A new generation of artificial intelligence could soon provide life-saving flood warnings and drought forecasts to some of the world’s most vulnerable and data-scarce regions.
According to new research published in the journal Machine Learning: Earth, AI “foundation models” trained on vast amounts of general time-series data can accurately forecast river flows even in areas with absolutely no local hydrological records.
In many parts of the developing world, physical river gauges are sparse, historical records are incomplete, and monitoring networks are notoriously difficult to maintain. Without reliable long-term datasets, local communities are often left completely blind to impending floods and lack the tools needed to manage severe droughts.
However, researchers from The University of Texas at Austin and Hydrotify LLC have demonstrated that AI could bridge this dangerous information gap.
The power of foundation models
The research team, which included Dr Alexander Sun and undergraduate student Albert Sun, evaluated several advanced AI systems known as time-series foundational models (TSFMs).
Rather than being trained strictly on water data, these models were originally trained using vast amounts of sequential data from entirely different sectors, such as energy, transport, and climate. The researchers tested these models on a massive US dataset comprising more than 500 different river basins to see if the AI could apply its general pattern-recognition skills to water flow.
The results were highly promising. One specific AI model, called Sundial, performed almost as well as a traditional Long Short-Term Memory (LSTM) model that had been painstakingly trained on decades of specific, local river flow records. The researchers noted that the AI models showed their absolute strongest performance in basins dominated by strong seasonal patterns, such as snowmelt-driven water flow.
Closing the data gap
As climate change accelerates and extreme weather events become more frequent, the ability to predict water hazards without relying on extensive local infrastructure is becoming increasingly critical.
Dr Alexander Sun said: “Reliable water information is essential for communities everywhere, but many regions still lack the long-term records needed to support traditional forecasting methods. Approaches like this show how new AI tools could help close that gap by giving more places access to data-driven predictions. While there is still progress to be made, especially in more complex river systems, this work points to a future where improved forecasting is possible even in areas that have been underserved for decades.”
The researchers concluded that the capacity of these AI models scales directly with the size of their training data. As future generations of the technology ingest more specific Earth science data, their ability to protect communities and manage real-world water resources is expected to grow exponentially.
