Assessing disasters in East Kalimantan: machine learning approaches for sustainable urban development

Abstract East Kalimantan, the designated site for Indonesia’s new capital, faces rising disaster risks due to rapid urban expansion, deforestation, and ecological degradation. These changes increase the likelihood of multiple, co-occurring hazards—posing serious challenges to sustainable development and disaster management. In response, this study developed a multi-hazard probability assessment using a Random forest algorithm, focusing on four major disasters: floods, landslides, forest fires (FFs), and droughts (DTs). By applying hazard-specific environmental variables, empirical threshold values, and bootstrap-based uncertainty analysis, the study produced spatially differentiated risk maps and identified areas with overlapping vulnerabilities. The results show that high-probability zones for floods and landslides are concentrated in the southern and central regions, while FF and DT risks are more prominent in the northern and western areas. Overlapping hazard zones, though relatively small in spatial extent, highlight critical regions where compound disaster risks may arise. These patterns suggest the need for region-specific mitigation strategies that reflect the dominant hazard characteristics in each area. Overall, the study provides baseline spatial information that can inform land-use planning and disaster management. The identification of environmental thresholds and uncertainty ranges supports more transparent and evidence-based decision-making.