MLRG Joins NASA Project on Prithvi-WxC-Based Probabilistic Precipitation Forecasting

Jul 11, 2026. The Machine Learning Research Group is pleased to join a newly awarded NASA-funded project under the Applied Sciences Program and its Earth Science to Action element: “Prithvi-WxC-Powered Ensemble Precipitation Forecasts (P-PFE): A Scalable and Cost-Efficient Framework for Decision Support.”

The project is led by Efthymios I. Nikolopoulos at Rutgers University and brings together a multi-institutional team with expertise in hydrometeorology, flood modeling, remote sensing, AI/ML, open-source software, and operational decision support. The senior team also includes Jie Gong from Rutgers University, Humberto J. Vergara from the University of Iowa, Timothy Mayer from the University of Alabama in Huntsville, and Georgios C. Anagnostopoulos from Florida Institute of Technology.

The project aims to develop a scalable and computationally efficient framework for generating high-resolution probabilistic precipitation forecasts for flood decision support. At its center is the adaptation of the NASA/IBM Prithvi-Weather and Climate (Prithvi-WxC) foundation model to produce 50-member precipitation ensembles that better characterize uncertainty in rainfall placement and intensity than deterministic guidance alone.

The broader technical effort combines harmonized meteorological and radar-derived precipitation datasets, foundation-model adaptation, generative ensemble modeling, hydrologic-model coupling, and cloud-deployable delivery tools. The initial demonstration environments are in New Jersey and Iowa, where project partners will evaluate how ensemble precipitation information can support flood-risk assessment, emergency response, and operational decision support.

MLRG will contribute to the project’s machine-learning and AI methodology, focusing on adaptation of the Prithvi-WxC backbone, development of a probabilistic generative forecasting framework, and related software support for scalable ensemble prediction. This work aligns closely with the group’s broader interests in probabilistic modeling, generative modeling, uncertainty quantification, and learning from high-dimensional spatiotemporal data.

Beyond its technical goals, the project is designed with operational relevance in mind. Through stakeholder engagement with the New Jersey Office of Emergency Management, the Iowa Flood Center, and local partners in Spencer, Iowa, the team aims to assess how probabilistic precipitation intelligence can be integrated into existing flood-forecasting and response workflows.