Assessing Four Decades of Land Use and Land Cover Dynamics in Mbeya, Tanzania Through Google Earth Engine and Random Forest Classification

Land use/land cover (LULC) change is a major environmental challenge affecting watershed ecosystems through agricultural expansion, urbanisation, and population growth. This study examined the spatiotemporal dynamics of LULC change within the watershed from 1990 to 2020 using multi-temporal Landsat imagery, Geographic Information Systems (GIS), remote sensing, and a Random Forest (RF) classifier. Seven LULC classes were mapped: bushland, cropland with scattered settlements, forest, grassland, tree plantation, urban area, and water bodies. Classification accuracy was assessed using a confusion matrix, producing an overall accuracy of 78.22% and a Kappa coefficient of 0.76, indicating substantial agreement between classified maps and reference data. The results revealed marked landscape transformation over the three decades. Cropland with scattered settlements increased from 63,745 ha in 1990 to 74,457 ha in 2020, while urban areas expanded nearly fourfold, from 2,047 ha to 7,800 ha. Conversely, forest cover declined from 76,461 ha to 68,753 ha, and grassland decreased from 37,087 ha to 31,641 ha. Sankey diagram analysis showed that the dominant land conversions were from forest, bushland, and grassland to cropland and urban areas. Agricultural expansion was the principal driver of change during 1990–2010, whereas urbanisation became increasingly influential during 2010–2020. These trends reflect growing anthropogenic pressure, leading to vegetation loss, landscape fragmentation, and reduced ecological stability. The study demonstrates the value of integrating remote sensing, GIS, and machine learning for long-term LULC monitoring and provides essential evidence to support sustainable watershed management and land-use planning.