Data-driven algorithms to estimate Maize Sap Flow Transpiration based on climatic and soil moisture data

Purpose Accurate estimation of crop transpiration is essential for optimizing irrigation management and improving water-use efficiency in precision agriculture. However, direct measurement of transpiration is often invasive, costly, and difficult to maintain at large scales. This study proposes a data-driven framework to estimate maize ( Zea mays L.) sap flow driven by transpiration using widely available climatic and soil moisture data combined with machine learning techniques. Methods Field experiments were conducted during the 2023 and 2024 growing seasons in central Italy under irrigated silage maize. Meteorological variables, soil water content, and crop growth indicators were used as inputs, while sap flow measurements served as reference outputs. Several machine learning models were evaluated, including Linear Regression, Support Vector Regression (SVR), Decision Tree Regressor, and Multi-Layer Perceptron Regressor (MLPR), using both Point Estimation and Temporal Estimation strategies. Temporal approaches incorporated short-term historical information through feature concatenation and previous-average windows. Results Results demonstrate that non-linear models, particularly MLPR and SVR, consistently outperform linear and tree-based approaches. The inclusion of short temporal windows (45 minutes to 2 hours) significantly improves predictive accuracy, enhancing reconstruction of the diurnal transpiration pattern. Feature concatenation proved more effective than averaging strategies in capturing soil–plant–atmosphere interactions. Model performance remained robust across two contrasting growing seasons, confirming good generalization capability under interannual variability and data discontinuities. Conclusion The proposed framework provides a reliable and minimally invasive solution for real-time estimation of maize transpiration, supporting precision irrigation management. These findings highlight the potential of machine learning models as practical decision-support tools for sustainable agricultural water management.