Influence of Groundwater Extraction Costs and Resource Depletion Limits on Simulated Global Nonrenewable Water Withdrawals Over the Twenty‐First Century

Abstract

Future rates of global groundwater depletion will depend on the economic and environmental viability of extracting water from increasingly stressed aquifers. Here we analyze global groundwater depletion by considering these factors explicitly. Global gridded groundwater availability and extraction cost data are aggregated to produce nonrenewable resource supply curves for 235 major river basins and geopolitical regions. These resources are then exposed to dynamically generated demands for water in a fully coupled, multisectoral, global simulation. As groundwater head levels drop, imposing greater capital and operating costs to bring water to the surface, modeled water use sectors are able to deploy a range of supply‐ and demand‐driven adaptive responses. Results demonstrate large sensitivity in global groundwater depletion rates to adjustments in resource exploitability. Extraction costs moderate demands for nonrenewable water substantially, resulting in the onset of a decline in global groundwater depletion rates within the twenty‐first century. New groundwater depletion hot spots may emerge as crop producers abandon overexploited basins and expand croplands in regions with cheaper, more plentiful water resources.

Introduction

Groundwater has played a pivotal role in the global agricultural economy, transforming vast swathes of arid territory into thriving croplands. Today, land irrigated by groundwater produces perhaps a fifth of the world's food (Siebert et al., ). But this state of affairs cannot continue indefinitely. An undesirable consequence of humanity's reliance on groundwater has been rapid depletion of water levels in many of the world's major aquifer systems (Famiglietti, ; Konikow, ; Rodell et al., ; Wada et al., ; Wada & Bierkens, ). Continued depletion of finite groundwater reserves raises globally significant threats, including abrupt agricultural decline (Aeschbach‐Hertig & Gleeson, ) and associated disruption to internationally traded food supplies (Dalin et al., ). The problem is likely to worsen as socioeconomic development increases global demands for water (Satoh et al., ). Sustainable management of groundwater is thus marked as a major global policy challenge, requiring risk analysis informed by long‐range projections of aquifer exploitation and exhaustion across the world's most important basins of irrigated agriculture (Ashraf et al., ; Connor, ).

Projecting global groundwater depletion into the twenty‐first century requires global‐scale models that inevitably rely on various simplifying assumptions. In the current literature, a common assumption is that there exists no feedback between water availability and water demand. Spatially varying demands for water are introduced as fixed trajectories, and nonrenewable groundwater is summoned where needed to supply the portion of that demand unmet by renewable water. Perhaps unsurprisingly, the resulting projections of groundwater depletion follow the assumed demands for water, increasing steadily through the century (e.g., Yoshikawa et al., ; Wada & Bierkens, ). But the underlying assumption of a fixed water demand trajectory is problematic, because it neglects the likelihood that users of groundwater will curb their withdrawals as resources draw down. There are two compelling reasons to expect this behavior: cost and government intervention. In order to use groundwater, users must first raise it to the land surface. This requires substantial capital investments (e.g., well construction, pumps) and operating cash flows (e.g., electricity costs), which tend to increase as reserves are drawn down (Dennehy et al., ; Fishman et al., ; Shah, ). Absent subsidies for water—which are widespread though not ubiquitous—water users will be less inclined to pump water when the costs of doing so eradicate their profit margins (Taylor, ). But even if the economics remain favorable, the environmental impacts may not. Pumping of groundwater may cause land to subside, allow contaminants to intrude drinking water supplies, or deplete streams, resulting in degraded aquatic ecology (Giordano, ; Zhu et al., ; MacDonald et al., ; Gleeson and Richter, ). Groundwater that is economically viable from the water user's perspective may therefore be left untapped as governments regulate to mitigate external, politically unacceptable environmental risks (the Sustainable Groundwater Management Act of California is a good example). In other words, economic and environmental limits to resource exploitation are likely to shape future global groundwater depletion. The question is not simply how much water is physically available; we need to know how much is economically and environmentally exploitable and then understand how those constraints would affect an assessment of when an aquifer will become unviable for human applications. To our knowledge, the scientific literature currently lacks a systematic analysis of these effects and their importance for modeling groundwater depletion at the global scale.

Here we explore the effects of water resource exploitability on global groundwater depletion using the Global Change Assessment Model (GCAM)—an integrated assessment model that allows for water users to adjust withdrawals in response to cost and availability constraints. We embed into GCAM a global data set specifying groundwater availability and associated cost of extraction for 235 water basins with global coverage. These resources are then subjected to endogenously generated sectoral demands for water. In other words, we make no assumptions as to future rates of groundwater depletion; rather, we specify the costs of continued depletion and allow the users of water to choose whether to continue depleting water as costs increase. We simulate six water exploitability scenarios to characterize uncertainty in spatiotemporal patterns of global groundwater depletion through the twenty‐first century. While previous integrated assessment modeling work has demonstrated sensitivity in global water withdrawals to costs of groundwater (Kim et al., ), this study is the first to employ realistic cost estimates and then track the global effects of water cost and availability constraints across the world's groundwater depletion hot spots.

Method

Justification

A possible approach to exploring the influence of exploitability on global groundwater depletion would be to constrain available groundwater within a global hydrological and water resource model (GHWRM), such as H08 (Hanasaki et al., ), WaterGap (Alcamo et al., ), or PCR‐GLOBWB (Van Beek et al., ). GHWRMs provide spatially distributed, fine temporal resolution hydrological process representation, including human water withdrawals and return flows. Advanced developments also incorporate dynamic demand feedback in response to limited water supply (de Graaf et al., ), which may allow one to investigate the sensitivity of global withdrawal and depletion rates to changes in available groundwater. A residual problem with this approach is a lack of internal model coupling with the socioeconomic drivers of groundwater depletion. Consider, for illustration, the example of groundwater withdrawals for irrigation. In an international economy, a loss of crop production leads to increased prices as demand exceeds supply. This should motivate the expansion of crop production in new lands. So the slowing of groundwater depletion in one region may coincide with quickening depletion in another. GHWRMs are presently ill‐equipped to incorporate international economic feedback of this nature.

A different approach would be to use an integrated assessment model. These models provide lumped regional representation of resource supply and demand. Few integrated assessment models include water resources, and those that do provide very simplified hydrological process representation as compared with GHWRMs. But integrated assessment models offer unique features that cannot be incorporated easily within a GHWRM structure, including international commodity markets, global price signals, and fully coupled integration of land, energy, climatic, and socioeconomic systems. An example is the MIT Integrated Global System Model framework, which has been coupled with a water resource system component (Strzepek et al., ) and used to assess effects of alternative climate change and mitigation policies on water stress (Blanc et al., ; Schlosser et al., ; Strzepek et al., ). Another example is the LPJml‐MAgPIE‐REMIND framework, which has been used to assess water demands under alternative socioeconomic and policy pathways (Mouratiadou et al., ). For the problem of exploring the sensitivity of global groundwater depletion to exploitability assumptions, one ideally needs an model that passes the costs of water extraction onto end users that are able to reduce their withdrawals through technology adoption (e.g., water efficiency improvements), development of unconventional supply sources (e.g., desalinated water), abandonment of unprofitable operations, and shifting of virtual water demands (e.g., crop irrigation water) toward regions with cheaper or more plentiful water resources. These capabilities are provided by the Global Change Assessment Model, adopted in this study.

Global Change Assessment Model

GCAM is a Representative Concentration Pathway (RCP) class, integrated assessment model designed for exploring the dynamics of global change in socioeconomic, energy, land, agricultural, and water systems (Calvin et al., ; Edmonds & Reilly, ; Kim et al., ; the software and associated documentation are publicly available at jgcri.github.io/gcam‐doc). Land, energy, and water resources are shared among competing users within prespecified geographical units that form markets, which are linked through trade in energy and agricultural commodities. Allocations of resources are determined by establishing equilibrium prices for all land, energy, and water markets in five‐year time steps. The model adopts and retires various technology options using variants of the logit choice function, which is implemented to avoid the unrealistic case of a single most profitable technology capturing the entire market share. Choices are driven by externally specified drivers (e.g., population and GDP growth) and coinciding intensification of competition for finite resources. All sectors are fully integrated within a single system and solved simultaneously for each time period.

In GCAM, water demand and supply are balanced in each of 235 water basins, delineated using a combination of catchment and geopolitical boundaries. Demands for water are represented across six distinct sectors (Hejazi, Edmonds, Clarke, Kyle, Davies, Chaturvedi, Wise, Patel, Eom, Calvin, & Moss, ; Hejazi, Edmonds, Clarke, Kyle, Davies, Chaturvedi, Wise, Patel, Eom, Calvin, ), allowing for a nuanced approach to demand adjustment in response to price and availability of supply (Kim et al., ). Modeled demand sectors are irrigation, livestock, electricity production, resource extraction (mining), industry, and domestic. Basin‐level irrigation demands are specified for 12 distinct crop classes (Chaturvedi et al., ). Water demands per unit crop produced are gradually reduced over the century to reflect projected water efficiency improvements (based on Bruinsma, ). All crops are traded freely on a global market, and so irrigated agriculture may respond to water price by shifting to new basins that have a comparative advantage in this respect (noneconomic barriers to trade are not incorporated). Substitution from irrigated to rain‐fed agriculture is also permitted depending on land availability. Electricity sector water demands can respond to water costs through adoption of efficient cooling technologies or even through shifts away from thermal power plants toward modes of generation that use less water. Domestic water demands—driven primarily by population and GDP—are elastic, forcing users to moderate consumption if prices increase (Hejazi et al., ). For all water demand sectors, technology costs, water use coefficients, and consumption efficiencies are specified at regional resolution, compiled from a wide range of data sets (Amarasinghe et al., ; Averyt et al., ; Davies et al., ; FAO, ; Gerdes & Nichols, ; IBNET, ; Macknick et al., ; Maheu, ; Mekonnen & Hoekstra, ; Rohwer et al., ; Shiklomanov & Balonishnikova, ; Solley et al., ; Vassolo & Döll, ).

Water supplies in GCAM come from three sources: renewable water, groundwater depletion (herein used interchangeably with “nonrenewable groundwater”), and desalinated seawater. Renewable water—which accounts for both direct surface water extraction and groundwater pumping that draws on recharged groundwater and captured streamflow—is the cheapest source of water in GCAM. An upper limit on renewable water is taken as the long‐term mean annual flow for each basin, computed by routing gridded runoff at 0.5° spatial resolution. The mean basin flow is taken as an upper limit because this represents a river system that is fully regulated—meaning water users can readily access the mean flow at all times. This limit is determined externally for each GCAM basin using Xanthos—a global hydrology model that accounts for surface and subsurface processes (Liu et al., ). Only a portion of the basin flow will be made available for use, depending on environmental flow requirements and installed infrastructure for capturing, transporting, and storing water. This accessible portion of renewable water is calculated for the majority of basins as volume of annual runoff that is potentially stable (meaning available even in dry years; Postel et al., ). This volume is determined by simulating the effects of both base flows and in situ storage reservoirs included in the Global Reservoir and Dams inventory (Kim et al., ; Lehner et al., ), with an allocation of 10% of streamflow for environmental purposes. For instances where estimates of groundwater depletion are available (approximately one fifth of basins), the accessible portion of renewable water is back‐calculated from the balance of total water withdrawals and supply from groundwater depletion observed over a historical calibration period (section ).

Nonrenewable groundwater may be tapped by all six demand sectors and is modeled using nonrenewable resource supply curves (examples in Figure S1 in the supporting information). Each basin is assigned a finite total volume of groundwater, which is split into separate grades of increasing cost. The cumulative volume represents the total available groundwater reserve. Depletion of a given grade of water will result in a sustained, irreversible price increase as the next grade is mobilized (data described in section ). Desalinated seawater is readily available to all nonirrigation demands, but at significant cost that accounts for energy use (see Kim et al., , for more details). GCAM water prices are passed onto the end users and ultimately reflected in the costs of their goods and services (e.g., cost of electricity or crops). Increasing costs of conventional water thereby promote nonconventional supply (i.e., desalination) and alternative, water‐efficient means of production. Dry cooling technologies may be installed in thermal power plants, rain‐fed crop land may replace irrigated crop land, and so on. In this way, groundwater depletion and coinciding price increases drives supply‐ and demand‐side adaptive response as well as potential regional shifts in groundwater depletion. To our knowledge, this study is the first to capture the influence of these behaviors on groundwater depletion at a global scale.

Nonrenewable Groundwater Cost and Availability

Physically exploitable groundwater reserves (i.e., total water before accounting for economic and environmental exploitability) are quantified for all major aquifers by combining a range of existing data sources. First, areal extents of aquifers are defined by geo‐referencing digital hydrogeological maps from the Worldwide Hydrogeological Mapping and Assessment Programme (Richts et al., ). Global data sets specifying porosity (Gleeson et al., ), aquifer thickness and permeability (De Graaf et al., ), and depth‐to‐groundwater (Fan et al., ) are then combined to produce estimates of total groundwater thickness at 50‐km spatial grids. From these data we must derive basin‐level estimates of environmentally exploitable groundwater, split into grades of available water that become more expensive as resources are used up (i.e., the format adopted in GCAM, described above). To estimate these nonrenewable resource supply curves, we apply an off‐line physics‐based extraction cost model (Naggar, ). This involves simulation of groundwater pumping at each 50‐km grid. Groundwater extraction costs comprise both capital and operating costs. Here annualized capital costs include well installation and maintenance as a function of depth and geological complexity (Richts et al., ). Operating costs are based on electricity associated with lifting groundwater to the surface, accounting for pump efficiencies, well yields, and country‐specific power costs (IEA, ). Lacking global data on groundwater quality, we neglect the costs of treating water to an acceptable standard for application. Also missing are the capital costs of conveyance and storage infrastructure. These costs may be substantial—in some cases comprising the dominant share of annualized groundwater costs (Palanisami et al., )—and should be addressed as and when suitable data products become available.

We develop three groundwater availability/cost scenarios by allowing for a maximum 5, 25, and 40% of the physical water availability in each grid cell. These limits are based on Korus and Burbach (), but are of course arbitrary to some extent and would depend on various local factors unavailable for global scale study. For each scenario, gridded well water production estimates are aggregated to GCAM basin delineations. Unit costs are normalized to USD/m³ and averaged over each basin. The three limits are introduced in the absence of data that might be deployed to define environmental limits to extraction for each basin. In other words, we lack a consistent data set with global coverage that would allow us to define or model for each basin the various environmental factors that would promote government regulation to halt groundwater depletion. This means that GCAM nonrenewable water withdrawal responds dynamically to extraction costs within each basin unless the assumed environmental limit has been breached, after which depletion must cease. Figure  shows how availability of water varies across the three scenarios. Inexpensive groundwater (<$0.10/m³) is distributed highly unevenly throughout the world—highlighting the importance of spatially varying cost assumptions for cost‐driven global groundwater depletion analysis. This inexpensive portion represents less than 2% of the groundwater available under the 25% exploitability scenario. A further 54% of the exploitable resource can be extracted for less than $1.00/m³, and 44% of available water in this scenario is considered expensive (>$1.00/m³). In the GCAM simulation, its viability will become highly dependent on global demand and price of agricultural commodities, as well as the relative costs of water across all basins where those commodities can be produced.

Fig. 1. Basin‐level exploitable groundwater volume (nonlinear color scale) available for extraction cost of less than $0.10/m3 and $1.00/m3 for (a) 5%, (b) 25%, and (c) 40% environmental exploitability scenarios.

Model Calibration

To ensure that groundwater depletion rates are simulated realistically in GCAM, we calibrate the model to match previously estimated historical groundwater depletion rates for each basin. This is achieved by adjusting the available proportion of renewable water that is accessible for human withdrawals (section ). First, the basin‐level upper limit of renewable water is defined by forcing Xanthos with WATCH reanalysis climate data (Weedon et al., ). The accessible fraction of this water is then iterated until historical period GCAM groundwater depletion matches target historical groundwater depletion rates. By calibrating the accessible renewable water fraction in this way, the model implicitly accounts for the range of factors affecting accessibility of water and its availability within each basin—including the regulation by storage infrastructure, internal water transfers, environmental limits to abstraction, and reuse of water, which has been found to be important in many regions that irrigate heavily (Grogan et al., ).

The target historical groundwater depletion rates in this calibration exercise are derived from prior volumetric estimates of groundwater abstraction and net inflow (recharge minus discharge to rivers) for the year 2000 in 0.5° grid resolution (Gleeson et al., ). In Gleeson et al. (), these abstraction data are downscaled from country‐level estimates (www.un‐igrac.org); recharge is simulated in a hydrological model PCR‐GLOBWB (van Beek et al., ) and aquifer discharge is taken uniformly as the 10th percentile of monthly flows. Here nonrenewable withdrawal is computed for each basin by aggregating abstraction minus net inflow across grids. Each GCAM basin is thus considered a distinct, hydrologically connected aquifer system.

Experimental Setup

We are concerned in this study with the influence of water resource cost and environmental exploitability on global groundwater depletion. We consider two main sources of uncertainty: first is the uncertainty in the cost and availability of environmentally exploitable groundwater within each GCAM basin (represented by the three scenarios described above) and second is the future rate of expansion of water storage infrastructure. Reservoirs regulate flows and thereby increase the reliable access to surface water resources. This potentially offsets reliance on groundwater resources and would therefore be expected to reduce rates of groundwater depletion. We possess neither reasonable projections nor a reliable cost model for future reservoir construction, so we incorporate this uncertainty by taking two extremes. A restriction scenario assumes that no additional reservoir construction will take place over the coming century—the fraction of basin flow that can be accessed is fixed over the entire GCAM simulation out to 2100. An expansion scenario assumes that all basins increase access to surface water linearly from current levels to maximum levels by 2100. This means that basins that are already well developed in terms of storage capacity will develop relatively little new storage infrastructure. In contrast, basins that have developed relatively little storage capacity are assumed to construct reservoirs to expand access to surface water rapidly.

We apply these two surface water assumptions to each groundwater availability scenario, giving six water resource exploitability scenarios to be simulated in GCAM. To avoid conflating the influence of resource exploitability with other uncertainties, we exclude possible influence of climate change and hydrology model uncertainty, details of which are reported extensively elsewhere (Schewe et al., ; Taylor et al., ; Haddeland et al., ; Elliott et al., ; Gosling & Arnell, ). Also excluded is uncertainty in the recent historical rates of groundwater depletion—against which the model is calibrated. Estimates of historical trends in terrestrial water storage vary widely across global hydrology models and have been shown to often depart significantly in both magnitude and direction from satellite‐derived estimates (Scanlon et al., ). Given the importance and scale of these uncertainties, the following results should not be read as predictions of global groundwater depletion and water use over the twenty‐first century. Instead, the results characterize the possible influence of groundwater cost and availability constraints on future rates of depletion globally, assuming that the users of water act rationally in economic self‐interest.

Results

Global Groundwater Depletion in the Twenty‐First Century

Simulated global groundwater depletion varies significantly across the six scenarios considered (Figure a). As expected, increased groundwater exploitability (i.e., lower costs and relaxed environmental limits) leads to increased rates of depletion. Interestingly, the marginal effect is much larger between the 5% and 25% scenarios than between the 25% and 40% scenarios, suggesting a diminishing effect of additional environmentally exploitable water on groundwater depletion rates. This effect is explained by two model behaviors. First, water users may not always want to withdraw additional environmentally exploitable water. This would occur in regions where the water has become too expensive for profitable use, or if the additional water is surplus to existing demands. This latter effect is evident in the collapse of global groundwater depletion rates under the renewable water expansion scenarios (i.e., solid bounds in Figure a). Second, an increase in environmentally exploitable water does not necessarily mobilize more groundwater in the simulation. Often the volume that can be physically pumped over the twenty‐first century will be less than the volume of environmentally exploitable water (such pumping limits are modeled in the gridded, physics‐based cost extraction model).

Fig. 2. Effects of resource exploitability on (a) global water withdrawals, (b) average cost of water withdrawn (from all sources), and (c) volumes of nonrenewable groundwater in storage. Results are compared across three groundwater exploitability scenarios (5, 25, 40%) and two renewable water expansion scenarios (expand, restrict).

When access to renewable water is restricted (i.e., no water infrastructure expansion), water users are more reliant on nonrenewable groundwater, which becomes increasingly expensive as resources are drawn down. This has a marked impact on the global average cost of water (Figure b). Under 5% exploitability, for instance, the global price of water in the latter half of the century is about double under the restrict scenario as compared to the expand scenario (although it should be noted that the costs of reservoir expansion are omitted from this analysis). Interestingly, costs begin to level off and decline toward the end of the century. This occurs in line with leveling population and improved crop yields (which improve gradually over the twenty‐first century); as total global water demands ease off, the most expensive resources drop out of use.

Projections of the global volume of remaining exploitable groundwater appear relatively flat over the simulation (Figure c). By 2100, global exploitable groundwater reserves are reduced by approximately 10 (±3)% relative to 2015 volumes. This unremarkable impact reflects the fact that a very large proportion of the world's exploitable groundwater lies in regions where it is not depleted, such as where climate conditions are unfavorable for crop growth (e.g., high northern latitudes), or where rainfall is sufficient to sustain crop production without depleting groundwater (e.g., tropical regions, including the Amazon basin). These global profiles of remaining exploitable water mask much more alarming patterns of resource exhaustion observed for certain regions.

Regional Patterns of Global Groundwater Depletion

Simulated groundwater depletion occurs predominantly across the northern tropics and subtropics (Figure a). The affected band encompasses the western United States, northwest Mexico, northern Africa, the Middle East, central and South Asia, and northern China. In all six simulated scenarios, more than four fifths of global depletion occurs within just 18 major crop producing basins, highlighting that irrigation is the primary driver of global groundwater depletion. The most rapid depletion occurs during the early twenty‐first century within the Indus river basin (~185 ± 3 km³/year in 2020) and the Sabarmati region (~160 ± 20 km³/year). By midcentury, depletion ceases in these basins and intensifies in the Arabian Peninsula (~105 ± 10 km³/year in 2050) and the Nile river basin (~72 ± 1 km³/year). This is an example of spatial redistribution of groundwater depletion in response to resource exhaustion. In the Indus, a large proportion of irrigated land is forced to drop out of service when groundwater is exhausted. Since GCAM is constrained such that global food demands must be met, loss of crop production in one region must coincide with increased crop production in another. Huge losses in crop production in the Indus thereby coincide with sharp expansion of irrigated lands in the Arabian Peninsula and Nile basins. These basins offer competitive yields (mass of crop production per unit land area) for the in‐demand crops and have relatively cheap and plentiful groundwater supplies.

Fig. 3. Simulated groundwater status across six scenarios for 18 highlighted basins (yellow fill, black border). Basins that experience low rates of groundwater depletion are shaded light yellow. All other basins (light gray) experience no depletion. (a) Rate groundwater depletion. (b) Extraction cost. (c) Viable water remaining (exploitable, but may not be economic) in five‐year time steps.

For some of the highlighted basins, maximum simulated rates of depletion are similar across all exploitability scenarios, although the timing of peak depletion may vary. In central Iran, for instance, all six scenarios indicate a steady increase in depletion rates from a current rate of ~25 km³/year to a peak of ~60 ± 10 km³/year, after which the resource becomes economically unviable. In the Indus basin, where there is almost no scope for expansion of surface water resources (currently ~95% of flow is withdrawn for irrigation—Laghari et al., ), further reservoir expansion has very little impact. This contrasts with a number of basins where groundwater depletion rates are more sensitive to adjustments in renewable water exploitability. In the Tigris‐Euphrates, for example, the expansion scenario results in a ceasing of groundwater depletion by midcentury, while the restriction scenario causes groundwater depletion to slow only once extraction costs become excessive (Figure b).

As with the globally aggregated results, there is a clear inverse relationship between extraction cost and depletion rate in each basin. When costs increase, depletion slows and often ceases (compare Figure a with Figure b for any basin). In some cases, such as Syr Darya and Southwest Caspian Sea, groundwater depletion may cease entirely despite substantial volumes of water remaining in the aquifer—a clear indication that costs have rendered the resource uneconomic (see Table S1 in the supporting information). In other cases, depletion will cease despite seemingly competitive costs of extraction—a clear indication that environmental limits have been reached. In central Iran, for instance, end‐of‐century depletion is zero (Figure c) even though the most expensive grade of water is priced at ~$0.3/m³.Geology plays a role in determining the rate of withdrawal cost increase with depletion. Basins that halt withdrawal due to cost are characterized by relatively large ranges of hydraulic properties, causing sharp extraction cost increases when water is depleted from permeable zones into less ideal geological conditions (e.g., Caspian Sea Southwest). In contrast, basins with relatively gentle cost curves, such as Rio Grande, California, and northwest Mexico, have more amenable geological properties. Extensive high‐permeability zones within these aquifers allow operators to draw on available resources till near exhaustion of the environmentally exploitable water (see examples in Figure S1 in the supporting information).

Discussion and Conclusions

It is not uncommon for analysts to assume that future groundwater depletion will increase steadily with global demands for water (Table ). Our findings challenge this preconception. Of course, a growing and increasingly wealthy human population will likely demand more water—most of which will go toward producing food crops. But groundwater depletion may only continue so long as aquifer water is exploitable. In reality, there are both environmental and economic limits to groundwater extraction. Overdepletion may be unacceptable for local human populations, and the costs of extracting groundwater are likely to increase with drawdown. Water users acting rationally in economic self‐interest may then respond by sourcing water from elsewhere, improving water efficiency, or abandoning operations that are no longer profitable. By allowing for these behaviors, our GCAM simulations provide new insights as to the possible trajectory and spatial distribution of groundwater depletion over the twenty‐first century.

Global Groundwater Depletion Rates (km³/year) for This Study, Two GHWM Studies—Using H08 (Yoshikawa et al., ) and PCRGLOB‐WB (Wada & Bierkens, )—and a Previous GCAM Study (Kim et al., )

Study	2000	2010	2050	2100
Yoshikawa et al. ()	332		775
Wada and Bierkens ()2		300		510–680
Kim et al. ()3	550	800	150–1750	60–1500
This study (restricted renewable water)3	550	640	580–910	180–480
This study (expanded renewable water)3	320–580	110–210

Note. All estimates are rounded to nearest ten km³/year. Uncertainty based on climate change using ensemble of five CMIP‐five GCMs. Uncertainty based on groundwater cost and availability assumptions.

Our analysis identifies three important effects that are absent from previous projections of global groundwater depletion. The first is a clear moderating effect on global groundwater depletion, which brings about a departure from the historically observed linear increase in depletion with demand. Yoshikawa et al. () project a doubling of global groundwater depletion rates between years 2000 and 2050. A projected rise in groundwater depletion is also present in Wada and Bierkens (), although the rate of increase is less, with an approximate doubling of current depletion rates by the end of the century (the discrepancy between the two studies is largely due to allowance for increased irrigated land area in the former). In contrast, we report here a significant decline in global groundwater depletion, the onset of which depends on the stringency of the exploitability assumptions. This general trend is also apparent in the Kim et al. () simulations—although it is less pronounced, owing to cheaper and more plentiful groundwater. Here we identify a clear slowing of global groundwater withdrawals, with end‐of‐century depletion rates exceeded by year 2010 depletion rates under all scenarios. This result suggests that it cannot be safely assumed that global groundwater depletion will increase as the demands for water increase.

The second important effect is relatively large uncertainty caused by groundwater exploitability assumptions. The uncertainty arising from our narrow range of possible exploitability assumptions (~300 km³/year uncertainty across the restriction scenarios) clearly eclipses the uncertainty resulting from climate change explored in Wada and Bierkens () (~170 km³/year). The uncertainty found in Kim et al. () is larger still (~1400 km³/year by 2100), although these estimates were derived from sensitivity around arbitrary groundwater cost curves as opposed to the physics‐based estimates applied here. By applying a narrow range of more realistic resource curves, we confirm that uncertainty in exploitability has a very important impact on global groundwater depletion that cannot be neglected if one is to understand likelihood of resource exhaustion of any aquifer.

The third related important effect is the impact of renewable water accessibility assumptions. Results indicate that the expansion of reservoir storage—providing additional access to renewable water resources—would roughly halve end‐of‐century depletion rates compared to totally restricted reservoir expansion. This suggests that for many basins a viable strategy for avoiding or adapting to groundwater depletion may to invest heavily in surface water infrastructure. Further work is needed to either project realistic rates of reservoir expansion or quantify the costs of water reservoir construction (and thus the marginal cost of reliable surface water supply), so that a model like GCAM might dynamically deploy these resources within the simulation.

The behaviors simulated in GCAM also provide unique insight into possible spatial dynamics of future groundwater depletion. The basin‐level results reported above are the first, to our knowledge, to highlight the possible effects of groundwater depletion shifting across regions as exploitability limits are reached. This reshuffling of virtual water demand allows global water withdrawals for irrigation to expand rapidly in the first half of the simulation. Between the years 2000 and 2050, the GCAM annual water withdrawal for agriculture increases by 1,100–1,300 km³, compared to 850–900 km³ in prior simulation studies that incorporate climate change impacts but not dynamically modeled irrigated crop land expansion (i.e., Wada & Bierkens, ). This result suggests that the combined effects of exploitability constraints and global economic forces—namely, commodities markets and trade—could define the spatial distribution of groundwater depletion in the twenty‐first century.

Our approach is not without limitations. Problems arising from lumped basin representation are exposed by abrupt shifts and directional changes in global groundwater depletion rates, particularly in the middle of the century (Figure a). These aberrations are caused by sudden exhaustion of exploitable groundwater in particular regions; if groundwater becomes uneconomic in a region of significant groundwater depletion, then cessation of pumping in that region may show up in the global total. It is unlikely that such abrupt shifts in global groundwater depletion would occur in reality. Instead, one would expect costs to rise nonuniformly across different areas of an aquifer, resulting in a more gradual slowdown in overall depletion within any basin. One would also expect a slower user response to extraction cost increases if, for example, sunk costs have already been incurred (e.g., major capital costs for borehole drilling and equipment) or if government policy supports further withdrawal through subsidies on pumping electricity costs for water users (as reported in Theesfeld, ).

Also problematic are the basin delineations and the assumption that these coincide with hydrologically connected aquifers—seldom the case in reality. The GCAM water basins are a practical solution for integrating multiple sectors (land, agriculture, water). Fidelity in the hydrological aspects of the model is traded off for improved representation of the interactions and feedback between these sectors. As we note in our introduction, hydrology models will better represent hydrology, but an integrated assessment model will better represent the drivers of water demand and how those drivers change in response to cost and availability of resources (in line with established international economic theory). Assumed frictionless trade in agricultural commodities is another difficulty. While trade is not constrained in GCAM, both production and demand of all agricultural products in all regions are calibrated, and there a degree of inertia in the trade outcomes, though perhaps less than is reasonable. This could have a nontrivial influence on the behavior of the agriculture sector and therefore the spatial distribution of groundwater depletion over time. The result that groundwater depletion may shift to new regions depends critically on the assumptions that global food demands follow a fixed trajectory and that crops can be easily traded across world regions.

Other limitations are common across all global studies of this nature. We rely on global data sets that are likely subject to significant bias. Given the likelihood of additional significant bias in the initial calibration data (documented in Scanlon et al., ), the large uncertainty bands depicted in our analysis represent a lower bound envelope of possible uncertainty in future global groundwater depletion. Other unexamined economic uncertainties that could affect simulated water withdrawals include: the cost of desalinated water and its future viability for supplying noncoastal water use applications with large demands; the level of government subsidies applied to groundwater extraction costs, which would have the effect of artificially expanding the volume of economically viable groundwater from the water user's perspective; and the possibility of dietary shifts in response to higher prices for water intensive crops. Political factors such as barriers to trade in agricultural products, or incentives to encourage water efficient technology adoption, may be important, and could be examined in future work.

Notwithstanding such limitations, our study shows the huge extent to which the cost and availability of resources could shape global groundwater depletion rates over the twenty‐first century. A major consequence may be a peaking of groundwater depletion in many important agricultural regions. This carries clear implications for groundwater‐dependent nations, which stand to suffer economically if their agriculture sector becomes uncompetitive. Governments may need to choose whether to support their agriculture sectors through increased subsidies for groundwater pumping, or infrastructure projects that extend access to renewable water resources. Either course of action could carry significant environmental risks. The alternative is to allow market forces to redefine the global distribution of irrigated agriculture in response to increasing extraction costs in the major food supply regions. Future work will examine more deeply the effects of groundwater depletion on regional agriculture and global trade in food commodities.

Acknowledgments

The groundwater cost and availability data applied in this study are described in the supporting information. Nonrenewable cost‐availability supply curves aggregated to GCAM basins are included in csv format. This research was supported by the Office of Science of the U.S. Department of Energy Biological and Environmental Research through the Integrated Assessment Research Program. PNNL is operated for DOE by Battelle Memorial Institute under contract DE‐AC05‐76RL01830.