ارزیابی خطر رخداد سیل در مناطق سیل‌خیز با بکارگیری مدل‌های احتمالاتی و یادگیری ماشین مطالعه موردی: حوزه آبخیز قره‌سو و گرگانرود استان گلستان

Flooding is one of the most destructive natural disasters with social, economic and environmental consequences, and machine learning and probabilistic methods have been developed to model and predict it. The objectives of this study are to prioritize the effective factors, zoning and predicting the sensitivity to flood events using support vector regression (SVR) and weighted Bayesian evidence (WOE) models and to introduce the most appropriate ones in the Gharesu and Gorganroud basins of Golestan province. A flood event distribution map was prepared based on 368 locations with a repetition of 782 flood events based on available information, field visits and Google Earth images in the period before and after the flood. Of this number, 70 percent (258 points) were randomly selected and divided as training data and 30 percent (110 points) as test data for model implementation and validation, respectively. In the next stage, 15 geological, hydrological, and morphometric factors of the basin were used along with climate data (as independent variables) for zoning and modeling. The relationship between the influencing factors and flood events was quantified and weighted using frequency ratios. In order to examine the information overlap of the influencing factors, the independence of the data was tested using linear multivariate regression analysis. The sensitivity zoning maps obtained from the implementation of the two models were evaluated, validated, and compared using the area under the receiver operating characteristic curve (AUC-ROC). The results of the investigation of the influencing factors in both models unanimously showed that land use, elevation, and vegetation factors had a significant impact on flood occurrence, respectively, and a large portion (more than 80%) of the floods occurred in the very high and high sensitivity classes. The results of the classification and validation of the models showed that the AUC-ROC for the success rate of the support vector regression and weighted control models were 0.92 and 0.88, respectively. Therefore, the results of the support vector regression model are more accurate than the weighted control model. The flood susceptibility zoning maps obtained from this study can be used as a basis for planning and crisis management due to flood events.