Rapid flood extent mapping and exposure assessment using SAR and machine learning in the Küçük Menderes Basin, Türkiye
Natural Hazards, cilt.122, sa.12, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 122 Sayı: 12
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s11069-026-08292-6
- Dergi Adı: Natural Hazards
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, IBZ Online, Environment Index, Geobase, INSPEC, Natural Science Collection (ProQuest), Earth, Atmospheric, & Aquatic Science Collection (ProQuest), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: Cloud computing, Flood extent mapping, GEE, LULC, ML, SAR, Sentinel-1/2
- Dokuz Eylül Üniversitesi Adresli: Evet
Özet
Floods remain among the most damaging natural hazards worldwide. This study develops a cloud-native, integrated remote sensing and machine learning (ML) workflow for rapid flood extent mapping and exposure assessment in the Küçük Menderes River Basin, Türkiye. Sentinel-1 SAR and Sentinel-2 optical data are processed on Google Earth Engine to (i) produce a high-resolution land-use/land-cover (LU/LC) baseline and (ii) delineate flood extent using complementary approaches: event-specific thresholding (SAR backscatter; MNDWI) and a supervised Random Forest (RF) classifier. The LU/LC model attains 98.39% overall accuracy, while the flood-mapping RF achieves 97.43% overall accuracy with a high F1 score for the water class. All adopted thresholds are reported and visualized via histograms, and a short sensitivity analysis demonstrates robustness to threshold choice. Population and cropland exposure are quantified by intersecting mapped inundation with high-resolution demographic and land-use layers, yielding 48,264 residents and 523 ha of cropland within the flood extent. Spatial consistency with official records (e.g., AFAD and ministerial reports) supports the plausibility of mapped patterns, particularly along known impact corridors. Limitations include the absence of in-situ validation, urban radar artefacts (e.g., layover/double-bounce), and potential temporal mismatch between LU/LC and the flood event; these are explicitly analyzed and positioned as priorities for future field campaigns and hybrid ML–hydrodynamic integration. The cloud-based design enables scalable, near-real-time mapping suitable for operational uptake. The outputs directly support SDG 11.5 (reducing disaster impacts) and SDG 13.1 (strengthening resilience) by informing targeted preparedness, drainage maintenance, and response planning.