Tracking Gold Mining Activity Behind Tropical Deforestation with Satellite Time Series

Gold mining is a major and underestimated driver of tropical deforestation.Most forest loss from gold mining comes from unrecorded mines, many of them artisanal and small-scale gold mining (ASGM) operations. These operations drive deforestation, carbon emissions, and mercury pollution, yet policymakers lack a record of where and when they are active. Satellite detection tells where mining is, but not whether a site is worked or idle. We propose the Activity Index, which turns yearly satellite imagery into a record of mining activity, revealing expansion, decline, and reactivation with a measure of confidence. To evaluate it, we build Pan-ASGM, a multi-temporal semantic-segmentation dataset of nine sites in five countries across three continents. Under leave-one-site-out evaluation, the index recovers the direction of change in pond activity, a key proxy for ASGM activity, on 74.9\% of changed pixels, rising to 90.2\% behind its reliability filter. It outperforms hard-label stacking and classical trend baselines. It also captures the decline and subsequent reactivation of mining activity following a documented enforcement action in Peru, supporting its use for policy evaluation.