Flood susceptibility zonation map using remote sensing and XGboost, Random Forest, Nearest neighbor models in GIS: a case study Tebessa city, Algeria.

Abstract

Abstract Predictive mapping of flooding zone occurrence at the regional scale was performed for the urban perimeter of Tebessa City, Algeria, using remote sensing by Landsat to detect the past effect of flood occurrences of the last 10 years and estimate the future potential flood disaster. Machine learning ML was instrumental in predicting the flooding zones of Tebessa City through the utilization of three models: XGboost, Random Forest, and Nearest neighbor models. To generate the Flood susceptibility zonation map of Tebessa City, approximately 495 flood locations, and 490 Non-flood locations were selected as a training data according to the data set recorded by The direction of civil protection of the wilaya of Tebessa, and about 15% of them were used as a validation set. To predict the 54,945 locations as a test dataset; The estimated accuracy values of prediction rates using the Accuracy score method for XGboost, Random Forest classified, and Nearest neighbor models were 93.92%, 93.91%, and 93.24%, respectively. Geospatial databases relevant to flood zone occurrence (hydrographic factors, topographic factors, climatic factors, and human factors) were analyzed in a geographic information system environment GIS.