Riche et al. (2026) Predicting LULC Changes and Assessing their Impact on Surface Runoff with Machine Learning and Remote Sensing Data

Identification

• Journal: Water Resources Management
• Year: 2026
• Date: 2026-03-01
• Authors: Abdelkader Riche, Ammar Drias, Riccardo Ricci, Boularbah Souissi, Farid Melgani
• DOI: 10.1007/s11269-026-04560-z

Research Groups

• Department of Geography and Territorial Planning, University of Sciences and Technology Houari Boumediene, Algiers, Algeria
• Department of Information Engineering and Computer Science, University of Trento, Trento, Italy
• Department of Telecommunication, University of Sciences and Technology Houari Boumediene, Algiers, Algeria

Short Summary

This study developed an approach integrating remote sensing and machine learning to predict future land use and land cover (LULC) changes and assess their impact on surface runoff in a semi-arid Mediterranean watershed. It found that urbanization significantly increases runoff, while forests mitigate it, with land factors having limited influence during intense rainfall events.

Objective

• To investigate the influence of urbanization-induced land use and cover (LULC) changes on surface runoff by integrating remote sensing, machine learning, and the SCS-CN method.
• To forecast future LULC changes from 2000 to 2040, incorporating the temporality of influencing factors, and dynamically update Curve Number (CN) values to estimate runoff volume.

Study Configuration

• Spatial Scale: Wadi El Harrach sub-watershed, Algiers region, Algeria, covering an area exceeding 800 square kilometres. The region spans latitudes from 36°25’11” N to 36°46’40” N and longitudes from 2°56’34” E to 3°19’39” E, with elevations ranging from -1 metre to 1256 metres above sea level.
• Temporal Scale: LULC analysis and prediction for the years 2000, 2005, 2010, 2015, 2020, 2025, 2030, and 2040. Rainfall data used for return period calculations spanned from 1970 to 2019.

Methodology and Data

• Models used:
  • LULC Classification: Support Vector Machine (SVM) algorithm with Radial Basis Function (RBF) kernel.
  • LULC Prediction: Random Forest (RF) algorithm (selected for highest accuracy among Ada-Boost, ANN-mlp, KNN).
  • Runoff Modeling: Soil Conservation Service - Curve Number (SCS-CN) method.
  • Rainfall Return Periods: Generalized Extreme Value (GEV) probability distribution.
  • Spatial Interpolation: Inverse Distance Weighting (IDW) for population density and rainfall variables.
• Data sources:
  • Satellite Images: Landsat-5 Thematic Mapper (TM) (2000, 2005, 2010) and Landsat-8 Operational Land Imager (OLI) (2015, 2020) from the United States Geological Survey (USGS) Earth Explorer.
  • Rainfall Data: Daily rainfall data (1970–2019) from 9 stations, provided by the National Agency of Water Resources of Algeria (ANRH).
  • Soil Lithology: Hydrogeological map of northern Algeria, obtained from ANRH, used to derive Hydrological Soil Groups (HSG).
  • Digital Elevation Model (DEM): One-arcsecond Shuttle Radar Topography Mission (SRTM) DEM from USGS.
  • Socio-economic Data: Population density data for 1998 and 2008 from the National Office of Statistics (ONS), with future densities forecasted based on annual growth rates.
  • Geospatial Factors: Distances to main roads, transport stations, universities, hospitals, industrial and activity areas, and coastline (calculated using ArcGIS 10.4.1).

Main Results

• The LULC prediction model, using the Random Forest algorithm, achieved an overall accuracy of 85.05%.
• Built-up areas in the study watershed are projected to increase significantly from 11.73% in 2000 to 32.96% in 2040.
• Conversely, cultivated land and grassland are projected to decrease substantially from 46.50% in 2000 to 26.67% in 2040, and bare soil from 2.54% to 0.01% over the same period.
• Runoff patterns show significant variations across different LULC classes and time periods, with higher rainfall return periods (5, 10, and 20 years) consistently leading to expanded runoff areas.
• Built-up areas exhibit a strong positive correlation with runoff, particularly in higher runoff ranges (R² values increasing from 0.20 for 0–25 mm to 0.80 for 75–100 mm).
• Cultivated land and grass also show a high positive correlation with runoff (R² values of 0.91 for 50–75 mm and 0.73 for 75–100 mm).
• Woods and trees demonstrate a strong negative correlation with runoff, indicating their mitigating effect (R² values decreasing from 0.89 for 0–25 mm to 0.01 for 100–150 mm).
• During intense rainfall events (exceeding 100 mm), land factors such as interception and permeability show limited influence on runoff due to capacity and saturation constraints, leading to a consistent upward trend in runoff depths.

Contributions

• This study introduces a novel integration of multi-temporal LULC prediction with dynamic Curve Number (CN) updating to quantify long-term (2000–2040) runoff responses under projected urbanization scenarios in a semi-arid Mediterranean watershed.
• It provides a forward-looking framework for watershed management and planning by combining machine learning-based LULC forecasting with scenario-based hydrological simulation, unlike previous studies that primarily focused on historical LULC impacts.
• The research incorporates the concept of temporality in the factors influencing LULC changes, enhancing the accuracy of future LULC predictions.
• It is the first study to apply this integrated methodology and assess these specific results in the Wadi El Harrach sub-watershed, Algeria.

Funding

• Erasmus+ program (funded by the European Commission)
• Internship at the University of Trento (funded by the Université of Sciences and Technology Houari Boumediene)

Citation

@article{Riche2026Predicting,
  author = {Riche, Abdelkader and Drias, Ammar and Ricci, Riccardo and Souissi, Boularbah and Melgani, Farid},
  title = {Predicting LULC Changes and Assessing their Impact on Surface Runoff with Machine Learning and Remote Sensing Data},
  journal = {Water Resources Management},
  year = {2026},
  doi = {10.1007/s11269-026-04560-z},
  url = {https://doi.org/10.1007/s11269-026-04560-z}
}