Linking almond trees’ transpiration to irrigation’s mineral composition by physiological indices and machine learning

Irrigation's mineral composition affects plants’ water utilization. However, limited data concerning minerals’ interactive effects on transpiration hinder the integration of the mineral matrix into irrigation planning. Hence, we set to (1) characterize the effects of mineral availability on almond trees’ transpiration and (2) integrate climatic variables and physiological measurements to predict irrigation requirements under various mineral compositions. We constructed 0.9 m3 lysimeters to measure trees’ transpiration, hydraulics, growth, and reproduction in a wide range of nitrogen, phosphorus, or potassium concentrations. At peak summer, almond trees’ with optimal fertilization transpired 6.5 mm day−1, meeting the potential environmental demands. Potassium had a negligible effect on transpiration, while low phosphorus (≤ 1 mg L−1), low nitrogen (≤ 10 mg L−1), and high nitrogen (≥ 100 mg L−1) concentrations in the irrigation reduced maximal transpiration to 3 mm day−1. Transpiration changes corresponded to the trees’ canopy and stem growth rates rather than their stomatal conductance or water potential. We tested multivariable interactions to select parameters for transpiration predictions. A general boosting model (GBM) exhibited better training and validation performances than three alternative computations (linear, nonlinear, or support vector machine), integrating physiological parameters through machine learning. The transpiration predictions were reinforced by their correlation to yield, implying that almond trees’ basic annual water requirements are 360 mm, yet they require 900 mm to reach full potential. Our work highlights that irrigation planning could account for fertilization’s effects on trees’ physiology through computational tools, meteorological data, and physiological parameters that are readily accessible.