The responses of ecosystem water use efficiency to CO2, nitrogen deposition, and climatic drivers across China

The responses of ecosystem water use efficiency (WUE) to environmental perturbations are non-linearly dependent on how strongly coupled are water and carbon cycling. Here, we evaluate the structure of these non-linearities using two high-fidelity explainable machine learning (XML) models to disentangle the confounding effects of the air temperature (Ta), precipitation (PR), downward shortwave radiation (SR), atmospheric CO2 concentration (Ca), and nitrogen deposition (Ndep) on WUE. The findings suggest that mean annual PR plays a dominant role in shaping the spatial pattern of WUE. Spatially, WUE responded positively to PR and Ca but negatively to Ta and SR, while Ndep had little impact on WUE. For temporal trends, the spatial pattern of the mean annual WUE determined the spatial pattern of the WUE trend. Furthermore, the spatial patterns of the Ndep and Ca trends also contributed substantially to the spatial pattern of the WUE trend. Notably, the WUE and Ca trends exhibited negative correlations, while the WUE and Ndep trends showed positive or negative correlations depending on the different nutrient constraints on vegetation. The WUE growth rate responded negatively to SR in both spatial patterns and temporal trends. Additionally, using the trend of 0.077 K yr(-1) and 0 mm yr(-1) as the threshold, the trends of Ta and PR shifted from positive to negative relationships on WUE growth. Our results help identify key sensitivities and thresholds in WUE to environmental controls over space and time.