AI ‘Foundation Models’ Revolutionize Flood Forecasting for Data-Scarce Regions Using Global Time-Series Data

UT Austin researchers show AI "foundation models" can forecast river flows in data-scarce regions, improving global flood and drought warnings.

By: AXL Media

Published: Mar 20, 2026, 8:41 AM EDT

Source: The information in this article was sourced from IOP Publishing

Closing the Information Gap in Global Hydrology

In many developing regions, the absence of reliable river gauges and long-term hydrological records has historically made accurate flood and drought forecasting nearly impossible. A study published in Machine Learning: Earth suggests that advanced artificial intelligence may finally provide a solution to this data scarcity. Researchers have demonstrated that "foundation models"—AI systems trained on massive, diverse datasets—can be adapted to forecast river flows with remarkable accuracy, even in basins where local sensors are sparse or non-existent.

The Rise of Time-Series Foundational Models

The research team evaluated several Time-Series Foundational Models (TSFMs), which were originally designed for sectors like energy, transport, and general climate science. Unlike traditional hydrological models that require decades of specific local data to function, TSFMs leverage patterns learned from vast quantities of global information. One specific model, named Sundial, was tested against a large United States river dataset covering over 500 basins. The results showed that Sundial performed almost as well as a specialized Long Short-Term Memory (LSTM) model that had been extensively trained on 30 years of local river records.

Performance Peaks in Seasonal Flow Patterns

The AI models exhibited their strongest predictive capabilities in river basins dominated by predictable seasonal cycles, such as those driven by annual snowmelt. By recognizing these broad temporal patterns, the foundation models could provide reliable forecasts that would typically require a dense network of physical sensors. Dr. Alexander Sun, lead researcher from UT Austin and Hydrotify LLC, noted that while complex river systems still pose challenges, these tools are a vital step toward providing data-driven predictions to regions that have been underserved by traditional meteorology for decades.