Introduction

Plants in terrestrial ecosystems on Earth assimilate atmospheric CO2 through photosynthesis, which is inherently accompanied with the loss of water through stomata that regulate the mass-energy exchange between the leaf and the atmosphere1,2. The rate of carbon uptake per unit of water lost, also called water-use efficiency (WUE), is an important parameter for understanding the metabolism of terrestrial ecosystems. Carbon and water fluxes of leaves are related to those of larger scale ecosystems, but fluxes at ecosystem scales are weakly constrained3. The question of how much water a plant uses relative to carbon gained has been examined in fields ranging from plant physiology to applied scientific disciplines such as irrigation science and agronomy4. Given ongoing climatic change and ecosystem degradation, a deeper understanding of whole ecosystem WUE will improve our ability to simulate and predict carbon and water cycles and to refine water management5,6.

Owing to measurement difficulties, few studies have systematically compared global patterns of WUE of terrestrial ecosystems across different vegetation types or have analyzed the seasonal variability of WUE in relation to meteorological conditions. Ecosystem WUE is slightly different from plant WUE. Plant physiologists usually consider WUE at leaf or stand scales and are mainly interested in relations between total or above-ground biomass, stem biomass or net CO2 uptake to transpiration or evapotranspiration (ET)7,8. Here, we use a whole ecosystem estimate of water use, evapotranspiration (ET), defined as the total water vapour flux between the canopy and the atmosphere consisting of evaporation from soil, plant transpiration and evaporation of the intercepted fraction. Major ecozones are often characterized with differing water-use efficiencies owing to inherent physiological variation in leaf gas exchange and environmental conditions. Our definition is similar to what ecologists commonly use for whole ecosystem WUE, which is the ratio of net primary production, net ecosystem production, or gross ecosystem production to water use or evapotranspiration4,9,10. While the exchange of both CO2 and water vapor is regulated by stomatal aperture for leaf-level WUE, ecosystem-level WUE is also affected by evaporation and vegetation morphology. This discrepancy complicates comparisons of WUE from different sources. Here we use the ecosystem-level definition, which is relevant for evaluating ecosystem models.

Further, variability in WUE can be evaluated at different time scales, ranging from diurnal, seasonal, to interannual11. The time scale of investigation needs to be determined primarily in order to quantify the different patterns of WUE and the underlying mechanisms in relation to vegetation types and meteorological conditions. Here, we analyzed the dynamics of WUE at both annual and seasonal time scales.

Also, WUE is dependent on the spatial unit of analysis. Water and carbon cycles usually occur heterogeneously over the land surface, which requires an appropriate upscaling methodology at regional and global scales. Although several studies have explored the interaction between water and carbon cycles12,13, few global-scale analyses have been performed till now. Better quantification of global patterns of terrestrial WUE is needed to further understanding of natural and human impacts.

The seasonal dynamics of WUE differ strongly depending on location, climatic factors, plant functional type, species composition and disturbance history, requiring consistent, temporally continuous and spatially distributed observations for accurate assessment of WUE. In addition to leaf-level measurements and inventory surveys14,15, in recent years, with the development of the long-term eddy covariance technique, tower-based monitoring of ecosystem carbon and water cycles has made global evaluation of productivity, respiration and evapotranspiration possible16,17,18. Data from hundreds of sites are cooperatively shared through the global network–FLUXNET19,20,21. Currently the FLUXNET community throughout the world has been running for more than two decades enabled scientists to assess terrestrial WUE and the determining environmental conditions at different time scales across numerous sites of diverse vegetation types precisely3,4,22. Although uncertainties associated with site-to-site variation in site quality criteria, flux measurement methods, calculations and data quality control still exist, ongoing standardization and quality assurance efforts enable global integration.

Satellite-based remote sensing of vegetation can be used to derive global WUE. NASA TERRA and AQUA MODIS-based estimates of gross primary production (GPP) and terrestrial evapotranspiration (ET) can be obtained to quantify large-scale WUE23,24. Tower-based measured WUE can be used to evaluate the reliability and accuracy of estimates of global WUE from satellite-based approaches.

The primary objectives of this study are 1) to investigate the latitudinal trend in WUE of terrestrial ecosystems across different plant functional types with data from FLUXNET; 2) to evaluate global patterns of terrestrial WUE using satellite remote sensing-based GPP and ET products, which provides a new framework for global research on carbon and water cycles.

In this study, we used measurements of CO2 and H2O fluxes combined with other hydrometric measurements of 32 sites covering six major plant functional types that represent the major ecozones of the northern Hemisphere. These sites range from 20°N to 70°N in latitude and provide 137 site-years in total (Table 1 and Fig. 1).

Table 1 FLUXNET sites used in this study
Figure 1
figure 1

Locations of the 32 flux tower sites (see also Table 1) providing water-use efficiency data also used for validation of the remotely-sensed WUE product.

The global land cover classification data is produced by Hansen et al. (2000) and can be downloaded from http://www.landcover.org/data/landcover/index.shtml This figure was produced using ArcGIS 10.1.

Despite the wide range of plant functional types and associated stand structures, soil types, stand age, site disturbance history and year-to-year variability, there is a clear and distinct latitudinal trend in WUE of terrestrial ecosystems (Fig. 2 and Table 2). The trends indicated that as the latitude rose from the subtropics to the northern high-latitudes, multi-year average WUE of all sites increased and reached a peak at approximately 51°N and then tended to decline at higher latitudes. Although the peak magnitude of WUE differed among plant functional types, all peaks occur at approximately the same latitude, suggesting a key zonal differentiation rule driven by radiation and water availability. However, the tendency was also affected by the non-random coverage of terrestrial ecosystems, highlighting biological adaptations of WUE to specific climatic conditions. Indeed latitude is not a phenomenological driving variable per se, but is a proxy for the complicated effects of a multiplicity of abiotic and biotic factors, which determine the major ecozones.

Table 2 Multiyear mean annual water use efficiency (WUE) and the standard deviation (SD) at the 32 sites for different plant functional types (PFT)
Figure 2
figure 2

Latitudinal trends in multiyear mean annual water use efficiency at the 32 sites for different plant functional types.

The eddy covariance sites used were listed briefly in Table 1. The interannual variability is represented by the standard error bar. This figure was produced using Origin 8.0.

While having similar latitudinal pattern, the magnitude of ecosystem WUE differed among plant functional groups (Fig. 3). Evergreen forests generally had higher WUE than deciduous vegetation types at similar latitudes. WUE of broad-leaved evergreen forest was higher than that of needleleaf evergreen forest. Both mixed forest and grassland had lower WUE than broad-leaved deciduous forest. However, the peak WUE value of grassland at 51°N was equivalent in magnitude to mixed forests. Crop sites had the lowest WUE among the terrestrial ecosystems, which indicated the high water utilization of food production systems relative to carbon gained and has implications for agronomic crop breeding in dry-land farming and management of irrigation water. The lower annual WUE of crops could also be affected by non-productive phases with higher soil evaporation compared to permanent vegetation.

Figure 3
figure 3

Comparison of multiyear mean annual water use efficiency among the main plant functional types.

Interannual variation of per plant functional type is represented by the standard error bar. This figure was produced using Origin 8.0.

To further investigate the global patterns of terrestrial water-use efficiency, we also used satellite remote sensing-based data from the recently revised MODIS GPP and newly developed ET products between 2000 and 2013. These were used to infer mean annual terrestrial WUE over a period that closely corresponds to tower-based measured WUE. Multiyear mean annual GPP, ET and WUE showed strong regional variations corresponding to climatic variations in water availability but still presented consistent latitudinal gradients similar to data from the tower network (Fig. 4). Spatially, terrestrial WUE increases from low latitudes to high latitudes in the North America and Eurasia continents and tends to decline after reaching the peak values in the boreal regions, though slightly further north than the towers showed.

Figure 4
figure 4

Global distribution of multiyear mean annual GPP, ET and WUE in the study period.

Bare ground and ocean areas were excluded from the model calculations. This figure was produced using ArcGIS 10.1.

Ecosystems dominated by evergreen PFT at low latitudes adjacent to Pacific West Coast and Atlantic West Coast have high WUE. Alpine plateaus in Northwest China, African deserts and the Cordillera that runs along the western edge of North and South America showed markedly low WUE. The trough of WUE for needleleaf evergreen forest was found in the southeastern region of the United States and China's central and western regions and the peak was found in Northern Eurasia and northwestern North America. For broad-leaved deciduous forest, the minimum WUE is located at China's central and western regions whereas the maximum WUE is located in Northwest Russia. Grassland is mainly distributed between 45° ~ 55°N and the peak of WUE is situated at Southeast Alaska, North Europe and Southeast Russia. Cropland maximum WUE appears in Central Europe, southern Canada and South Russian.

We also compared the global MODIS-based calculation of GPP, ET and WUE with other proxies25,26,27. Compared with the global estimates of GPP (123 ± 8 Gt C/year) by Beer et al. (2010)25 using eddy covariance flux data and various diagnostic models, the present study showed multiyear mean annual GPP (during 2000 ~ 2013) of 156.82 Gt C/year. Jung et al. (2011)26 estimated a mean annual global land-surface ET value from 1982 to 2008 of 65 ± 3 × 103 km3/year and Miralles et al. (2011)28 found that annual land evaporation was estimated as 67.9 × 103 km3 during the period 2003 to 2007, 80% corresponding to transpiration, 11% to interception loss, 7% to bare soil evaporation and 2% snow sublimation, whereas multiyear mean annual ET from MODIS during 2000 to 2013 was 90.8 ± 1.2 × 103 km3/year. It could be partly explained by the coarse resolution of previous studies (1/2 degree in Jung et al. 2011 and 1/4 degree by Miralles et al. 2011) which might lead to small-scale features being neglected. Although our global GPP and ET estimates are significantly larger than previous studies, WUE from MODIS and by combining the GPP and ET estimates above were quite similar, globally at 1.89 g C kg−1 H2O (2.83 mmol C mol−1 H2O) from the tower upscaling and 1.71 g C kg−1 H2O (2.56 mmol C mol−1 H2O) by MODIS remote sensing. Therefore, our results showed the consistent estimates between tower- and remote-sensing approaches of global WUE.

To evaluate the reliability and accuracy of the satellite-based global WUE product, tower-based measured WUE data at these 32 sites were used for validation. Although the number of flux sites was limited for each vegetation type and occasionally large differences occurred between remotely-sensed WUE and measured WUE for individual points, the satellite-based WUE product was relatively accurate on a global basis (Fig. 5). A well-calibrated model should have a root mean square error (RMSE) that is small relative to the total observed variation and an R2 close to 1. Remotely-sensed WUE of mixed forest compared closely with the tower-measured WUE (RMSE = 0.178 g C kg−1 H2O; R2 = 0.961), while the overall bias of grassland and evergreen needleleaf forest was worse. Accuracies for crop, evergreen broadleaf forest and deciduous broadleaf forest were intermediate between these two extremes.

Figure 5
figure 5

Comparisons between tower-based measured WUE and remotely-sensed modeled WUE product at the 32 sites for the plant functional types.

This figure was produced using Origin 8.0.

We also compared our estimates of ecosystem WUE with estimates of plant WUE. Jasechko et al. (2013)29 used literature estimates of WUE and a watershed isotope estimate of transpiration and developed a global map of plant water use efficiency (GPP/transpiration). This map shows that evapotranspiration exceeds transpiration by 2/3 in most terrestrial ecosystems and patterns of global WUE were similar to our estimates of ecosystem WUE based on satellite data. This comparison shows that our estimates of ecosystem WUE can serve as a proxy for plant WUE. Ecosystem WUE also provides a more general measurement of total water use by an ecosystem, as to some extent plant canopy cover influences land surface evaporation and adaptive processes may lead to plant canopy characteristics that minimize evaporation by limiting soil radiation exposure, for example.

There are some differences between the satellite estimate of the WUE-latitude trend and the tower observations (Fig. 6). Unlike the flux tower latitudinal gradient, however, remote sensing analysis revealed two peaks in WUE trends for four vegetation types. For needleleaf evergreen forest, grassland and crop, the first peak occurred at 20° ~ 25°N while this peak was a little farther north (25° ~ 30°N) for broad-leaved deciduous forest. Apart from that, the tendency of ecosystem-level WUE was similar to latitudinal zonality found using flux measurements. The northern peaks of WUE in the satellite observations were located at 60° ~ 65°N for needleleaf evergreen forest, 55° ~ 60°N for broad-leaved deciduous forest and grassland and 50° ~ 55°N for cropland, respectively. The distinct differences in spatial patterns of WUE among various ecosystems imply an additional biological constraint over latitudinal radiation availability conditions.

Figure 6
figure 6

Latitudinal evolution of global WUE for the main plant functional types in the Northern Hemisphere.

Every 5 degrees in latitude was divided into an independent evaluation unit. The spatial variability in each unit is represented by the standard error bar. This figure was produced using Origin 8.0.

The seasonal pattern of WUE by site reveals how mechanisms that lead to long-term annual WUE vary by plant functional type (Fig. 7). WUE was higher at the beginning and end of the year and lowest in summer in broad-leaved evergreen forest except for an Italian Mediterranean site (IT-Cpz) where the relatively low temperature during winter months hindered the growth of vegetation. Similarly, needleleaf evergreen forest showed significant peaks during early growing season (spring) and late growing season (late summer). In contrast, a consistent tendency was found in both mixed forest and broad-leaf deciduous forest of a singe WUE peak in summertime. The seasonal dynamics in WUE of grassland lacked consistent patterns owing to the broadly-distributed location of grasslands among various climate patterns and both C3 and C4 photosynthetic pathways. Crop WUE is likely also sensitive to the variety of farming systems.

Figure 7
figure 7

Seasonal dynamics of terrestrial water use efficiency of the FLUXNET sites used in Table 1 according to the plant functional types at 8-day time scale.

These EC sites were listed briefly in Table 1. (a) broad-leaved evergreen forest; (b) needleleaf evergreen forest; (c) mixed forest; (d) broad-leaved deciduous forest; (e) grassland and (f) cropland. This figure was produced using Origin 8.0.

The northern peak value of WUE appears to be driven by the effect of evaporation and transpiration, which both decrease with latitude (driven by net radiation) while vegetation productivity peaks for boreal forests in mid- and high latitudes (driven by solar radiation in summer). Boreal systems are in the ‘sweet spot’ of cool conditions that limit water loss and high incoming radiation in summer to maximize photosynthetic uptake. To identify the main factors controlling WUE for each plant functional type and the underlying mechanism, the relationships between terrestrial WUE with the corresponding global radiation (Rg), air temperature (Ta), vapour pressure deficit (VPD), soil temperature (Ts), precipitation (Pr) and soil water content (SWC) were compared (Table 3). Radiation and temperature were negatively correlated with WUE for broad-leaved evergreen forest, whereas soil water content was positively correlated. However, for other ecosystems, the meteorological conditions including radiation, temperature and vapor pressure deficit have a strong positive correlation with water-use efficiency, which partly explains differences among tropics (operating at near optimum temperature and radiation conditions) and temperate systems (temperature and light limited). For broad-leaved deciduous forest and grassland, SWC showed significantly negative correlation with WUE, whereas it was weakly negative for crop and needleleaf evergreen forest. Temperature and vapor pressure deficit are the most important meteorological factors that drive WUE.

Table 3 Correlation analysis between water use efficiency and the controlling environmental factors in terrestrial ecosystems according to plant functional types

Given increases in urbanization land accompanying by reduction in vegetation cover over the world, we would predict that continued land-cover and land-use change on Earth leads to a decline in global WUE. However, the temporal trend in WUE is affected by a suite of co-varying natural factors and human activities. Keenan et al. (2013)3 found a substantial increase in WUE in temperate and boreal forests of the Northern Hemisphere as atmospheric carbon dioxide concentrations rose over the past two decades. However, these analyses were based on site-level measurements. Our analysis of the 2000–2013 MODIS time-series of global annual mean WUE indicated a distinctly decreasing tendency that might be correlated with land-use changes (Fig. 8). But the declining trend of WUE has changed direction in the past 4 years, perhaps reflecting CO2 fertilization and climate warming effects or just interannual variability. Donohue et al. (2013)30 also found the significant impact of CO2 fertilization on maximum foliage cover across the globe's warm, arid environments and this effect was occurring alongside ongoing anthropogenic perturbations to the carbon cycles.

Figure 8
figure 8

Trend analysis of time-series global annual mean WUE for 2000–2013.

The dashed line refers to multiyear mean annual value of WUE during this period.

Silva & Anand (2013)31 investigated the influence of atmospheric CO2 concentration and climate warming on tree growth and intrinsic water use efficiency (iWUE) based on a synthesis of the literature of site-level measurements and found that responses in growth are latitude dependent with temporal changes in iWUE. Moreover, they identified positive relationships between iWUE and tree growth in boreal and Mediterranean forests located in latitudes greater than 40°N, with progressively lower responses toward lower latitudes. In this respect, these research results are consistent with our global time-series trend analysis. In addition, Silva & Horwath (2013)32 found that previously reported trend in iWUE cannot reflect a coherent global response to rising atmospheric CO2 and our present study indicated a distinctly decreasing tendency in global WUE during the past decade in spite of rising CO2 concentration owing to natural and human activities.

In addition, WUE trends in C3 and C4 plants differ. Plant water use efficiency has been shown to vary between photosynthetic pathways and C4 vegetation showed increasing water use efficiency in response to rising levels of atmospheric CO2 with consistent reductions in stomatal conductance and transpirational water loss compared to C3 species. Still et al. (2003)33 developed the C3/C4 distribution by combining remote sensing products, physiological modeling, a spatial distribution of global crop fractions and national harvest area data for major crop types. Therefore, by comparing with global patterns of WUE, distribution of C4 vegetation and vegetation type map in Fig. 1, we can find that 1) the C4 vegetation mainly occurs in the vast tropical and subtropical grassland and savanna regions. Temperate grassland regions in North and South America and Africa also contain high fractional C4 coverage whereas it is only found in a very small fraction throughout temperate Eurasia and in the upper Great Plains region of North America; 2) although the C4 vegetation have impact on predictions of GPP and WUE with changing climate, global trend analysis is affected by a suite of natural factors and human activities and the mix of vegetation, which is dominated by C3 responses. Currently it appears that recent trends of global WUE were more controlled by land use and land cover changes from human disturbances.

Flux observations in terrestrial ecosystems across different plant functional types reveals a consistent latitudinal zonality in WUE from the subtropics to the northern high-latitudes, peaking at approximately 51°N and then tending to decline at higher latitude. Though WUE from individual site shows large variation, the overall latitudinal trend observed in each vegetation type proved to be a persistent one. The latitudinal gradient in WUE was consistent among various estimates of WUE, whether by satellite or tower, though subtle difference in location of the peak WUE and regions of very high or low WUE differed. Satellite remote sensing provides a new tool for the global research on role of WUE and monitoring how WUE responds to changing environmental conditions. Decreasing WUE globally in the last decade may reflect land use change as well as other poorly understood factors. Our understanding of the large-scale processes determining terrestrial WUE is still insufficient. The water and carbon cycles usually occur heterogeneously over the land surface, which requires an appropriate upscaling methodology at the regional and even global scales. Although several model studies have explored the interaction between water and carbon cycles, few global-scale analyses have been performed. Our analysis of global patterns of terrestrial WUE will be helpful to analyze ecosystem responses to natural and human impacts.

Methods

Data acquisition

Our analysis is based on continuous observations of land surface exchanges from 32 flux towers situated in biomes of different plant functional types distributed around the Northern Hemisphere from 20°N to 70°N (Fig. 1 and Table 1), giving a total of 137 site-year dataset to reduce the uncertainty of year-to-year variation. Specifically, these flux sites encompassed broad-leaved evergreen forest (3 sites), needleleaf evergreen forest (9 sites), broad-leaved deciduous forest (6 sites), mixed forest (3 sites), grassland (5 sites) and crop (6 sites). The FLUXNET level 4 product provides measurements of canopy-scale water vapour flux, CO2 flux, meteorological variables (including radiation, air temperature, relative humidity, soil temperature and soil water content ) and estimates of gross primary photosynthesis derived from the measured NEE fluxes34,35,36. These data were quality checked and data gaps due to system failure or data rejection were filled using standardized methods to provide complete and standardized data sets20,37,38.

To estimate and analyze the global patterns of terrestrial water-use efficiency, we also used the remotely-sensed data from MODIS gross primary production (GPP) and developed ET at the global scale. The old Collection 4 GPP product was found to have considerable errors due to problems in the inputs. Zhao et al (2005)23 rectified these products by improving the data processing methods and modifying parameters in the algorithm used, which generated the improved Collection 5 (C5) MOD17 estimates. The 8-day composite 1-km fraction of photosynthetically active radiation (FPAR) and leaf area index (LAI) data from the MODIS sensor were used as remotely sensed vegetation property dynamic inputs to the algorithm. Data gaps in the 8-day temporal MODIS FPAR/LAI caused by cloudiness were filled with information from accompanying quality-assessment flags. For daily meteorological data required to drive the algorithm, 6-hourly National Center for Environmental Prediction/Department of Energy (NCEP/DOE) reanalysis II data were implemented. Monthly and annual GPP averages were derived by summing up each 8-day period. In addition, Mu et al. (2011)24 improved a satellite remote sensing-based ET algorithm to assess global terrestrial ET using MODIS and global meteorology data by 1) simplifying the calculation of vegetation cover fraction; 2) calculating ET as the sum of daytime and nighttime components; 3) adding soil heat flux calculation; 4) improving estimates of stomatal conductance, aerodynamic resistance and boundary layer resistance; 5) separating dry canopy surface from the wet; and 6) dividing soil surface into saturated wet surface and moist surface. These data can be obtained freely from http://www.ntsg.umt.edu/project/mod16. Global land cover classification product at 1 km resolution developed by the University of Maryland was used to assist in extracting the latitudinal evolution of WUE for each biome (http://glcfapp.glcf.umd.edu:8080/esdi/index.jsp)39.

Data treatment and statistical analysis

Water-use efficiency is calculated in various ways according to different scientific disciplines4,6. To systematically analyze the global patterns of terrestrial WUE on Earth and explore the key zonal differentiation rule along latitude across different plant functional types, we defined the WUE at the ecosystem scale as: WUE = GPP/ET. The seasonal trend of WUE was characterized by the weekly WUE (g C kg−1 H2O). The measured latent heat (LE, w/m2) fluxes were used to obtain water loss (ET, mm/day) by multiplying a factor of 0.035 which was converted by the formula ET = LE/λ (λ represents amount of energy to evaporate a unit weight of water; 2454000 J kg−1). The mean annual WUE value of each flux tower site was calculated as a proxy. The interannual variability is represented by the standard error. R2 and RMSE were used to test the goodness of the satellite product at tower sites.