Despite efforts to improve estimations of terrestrial carbon and water fluxes, considerable uncertainties remain in their variations, which can be largely attributed to a poor understanding of water constraints in models. Here, we investigate the effects of three water indicators (atmospheric vapor pressure deficit, VPD; soil water content, SWC; and plant water of land surface water index, LSWI) on interannual and annual variations of carbon and water fluxes (gross primary productivity, GPP; evapotranspiration, ET; and water use efficiency, WUE) in dryland forest, shrubland, grassland, and cropland. We find that LSWI has higher significant Pearson correlations with GPP (r = 0.47), ET (0.43), and WUE (0.33) than SWC (GPP: 0.33; ET: 0.42; WUE: 0.11) and VPD (GPP:-0.34; ET:-0.31; WUE:-0.17), which are almost unaffected by SWC or VPD according to partial correlations. This may be due to that plant water directly controls stomatal conductance, whereas soil water only provides the maximum moisture boundary and atmospheric water further limits the carbon dioxide available for photosynthesis. The correlations of LSWI with GPP, ET, and WUE strengthen 1) from wet to dry conditions, 2) from woody to herbaceous ecosystems, and 3) from daily to 8-day and monthly scales. The ecosystem energy closure and differences in canopy structures and root depths of plants may collectively affect these relationships at various temporal scales and ecosystems. Our findings highlight the higher potential of plant water than atmospheric and soil water in current models for capturing variations of carbon and water fluxes in drylands.