Identification of the Most Important Environmental Variables in Spatial Prediction of Flood Prone Areas using the Maximum Entropy Model in Parts of Golestan Province

Ehsan Moradi¹

1) PhD in Water Science and Engineering

IOI Code : AAAB-ABZBCE
Publication : 2nd International Congress on Engineering Sciences and Sustainable Urban Development(dtuconf.com)
Abstract :
    Flooding is a serious and devastating natural hazard in recent years. In this study, the maximum entropy model, which is one of the advanced data mining models, was used to model and prepare a spatial map of flood-prone areas in the Salian Tepe watershed with an area of ​​4515.47 km2 in Golestan province. First, a flood distribution map was prepared based on existing reports and field surveys.For maximum entropy modeling, thirteen influential variables were identified as predictive factors including elevation classes, slope percentage, slope direction, rainfall, distance from drainage network, land use, lithology, soil texture, design curvature, profile curvature, topographic moisture index, drainage density, and flow power index and introduced into the model.Three different sets of flood hazard points (ds1, ds2, ds3) including 70% for model training and 30% for validation were randomly prepared to evaluate the accuracy and integrity of the model. The accuracy of the model was evaluated based on the ROC index, and the maximum entropy model showed excellent prediction accuracy (above 90%).Also, in this study, the degree of importance of the variables was examined by the model, and the factors of drainage density (about 49% importance), distance from the stream (about 15% importance) were identified as the most important environmental factors affecting flooding in the study area, and were identified as the most important predictive factors in this area, respectively.

Keywords : salian tepe watershed robustness (model integrity) flooding roc index predictive factors data mining model