نوع مقاله : مقاله علمی پژوهشی
نویسندگان
1 عضو هئیت علمی دانشگاه اصفهان
2 دانش آموخته کارشناسی ارشد سنجش ازدور و سیستم اطلاعات جغرافیایی دانشگاه اصفهان
3 دانش آموخته کارشناسی ارشد سنجش از دور و سیستم اطلاعات جغرافیایی دانشگاه اصفهان
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Flooding, as one of the most destructive natural disasters, causes significant loss of life and property annually in various regions worldwide, particularly in the semi-arid areas of Iran. In this context, accurate flood risk modeling plays a decisive role in the planning and risk management of this phenomenon. This research aims to assess and map flood risk in the Darab watershed, utilizing the Support Vector Machine (SVM) and Random Forest (RF) machine learning algorithms within the GIS environment and the R programming language, to compare the performance of these two models in spatial analysis. For this purpose, a comprehensive spatial database was compiled, including environmental indicators (soil type, geology, distance to road, NDVI), hydrological indicators (drainage density, distance to river, stream power, NDWI, TWI, SPI, and WRI), and topographical indicators (elevation, slope, aspect, and slope length). The results of the model performance evaluation indicated that both algorithms possess a suitable accuracy in identifying spatial flood patterns; however, the RF model demonstrated higher accuracy compared to SVM (R² = 0.8906 vs. R² = 0.875). The RF model identified approximately 71% of the watershed area as very low and low-risk zones, which aligns with the actual flood occurrence patterns (concentrated around the river channels and lower-lying areas of the watershed). In contrast, the SVM model, emphasizing the correlation between hydrological indicators and remote sensing data, classified nearly 77% of the watershed area into medium to very high-risk categories, primarily in the central and eastern parts of the basin. These findings highlight the high efficiency of machine learning algorithms, particularly the RF model, in flood risk modeling, confirming their superiority over traditional physical and semi-empirical methods like HEC-RAS in terms of performance, speed, and reduced need for field data.
کلیدواژهها [English]
)