UNIST Unveils FWI-Net: Global AI Model Predicts Wildfire Risk 31 Days Ahead, Closing Africa's Forecasting Gap

A research team led by Professor Lim Jeong-ho at Ulsan National Institute of Science and Technology (UNIST) in South Korea published FWI-Net on June 25, 2026 — a global deep-learning model that forecasts the Fire Weather Index daily up to 31 days in advance. The model cut root-mean-square error by 6.6% across the full forecast window compared to existing numerical approaches, and by as much as 12.4% during the first week, while reducing prediction bias in 85% of areas where wildfire exposure and socioeconomic vulnerability are both high. Most notably, in low-income regions of Africa where ground-based forecasting infrastructure is severely lacking, FWI-Net maintained meaningful predictive skill for an average of 22 days — more than three weeks — demonstrating it can help close the 'disaster information gap' between wealthy and vulnerable nations that typically lack their own weather forecasting capacity.