Climate

Cemaden Develops FloodcastingXAI, an Open-Source AI That Forecasts River Rises Up to 4 Hours Ahead

· 6 min read · Climate

Cemaden Develops FloodcastingXAI, an Open-Source AI That Forecasts River Rises Up to 4 Hours Ahead

Researchers at Brazil’s National Centre for Monitoring and Early Warning of Natural Disasters (Cemaden) have developed FloodcastingXAI, an artificial intelligence system that forecasts rises in river levels 1 to 4 hours in advance, TechnologyNews reported on November 10, citing the Ministry of Science, Technology and Innovation (MCTI). The ministry announced the tool on Monday. Cemaden is a unit of the MCTI, and the system is meant to give civil defence teams more time to act before water reaches homes and streets.

FloodcastingXAI combines data from existing rain gauges and river-level sensors, updated every 10 minutes, with AI models trained to recognise patterns and estimate how the water level will evolve. It uses explainable AI, which lets researchers see which inputs, such as rainfall intensity or river conditions, most influenced each forecast. That matters for the officials who must decide whether to issue a warning. In operational tests, the performance of forecasts in the 1-to-4-hour window improved by about 40%, according to the ministry.

What it means in practice

Cemaden Develops FloodcastingXAI, an Open-Source AI That Forecasts River Rises Up to 4 Hours Ahead — contextual photo

The code is open source and was developed collaboratively without outside contracting, with participation from the National Institute for Space Research (Inpe), Unesp, Unifesp, city governments including São Paulo, and the iFast project, which brings together more than 60 researchers from 15 Brazilian and foreign institutions. The ministry stressed that the system supports, but does not replace, the technical teams and Civil Defence officers responsible for monitoring and issuing alerts. Cemaden also runs GeoRisk, which forecasts landslide risk up to 72 hours ahead.

The project joins other public uses of AI against environmental hazards in Brazil, such as AI cameras watching for fires in the Pantanal and Inpe’s AI deforestation alerts backed by the Amazon Fund. What to watch is how quickly municipalities beyond São Paulo plug the open-source model into their own sensor networks, and whether its forecasts reach residents in time during the coming rainy season.

← Back to AI4Brazil