Flood, landslides, forest fire, and earthquake susceptibility maps using machine learning techniques and their combination

Protection against natural hazards (i.e., floods, landslides, forest fires, and earthquakes) is vital in land-use planning, especially in high-risk areas. Multi-hazard susceptibility maps can be used by land-use manager to guide urban development, to minimize the risk of natural disasters. The objective of the present study was to use four machine learning models to produce multi-hazard susceptibility maps in Khuzestan Province, Iran. In this work, four different natural hazards (flood, landslides, forest fire, and earthquake) using support vector machine (SVM), boosted regression tree (BRT), random forest (RF), and maximum entropy (MaxEnt) techniques were created. Effective factors used in the study include elevation, slope degree, slope aspect, rainfall, temperature, lithology, land use, normalized difference vegetation index (NDVI), wind exposition index (WEI), topographic wetness index (TWI), plan curvature, drainage density, distance from roads, distance from rivers, and distance from villages. The spatial earthquake hazard in the study area was derived from a peak ground acceleration (PGA) susceptibility map. The second step in the study was to combine the model-generated maps of the four hazards in a reliable multi-hazard map. The mean decrease Gini (MDG) method was used to determine the level of importance of each effective factor on the occurrence of landslides, floods, and forest fires. Finally, “area under the curve” (AUC) values were calculated to validate the forest fire, flood, and landslide susceptibility maps and to compare the predictive capability of the machine learning models. The RF model yielded the highest AUC values for the forest fire, flood, and landslide susceptibility maps, specifically, 0.81, 0.85, and 0.94, respectively.