Rabie et al. (2025) Remote Sensing, GIS, and Machine Learning in Water Resources Management for Arid Agricultural Regions: A Review

Identification

• Journal: Water
• Year: 2025
• Date: 2025-10-31
• Authors: A. Rabie, Mohamed Elhag, Ali M. Subyani
• DOI: 10.3390/w17213125

Research Groups

• National University of Science and Technology, Moscow Institute of Steel and Alloys (MISIS), College of Computer Sciences, Department of Automated Control Systems, Russia
• Russian Academy of Sciences, Institute of Geography, Cartography and Remote Sensing Department, Russia

Short Summary

This study optimized geospatial data pipeline automation for landscape monitoring in Italy using GeoAI and machine learning on Landsat imagery, demonstrating that the Support Vector Machine (SVM) algorithm achieved the highest classification accuracy for detecting land cover changes over a five-year period.

Objective

• To assess and improve the efficiency of geospatial data analytics in GRASS GIS using AI applications.
• To quantify land cover changes in Italy using AI-processed satellite imagery and compare the performance of various machine learning algorithms (Support Vector Machine, Decision Tree Classifier, Random Forest, Multilayer Perceptron Classifier).
• To promote advancements in remote sensing data processing through AI application and analyze its role in geospatial data management.

Study Configuration

• Spatial Scale: Central Apennines, Italy, with a spatial resolution of 30 meters per pixel.
• Temporal Scale: Short-term time series analysis over a 5-year period (2018-2023), covering both spring and autumn seasons.

Methodology and Data

• Models used:
  • Machine Learning (ML) algorithms from Python's Scikit-Learn library: Support Vector Machine (SVM), Decision Tree Classifier (DTC), RandomForest (RF), and Multilayer Perceptron Classifier (MLPC) of Artificial Neural Network (ANN).
  • GRASS GIS (Geographic Resources Analysis Support System) for image processing and integration of ML algorithms via r.learn.train, r.learn.predict, and r.learn.ml modules.
  • K-means clustering for signature file generation and i.maxlik for initial classification.
  • Statistical analysis in R for accuracy assessment (Cohen’s Kappa and F-1 scores).
• Data sources:
  • Eight multispectral Landsat-8-9 OLI/TIRS satellite images (four for spring and four for autumn across 2018, 2019, 2022, 2023).
  • Data obtained from the United States Geological Survey (USGS) EarthExplorer repository.
  • Seven multispectral Landsat bands (Aerosol, Blue, Green, Red, Near-Infrared, Shortwave Infrared-1, Shortwave Infrared-2).
  • CORINE classification data used for training the ML models.

Main Results

• Ten land cover classes were identified based on the CORINE classification: urban areas, arable lands and agricultural fields, water areas, needle-leaved trees, bare lands and soils, pastures, permanent herbaceous vegetation, croplands, broadleaved trees, and shrubland.
• A comparative analysis of four ML algorithms (SVM, RF, DTC, MLPC) for image classification demonstrated that the Support Vector Machine (SVM) algorithm consistently outperformed the others in terms of accuracy.
• Statistical evaluation using Cohen’s Kappa agreement coefficient showed SVM achieving the highest values (ranging from 0.87 to 0.91 across the study period), followed by RF (0.82-0.85), DTC (0.74-0.79), and MLPC ANN (0.74-0.81). The F-1 score also confirmed SVM's superior performance.
• Observed land cover dynamics in the Central Apennines (2018-2023) indicated an increase in 'pastures' (15%) and 'herbaceous vegetation' (17%), a decrease in coniferous forests (5%), and a significant increase in shrubland (29%), suggesting a gradual replacement of other vegetation types.
• SVM was highlighted for its effectiveness with high-dimensionality data, suitability for complex pattern recognition due to kernel functions, and lower susceptibility to overfitting compared to other ML and ANN approaches.

Contributions

• Developed and demonstrated an improved, automated workflow for satellite image analysis by integrating multiple AI capabilities (specifically Scikit-Learn ML algorithms) within the GRASS GIS environment.
• Provided an experimental benchmark of the effectiveness of AI-supported analytical tools in remote sensing, highlighting the advantages of ML for rapid, accurate, and multidimensional geospatial data processing.
• Quantified land cover changes in the Central Apennines, Italy, over a 5-year period (2018-2023), offering valuable insights for environmental monitoring and land management.
• Identified Support Vector Machine (SVM) as the most effective algorithm among the tested ML methods (SVM, RF, DTC, MLPC) for land cover classification in heterogeneous landscapes, supported by robust statistical evaluation (Cohen's Kappa and F-1 scores).

Funding

Funding information for this research is not explicitly provided in the paper.

Citation

@article{Rabie2025Remote,
  author = {Rabie, A. and Elhag, Mohamed and Subyani, Ali M.},
  title = {Remote Sensing, GIS, and Machine Learning in Water Resources Management for Arid Agricultural Regions: A Review},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17213125},
  url = {https://doi.org/10.3390/w17213125}
}

Generated by BiblioAssistant using gemini-2.5-flash (Google API)