From Gaps to Storms: Leveraging GAIA Representations for Global Precipitation Estimation and Extreme Weather Segmentation

Olivia Alexander; Jad Makki; Ata Akbari Asanjan; Tom Berg; Matt Yang; Cara Crawford; Disha Shidham; Srija Chakraborty; Stephen Peng; Arun Ravindran; Olivier Raiman; David Potere; David Bell

doi:10.22541/essoar.176858444.46117676/v1

From Gaps to Storms: Leveraging GAIA Representations for Global Precipitation Estimation and Extreme Weather Segmentation

2026 Preprint

View at publisher

Authors: Olivia Alexander, Jad Makki, Ata Akbari Asanjan, Tom Berg, Matt Yang, Cara Crawford, Disha Shidham, Srija Chakraborty, Stephen Peng, Arun Ravindran, Olivier Raiman, David Potere

+1 more

David Bell

DOI: 10.22541/essoar.176858444.46117676/v1

Cited by: 0

Abstract

Foundation models pretrained on Earth observation archives offer broad utility across geoscience tasks. Building on our DINO-MAE model, named GAIA, trained on globally harmonized longwave infrared (IR, ~11 μm) imagery from geostationary satellites (2000–present), we evaluate its transferability to key downstream applications: gap filling , precipitation estimation, and segmentation of tropical cyclones (TCs) and atmospheric rivers (ARs).

Fields of Study & Subjects

Precipitation Measurement and Analysis Meteorological Phenomena and Simulations Advanced Neural Network Applications

Publication details

Publication date: 16 January 2026
Document type: Preprint