# Prenúncio API, instructions for agents Prenúncio forecasts deforestation in the Brazilian Amazon. Every 500 m forest cell (about 14 million) gets the chance that it loses at least one hectare in the PRODES year 2027 (1 August 2026 to 31 July 2027). The forecast was made with information up to 31 July 2026. Sums are given by municipality, state, Indigenous land and conservation unit. ## How to connect Plain HTTPS GET, static JSON, no key, no login, CORS open. Base: https://prenuncio.com/api/v1 Start with https://prenuncio.com/api/v1/index.json, which lists every endpoint with a description. If prenuncio.com answers 403 with a browser check (its host does that to some clients that are not browsers), use the mirror, which holds the same files: https://raw.githubusercontent.com/gdavidss/prenuncio/api/v1 for example https://raw.githubusercontent.com/gdavidss/prenuncio/api/v1/index.json ## Endpoints - amazon.json the whole Amazon: forecast total, distribution of cell chances, yearly and monthly history - states.json the nine states, full detail - municipalities.json one summary row per municipality (id is the 7 digit IBGE code), sorted by rank - municipality/.json full detail for one municipality, for example municipality/1507300.json (São Félix do Xingu, rank 1) - indigenous-lands.json, conservation-units.json the same for protected areas - priority.json, priority.csv the 500 blocks of 5 km ranked highest, where to look first - permits.json clearing permits (SINAFLOR), with dates, point and hectares - monthly-forecast.json km2 per month for the next 12 months, by hazard, Amazon and states, with ranges - enforcement-effect.json clearing observed against clearing implied with enforcement held at its 2021 to 2022 level - calibration.json the check of the chances against what happened in 2024 to 2026 - benchmark.json the open benchmark's leaderboard - cells/grid.json the grid, and how to read the chance of any cell ## The chance of one place row = floor((north - lat) / cell_degrees), col = floor((lon - west) / cell_degrees), from cells/grid.json. GET cells/2027/_.bin, a row-major array of bytes, 100 per row. byte = bytes[(row % 100) * width + (col % 100)], where width is 100 except in the last column of tiles. chance = (byte / 255) ** 2. Byte 0 means not forest. A 404 means no forest in that tile. ## Reading the numbers - km2 figures by year are PRODES years (August to July), named by the year they end in. - Monthly figures are what INPE's DETER detects; the yearly PRODES count is larger. - "expected" values are forecasts. "cleared" and "observed" values are measurements by INPE. - The chances were checked: in 2024, cells given about 3% lost a hectare 3% of the time (calibration.json). - A chance is not a verdict. Most cells with a 10% chance are not cleared. - Fire amounts are not forecast; they depend on a drought that has not happened yet. - enforcement-effect.json is an association and a floor, not a causal estimate. - Permit points are property centres, not the clearing itself. Mato Grosso permits may have entered the federal register after their issue date. ## Licence and credit Data CC BY-SA 4.0, derived from INPE PRODES and DETER, IBAMA and Banco Central open data. Credit INPE and Prenúncio. Code and the open benchmark: https://github.com/gdavidss/prenuncio