Floods, as a disastrous phenomenon, have a significant impact on natural ecosystems and human lives. Given the increased likelihood of frequency and intensity of floods in the future due to climate change, forecasting and planning in advance in flood-prone areas is essential to reduce damage. In recent years, Golestan province has experienced several floods. In this study, advanced remote sensing methods and machine learning were used to identify and predict areas with potential for flooding in Golestan province. Specifically, using Sentinel-1 radar images (SAR), flood occurrence areas were identified between 2015 and 2022. The variables of digital elevation model, normalized difference vegetation cover index, topographic slope, vegetation moisture index, flow accumulation, flow distance, monthly precipitation, flow direction, which did not show any signs of collinearity between them according to the VIF test, were selected as environmental variables affecting flood. Then, with seven machine learning algorithms including GLM, GAM, BRT, RF, MARS, FDA and CART, flood-prone areas were predicted. Then, by combining the single models, the uncertainty in the results was reduced. The performance of the models was evaluated using four indices AUC, COR, TSS and Deviance. The results of this study showed that the variables of digital elevation model, flow accumulation and rainfall are the most important variables in predicting flood potential in this region. All single models showed high performance and among the models RF had the best performance. Based on the combined model, Gomishan city, Agh Ghala city, Bandar Turkmen city, western parts of Gonbad Kavous city, northern part of Bandar Gaz city and a small part in the north of Ramian city in Golestan province have the highest potential for flood occurrence. The results of this study will help planners, decision-makers and managers of Golestan province to take appropriate measures to prevent and reduce flood occurrence in predicted spatial situations. |
