چكيده به لاتين
Floods are one of the most devastating natural disasters that cause significant loss of life and infrastructure damage every year. The change in climate patterns caused by the increasing emission of greenhouse gases will increase the sea level and increase heavy rainfall. The runoff resulting from heavy rains and melting snow, in the absence of proper infrastructure, including the Best Management Plans and the right choice of land use, will quickly turn into a flood, and as a result, there are many risks. It is very difficult to predict the location of floods due to sudden climatic changes and human factors. However, identification of flood sensitive areas with the help of machine learning and remote sensing methods can help in effective flood susceptibility mapping. The use of satellite information techniques and machine learning with the help of historical data can provide accurate predictions of flooding. In this research, seven effective machine learning models named logistic regression, decision tree, random forest, multi-layer perceptron, support vector machine, Adaboost and extreme gradient boosting were modeled to prepare a flood susceptibility map of Golestan province, located in the north of Iran. For this purpose, a geo-spatial database including 142 flooded locations mentioned in the study (Rahmati et al., 2016), and eleven flood influencing factors including: slope, elevation, distance from the river, River density, profile curvature, plan curvature, topographic wetness index, NDVI, land use, soil type and lithology of the area have been prepared and produced. The results obtained in the present study show that topographic wetness index, profile curvature and plan curvature do not contribute significantly to the flood susceptibility of Golestan province. Statistical criteria were used to evaluate the performance of the models, such as the confusion matrix and Receiver Operating Characteristic curve, to validate and compare the performance ability of the models. Then, for the validation data set, a flood prediction map was produced. The flood susceptibility map as the main output of this research was classified into five classes with the help of machine learning algorithms. The results show that the random forest model has the best performance in predicting flood susceptibility in the target area. According to the map output from the models, 5310 square kilometers equal to 55.89% of the area are high to very high susceptible to flooding.