Land Use and Land Cover Change Detection in Bogura District Using Random Forest and Google Earth Engine

Abstract

Land Use and Land Cover change analysis is essential for understanding environmental dynamics and supporting sustainable land management. This study determines changes in the Bogura District between 2009 and 2025 using Landsat imagery within the Google Earth Engine environment. A Random Forest classifier was employed to generate LULC maps based on spectral bands and indices, including NDVI, MNDWI, and EVI. Five major land cover classes were identified: urban, bare land, water bodies, vegetation, and cropland. The classified maps achieved overall accuracies of 88% and 91% for 2009 and 2025, respectively, indicating a high level of classification reliability. The results further demonstrate that cropland remained the dominant land use throughout the study period, expanding from 38.6% (1431.58 km²) in 2009 to 42.3% (1623.66 km²) in 2025. Vegetation covers also increased significantly from 24.3% to 39.3%, while bare land decreased markedly from 20.2% to 5.7%. Urban areas expanded from 3.7% to 6.5%, reflecting rapid urbanization. In contrast, water bodies declined from 13.3% to 6.2%, indicating potential hydrological and environmental concerns. Change detection analysis using a post-classification comparison approach reveals that major land transformations occurred from bare land and vegetation to cropland, and from multiple land classes to urban areas. Significant portions of water bodies were also converted into cropland and bare land. These transitions highlight the combined effects of agricultural expansion, urban growth, and river dynamics on land transformation. The study demonstrates the effectiveness of integrating cloud-based geospatial platforms with machine learning techniques for accurate LULC mapping and change detection. The findings provide valuable insights for land use planning, environmental management, and sustainable development in rapidly changing regions.