| Titre : | Hybrid framework of physics-inspired optimization and explainable ensemble learning for irrigation classification mapping in Morocco (2026) |
| Auteurs : | C. Singha ; S. Sahoo ; A. Govind |
| Type de document : | Article |
| Dans : | Environmental Research Communications (vol. 8, n. 1, January 2026) |
| Article en page(s) : | p. 015022 |
| Langues : | Anglais |
| Langues du résumé : | Anglais |
| Catégories : |
Catégories principales 07 - ENVIRONNEMENT ; 7.3 - Eau. Gestion de l'EauThésaurus IAMM CULTURE IRRIGUEE ; IRRIGATION ; GESTION DES EAUX ; CARTOGRAPHIE ; TELEDETECTION ; MODELE ; MAROC |
| Résumé : | Accurate mapping of irrigated areas is essential for effective water resource management in regions increasingly affected by climate change, water scarcity, and intensified agricultural demands. In this study, we present a novel integrated ensemble classification framework to generate high-resolution irrigation maps for the Moroccan region in Africa. The methodology leverages remote sensing (RS) data, field surveys, and advanced machine learning (ML) techniques. We employed a hybrid ensemble approach combining physics-inspired ML algorithms PSA (Photon Search Algorithm), QANO (Quantum-Behaved Avian Navigation Optimizer), KOA (Kepler Optimization Algorithm) NRO (Nuclear Reaction Optimization), with Extreme Gradient Boosting (XGB). These models were trained using a diverse set of input variables, including topographic, hydro-meteorological, and RS indicators. A stratified 10-fold cross-validation strategy was implemented to ensure robust evaluation. Ground truth datasets for 2024 included 754 training and 324 testing samples for drip irrigation, 288 and 120 for flood irrigation, and 953 and 411 for sprinkler irrigation. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied. Feature selection using the graph centrality Laplacian score identified vapor pressure deficit (VPD) and land surface temperature (LST) as the top two influential variables, significantly enhancing classification performance. Drip irrigation was the most accurately classified type across all models, with XGB-QANO (F1-score of 0.9639, and AUC of 0.99) showing the best overall performance. Flooding irrigation was the most challenging to classify, with XGB-NRO achieving the highest F1-score (0.7029) and AUC (0.934). Sprinkler irrigation classification was highly consistent, with XGB-NRO achieving top results (F1-score: 0.9071 and AUC: 0.967). Therefore, SHAP analysis identified VPD and AET as two key features influencing irrigation classification. Spatially, irrigation has notably expanded upstream in the Upper Rabat-River Basin, especially near tributaries, where new croplands remained irrigated during droughts, unlike downstream areas. This trend underscores the long-term sustainability concerns for Moroccan agriculture. |
| Cote : | En ligne |
| URL / DOI : | https://doi.org/10.1088/2515-7620/ae2449 |
Accueil
Sélection de la langue
Adresse
ÉQUIPE
Nous contacter
Informations pratiques
contact
