A Near Real-Time Mine Alert System using Sentinel 1 data

Abstract:

Illegal gold mining in the Amazon has become a key driver of deforestation and forest degradation, promotes additional illegal activities and has significant negative impacts on public health. Traditional remote sensing based monitoring systems can detect gold mining activity but are limited by frequent cloud cover and lag time between detection and reporting. Accuracy of detection is also a significant issue, as confidence in the alerts determines the use of end users and can affect the overall confidence in the data for stakeholders wanting to pursue legal steps on the ground to deter illegal mining. Further, automated mapping processes all have errors, therefore it is important to verify that a mapped event is accurate before incurring the costs (and risk) associated with sending enforcement officials into the field to curtail illegal mining.

Here, we present a mine monitoring tool, named RAMI--Radar Mining Monitoring-- a system to detect the advance of gold mining deforestation based on radar (Sentinel-1) products, which creates a seamless user interface to bridge the gap between near real-time map products, high resolution imagery, and enforcement officials. We will outline the flexible web-based user interface, underlying system architecture, built on Django framework for python and Google Earth Engine (GEE), and the verification system that seamlessly communicates with Collect Earth Online (CEO). Collect Earth Online is an open-source, satellite image viewing and interpretation system for use in projects that require land cover and/or land use reference data. The integrated framework alerts users when events are mapped, then they are able to verify it using high resolution images, such as Planet WMS feed, prior to making an enforcement decision. The subsequent results can also be used to continue to improve maps by supplying additional data for calibration of the mapping algorithms. We investigate the use of three detection change algorithms for monitoring mining activities: a Q-test; a Change Point Detection; and Radar Change Ratio. Early results suggest a significant improvement on illegal gold mining classification compared to general forest cover loss near real-time systems.