| Urban Flood Susceptibility Mapping with Convolutional Neural Network and Explainable AI: Tehran Case Study |
| کد مقاله : 1154-IWWA |
| نویسندگان |
|
مهدی آمره ئی1، محمد حسین نیک سخن2، بنت الهدی اصل روستا *2 1دانشکده محیط زیست _دانشگاه تهران 2دانشکده محیط زیست_دانشگاه تهران |
| چکیده مقاله |
| Urban flooding presents a major challenge to infrastructure resilience and public safety in rapidly growing cities including Tehran, the capital city of Iran. This study introduces an explainable deep learning framework for flood susceptibility mapping using Convolutional Neural Networks (CNNs) trained on multi-source geospatial data. Thirteen spatial input layers—including DEM, slope, aspect, curvature, TWI, TRI, TPI, SPI, SLF, land use, distance to rivers and drainage networks, NDVI, and precipitation—were used to extract 64×64 pixel patches for model training. Flood inundation maps generated from Sentinel-1 imagery were used as reference labels across multiple flood and non-flood events. The model was optimized with focal loss to address class imbalance. Its performance was evaluated using the Area Under the ROC Curve (AUC), achieving values above 0.95, indicating excellent predictive capability. Model interpretability was enhanced using SHAP and Grad-CAM, which revealed the spatial importance of input features. Finally, the flood susceptibility map of Tehran’s administrative boundary was produced, providing a valuable tool for flood risk assessment and urban planning. |
| کلیدواژه ها |
| Keywords: Flood mapping, CNN, Deep Learning, Explainable AI, Google Earth Engine |
| وضعیت: پذیرفته شده برای ارائه شفاهی |
