Use of explainable artificial intelligence with high-resolution satellite imagery to assess water status and vine shoot growth in grapevine (Vitis vinifera L.)

Abstract Grapevine is an essential socio-economic crop in the Mediterranean regions. However, its cultivation is threatened by climate change, particularly due to the increase in drought events. Traditional indicators of plant water status, i.e., predawn leaf water potential (Ψ pd ), are helpful but labor-intensive. Furthermore, another index based on visual assessment of the shoot apex (i.e., iG-Apex), has been recently used to evaluate the water stress of grapevine plants. The application of remote sensing technologies and machine learning models gives the possibility to monitor grapevine spatio-temporal variability of plant physiological indicators in a non-destructive way. This study aims to develop an explainable artificial intelligence (XAI) framework to predict Ψ pd and iG-Apex in two rainfed vineyards using high-resolution multispectral imagery from the Planet SuperDove constellation. Five machine learning models—extreme gradient boosting, random forest, support vector regressor, ElasticNet, and Linear Regression—were trained and compared. Extreme gradient boosting achieved the best performance for Ψ pd prediction (R² = 0.778), while random forest outperformed for iG-Apex prediction (R² = 0.615). SHAP and ICE analyses were carried out for models’ explainability. The analyses revealed that visible and NIR bands were the most important predictors, influencing both Ψ pd and iG-Apex prediction. Our results highlight the potential of combining XAI and remote sensing to support grapevine management under Mediterranean conditions, providing a useful tool for early water stress detection.