ارزيابي كارايي الگوريتم هاي يادگيري ماشين در تهيه نقشه پتانسيل سيل گيري )منطقه مورد مطالعه: حوزه آبخيز هليل رود جيرفت)

eva‎luation of the Performance of Machine Learning Algorithms in Generating Flood potential Maps (Study Area: Halil Rud Jiroft Watershed)

جغرافياي طبيعي

Flooding is considered one of the most significant natural hazards, an‎d the Helilroud watershed is always at high risk of occurrence due to its climatic conditions, topographical features, an‎d geological structure. This study aimed to create a flood susceptibility map for this watershed by employing three advanced machine learning algorithms: XGBoost, Ran‎dom Forest, an‎d CART. All modeling stages were conducted using R software. In this study, actual flood occurrence points were extracted through a combination of field data, official reports, an‎d satellite image analysis, an‎d were entered into the model as dependent variables. Then, 30% of these points were used for training an‎d 70% for eva‎luating the model performance. The environmental data included a set of layers related to precipitation, slope, lan‎d use, drainage density, an‎d high-resolution digital elevation models. After removing multicollinearity factors, 15 main variables were selec‎ted as final inputs. The results indicated that the XGBoost model performed more accurately than the other two models, classifying approximately 16.87% of the watershed area into the "very high" susceptibility class; this value was 9.94% for the Ran‎dom Forest model an‎d about 18.5% for CART. Based on the final maps, the central areas of the watershed an‎d parts of the sub-basins of Jiroft, Anbarabad, Faribab-e Sharghi, Esfan‎degheh, Dashtab, an‎d the Ramon–Bahr-e Aseman region showed the highest flood susceptibility. This may be due to these areas being located in low-lying lan‎ds an‎d in proximity to major rivers. The findings of this research can provide practical support for hazard management, urban planning, an‎d flood damage reduction in the Helilroud watershed an‎d similar regions.

5

154541