EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Spatial prediction of flood susceptibility using gradient boosting machines in the western mountainous region of Ha Tinh Province, Vietnam

Duc Loc Nguyen, Khac Nang Nguyen, Trong Phuong Tran

Environmental Research Communications · 01.07.2026 · 8 (7) · 075027

DOI: 10.1088/2515-7620/ae8aba ↗

Kurz gefasst

Fachartikel in der Zeitschrift Environmental Research Communications, online erschienen am 22.07.2026. Verfasst von Duc Loc Nguyen und 2 weiteren Autor:innen. Die Arbeit nennt 59 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Vietnam National University of Agriculture.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Flooding remains a catastrophic natural hazard causing severe socio-economic damage globally and within Vietnam. Accurate flood susceptibility mapping (FSM) is critical for mitigation strategies, yet conventional machine learning applications often suffer from multicollinearity, sub-optimal hyperparameter tuning, and lack of model interpretability. This study develops a robust, interpretable, and automated FSM framework by integrating gradient boosting machines, including XGBoost, LightGBM, and categorical boosting (CatBoost), with the Optuna optimization framework and SHapley Additive exPlanations (SHAP). Focus was directed at the flood-prone mountainous districts of western Ha Tinh Province, North-Central Vietnam. A flood inventory was constructed using a Sentinel-1 radar change-detection approach via Google Earth Engine (GEE) across three historical flood events (in 2018, 2019, and 2020). Ten conditioning factors were validated using Pearson correlation and variance inflation factor assessments to prevent data redundancy. Experimental results showed that all three models achieved exceptional predictive power, with testing accuracies exceeding 0.95 and AUC values above 0.97. Notably, the Optuna-optimized CatBoost model outperformed the other classifiers, securing the highest testing accuracy (0.9583) and AUC (0.9807). Global and local SHAP analyses decoded the models’ black-box decisions, revealing elevation (ELE) as the overarching governor of flood distribution, while highlighting that normalized difference built-up index and density of road significantly amplify runoff. Conversely, normalized difference vegetation index acts as a critical mitigator. The generated maps isolated high to very high risk zones across 24%–26% of the territory, concentrated in low-lying, urbanized plains. These highly accurate outputs provide transparent, actionable spatial intelligence to support sustainable land-use zoning and resilient disaster risk governance.

Förderung

  • Vietnam National University of Agriculture

Zitieren

APA

Nguyen, D. L., Nguyen, K. N., & Tran, T. P. (2026). Spatial prediction of flood susceptibility using gradient boosting machines in the western mountainous region of Ha Tinh Province, Vietnam. Environmental Research Communications, 8(7), 075027. https://doi.org/10.1088/2515-7620/ae8aba

Vancouver

Nguyen DL, Nguyen KN, Tran TP. Spatial prediction of flood susceptibility using gradient boosting machines in the western mountainous region of Ha Tinh Province, Vietnam. Environmental Research Communications. 2026;8(7):075027. doi:10.1088/2515-7620/ae8aba

BibTeX
@article{Nguyen2026,
  author = {Nguyen, Duc Loc and Nguyen, Khac Nang and Tran, Trong Phuong},
  title = {Spatial prediction of flood susceptibility using gradient boosting machines in the western mountainous region of Ha Tinh Province, Vietnam},
  journal = {Environmental Research Communications},
  year = {2026},
  volume = {8},
  number = {7},
  pages = {075027},
  doi = {10.1088/2515-7620/ae8aba}
}
Kostenlos starten →