skill gains relative to publicly available near-real-time products, such as IMERG-Early.
• The incorporation of microwave observations, particularly from the commercial Tomorrow.io sounders, significantly benefited precipitation retrievals, as illustrated by data-denial experiments during impactful 2024 hurricanes.
• The system produces spatially and temporally complete precipitation rate trajectories across the hour-long input window, accommodating sparse data from individual microwave swaths.

Contributions

• Introduction of a novel satellite-based precipitation retrieval that uniquely combines publicly available geostationary imagers and polar-orbiting microwave sensors with commercially launched microwave sounders.
• Development of a convolutional neural network capable of handling sparse microwave data to produce spatially and temporally complete precipitation estimates.
• Demonstrated significant improvements in precipitation estimation accuracy (5–10% over previous versions, 30–50% over IMERG-Early), enhancing real-time flood monitoring and water resource management capabilities globally.

Funding

Not specified in the abstract.

Citation

@article{Cannon2026Deep,
  author = {Cannon, Forest and Pfreundschuh, Simon and Taylor, Brandon and Munchak, S. Joseph and Chase, Randy J. and Nelson, Ethan and L’Heureux, John and Owens, C. and Conibear, Luke and Flampouris, Stylianos and Chawla, Arun},
  title = {Deep Learning for Multi-Satellite Precipitation Retrievals: Impact of Tomorrow.io’s Microwave Sounders},
  journal = {Journal of Hydrometeorology},
  year = {2026},
  doi = {10.1175/jhm-d-25-0007.1},
  url = {https://doi.org/10.1175/jhm-d-25-0007.1}
}