Local flow regulation and irrigation raise global human water consumption and footprint

Science  04 Dec 2015:
Vol. 350, Issue 6265, pp. 1248-1251
DOI: 10.1126/science.aad1010

Local decisions with global consequences

Some estimates suggest that humanity has already exceeded our sustainable global water footprint: the balance between fresh water use and supply. It seems that the situation may be more unsustainable than we realize. Jaramillo and Destouni analyzed hydroclimatic data for 100 large basins dating back to 1901. Better accounting of local water use revealed larger than anticipated effects on the global water cycle. For example, local regulation of surface water flow and expanded regional irrigation activities have increased global evapotranspiration rates.

Science, this issue p. 1248


Flow regulation and irrigation alter local freshwater conditions, but their global effects are highly uncertain. We investigated these global effects from 1901 to 2008, using hydroclimatic observations in 100 large hydrological basins. Globally, we find consistent and dominant effects of increasing relative evapotranspiration from both activities, and decreasing temporal runoff variability from flow regulation. The evapotranspiration effect increases the long-term average human consumption of fresh water by 3563 ± 979 km3/year from 1901–1954 to 1955–2008. This increase raises a recent estimate of the current global water footprint of humanity by around 18%, to 10,688 ± 979 km3/year. The results highlight the global impact of local water-use activities and call for their relevant account in Earth system modeling.

Hydroclimatic changes on land determine the availability of freshwater resources required for human societies and ecosystems on Earth. However, the magnitude and key drivers of such changes historically (1, 2) and in the future (3) are highly uncertain, especially regarding the global role of human drivers and the magnitude of their related freshwater consumption. Both changes in the atmospheric climate and in the landscape may drive freshwater change (4, 5) (fig. S1). Among landscape changes, human-controlled flow regulation and irrigation (FRI) affect inter- and intra-annual freshwater conditions locally, but recent results indicate possible important effects on larger scales as well (6).

FRI developments over the past century have either moderately or strongly affected 59% of the world’s largest river systems (7). They include around 45,000 large dams and many other smaller ones, spread over 140 countries around the world (8) and constructed mostly over the past century to store water for irrigation, flood control, urban water supply, hydropower, or a combination of such purposes. These developments are linked with approximately 12 to 16% of the current global food production and 19% of the world’s electricity supply (8), even though they only cover 0.3% (9) and 2% (10) of the global land area, respectively. Regarding the environmental impacts of FRI, attention so far has focused on ecosystem effects of river fragmentation and diversion (11) and water storage (12). More recently, studies at local to regional scales have found an FRI-related enhancement of the ratio of actual evapotranspiration (AET) to precipitation (P); i.e., of AET/P (6, 13, 14). For flow regulation, a concurrent decrease is also found in the short-term (daily and monthly) variability of runoff (R) (6, 14, 15). A combination of these effects on AET and R can be then used to distinguish the impacts of FRI developments from those of other drivers of freshwater change (6). At a global scale, some studies have addressed at least one of these FRI-related effects in global-scale modeling (1621) but have not provided observation-based evidence of the global importance of FRI as a driver of freshwater change.

To fill this key observation gap, we analyzed global hydroclimatic data from 1901 to 2008 for 100 large hydrological basins (Fig. 1). For these basins, we computed hydroclimatic changes (supplementary materials) between the 54-year periods 1901–1954 and 1955–2008 and compared them with previously categorized impact levels (7, 11) and parameterized developments (9, 10) of FRI (table S1). From the results, we further quantified the magnitude of the FRI-driven hydroclimatic changes in each basin and assessed their implications for global human consumption of fresh water.

Fig. 1 Impact-level of (FRI) in 100 large hydrological basins.

The global distribution of the 100 hydrological basins investigated in this study is shown. Colors differentiate the basins according to independent categorization of impacts of FRI (7, 11): non-affected (NA) in blue (n = 17 basins; 2,022,050 km2 and 4% of total area); moderately affected (MA) in yellow (n = 30 basins; 20,695,803 km2 and 46% of total area); and strongly affected (SA) in red (n = 53 basins; 22,578,466 km2 and 50% of total area). The 100 basins cover a total area of 45,296,318 km2 or 35% of the global land area, excluding Antarctica.

Globally, the quantified hydroclimatic changes reveal consistent characteristic signals of increased AET/P and decreased relative intra-annual variability of monthly runoff (CVR) with higher FRI impact level (Fig. 2 and fig. S3). Further study of the distribution of AET/P changes among basins shows large variability (Fig. 3A and fig. S4), but still a significant ΑET/P increase with increasing basin measures of FRI development (Fig. 3, B and C). The latter measures are quantified from previous basin parameterization of total reservoir storage capacity (9) relative to basin area, specifically its change between 1901–1954 and 1955–2008 (ΔRES), and area equipped for irrigation (10) relative to basin area (IA).

Fig. 2 Consistent FRI-related patterns of change in relative evapotranspiration and temporal runoff variability.

Shown are the area-weighted mean (colored bars) and spatial standard deviation around the mean (whiskers) (supplementary materials) of changes in (left) the ratio of actual evapotranspiration to precipitation (ΔAET/P) and (right) the relative intra-annual variability of monthly runoff (ΔCVR), between the periods 1901–1954 and 1955–2008, in each FRI impact level.

Fig. 3 Dominant human effects of FRI on relative evapotranspiration.

(A) Kernel density estimate curve of the distribution of changes in the ratio of actual evapotranspiration to precipitation (ΔAET/P) between the periods 1901–1954 and 1955–2008 for 100 large basins in terms of (B) the change in total reservoir storage capacity relative to basin area between the two time periods (ΔRES; in mm); (C) the area equipped for irrigation relative to basin area (IA); (D) the ratio of potential evapotranspiration to precipitation (PET/P); and (E) the change in PET/P; i.e., ΔPET/P. The distribution of ΔAET/P is divided into five subgroups with 20 basins each: G1, G2, G3, G4, and G5. The groups include basins with changes between G1, the minimum value of ΔAET/P (Qmin = –0.09) and the 20% quantile (Q20 = –0.02); G2, Q20 and the 40% quantile (Q40 = –0.01); G3, Q40 and the 60% quantile (Q60 = 0.01); G4, Q60 and the 80% quantile (Q80 = 0.03); and G5, between Q80 and the maximum value of the distribution (Qmax = 0.10). The box plots thus show the distribution of ΔRES, IA, PET/P, and ΔPET/P values among the five subgroups. Box plot statistics include the arithmetic mean (blue triangles), median (thick vertical black lines), interquartile range (IQR) (boxes, colored blue or green depending on significance), whiskers (confidence interval of Embedded Image), and outliers (black circles). Values of outliers falling outside of the plot scale are also shown on the inner margin of each plot. The statistical significance (P < 0.05; by a one-sided unpaired Wilcox rank sum test) of the variable outcome distribution among the subgroups is shown by green box color when the variable values within a subgroup of the distribution are significantly greater than the variable values in at least one of the lower ΔAET/P subgroups (materials and methods of the supplementary materials). The P values of the statistical test are shown in table S4A.

We also tested the possibility of the AET/P changes being explained by geographic basin location or atmospheric climate change. Specifically, we checked the relationship of AET/P change with relative potential evapotranspiration (PET/P, expressing water-relevant climate conditions in each basin) and change in PET/P (expressing water-relevant climate change occurring in the basin) (22). We did not find these explanatory patterns between AET/P change and PET/P or PET/P change (Fig. 3, D and E). Regarding PET/P, the water-limited basins (PET/P > 1) should have less water available for AET/P increase than the basins with mostly energy-limited conditions (PET/P < 1) (5, 22, 23). Rather, the relatively large AET/P increases in water-limited basins are consistent with irrigation developments occurring preferentially in their arid and semi-arid climates. Overall, changes in AET/P among the investigated basins are better explained by differences in the basin characteristics of reservoirs and irrigation than by differences in atmospheric climate conditions or their changes.

Changes in CVR among the 100 basins are also variable (Fig. 4A), yet a dominant change pattern is seen as CVR decreases with higher increase in the relative storage capacity of reservoirs, ΔRES (Fig. 4B). Such a pattern should not be surprising as reservoir water management commonly aims at smoothing runoff variability. In general, CVR changes are better explained by differences in the reservoir water storage than by differences in the relative area equipped for irrigation among basins (Fig. 4C). This is understandable since irrigation does not per se imply the use of reservoir water storage that decreases CVR; the water used for irrigation can also be taken from groundwater and not all of the water reservoirs that decrease CVR are used for irrigation. The CVR changes are further not explained by atmospheric input conditions in terms of the relative intra-annual variability of monthly precipitation (CVP) or the change in CVP (Fig. 4, D and E). These results support previous regional findings of CVR decrease as a flow regulation effect (6, 15) rather than an effect of irrigation per se, or of atmospheric input conditions or their changes.

Fig. 4 Dominant human effect of flow regulation on temporal runoff variability.

(A) Kernel density estimate curve of the distribution of changes in relative intra-annual variability of monthly runoff (ΔCVR) between the periods 1901–1954 and 1955–2008 for 100 large basins in terms of (B) the change in total reservoir storage capacity relative to basin area between the two time periods (ΔRES, in millimeters); (C) the area equipped for irrigation relative to basin area (IA); (D) the relative intra-annual variability of monthly runoff (CVP); and (E) the change in CVP; i.e., ΔCVP. The distribution of ΔCVR is divided into five subgroups, and box plot statistics are shown similarly as in Fig. 3, where Qmin = –0.91, Q20 = –0.15, Q40 = –0.05, Q60 = –0.01, Q80 = 0.02, and Qmax = 0.19. Statistical significance is also tested and shown as in Fig. 3 but in this case regards whether the variable values within a subgroup of the distribution are significantly greater (P < 0.05) than the variable values in at least one of the higher ΔCVR subgroups. The P values of the statistical test are shown in table S4B.

Changes in AET/P and CVR may also be driven by additional atmospheric climate conditions and changes to those we investigated [e.g., (24, 25)] and by other landscape drivers than FRI alone, such as deforestation, non-irrigated agricultural development, and/or other changes in landscape conditions for water storage and water phase [(5) and references therein]. The effects of such additional change drivers may explain more of the total variability of AET/P and CVR changes among basins (Figs. 2 to 4 and figs. S4 and S5). Nevertheless, the FRI-related effects explain a large part of that variability. Overall, the FRI-related areas of surface water reservoirs and irrigation cover less than 3% (9) and 25% (10) of total basin area (maximum values for the Riviere aux Outardes and San Joaquin River basins), respectively. It is thus more as proxies for associated human water use, rather than as extensive land-use areas per se, that reservoir and irrigation extent measures can explain the FRI-related increase in AET/P and decrease in CVR that were found on a global scale.

In order to estimate the absolute global FRI-related increase in AET, we assumed that, on average across the different world conditions and changes that are spanned by the basins of each FRI impact category, the change of area-normalized AET may be approximately similar among the three basin categories (Fig. 1), except for the FRI-related change component that distinguishes these categories. We applied this assumption in two different methods (see the supplementary materials) for estimating the FRI-related global AET change. Method 1 assumes similarity on average in all types of changes (except the FRI-related ones), whether they are driven by atmospheric climate change or by various (non-FRI) changes in the landscape. Method 2 assumes that only the landscape-driven changes are, on average, similar among basin categories. In the estimation of global FRI-related AET increase, we also considered FRI-related changes in basin water storage that can be expected to be consumptive (7) because of the filling of constructed reservoirs and the use of groundwater for irrigation (26).

The combined estimates of methods 1 and 2 imply a global FRI-related increase in long-term average volumetric AET flow of 3563 ± 979 km3/year from 1901–1954 to 1955–2008. This implies an increased freshwater loss from the landscape to the atmosphere and thus a corresponding increase in the global human consumption of fresh water. Adding this FRI-related increase to previous estimates of global human freshwater consumption for various other sectors [in total, 807 km3/year (1) for non-irrigated agriculture, deforestation, industry, and municipalities] yields a total global human freshwater consumption of 4370 ± 979 km3/year. This long-term average consumption over 1955–2008, including most of its uncertainty range, is above a proposed freshwater planetary boundary of 4000 km3/year (27) (fig. S6).

Furthermore, the FRI-related consumption increase from 1901–1954 to 1955–2008 corresponds to 39% of a recent estimate of 9087 km3/year for the current global water footprint of humanity (28). Updating the irrigation component of 1962 km3/year (with no flow regulation part) in the latter (29) by the present estimate of FRI-related water consumption raises the total global water footprint of humanity by 18% to 10,688 ± 979 km3/year; i.e., to a considerably more unsustainable level (30).

Even though coarse, the present estimates use a wide range of available long-term hydroclimatic observations for the quantification of FRI-related changes in global freshwater consumption. The results expand the uncertainty range of both the FRI-related and the total global human consumption of fresh water. They also show that worldwide observation data can and should be used to quantify global freshwater consumption effects of FRI, in addition to only attempting to model such effects. Modeling alone may lead to considerable effect underestimation (9, 16, 21, 27, 31, 32) when compared with the present observation-based results. Finally, these results stress the importance of considering local water use as a key change driver in Earth system studies and in the modeling of global hydroclimatic change.

Supplementary Materials

Materials and Methods

Figs. S1 to S7

Tables S1 to S4

References (3362)


  1. ACKNOWLEDGMENTS: The Swedish Research Council (VR, project 2009-3221), the strategic environmental research project Ekoklim at Stockholm University, and the BECC strategic research area of Lund University and the University of Gothenburg have funded this study. All data used in our analysis are included in the supplementary materials available on Science Online. The authors thank the anonymous reviewers for their suggestions and recommendations, which have considerably improved this manuscript, as well as J.-O. Persson from the Department of Mathematics at Stockholm University for his statistical advice.
View Abstract


Navigate This Article