Assessing change point detection methods to enable robust detection of early stage Artisanal and Small-Scale mining (ASM) in the tropics using Sentinel-1 time series data

Artisanal and Small-Scale mining (ASM) provides essential livelihoods for many in developing countries but often lacks regulation, leading to environmental issues such as water pollution and deforestation. Timely and accurate mapping of ASM activities is vital for responsible mining that benefits the environment and local communities. Synthetic Aperture Radar (SAR) is crucial for regular ASM monitoring in cloudy regions due to its ability to penetrate clouds. However, atmospheric effects can limit its effectiveness, particularly with shorter wavelengths in wet tropical areas during the rainy season. This study utilised a time series smoothing technique to improve Sentinel-1 (S-1) SAR time series data, reducing SAR noise and atmospheric effects from heavy rainfall for early ASM activity detection. We tested three change point detection (CPD) methods, including cumulative sum (CuSuM), pruned exact linear time (PELT), and binary segmentation (BinSeg) in the Western and Ashanti wet regions in southern Ghana using the smoothed S-1 data for early ASM detection. We observed a relatively fast response of ASM activity tracking when utilising smoothed S-1 data at both sites for VV and VH polarizations during the rainy seasons. However, VH polarization is more effective than VV polarization during rainy seasons. While all CPD algorithms showed similar performance, CuSuM had the shortest lag time of up to 9 days, compared to 11 days for PELT and BinSeg. This method significantly reduces ambiguity caused by heavy rainfall when identifying change points due to ASM activity, making it a viable option for near real-time monitoring in wet tropical regions.