Mapping forest loss and encroachment drivers using remote sensing data and random forest classification

• Published in: Spatial information research (Online) Vol. 33; no. 3; p. 17
• Main Authors: Ghimire, Bhoj Raj, Maharjan, Bijaya, Sharma, Bharat Kumar, Poudel, Shobha, Mishra, Bhogendra, Shahi, Tej Bahadur
• Format: Journal Article
• Language: English
• Published: 대한공간정보학회 01.06.2025
• Subjects: 공학일반
• ISSN: 2366-3286; 2366-3294
• DOI: 10.1007/s41324-025-00616-1

Forest conservation is imperative to safeguard biodiversity and ensure sustainability of ecosystems providing a wide range of ecological and economic benefits. Forests are crucial in carbon sequestration, water regulation, and soil preservation. In Nepal, a stretch of forest has been lost or deteriorated due to human activities, with forest encroachment being a major factor that converts forested land to other uses, such as agriculture, settlements, and commercial zones. Here, we develop a framework to identify drivers of forest encroachment in Sudurpaschim Province (SP) of Nepal. We identified forest loss areas between 1990 and 2020, conducted field data collection for four identified encroachment drivers within the forest loss areas, trained six different machine learning classifiers, and compared their accuracy using field based dataset. The Random Forest classifier outperformed other classifiers with an overall accuracy of more than 81%, a recall of 58%, an F-score of 65%, and a precision of 81%. Therefore, we used the Random Forest model to map these drivers across the study area. Agricultural expansion was the dominant driver (78.21%), followed by settlement expansion (15.97%). Furthermore, the forest change analysis showed that a total of 1,634 $$\hbox {km}^{2}$$ km 2 of forest was converted to non-forest areas during the study period. Finally, the study provides recommendations to prevent future forest encroachments.