Irrigation detection in Mediterranean olive orchards using Sentinel-2 vegetation indices and machine learning

The Mediterranean basin is affected by water shortages, particularly during the summer dry period. In olive orchards, this situation has increased the use of irrigation. Given the regional water scarcity, irrigation detection in perennial crops is needed to support policy development and sustainable water management. This study evaluates the ability of vegetation indices derived from Sentinel-2 data to distinguish irrigated and rainfed olive orchards in five Mediterranean countries (Greece, Italy, Morocco, Portugal, and Spain). A machine learning approach was applied to assess the discriminatory capacity of vegetation indices computed for winter (January, baseline) and summer water-stress conditions (August). Field-level median values were extracted and evaluated using eleven supervised classification algorithms. Ensemble-based models demonstrated higher classification performances than linear and probabilistic approaches. The highest global F1-score was obtained with Extremely Randomized Trees (ET) with F1-score = 0.843, followed by extra trees, and gradient boosting (GB), (F1-score ≈ 0.83). Within the single-index training approach, the moisture stress index (MSI) proved to be the most discriminative index (F1 = 0.771 with AdaBoost). The Normalized Difference Moisture Index also showed a strong performance with SVM (F1 = 0.772), highlighting the importance of shortwave-infrared moisture-sensitive indices for irrigation detection. These results indicate that irrigation discrimination in Mediterranean olive orchards is mainly related to canopy water status rather than greenness alone. The proposed approach combines physiological interpretability with computational scalability and is transferable to large-scale detection in heterogeneous agro-climatic conditions.