INPE Gets R$147.6 Million From the Amazon Fund for AI-Driven Deforestation Alerts
Brazil’s National Institute for Space Research (INPE), which produces the country’s official deforestation figures, will use artificial intelligence to speed up detection of native vegetation loss and flag in advance the areas most at risk of clearing, degradation and fire, g1 reported on October 16. The project, called SIM-Biomas, will receive R$147.6 million in non-reimbursable support from the Amazon Fund. It was announced the same day the federal government released Amazon deforestation data showing the lowest rate since 1988.
Today the annual rate and the near-daily alerts come from separate systems: Prodes, which is more precise, and Deter, which is faster. SIM-Biomas aims to merge them, using machine-learning models trained on thousands of examples already classified by researchers to automate image analysis and alert generation, so that Prodes data can be produced more than once a year. Deter alerts, now limited to the Amazon, Cerrado and Pantanal, will be extended to the Caatinga, Atlantic Forest and Pampa.
Burn-scar mapping outside the Cerrado still works with squares 1 kilometre on a side; the project will bring 30-metre resolution to every biome. A territorial intelligence subsystem will cross vegetation data with land-tenure, socioeconomic and climate information to produce short-term risk indicators by river basin or Indigenous territory, and a new TerraBrasilis++ platform will gather maps, alerts and scenarios. INPE director Antônio Miguel Vieira Monteiro said the software and methods will be open so other institutions can replicate and verify them.
What it means in practice
The money sits outside INPE’s regular budget, which fell to R$79 million in 2021 and stood at R$102 million in 2025. The Amazon Fund, managed by BNDES with the environment ministry, is financed by donors including Norway, Germany and Japan. Federal science computing is moving in parallel, from LNCC’s billion-real supercomputer plan in Macaíba to the delayed national AI supercomputer tender, while USP’s TITAN-LSTM, which tracks storms 30 minutes ahead shows Brazilian teams already applying machine learning to climate risk.
For enforcement agencies, faster and finer maps matter only if field teams can act on them; g1 notes that turning an alert into action still depends on inspectors on the ground. Planned flights to sample carbon dioxide, methane, nitrous oxide and carbon monoxide over the Amazon should also tighten the emissions inventories Brazil reports under the Paris Agreement.
What to watch: the first intra-year Prodes release, the debut of Deter alerts in the three new biomes, and whether state environment agencies adopt the open code.