Brazil's forest enforcement system has been operating with blurry glasses for sixteen years. On Tuesday, Google released a 5-meter satellite dataset of Brazil's forests as they existed in 2008, an upgrade from the 30-meter resolution that had been the legal standard since the Brazil Forest Code was first applied. The dataset, built in partnership with Brazil's Ministry of Management and Innovation (MGI) and Ministry of Environment (MMA), processes more than 6,000 high-resolution SPOT satellite scenes from the CNES historical archive through Google Earth Engine to produce a cloud-free mosaic covering 68 percent of Brazil's territory, rising to 93 percent in the Amazon arc states of Maranhão, Mato Grosso, Pará, Rondônia, and Tocantins.
The Brazil Forest Imagery Dataset 2008 is not an AI product. It is infrastructure. The ML in the pipeline handles cloud removal and radiometric correction, which is useful but incidental. The actual output is a sharper picture of what the land looked like on July 22, 2008, which is the legal baseline date against which every rural property owner in Brazil's Rural Environmental Registry (CAR) must measure their compliance obligations under the Forest Code.
Those obligations are not abstract. The Forest Code requires landowners to protect or restore a percentage of native vegetation on their properties, with the required percentage determined by land use as of July 2008. The CAR system tracks which properties are in compliance and which owe restoration or face fines. Brazil's deforestation rate for the twelve months ending July 2025 was 5,796 square kilometers, the lowest in eleven years, down 11 percent from the prior year, according to Mongabay. That number will be easier to argue about in both directions now that the underlying evidence is sharper.
The limitation of 30-meter Landsat imagery was not theoretical. At that resolution, a selective logging operation could be invisible. A narrow riparian buffer violation could look like a shadow. Small-scale land clearing could be lost entirely in a pixel representing 900 square meters of ground. The new 5-meter dataset brings that footprint down to 25 square meters per pixel, making discrepancies between what the old imagery showed and what actually existed on the ground legally actionable in a way they were not before.
"The improvement in clarity unlocks the ability to identify critical small-scale environmental features, like forest patches and riverbanks, essential for Brazil Forest Code enforcement but previously indistinguishable," Google wrote in its announcement of the dataset.
The dataset ships in two versions: a Visual Basemap optimized for human interpretation and verification, and an Analytic Basemap ready for machine learning pipelines. Both are accessible through Google Earth Engine and the Google Earth Data Catalog. Brazilian authorities will be able to use the visual layer in SICAR (the Rural Environmental Registry System) public consultation module, meaning the underlying imagery will be available for broad public oversight, not just government analysis.
This is where the evidentiary analogy holds. A blurry photograph of a license plate is not useful in court. A sharp one is. The same applies to property boundary disputes and Forest Code compliance assessments: the precision of the source data determines whether a discrepancy is visible and actionable.
The dataset covers v1.0 with known gaps. Cloud cover in the 2007-2009 window created coverage holes, and the team extended the temporal window where 2008 data was unavailable, filling in from adjacent years. That methodological choice will matter in practice: any compliance dispute that hinges on imagery from a year other than 2008 will require careful treatment of what that pixel actually shows.
The partnership fits into a broader push around Brazilian environmental governance. The Tropical Forests Forever Facility, launched at COP30 with $6.7 billion in initial pledges, represents the largest international financing mechanism specifically directed at preventing tropical deforestation, and the improved baseline dataset is designed to work within that framework by giving both government and private landowners a clearer picture of their obligations and assets.
Google has a track record building analysis-ready geospatial datasets, including the Dynamic World 10-meter land cover dataset and the Satellite Embedding dataset powered by its AlphaEarth Foundations model. The Brazil Forest Imagery Dataset 2008 fits that pattern: the company is not conducting the enforcement analysis itself but providing the foundational layer that makes independent analysis possible.
The dataset is available now in the Earth Engine Data Catalog for approved researchers and government partners. Visual access through Google Earth is open to everyone. What happens next is a property-by-property reckoning: the compliance obligations have always existed, but for the first time the evidence to assess them is sharp enough to use.
