Triaging Deforestation Alerts: Clustering Alerts for Review

Advances in the automatic detection of deforestation and forest disturbance now provides alerts on disturbances as small as 30x30m pixels on an 8 to 16 day update schedule. Global Land Analysis & Discovery (GLAD) Alerts, developed by the University of Maryland are publicly available from Global Forest Watch. These provide valuable new insights into near real-time forest canopy loss. For agencies tracking deforestation, this introduces a new challenge to deal with the large volume of alerts. A technique is presented to triage deforestation alerts. Neighbouring alerts are processed into clusters. Combining many adjacent and nearby disturbances into clusters results in fewer cases to investigate. Cases can be forwarded to a crowd of volunteers for a quick visual check before being sent to locals to investigate. Both the original alert coordinates and their cluster perimeters may be encoded in the same GeoJSON Feature Collection. This aids both visualization and coordinating the response. Clustered alerts add value to the original alerts to aid local decision support activities. A prototype is demonstrated using Bunjil Forest Watch, using alert data from Global Forest Watch and processing by Google Earth Engine.