نوع مقاله : پژوهشی
نویسندگان
1 دانشجوی دکتری مهندسی عمران، گروه مهندسی عمران، واحد تهران مرکزی، دانشگاه آزاد اسلامی، تهران، ایران
2 استادیار گروه مهندسی عمران، واحد رودهن، دانشگاه آزاد اسلامی، رودهن، ایران
3 استادیار گروه مهندسی عمران، واحد تهران مرکزی، دانشگاه آزاد اسلامی، تهران، ایران
چکیده
کلیدواژه‌ها
عنوان مقاله [English]
نویسندگان [English]
Introduction and goal
The phenomenon of flood is one of the frequent hazards that have caused financial losses and many lives in Iran in the last decade. One of the three main natural hazards of Iran is the occurrence of floods, and it is safe to say that a large flood occurs at least once a year in some part of this country. Floods are considered as a great threat to human life by damaging or killing people and animals, especially buildings and houses, agricultural land and crop production, urban infrastructure, bridges and roads. Floods in Iran cause a lot of damage from an economic point of view; it causes destruction of the environment, natural and residential resources and loss of life. In the past years, about 70% of the annual credits of the plan to reduce the effects of natural disasters and the headquarters of unexpected events have been used to compensate for the damages caused by floods. In this research, by using flood location data and using machine learning models and random forest data mining and the weight of evidence and their combination, the spatial prediction of flood prone areas has been discussed.
Materials and methods
In this research, combined models and 11 predicting variables of flood probability in Karkheh watershed located in Lorestan province have been used. These variables include maps of geomorphometric indicators, including topographic humidity index, relative slope position and topographic position index, hydrological maps including: drainage density and distance from the drainage network, for this purpose, first, data mining models for the initial analysis of the relationships between environmental variables and events Past risks are used and their results are used as input data for machine learning models. Flood data were randomly divided into two groups: training (70%) and validation (30%). Prediction accuracy was evaluated using operating characteristic curve (ROC) method.
Results and discussion
According to the results of the random forest model, the accuracy was 0.904, the weight of evidence was 0.886, and the combined model of random forest - the weight of evidence was 0.978. Based on the combined model of random forest and the weight of evidence as the superior model, 20.49% of the surface has medium upward potential. Based on the random forest model, drainage density, distance from waterways, height and land use were the most important factors affecting flood potential.
کلیدواژه‌ها [English]
)