Mapping urban expansion and land transformation in Dhaka, Bangladesh by fusing night-time lights, thermal, and spectral data via machine learning approaches

This study evaluates urban expansion and land transformation in Dhaka District, Bangladesh, between 2016 and 2024 by integrating multi-source geospatial data with machine learning. The aim was to map urban growth more precisely than conventional spectral indices allow and to characterize its spatial patterns through landscape metrics and change detection. Landsat 8 thermal and spectral data were fused with Visible Infrared Imaging Radiometer Suite (VIIRS) night-time lights, a suite of spectral indices (NDVI, NDBI, SAVI, MNDWI), Gray-Level Co-occurrence Matrix (GLCM) texture features, and road-network data. Three supervised classifiers, Random Forest (RF), Support Vector Machine (SVM), and Gradient Tree Boosting (GTB), together with a majority-voting ensemble, were compared. Random Forest performed best in every study year (overall accuracy 99.57–99.62%; Kappa 0.991–0.992) and was adopted as the urban layer for all subsequent analyses. The urban area expanded by 39.53 km 2 over the period. Vegetation and agricultural land were the predominant source of new urban land (75.6%), whereas direct water-to-urban conversion was minor (below 2%); total water-body area nonetheless declined by roughly 25% (from 79.0 to 58.9 km 2 ), largely through indirect conversion to vegetation and seasonal land rather than through direct urbanization. Landscape metrics indicate increasingly fragmented growth, with patch density rising from 2.46 to 7.82 per 100 ha and Shannon's entropy from 0.45 to 0.49. Zonal analysis identifies the northeastern and southeastern peri -urban fringes as the fastest-growing, dominated by leapfrog and edge development. The findings depict an unsustainable expansion trajectory whose environmental costs stem from the direct loss of vegetation and agricultural land alongside indirect pressure on floodplain water bodies, and they offer a transferable multi-modal framework for monitoring sprawl in Global South megacities. The application of the machine learning and geospatial approach in this study will help urban planners, policymakers, and stakeholders manage Dhaka's rapid and fragmented urban growth.