Ecosystem water use efficiency (eWUE = gross primary production (GPP)/evapotranspiration (ET)) is widely used to characterize the coupling of ecosystem water and carbon processes. To investigate how eWUE responds to environmental changes, we compared environmental controls on annual, monthly, and daily GPP, ET, and eWUE from diverse ecosystems and climate regimes and quantified their daily relative impacts with machine learning techniques. Similar to GPP and ET, eWUE was strongly related to environmental variables at daily and monthly scales than at annual scales, indicating the tighter interplays of ecosystem processes with surroundings at shorter timescales. More critically, daily GPP and ET variations were primarily driven by net radiation (Rn) at most sites; whereas, vapor pressure deficit (VPD) dominated daily eWUE variations from humid to semi-arid sites, leaf area index (LAI) controlled eWUE variations at arid sites. It was largely attributed to the asynchronous responses of daily GPP and ET to environmental variables: the positive responses (though with different degrees) of daily GPP and ET to Rnweakened the Rnimpact on eWUE; whereas, the opposite responses of daily GPP (negative) and ET (positive) to VPD enhanced the VPD impact on eWUE. The greater LAI impact on daily eWUE at arid sites was due to the dominant control of LAI on GPP variations under arid conditions. Unlike early eWUE models that incorporate VPD, our data showed that Rncould significantly improve eWUE models. This work provides valuable insights into understanding the controlling mechanisms of eWUE and ameliorating the representation of GPP and ET coupling.