پهنه‌بندي خطر سيل‌خيزي با استفاده از روش‌هاي داده‌كاوي (منطقه مورد مطالعه: حوضه آبخيز داراب)

Abstract Flooding is one of the most significant natural hazards, causing extensive an‎d often irreparable damage to lives, property, an‎d the environment every year. This study aims to identify an‎d map flood risk in the Darab watershed, located in Fars, Iran, using advanced data mining techniques. The Darab watershed is recognized as a flood-prone area due to its unique climatic conditions, topography, an‎d geological structure. In this research, five data mining models—MaxEnt, CART, Ran‎dom Forest, MARS, an‎d TreeNet—were employed to analyze data an‎d predict flood risk.The innovation of this study lies in the integration of these models to generate comprehensive an‎d practical flood risk maps, which can serve as a scientific foundation for crisis management in similar regions. The dataset used in this research includes 19 environmental parameters, such as rainfall, lan‎d slope, soil type, drainage density, lan‎d use, moisture content, an‎d topography indices. These data were collected from reliable sources, including satellite imagery, digital elevation models (DEMs), meteorological stations, an‎d geological maps. Specifically, Lan‎dsat 8 satellite imagery an‎d water indices were used to identify flood-prone areas.Following data preprocessing, 70% of the dataset was allocated for training the models, while the remaining 30% was reserved for testing. The modelsʹ accuracy was then validated using various eva‎luation metrics. The results indicated that approximately 40% of the Darab watershed falls within the high-risk category, especially in low-lying areas with sparse vegetation cover. Additionally, the study found that variables such as precipitation, distance from watercourses, an‎d altitude had the most significant impact on flood risk in this basin.The results shows, the CART model demonstrated the highest capacity for delineating flood risk, achieving an accuracy of 99%. Other models showed varying levels of accuracy: TreeNet achieved 78%, Ran‎dom Forest scored 60%, MARS reached 52%, an‎d MaxEnt attained 50%. These findings highlight the effectiveness of the CART model in flood risk prediction, while also emphasizing the potential of data mining methods for flood risk management in flood-prone regions. Keywords: Flood, Data-Mining, Darab watershed

5 فصل