The domain of this study includes global land areas and inland water bodies that are not permanently covered by snow/ice, covering 89% (132.05 × 10 ⁶ km ²) of the global land area. The multi-year mean of estimated global annual ET is 563.4 mm yr ⁻¹ (74.3 × 10 ³ km ³ yr ⁻¹) with an inter-annual variability of 10.4 mm yr ⁻¹ (1.3 × 10 ³ km ³ yr ⁻¹) from 1982 to 2013, which is similar to other reported global estimates 3 10 11, considering differences in the extent of study areas. The modeled monthly ET compares favorably with global in situ tower measurements at the 1-km flux tower footprint scale and with independent ET estimates inferred from long-term water balance measurements from 284 globally distributed basins covering 65% of the global vegetated area (see Supplementary Information Section 2 and Fig. S1). It also agrees reasonably well with three independent ET data sets (see Supplementary Information Section 2 and Fig. S2). As a whole, the global land area shows a significant upward ET trend of 0.88 mm yr ⁻² ( P <  0.001) during the 32-year period ( Fig. 1a). The upward trends occur widely across the globe with 42% of the global land area experiencing significant ( P <  0.1) positive trends in ET ( Fig. 1b) and exceeding areas (13% of global land area) with significant negative trends ( Fig. 1b). The rising global ET is consistent with the upward trend seen in a data-driven global ET record from 1982 to 2008 in a previous study 3 (the black line in Fig. 1a). The global ET record derived in the current study shows similar inter-annual variability (r = 0.75; P <  0.01) with the ET record in the previous study 3 for the overlapping 1982–2008 period. Both the current results and ET records of the previous study 3 show slowing or ceased growth rates from 1998 to 2008, which was primarily attributed to moisture limitation in the Southern Hemisphere 3. However, our extended record shows a recovered ET growth rate after 2008 ( Fig. 1a). This suggests that the lapse in the ET growth rate from 1998 to 2008 is likely an episodic phenomenon of the Earth’s climate system. This finding is also supported by another recent study, which suggests that the ET declines from 1998 to 2008 reflect transitions to El Niño conditions and are not the consequence of a persistent reorganization of the terrestrial water cycle 12.

The upward trend of global ET coincides with positive trends in annual land surface air temperature from ground observations (CRUTEM4.3 13: 0.03 °C yr ⁻¹; P <  0.001) and the reanalysis record (0.023 °C yr ⁻¹; P <  0.001), and satellite-observed ensemble-mean Normalized Difference Vegetation Index (NDVI) greenness (0.018% per year; P <  0.001) ( Fig. 1a). Global annual ET is highly correlated with NDVI (r  = 0.91; P <  0.001) and annual mean air temperature (r  = 0.69; P <  0.001). These parameters explain 84% of the variation in annual ET and have the similar positive trends as global annual ET, indicating that the global ET trend is mainly driven by increases in vegetation density and warming. A temporary global cooling and corresponding vegetation browning event occurred after the June 1991 Mount Pinatubo eruption 14, resulting in a temporary decrease in ET ( Fig. 1a). The El Niño-Southern Oscillation (ENSO) does not correspond strongly with variations of annual land-surface air temperature, NDVI, and ET at the global scale ( Fig. 1a). This is likely due to complex, lagged responses of the entire troposphere to the ENSO activities 15. In addition, the ENSO is associated with dry-wet cycles in many land areas 16, leading to more complex ENSO linkages with vegetation states and vegetation-impacted ET.

We constructed a global map of relative contributions of the major climatic control factors (demand, supply and energy) influencing ET (see Methods). Water supply most strongly influences ET over 49% of the global domain, whereas available energy and atmospheric water demand are dominant influences on ET over 32% and 19% of the global domain, respectively ( Fig. 2a). The atmospheric demand for ET shows the most extensive changes globally over the 32-year record ( Fig. 2b). Approximately 29% of the global domain shows significant ( P <  0.1) increases in ET demand from 1982 to 2013, mainly driven by a general global warming trend and associated increases in air vapor pressure deficit ( Supplementary Fig. S3), while 8% of the domain shows significant decreases in ET demand for the period, including western South America, and resulting from regional cooling and/or reduction of vapor pressure deficit ( Supplementary Fig. S3e). The available energy for ET shows the second most extensive significant changes over the global domain ( Fig. 2c). Approximately 8% of the domain shows significant increases in available energy for ET, mainly over tropical and subtropical regions ( Fig. 2c) due to reduction in total cloudiness 17; about 18% of the domain shows significant reductions in available energy for ET, mainly located over eastern Asia, central North America, Sudanian Savanna, and Australia ( Fig. 2c), which are mainly caused by reduced incoming solar radiation over the study period due to emission of anthropogenic aerosols 18. The available water supply for ET shows the smallest trends ( Fig. 2d). Only small portions of the global domain show significant changes in water supply (9% and 6% for increasing and decreasing supply, respectively). Regions with significant positive water supply trends are mainly located in the subtropics and high latitudes, including the African Sahel, while areas with significant negative water supply trends are predominantly located in the mid-latitudes. The observed changes in climatic control factors predominantly favor increasing ET, consistent with the finding of rising global land ET over the same period. However, increases in water supply are lagging the coincident rise in atmospheric moisture demand with warming, indicating that ET may become more supply limited on a global basis if these trends persist in the future.

The factorial analysis of variance to the ET estimates show that changes in vegetation greenness, air temperature and vapor pressure, and the two-way interactive effect of air temperature and vapor pressure are the dominant factors contributing to the multi-decadal changes in global ET ( Fig. 3a). The direct effects of varying air temperature and vapor pressure have opposite impacts on ET relative to their two-way interactive effect ( Fig. 3a), leading to a small net effect on ET. Changes in solar radiation impose a slight positive effect on ET, while wind speed and atmospheric CO ₂ exert slight negative effects on ET due to CO ₂ fertilization 19 and recent declining wind speed 20, respectively; however, these effects are predominantly outweighed by the impact of vegetation change ( Fig. 3a). As a result, the net effect of climate change and CO ₂ fertilization is positive but much smaller than the effect of vegetation change ( Fig. 3b), indicating that vegetation change is the primary driver of the global ET trend over the study period. Considering that vegetation change is also a response of the Earth’s terrestrial ecosystems to climate change, climate change imposes an indirect impact on ET by influencing global vegetation.

To determine the combined impacts of solar radiation, air temperature, wind speed and air vapor pressure on atmospheric demand for ET, we calculated PET using the Penman equation and used this as a surrogate measure of atmospheric moisture demand. The resulting global PET record shows a strong, increasing trend of 0.75 mm yr ⁻¹ ( P = 0.016) over the 32-year record ( Fig. 4) coinciding with global warming and consistent with the exponential increase in atmospheric capacity to hold moisture with increasing temperature as defined by the well-known Clausius-Clapeyron relationship. Global P for the domain shows large temporal variability with a positive trend of 0.66 mm yr ⁻² ( P = 0.08) ( Fig. 4). The resulting CWD shows a weak, negative trend of −0.12 mm yr ⁻² ( P = 0.83) ( Fig. 4) and indicates a slight tendency to growing deficit between atmospheric moisture demand and available water supply for evaporation. This general increase in climate aridity is supported by a significant positive trend in global annual vapor pressure deficit (1.12 Pa yr ⁻¹; P = 0.02) over the same period ( Supplementary Fig. S4). The TWA also shows a similar slight downward trend (−0.07% per year; P = 0.59) because a positive ET trend is offsetting increases in P over the 32-year record. This implies that any increases in P are being lost to ET rather than being allocated to other components of the global water cycle. The ENSO driven climate oscillations impart clear impacts to global P, indicated by significant negative correlation between the annual MEI (Multivariate ENSO Index) and annual P ( r = −0.71; P <  0.001). During strong, positive (i.e. El Niño) ENSO phases, the global land area as a whole receives less-than-normal P as shown in Fig. 4, despite that precipitation responses to El Niño events may differ in different regions. On the other hand, the global land area has lumped more-than-normal P during strong, negative (i.e. La Niña) ENSO phases ( Fig. 4). Due to the close relationship between global P and ENSO activities, the global CWD and TWA are also correlated with ENSO (r  = −0.43, P <  0.01; r  = −0.62, P <  0.001) ( Fig 4).

Natural ENSO-driven climate oscillations are associated with periodic wetting and drying cycles for terrestrial precipitation and are superimposed on a longer-term, positive P trend. If positive PET and ET trends continue to outpace growth in P ( Fig. 4), ET and associated ecosystem processes will become more supply-limited, especially in these areas closely responding to "natural" El Niño dry cycles. Global climate projections also indicate future changes in ENSO characteristics and increasing occurrence of El Niño events 21. These changes are likely to exacerbate global water deficits and the frequency, extent and severity of drought. Severe droughts serve as environmental triggers for other vegetation disturbances, including productivity decline 22, mortality 23 24, insect epidemics 25 26 and fire 27 28. These events may stimulate environmental tipping points toward large-scale ecosystem adjustments and biome conversions 1 29 with continued climate warming.