Deciphering Olive Yield Determinants under Contrasting Water Regimes: A Multi-Site Machine Learning Approach in Morocco Agro-Ecosystems

Improving olive yield in Moroccan agro-ecosystems requires a better understanding of the interactions between water availability, soil properties, and management practices. The complexity and non-linear nature of these interactions limit the effectiveness of conventional analytical approaches. This study applies machine learning methods to predict olive yield and to assess how the importance of yield determinants varies under contrasting water regimes. A multi-site dataset from Moroccan olive groves, including more than 2,000 observations, was analyzed. Machine learning models showed high predictive accuracy across water regimes. Under rainfed conditions, CatBoost achieved the best performance (R² = 0.845), indicating that yield variability is mainly driven by soil properties and spatial context. Under irrigated conditions, XGBoost provided the highest accuracy (R² = 0.855), highlighting the increasing role of management practices such as planting density and nitrogen fertilization. Under intensive irrigation, fruit-related variables, particularly 100-fruit weight, became the dominant predictors, while the influence of edaphic constraints decreased.Overall, the results demonstrate that irrigation does not simply increase olive yield but fundamentally alters the hierarchy of factors controlling production. These findings emphasize the need for data-driven, site-specific management strategies to enhance the sustainability and efficiency of olive production in Morocco.