Wildfire AI

Wildfire AI is Terrafuse AI's machine-learning model that predicts the likelihood of a wildfire occurring at a given location over the next year. It is aimed chiefly at property insurers: as wildfire losses have grown, insurers have often fallen back on blanket pricing or refused to insure properties, particularly in California.

Terrafuse sources billions of Earth observation data points from the scientific community and trains its models on the interdependencies between more than 50 environmental conditions, such as building density, dead fuel moisture and vegetation, and embeds atmospheric physics through machine learning. The output is a wildfire risk score from 1 to 10 with over 250 possible values, available at 30 square meter spatial resolution, together with a rank-sorted list of 27 factors contributing to risk to guide mitigation. Users can view risk at a single location or run a portfolio analysis in a web platform, or integrate a REST API with their own software to receive geospatial visualizations or tabular data; Terrafuse works with customer teams to embed the scores in existing systems. A free public Wildfire AI map also shows daily risk scores for any location in California.

Terrafuse states the model is validated against a portfolio with over $400B total insured value and historical wildfire claims, and reports ~33% better performance than the incumbent model by Gini coefficient. A published case study with a California insurer reports an estimated ~10% gross margin improvement for the portfolio, that the model would have prevented 71% of the portfolio's losses had it been used, and a lift of 113 for a top-10 primary insurer. Terrafuse, a spin-out of Lawrence Berkeley National Laboratory holding four patents, was acquired by FutureProof Technologies in November 2025.