Mohajane et al. (2026) Machine Learning-Based Flood Susceptibility Mapping Using Geoenvironmental Factors in Central Morocco

Identification

• Journal: Earth Systems and Environment
• Year: 2026
• Date: 2026-02-01
• Authors: Meriame Mohajane, Sk Ajim Ali, Sliman Hitouri, Renata Pacheco Quevedo, Tadesual Asamin Setargie, Costanza Fiorentino, Ismail ElKhrachy, Paola D’Antonio, Meriam Lahsaini
• DOI: 10.1007/s41748-025-01019-w

Research Groups

• Department of Agriculture, Forest, Food, and Environmental Sciences, University of Basilicata, Potenza, Italy
• Department of Geography, Faculty of Science, Aligarh Muslim University (AMU), Aligarh, India
• Dr. Bhupendra Nath Dutta Smriti Mahavidyalaya, Hatgobindapur, Bardhaman, India
• Geosciences Laboratory, Department of Geology, Faculty of Sciences, Ibn Tofail University, Kenitra, Morocco
• Mohammed VI Polytechnic University, Geology & Sustainable Mining Institute, Ben Guerir, Morocco
• ENGAGE Research Group, Department of Geography and Regional Research, University of Vienna, Vienna, Austria
• Faculty of Civil and Water Resources Engineering, Bahir Dar Institute of Technology, Bahir Dar University, Bahir Dar, Ethiopia
• Research Institute for Geo-hydrological Protection, National Research Council (CNR-IRPI), Perugia, Italy
• Civil Engineering Department, College of Engineering, Najran University, Najran, Saudi Arabia
• Institute of Geosciences and Earth Resources (IGG), National Research Council of Italy (CNR), Pisa, Italy

Short Summary

This study evaluates and compares three machine learning models (CART, SVM, XGBoost) for flood susceptibility mapping in the Tensift Watershed, Central Morocco, identifying Classification and Regression Trees (CART) as the most accurate model with an Area Under the Curve (AUC) of 0.882.

Objective

• To assess the applicability and evaluate the performance of Classification and Regression Trees (CART), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost) models for flood susceptibility mapping in the Tensift Watershed, Central Morocco, using sixteen geoenvironmental conditioning factors.

Study Configuration

• Spatial Scale: Tensift Watershed, central-western Morocco (Marrakech province), covering an area of 10,819.9 square kilometers.
• Temporal Scale: Historical flood events (e.g., 1995, 2023) were used to create the flood inventory map, informing the prediction of future flood occurrences.

Methodology and Data

• Models used: Classification and Regression Trees (CART), Support Vector Machines (SVM), Extreme Gradient Boosting (XGBoost).
• Data sources:
  • Flood Inventory: 228 flood inventory points (114 flood, 114 non-flood locations) derived from field surveys, Google Earth Pro imagery, GIS processing (ArcGIS 10), meteorological records, and satellite imagery.
  • Conditioning Factors (16 geoenvironmental factors):
    • Topographic: Elevation (Digital Elevation Model, 25 m spatial resolution), Slope, Aspect, Plan Curvature, Terrain Ruggedness Index (TRI), Sediment Transport Index (STI), Stream Power Index (SPI), Topographic Wetness Index (TWI).
    • Hydrologic: Distance to streams (from geossc.ma).
    • Climatic: Rainfall (from study area stations).
    • Geologic: Distance to faults, Lithology (from geological map of Morocco, 1:1,000,000 scale).
    • Land Cover/Use: Soil (from Food and Agricultural Organization (FAO) data, 25 m resolution), Land Use/Land Cover (LULC) (from Landsat 8 OLI, 25 m resolution), Normalized Difference Vegetation Index (NDVI) (from Landsat 8 OLI).
    • Anthropogenic: Distance to roads.
  • Data Split: 70% (160 points) for training, 30% (68 points) for validation.
  • Multicollinearity Assessment: Variance Inflation Factor (VIF) and Tolerance (TOL).
  • Accuracy Assessment: Receiver Operating Characteristic (ROC) curve analysis, Area Under the Curve (AUC).

Main Results

• Multicollinearity analysis confirmed no serious multicollinearity among the 16 predictor variables (TOL > 0.1, VIF < 10).
• The CART model achieved the highest predictive accuracy with an AUC of 0.882 (95% confidence interval: 0.833–0.930, standard error: 0.025).
• The SVM model followed with an AUC of 0.860 (95% confidence interval: 0.807–0.912, standard error: 0.027).
• The XGBoost model had an AUC of 0.833 (95% confidence interval: 0.777–0.889, standard error: 0.029).
• All three models demonstrated "excellent discrimination" based on AUC values (between 0.8 and 0.9).
• Feature importance ranking (using the best-performing CART model) identified elevation (1.86), distance to streams (1.72), rainfall (1.66), slope (1.50), and lithology (1.48) as the most influential factors. Soil type (0.46) and NDVI (0.50) were ranked as the least important.
• The CART model predicted the highest percentage of the study area as "high" (20.61%) and "very high" (20.97%) flood susceptibility classes compared to SVM and XGBoost.
• Most flood-prone areas were identified along tributaries originating in the High Atlas or Jbilet regions, consistent with historical flood events.

Contributions

• This study represents the first-time application and comparative evaluation of CART, SVM, and XGBoost for flood susceptibility mapping in the Tensift watershed, Morocco.
• It provides novel insights into regional flood risk assessment using machine learning approaches in a region characterized by distinct geological and climatic conditions.
• The research identifies CART as the most effective machine learning algorithm for flood susceptibility mapping within this specific context.
• It generates comprehensive flood susceptibility maps for the entire Tensift watershed and Marrakech region, addressing a gap in existing literature.
• The findings are expected to support flood risk management activities, guide infrastructure planning, enhance community preparedness, aid in early warning systems, contribute to ecosystem preservation, facilitate efficient resource allocation, and inform evidence-based policy development in the Marrakech province.

Funding

• Not explicitly mentioned in the provided paper text.

Citation

@article{Mohajane2026Machine,
  author = {Mohajane, Meriame and Ali, Sk Ajim and Hitouri, Sliman and Quevedo, Renata Pacheco and Setargie, Tadesual Asamin and Fiorentino, Costanza and ElKhrachy, Ismail and D’Antonio, Paola and Lahsaini, Meriam},
  title = {Machine Learning-Based Flood Susceptibility Mapping Using Geoenvironmental Factors in Central Morocco},
  journal = {Earth Systems and Environment},
  year = {2026},
  doi = {10.1007/s41748-025-01019-w},
  url = {https://doi.org/10.1007/s41748-025-01019-w}
}