Mapping flood susceptibility with a stacking ensemble approach: an exemplary case study in Yemen
Acta Geophysica, cilt.74, sa.3, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 74 Sayı: 3
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s11600-026-01915-3
- Dergi Adı: Acta Geophysica
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Compendex, Geobase, INSPEC, Academic Search Ultimate (EBSCO), Natural Science Collection (ProQuest), Earth, Atmospheric, & Aquatic Science Collection (ProQuest), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: Flash flood, GIS, Machine learning, SAR
- Kocaeli Üniversitesi Adresli: Evet
Özet
Flood susceptibility mapping (FSM) provides a practical basis for land use planning and disaster risk reduction in data-scarce and flood-prone regions such as the Qaa’Jahran watershed in Dhamar, Yemen. Many FSM studies still rely on single learners and rarely pair rigorous ensemble validation with transparent inventory verification (e.g., using SAR observations to check reported inundation extents), leaving a methodological gap for arid and semi-arid basins. This study introduces a stacking ensemble machine learning approach for flood susceptibility mapping in the Qaa’Jahran watershed, addressing the need for more accurate flood prediction in data-scarce regions. A flood inventory was compiled from documented historical flood information and refined using SAR imagery as a supplementary verification layer, and 15 conditioning factors were compiled in a GIS environment. Five base classifiers (Artificial Neural Networks (ANN), Random Forest (RF), K-Nearest Neighbors (kNN), Support Vector Machines (SVM), and Logistic Regression (LR)) were combined with a LR metamodel, and the resulting historical flood inventory (verified using SAR imagery) was used for model training. The stacking ensemble outperformed all individual models (AUC = 0.92–0.97) and two current benchmarks (TPOT and ABRBF) in discrimination (AUC = 0.98) and classification (accuracy = 98.75%). Event-based comparison against the 2022 flood event indicates strong spatial agreement between observed inundation and mapped high-susceptibility zones. The proposed workflow is transferable to other data-scarce basins and provides a reproducible decision support tool for prioritizing flood mitigation and preparedness under evolving hydroclimatic conditions.