A machine learning approach to understand how accessibility influences alluvial gold mining expansion in the Peruvian Amazon

Abstract

Alluvial small-scale gold mining (ASGM) mining in the Amazon is expanding fiercely, generating severe environmental degradation, which includes the fast disappearance of primary forests in a highly biodiverse area of the world. Different factors motivate the growth of mining in the areas and understanding this expansion is important to safeguard protected areas or implement strategies to mitigate the related social and environmental impacts. Thus, the goal of this study is to apply machine learning techniques to explore gold mining expansion in Madre de Dios, in the Peruvian Amazon, and to identify possible future hotspots of these activities. Using an unsupervised learning algorithm and a random forest classification model, past expansion trends were analyzed and an explicit geo-spatial model was built. Results demonstrate that proximity to infrastructure is not always indicative of high mining probability. In fact, when analyzing the spatial distribution of model accuracy, it is observed that model performance decreases in clusters where accessibility and mining activity showed opposite trends. In contrast, the models yield accuracies greater than 0.9 when accessibility-related variables stand out as the most important. The model, which is flexible and reproducible, demonstrates to be useful to enhance decision making when implementing geo-spatial policies to address the problem of ASGM expansion in the Amazon.