Dynamic crop-atmosphere water feedbacks challenge isotopic equilibrium knowledge in the lower atmospheric boundary layer

Understanding water fluxes in agricultural ecosystems is fundamental to securing food production under increasing water scarcity. This study investigates the dynamics of evapotranspiration (ET) and its stable water isotopic composition (d ¹⁸ O, d ² H) in a heterogeneous cropland cultivated with winter oil rapeseed and winter wheat. By coupling a fully automated chamber system with high-resolution in-situ laser spectroscopy, we implemented a novel observational approach that achieved the continuous monitoring of ambient water vapor, precipitation, and ET isotopic compositions at an unprecedented temporal resolution over two growing seasons. We found the isotopic composition of ET to serve as a sensitive indicator of crop phenology and physiological response to changing environmental conditions and observed a distinct seasonal shift from evaporation-dominated to transpiration-dominated fluxes. High transpiration rates strongly affected the ambient water vapor diurnally and seasonally, challenging common ecohydrological assumptions. Furthermore, we demonstrate that a data-driven Random Forest model can successfully predict the complex, non-linear dynamics of isotopic compositions using standard micrometeorological variables and easily measurable ambient water vapor compositions. We conclude that high-resolution monitoring linked with machine learning offers a robust pathway to disentangle spatio-temporal dynamics of soil-plant-atmosphere interactions and improves our ability to predict crop water dynamics under changing environmental conditions.