AI system
- System
-
Amazon Mining Watch gold-mine detector (ensemble of convolutional neural networks on Sentinel-2 imagery)
- Model
-
Not assessed
- Release
-
48px_v4.10b-18d-20g-21a-22bc-ensemble.
- Application developer
-
Earth Genome (model), with Amazon Conservation Association and the Pulitzer Center's Rainforest Investigations Network
- Role
-
The models detected mining activity in satellite images and calculated its extent.
- Configuration
-
The repository identified the July 2026 production model as 48px_v4.10b-18d-20g-21a-22bc-ensemble.
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
Free Sentinel-2 imagery composited on Google Earth Engine. Hand-labeled training set of 23,463 patches built from earlier human surveys with journalists. Ensemble CNN detector run over the whole basin yearly and quarterly. Public release of detections and code for journalists and NGOs.
- Human oversight
-
Not assessed
- Relation to AI
-
The AI system produces the maps and area statistics themselves. The makers' repository describes an ensemble of convolutional neural networks on Sentinel-2 patches, with reported holdout specificity 0.976 and sensitivity 0.991. Mongabay independently describes the platform as training AI algorithms to recognize mining scars. No cited source links a detection to an enforcement action. Mongabay remarks that the platform is useful for cross-border work, citing 2025 Brazil-Colombia operations on the Pure River. It does not state that those operations used its detections. MAAP's Operation Mercury analysis, by the same maker, used satellite imagery without machine learning.
- AI contribution
-
The AI system produces the maps and area statistics themselves. The makers' repository describes an ensemble of convolutional neural networks on Sentinel-2 patches, with reported holdout specificity 0.976 and sensitivity 0.991. Mongabay independently describes the platform as training AI algorithms to recognize mining scars. No cited source links a detection to an enforcement action. Mongabay remarks that the platform is useful for cross-border work, citing 2025 Brazil-Colombia operations on the Pure River. It does not state that those operations used its detections. MAAP's Operation Mercury analysis, by the same maker, used satellite imagery without machine learning.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
496,000 hectares affected between 2018 and 2025
- Category
-
Improved Knowledge Discovery
- Status
-
Not assessed
- Beneficiary groups
-
People. Organizations. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
496,000 hectares affected between 2018 and 2025
- Comparison
-
hand-labeled Sentinel-2 patches (Venezuela geographic holdout)
- Results
-
Result: specificity 0.976, sensitivity 0.991.
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
Not assessed
- Amount
-
Not assessed
- Currency and prices
-
Not assessed
- Period and unit
-
Not assessed
- Realization
-
Not assessed
- Recipients and payers
-
Not assessed
- Gross or net
-
Not assessed
- Included costs
-
Not assessed
- Comparison
-
Not assessed
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer: Earth Genome (model), with Amazon Conservation Association and the Pulitzer Center's Rainforest Investigations Network. Operator: Amazon Mining Watch partnership (public web platform).
- Jurisdictions
-
Benefit: Amazon basin (
Brazil, 
Peru, Guyana, 
Venezuela and other basin countries). Developer: 
United States. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Cloud-free composites over the period monitored. Mining scars visually distinct from sandbars, braided rivers and aquaculture ponds. Continued labeling and corroboration across years to limit false detections. A user, agency or newsroom willing to act on detections, which no cited source documents.
- Replication attempts
-
2 follow-up entries. Outcome unresolved: 1. Benefit recurred: 1.
- Later findings
-
1 August 2026. Earth Genome published a full rebuild of the Amazon Mining Watch models and data in August 2026, continuing the platform after the April 2026 report.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0348-source-0001–0006
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Independent, Developer
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
People. Organizations. Society.
- Depth
-
Not assessed
- Reasons
-
People: No individual beneficiary documented. Organizations: Journalists and non-governmental organizations (NGOs) use the maps, but no organization's decision is documented as changed by a detection. Society: Basin-wide statistics exist, but no enforcement or policy outcome is attributed to the tool in a cited source.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently reported
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Environmental monitoring: illegal gold mining in tropical forest. Non-profit remote-sensing platform for journalism and conservation.
- Purpose
-
Automated detection and mapping of artisanal gold-mining scars in satellite imagery. The users are journalists, researchers, civil society and policymakers. The mapping is the intended use.
- Dates
-
Date basis unassessed: 2026 (precision unassessed). Updated: 21 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: 65329a090a51…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
6 revisions. Latest: 21 October 2026.
Sources
-
Report Developer
Finer, M. et al. (2019). MAAP #104: Major Reduction in Illegal Gold Mining from Peru's Operation Mercury. Monitoring of the Andean Amazon Project, Amazon Conservation, 4 August 2019 (no mention of machine learning or AI).
The gold mining deforestation decreased 92% between 2018 (900 hectares) and 2019 (67 hectares)
-
News report Independent author
Malleret, C. (2026). AI tool tracks spread of illegal gold mining in Amazon protected areas. Mongabay, 24 April 2026.
496,000 hectares (1.2 million acres) of Indigenous territories and protected areas were affected by mining between 2018 and 2025.
Amazon Mining Watch trains artificial intelligence algorithms to recognize mining scars in satellite images.
-
Press release Developer
Amazon Mining Watch (n.d.). About. amazonminingwatch.org.
Amazon Mining Watch is a collaborative platform developed through a partnership between the Pulitzer Center's Rainforest Investigations Network, Amazon Conservation Association, and Earth Genome.
-
Other Developer
The mine detector is an ensemble of convolutional neural networks trained to discriminate mines from other terrain using hand-labeled examples in Sentinel-2 L1C imagery.
Venezuela holdout: specificity 0.976, sensitivity 0.991