Predicting flood-prone areas using generalized linear and maximum entropy machine learning models

<p>The purpose of this study is to identify the effective factors, prepare flood risk prediction maps using machine learning models, and finally evaluate the efficiency of these models in the Zive watershed of Urmia. For this purpose, environmental and human factors including morphometric indices; Waterway Power Index (SPI), Slope Length Index (LS), Topographic Wetness Index (TWI), Topographic Position Index (TPI), Land Roughness Index (TRI), Mass Balance Index (MBI), Profile Curvature Index and The surface curvature index (Plan Curvature), rainfall, basin height, slope degree, slope direction, lithology, land use, normalized difference index of vegetation cover (NDVI), distance from waterway, distance from village and distance from fault were used. For this purpose, 96 flood spots were identified in the basin by using field visits and Google Earth images and sources received from the offices. Layers related to morphometric indices from the digital height model (12.5 x 12.5) meters and in the SAGA_GIS environment; And maps of environmental and human factors were prepared and digitized in the ArcGIS geographic information system. The evaluation results of two models using the ROC curve for machine learning (ML) models showed that the maximum entropy model with AUC=0.916 and the generalized linear model with AUC=0.902 have excellent performance in the field The results of the Kappa index for the superior model showed that environmental factors including geology, distance from waterways, height and slope have the greatest impact and the least impact related to profile curvature index factors. , land use, and mass balance index. Identifying high-risk areas and determining factors affecting the occurrence of floods in this basin can be very efficient in reducing possible damages.</p>