Costache et al. (2026) Intelligent geospatial modelling of flash flood susceptibility using ICO-optimized hybrid ensembles

Identification

• Journal: Geomatics Natural Hazards and Risk
• Year: 2026
• Date: 2026-09-25
• Authors: Romulus Costache, Costel Pleşcan, Javed Mallick, Hoang Thi Hang, Hazem Ghassan Abdo
• DOI: 10.1080/19475705.2026.2736924

Research Groups

• Research Institute of the University of Bucharest, Bucharest, Romania
• Department of Civil Engineering, Faculty of Civil Engineering, Transilvania University of Brasov, Brașov, Romania
• Department of Ecological Restoration and Species Recovery, Danube Delta National Institute for Research and Development, Tulcea, Romania
• National Hydrological Forecast Center, National Institute of Hydrology and Water Management, Bucharest, Romania
• Department of Civil Engineering, College of Engineering, King Khalid University, Abha, Kingdom of Saudi Arabia
• Geography Department, Faculty of Arts and Humanities, Tartous University, Tartous, Syria

Short Summary

This study proposes an integrated framework for flash flood susceptibility assessment combining classical statistical models with machine learning algorithms optimized through the iterative classifier optimizer (ICO). The analysis was carried out in the Comana Basin (Romania) using ten relevant conditioning factors.

Objective

• Assess the susceptibility to flash floods in a study area characterized by pronounced hydrological dynamics, using an integrated methodology based on bivariate statistics, machine learning algorithms, and an iterative classifier optimizer.

Study Configuration

• Spatial Scale: Local scale, Comana River Basin (Romania)
• Temporal Scale: Not specified

Methodology and Data

• Models used:
  • Weights of Evidence (WoE)
  • K-Star (K*)
  • Logistic Model Tree (LMT)
  • Rotation Forest
• Data sources:
  • Satellite images from Google Earth
  • Digital Elevation Model (DEM) from SRTM 30 m database
  • CORINE Land Cover 2018

Main Results

• The models demonstrate excellent performance in predicting flash flood susceptibility, with ICO-RF-WOE performing the best.
• High-flood-risk areas are concentrated mainly on steep slopes.

Contributions

• This study proposes an integrated methodological framework for mapping flash flood susceptibility, combining the advantages of explanatory statistical analysis with those of advanced predictive modelling.
• The use of iterative classifier optimizer (ICO) improves the performance of machine learning models by iteratively adjusting hyperparameters and optimizing their structure.

Funding

• Not specified

Citation

@article{Costache2026Intelligent,
  author = {Costache, Romulus and Pleşcan, Costel and Mallick, Javed and Hang, Hoang Thi and Abdo, Hazem Ghassan},
  title = {Intelligent geospatial modelling of flash flood susceptibility using ICO-optimized hybrid ensembles},
  journal = {Geomatics Natural Hazards and Risk},
  year = {2026},
  doi = {10.1080/19475705.2026.2736924},
  url = {https://doi.org/10.1080/19475705.2026.2736924}
}