An integrated machine learning and remote sensing approach for monitoring forest degradation due to Rohingya refugee influx in Bangladesh

Abstract

This paper aims to monitor and predict forest degradation due to the Rohingya refugee influx in Bangladesh using machine learning and remote sensing techniques. Close to a million refugees arrived between 2016 and 2018 and they were hosted in settlement camps by clearing forest land. Forest degradation is a continuous process as refugees venture into the peripheral areas for firewood collection. Therefore, monitoring and predicting forest degradation at a large scale or at the district level is important. Three land use land cover change indices were used to monitor forest degradation trend. Machine learning technique was used to reaffirm the forest degradation trend nearby refugee camps and to predict land use for 2030. The finding is that deep vegetation has decreased by an estimated 74.28 km2 between 2013 and 2019, which has been compensated by increase in shallow vegetation. Additionally, using random forest machine learning technique, the findings suggest that Teknaf and Ukhia, the two biggest concentrations of refugee camps have experienced the highest forest degradation. Finally, the predicted land use suggests doubling of shallow vegetation near refugee camps by 150.92 km2 between 2013 and 2030. The integrated machine learning and remote sensing approach of this study can thus be a cost-effective and efficient technique for continuous forest degradation monitoring.