Leveraging machine learning and Landsat time series for high-resolution mapping of mining-induced vegetation changes in Ouagadougou, Burkina Faso

Abstract

Quarries are open pit mines, where construction materials such as granite and clay are extracted from the earth's surface, leading to a major disturbance of vegetation and subsequent degradation of ecosystem services. Assessing mining-associated degradation over time remains a challenge. This paper uses a temporal series of Landsat satellite image extracted from Google Earth Engine (GEE) for monitoring mining effects on land cover around the West African city of Ouagadougou, Burkina Faso. We conducted a binary classification of quarry and non-quarry using Landsat 9 images. To account for varying quarry patterns, we applied the LandTrendr algorithm to analyze images from 1990 to 2022, mapping annual vegetation disturbance and recovery. This generated high-resolution land cover maps with over 90 % accuracy across 257 quarries (38 km²). LandTrendr identified disturbance years with 87 % accuracy and assessed recovery in abandoned quarries. Remote sensing accuracy was validated through fieldwork and analysis of historical aerial photos. Our findings revealed that major disturbances coincided with new quarrying activities, particularly during urban expansion from 2005 to 2019, with NDVI decreasing by 0.3 in active mining areas. Vegetation recovery was observed in 23 % of the abandoned quarries, with NDVI increases by 0.15 within five years post abandonment, particularly in clay quarries. The strongest recovery occurred in given up abandoned from 2015 to 2019, aligned with increased conservation efforts. Our approach offers valuable insights into mining impacts and vegetation resilience in semi-arid environments, informing sustainable urban planning and environmental management in rapidly transforming regions of West Africa.