Along with the development of remote sensing technology, the spatial–temporal variability of vegetation productivity has been well observed. However, the drivers controlling the variation in vegetation under various climate gradients remain poorly understood. Identifying and quantifying the independent effects of driving factors on a natural process is challenging. In this study, we adopted a potent machine learning (ML) model and an ML interpretation technique with high fidelity to disentangle the effects of climatic variables on the long-term averaged net primary productivity (NPP) across the Amazon rainforests. Specifically, the eXtreme Gradient Boosting (XGBoost) model was employed to model the Moderate-resolution Imaging Spectroradiometer (MODIS) NPP data, and the Shapley addictive explanation (SHAP) method was introduced to account for nonlinear relationships between variables identified by the model. Results showed that the dominant driver of NPP across the Amazon forests varied in different regions, with temperature dominating the most considerable portion of the ecoregion with a high importance score. In addition, light augmentation, increased CO₂ concentration, and decreased precipitation positively contributed to Amazonia NPP. The wind speed for most vegetated areas was under the optimum, which benefits NPP, while sustained high wind speed would bring substantial NPP loss. We also found a non-monotonic response of Amazonia NPP to VPD and attributed this relationship to the moisture load in Amazon forests. Our application of the explainable machine learning framework to identify the underlying physical mechanism behind NPP could be a reference for identifying relationships between components in natural processes.

Inter-basin water transfer (IBT) is widely used to mitigate water shortage at the cost of compromising water availability in water-exporting regions. Yet, we do not know how efficient are the IBTs in alleviating inter-regional water stress in a changing climate and water supply-demand context. From a socio-hydrological perspective, we here quantify the efficiency of more than 200 IBTs across the United States by a Stress Relief Index that measures the impact of water redistribution on the overall water stress level. Based on the assumption that an IBT-induced increase and reduction in water availability would respectively constitute a positive and negative impact on regional water security, we show that 29% of the IBTs could be considered socially inefficient by 2010 as they shift water stress from water-receiving to water-exporting and downstream regions. Future stress escalations induced by growing population, declining runoff, and increasing demands for energy production and irrigation will alter IBT efficiency disproportionately. The inefficient IBTs would amount to 32% and 35% by the end of the 21st century under the scenarios of representative concentration pathway (RCP) 4.5 and RCP8.5, with 7 ∼ 16 IBTs reaching a tipping point that their role in the water system could switch from alleviating to aggravating the overall water stress. Our results indicate that the evolving climatic and socioeconomic status can largely affect transfer efficiency, highlighting the need of basin-level adaptation strategies for sustainable use of the IBTs.

This study examines the relative roles of climatic variables in altering annual runoff in the conterminous United States (CONUS) in the 21st century, using a monthly ecohydrological model (the Water Supply Stress Index model, WaSSI) driven with historical records and future scenarios constructed from 20 Coupled Model Intercomparison Project Phase 5 (CMIP5) climate models. The results suggest that precipitation has been the primary control of runoff variation during the latest decades, but the role of temperature will outweigh that of precipitation in most regions if future climate change follows the projections of climate models instead of the historical tendencies. Besides these two key factors, increasing air humidity is projected to partially offset the additional evaporative demand caused by warming and consequently enhance runoff. Overall, the projections from 20 climate models suggest a high degree of consistency on the increasing trends in temperature, precipitation, and humidity, which will be the major climatic driving factors accounting for 43–50, 20–24, and 16–23 % of the runoff change, respectively. Spatially, while temperature rise is recognized as the largest contributor that suppresses runoff in most areas, precipitation is expected to be the dominant factor driving runoff to increase across the Pacific coast and the southwest. The combined effects of increasing humidity and precipitation may also surpass the detrimental effects of warming and result in a hydrologically wetter future in the east. However, severe runoff depletion is more likely to occur in the central CONUS as temperature effect prevails.