Global expansion of sustainable irrigation limited by water storage

Significance

Expansion of sustainable irrigation (i.e., using sustainable water resources to irrigate water-limited croplands) can increase food production, while neither depleting water stocks nor encroaching upon nature. Yet, there is a mismatch in timing of water availability and of irrigation needs in many geographies, necessitating temporary water storage. We quantify global volumes of water that requires temporary storage to be leveraged for an expansion of sustainable irrigation and discuss options to provide that storage. While dammed reservoirs are crucial for today’s irrigation, dams alone will not suffice to fully leverage sustainable water resources in the future and while creating major impacts on nature and people. This highlights the urgent need for alternative solutions to water storage and demand side approaches to food security.

Abstract

Providing affordable and nutritious food to a growing and increasingly affluent global population requires multifaceted approaches to target supply and demand aspects. On the supply side, expanding irrigation is key to increase future food production, yet associated needs for storing water and implications of providing that water storage, remain unknown. Here, we quantify biophysical potentials for storage-fed sustainable irrigation—irrigation that neither depletes freshwater resources nor expands croplands but requires water to be stored before use—and study implications for food security and infrastructure. We find that water storage is crucial for future food systems because 460 km³/yr of sustainable blue water, enough to grow food for 1.15 billion people, can only be used for irrigation after storage. Even if all identified future dams were to contribute water to irrigation, water stored in dammed reservoirs could only supply 209 ± 50 km³/yr to irrigation and grow food for 631 ± 145 million people. In the face of this gap and the major socioecologic externalities from future dams, our results highlight limits of gray infrastructure for future irrigation and urge to increase irrigation efficiency, change to less water-intensive cropping systems, and deploy alternative storage solutions at scale.

While 1 in 10 people still lack access to sufficient and reliable food (1), current agricultural practices deplete water resources, impact water quality (2–4), impair natural landscapes (5), generate one-fourth of the world’s greenhouse gas emissions (6, 7), and thus open existential risks for future generations (8, 9). Major increases in food demands are projected by 2050 (10–12) and meeting those demands while securing the climate, biodiversity, and nature’s benefits to people is perhaps the greatest challenge of the 21st century (12–14).

Global societies will need to adopt a broad portfolio of policies to provide nutritious, sufficient, and affordable food to all without exceeding planetary boundaries (10, 12). While 3.3 billion people are currently fed from unsustainable agricultural practices, it has been estimated that up to 10.2 billion, an upper projection of the global population in 2050 (11), could be fed sustainably (12) through a transformation on both demand and supply sides of the global food sector (14, 15). On the demand side, reducing food waste and promoting sustainable diets are of outstanding importance (16–19) to feed up to 2.4 billion more people (12). Moreover, ambitious supply-side solutions, such as improved land, water, and nutrient management paired with expansion of irrigation, fertilizers, and croplands can potentially feed up to 4.4 billion people (12).

On the supply side, two important pathways prevail, with respective opportunities and challenges. Extensification (i.e., expanding croplands) would risk encroaching upon natural lands and their vital contributions to people, nature, and food systems themselves (e.g., through pollination, pest control, and genetic resources) (20). Intensification (i.e., increasing productivity on existing croplands) avoids impacts of land conversion but might cause others, for example through unsustainable water (21, 22) and fertilizer use (23). Combination of both pathways is possible, for example by retiring and restoring unproductive land with high biodiversity value, and expanding and intensifying agriculture on natural lands that are less critical for planetary health (24). Ideally, both pathways are embedded with increased resource use efficiency, for example by implementing efficient irrigation and fertilizer application.

Increasing food supply requires careful assessments of opportunities and risks to avoid impacting planetary health and depleting resources, particularly with regards to water. Today, productivity on two-thirds of global cropland is constrained by available rainfall (21) and existing irrigation is often fed from non-renewable water resources (22, 25). At the same time, there are still major freshwater resources available to expand irrigation to water limited croplands without depleting freshwater stocks and environmental flows (3, 26, 27). These water resources could drive a sustainable expansion of irrigation, which could potentially produce food for up to 1.4 billion people (21) while neither depleting water resources nor encroaching upon natural areas (28).

While quantitative limits for expanding irrigation have been studied, broader socioenvironmental implications remain unknown (29). For example, in many geographies, water is available to expand irrigation but water availability and water demands for irrigation do not match in time (2, 30, 31). This mismatch necessitates temporary water storage to leverage sustainable water resources for food security. Water storage for irrigation can be provided through water harvesting with small dams (32), managed aquifer recharge (33), and better management of soil moisture (34). Yet, existing irrigation systems often rely on gray infrastructure in the form of dammed reservoirs (30, 31) to provide water storage (30, 35–37).

Previous studies highlighted the contribution of existing dams and decentralized storage to current irrigated agriculture (30–32, 38, 39), while research on impacts (40, 41) and potential alternatives to future dams (42, 43) has overwhelmingly focused on hydropower dams, for which 3,700 potential sites have been identified (44). However, little is known about the role of water storage in future food systems. Studies on a future expansion of irrigation often assume that water storage is possible without providing further specification (12, 21). This is a significant knowledge gap for three reasons. First, assuming that water storage is feasible likely leads to a major overestimation of how much sustainable water resources remain for future irrigation (26). Second, if gray infrastructure was to remain equally important for future irrigation as it is today, future irrigation dams could create significant economic burdens and negative environmental externalities of future food systems. Third, water storage will be crucial for many aspects of water–food–energy systems (e.g., for flood protection and buffering fluctuating renewable energy sources). Estimating future needs for irrigation storage is thus important for identifying potential synergies and conflicts between sectors that rely on water storage in the water–energy–food nexus.

To overcome these limitations, we herein aim to quantify the role of storage-fed irrigation to leverage sustainable water resources to increase food supply. Our analysis aims to highlight global challenges and opportunities as a primer to study infrastructure implications of future food systems, and as a contribution to a more holistic understanding of future roles and alternatives for large dams in the water–energy–food nexus. We analyze basin-level monthly blue water availability (freshwater in surface and groundwater bodies generated and renewed from the annual hydrological cycle, without considering for nonrenewable groundwater) and water demands of current crop mixes on currently irrigated and rainfed lands (21, 26). In months in which water availability exceeds demands, surplus water can be stored and used later when demands exceed water availability. In those cases, water storage is required to fully leverage sustainable blue water resources for food production. Otherwise, crops would either not attain full yields in months with a water deficit, or water demands would need to be met from unsustainable blue water or nonrenewable groundwater (25).

We calculate annual volumes of storage-fed irrigation on currently irrigated lands (current conditions) and after expanding irrigation to currently rainfed and water-limited lands (future expansion of irrigation for a sustainable intensification of agriculture). This first analysis delineates the biophysical potentials for storage-fed irrigation. We then compare current and future potentials for storage-fed irrigation to how much water storage could be provided by current and future dammed reservoirs. This combination of current and future land use and current and future reservoir storage results in two scenarios: 1) current (demand for storage-fed irrigation on currently irrigated land, storage in existing irrigation reservoirs) and 2) future (demand for storage-fed irrigation on currently irrigated land and where irrigation can be expanded sustainably, storage in existing irrigation reservoirs). There is no global information about future dams for irrigation. To estimate the technical potential for reservoir-based storage in the second scenario, we assume that 3,700 potential dam sites, mapped for their hydropower potential (44), would also provide some irrigation storage. All infrastructure analyses include a Monte Carlo analysis to account for uncertainty in how current and future dams will be operated, if and which fraction of their total storage will be allocated to irrigation, and how much water will reach crops “on the field.”

Based on that framework, we first investigate potential annual volumes of storage-fed irrigation. Second, we quantify for how many people storage-fed irrigation could provide food, and thus the contribution of water storage to intensifying agriculture and thus global food security. Third, we compare the potentials for storage-fed irrigation to storage in current and future dammed reservoirs and estimate the potential contribution of these gray storage infrastructures to future food systems.

Results

Biophysical Storage Needs for Water-Sustainable Agriculture.

Based on our basin-level water balance analysis (Methods and SI Appendix, Box S1), we find that 265 km³/yr of the 469 km³/yr of water that can be used for sustainable irrigation on irrigated lands require temporary storage. Only where there are remaining sustainable water resources did we then model expanding irrigation onto rainfed croplands. This sustainable expansion of irrigation would enable using additional 540 km³/yr of sustainable water resources for crop production worldwide. Of those 540 km³/yr, 195 km³/yr would require temporary storage. Thus, when using sustainable water resources on currently irrigated and currently rainfed lands, 460 km³/yr would be from storage-fed irrigation: that is, using sustainable blue water which has undergone temporary storage (Fig. 1A).

Fig. 1.

Future irrigation will require 460 km³/yr of water from temporary storage (storage-fed irrigation), 265 km³/yr on currently irrigated land, and additional 195 km³/yr when expanding irrigation onto rainfed lands. (A) Potentials for storage-fed irrigation on a basin scale, accounting for storage-fed irrigation on currently irrigated and rainfed croplands. (B) Breakdown of basin-scale totals into storage-fed irrigation on currently irrigated land (hatched bar area) and on currently rainfed croplands to which irrigation could be extended in the future (nonhatched bar). Square markers indicate total potentials for sustainable irrigation: that is, sustainable irrigation both with and without storage (small arrows and numbers indicate values outside of axis limits). The selected 30 river basins with the greatest future totals of storage-fed irrigation account for 60% of the total potential for storage-fed irrigation. Colors for total storage-fed irrigation match between A and B (SI Appendix, Fig. S1 indicates location of highlighted basins).

These volumes of storage-fed irrigation are the biophysical requirements on the field. They do not account for water losses in storage and conveyance, percolation, and evaporation from reservoirs (45), and that active storage volumes are often much smaller than the total storage. Thus, providing any volume of stored water on the field will require significantly more storage volume in a reservoir (an effect that we analyze in the next sections).

The Ganges-Brahmaputra, Mississippi, Tigris-Euphrates, and Niger basins stand out regarding the potential volume of storage-fed irrigation (Fig. 1). Fig. 1B distinguishes between storage-fed irrigation on currently irrigated land (hatched part of bars) and additional storage-fed irrigation on currently rainfed land (nonhatched part of bars). Fig. 1B also highlights how storage-fed irrigation compares to the total water resources available for sustainable irrigation (diamond markers). Of the world’s large rivers, the Ganges-Brahmaputra Basin has by far the largest potential for storage-fed irrigation, nearly three times more than the second-ranked Mississippi. In the Ganges-Brahmaputra, 115 km³/yr of blue water are available for sustainable irrigation (Fig. 1B, square marker). Only 48 km³/yr (difference between bar and square marker in Fig. 1B) can be used for irrigation from instantaneous withdrawals: that is, using sustainable blue water that is available in each month (see SI Appendix, Box S1 for how values were calculated for the Ganges basin). The remainder (67 km³/yr) would require temporary storage before being used for irrigation. Of the 67 km³/yr, 61 km³/yr of stored water resources would supply irrigation on currently irrigated cropland (hatched bar area in Fig. 1B). An additional 6 km³/yr of storage is required for expanding storage-fed irrigation to rainfed croplands (nonhatched bar area in Fig. 1B). It should be noted that 61 km³/yr are the current biophysical potential for storage-fed irrigation on currently irrigated lands. However, in the Ganges-Brahmaputra and elsewhere, only part of that storage-irrigation might be implemented for lack of storage and conveyance (46), current irrigation might be partly unsustainable (e.g., depleting groundwater), or irrigated crops might be grown under a water deficit (25, 26).

Linking Storage-Fed Irrigation and Food Security.

We found that storage-fed irrigation could supply enough food for 1.15 billion people. Of that, food for 719 million people is from storage-fed irrigation on currently irrigated lands and an additional 431 million is from expanding storage-fed irrigation to currently rainfed land (Fig. 2A). Because of different crop mixes, storage-fed irrigation volumes (Fig. 1) and the associated food production (Fig. 2) are not directly correlated. However, the Ganges-Brahmaputra also stands out in terms of potential food production from storage-fed irrigation. Storage-fed irrigation in the Ganges-Brahmaputra can provide food for 189 million people, 158 million from storage-fed irrigation on currently irrigated lands (Fig. 2B, hatched bar area) and an additional 31 million when expanding storage-fed irrigation onto rainfed lands (Fig. 2B, nonhatched bar). All sustainable irrigation—that is, from both stored water and instantaneous withdrawals in the basin—can produce enough food for 317 million people (Fig. 2B, diamond marker).

Fig. 2.

Storage-fed irrigation has the potential to supply food for 1.15 billion people: 719 million on currently irrigated lands and additional 431 million people after expanding irrigation to rainfed lands. (A) Global distribution of people supported from storage-fed irrigation if storage-fed irrigation is fully implemented on currently irrigated and rainfed lands. Basin scale totals depend on water availability (Fig. 1) and the caloric value of local crop mixes. (B) Breakdown of basin-scale totals shows how many people can be supported from storage-fed irrigation on currently irrigated land (hatched) and after expanding storage-fed irrigation to rainfed land for the 30 basins with the greatest future total food potential from storage-fed irrigation. Diamond markers indicate how many people can be fed from irrigated agriculture (with and without storage, small arrows and numbers indicate values outside of axis limits). The 30 large river basins shown in B accounts for 64% of the total food from storage-fed irrigation. Colors for future total storage-fed irrigation potentials correspond between A and B. SI Appendix, Fig. S1 indicates location of highlighted basins.

In the Niger Basin, storage-fed irrigation on currently irrigated cropland can produce food for only a million people today (Fig. 2B hatched bar) because it is sparsely irrigated. Yet, there is a potential for a major increase in storage-fed irrigation on currently rainfed croplands. In total, storage-fed irrigation in the Niger could provide food for an additional 33 million people, the greatest increase in any basin.

Global Gaps in Reservoir-Based Water Storage.

Based on the major biophysical potential for storage-fed irrigation, we studied the role of gray infrastructure in the form of dammed reservoirs to realize this potential. Currently dammed reservoirs have a reported total volume of 5,500 km³ (i.e., without accounting for storage loss from sedimentation), with the storage volume of irrigation reservoirs being nearly 2,000 km³ (47). The total volume of irrigation reservoir is thus already greater than the 460 km³/yr needed for storage-fed irrigation.

Total reservoir storage for irrigation could increase further if some of the identified 3,700 hydropower dam sites (44) would be designed and operated to provide not only hydropower but also water storage for irrigation, as was often the case in the past (35, 37). However, only a fraction of total reservoir storage will be available to meet irrigation water demands on the field. In fact, the amount of stored water from a reservoir that is available to meet irrigation water demands depends not only 1) on the reservoir’s storage volume, but also 2) on the fraction of storage allocated to irrigation and 3) on the efficiency of the receiving irrigation system and water losses during storage, conveyance, and application (2, 48, 49). In turn, this implies that any volume of storage-fed irrigation will require a significantly larger amount of storage volume in a dammed reservoir (2, 48). As a result, the available water storage might not be sufficient to meet storage-fed irrigation water requirements, even in basins where total reservoir storage exceed storage-fed irrigation volumes.

We developed a Monte Carlo analysis to account for uncertainties in how much stored water from each reservoir reaches crops on the field. We then calculated the difference between storage-fed irrigation and water availability from dammed reservoirs: that is, storage deficits (reservoirs volumes in a basin are not sufficient to store all sustainable water resources that could be used for irrigation) or storage surpluses (reservoir volumes exceed what is needed to store all sustainable water resources that could be used for irrigation).

We estimate that currently 98 major basins have storage deficits totaling 149 ± 2.7 km³/yr (where the interval represents the SD over 100,000 Monte Carlo runs). Thus, 149 km³/yr of sustainable blue water could be used for food production on currently irrigated lands but are not accessible for irrigation because of a lack of storage (Fig. 3). In contrast, there are 102 major basins with a storage surplus, totaling 353 ± 26 km³/yr. In those basins, the storage volume of irrigation dams is sufficient to fully supply irrigated agriculture. Storage deficits are found in South Asia, while basins in Europe and the Unites States currently have storage surpluses (Fig. 3A). Even if all potential dam sites were developed and part of the associated reservoir storage was available for storage-fed irrigation, future storage deficits would increase to 209 ± 50 km³/yr. This is because increases in storage-fed irrigation on currently rainfed land would outpace the growth in reservoir storage (Fig. 3B). At the same time, the storage surplus in basins with a surplus will increase to 592 ± 48 km³/yr, as many potential dam sites have been identified in basins with limited future storage-fed irrigation (e.g., the Yangtze, Mekong, or in the Balkans) (see also SI Appendix, Fig. S2).

Fig. 3.

Relying only on dammed reservoirs for sustainable irrigation would forego the use of 209 ± 50 km³/yr of sustainable blue water. (A) Global map of storage surplus/deficit for current conditions (i.e., considering for existing reservoirs and currently irrigated croplands). Red colors indicate a storage gap: that is, where dammed reservoirs are not sufficient to store all water available for storage-fed irrigation. (B) Global map of storage surplus/deficit for future conditions (i.e., with current and future dams and expansion of irrigation to currently rainfed lands). Colors correspond to means over 100,000 Monte Carlo runs to estimate uncertainty from factor controlling water allocation from dams to irrigation. (C) results of Monte Carlo run for 15 large basins for current (blue, corresponding to A) and future (purple, corresponding to B) conditions. Variability in C is created by uncertainty in irrigation efficiencies and water allocations to irrigation, and the volume of future dammed reservoirs (box center line: median, box limits: interquartile range, whiskers: 1.5 times interquartile range; over 100,000 simulations). SI Appendix, Fig. S1 indicates location of highlighted basins.

Different mechanisms are responsible for the storage deficits in different regions. Most European basins have potentials to expand irrigation, but very limited existing irrigation storage and no foreseen additions of future dammed reservoirs (basins with potential dam sites (44) are highlighted in Fig. 3B). Basins in Sub-Saharan Africa have currently no or only small storage deficits (Fig. 3A) because there is some reservoir storage but very little irrigation. Expanding irrigation in Sub-Saharan Africa would create storage deficits even with future dams. In the Tigris-Euphrates and Niger (Fig. 3B) the median storage deficit increases even if future dams could be used for irrigation (Fig. 3C) because expanding irrigation would require more storage than what can be supplied by future dams. In both basins the distribution of future storage deficits/surpluses also includes positive values (see boxplots in Fig. 3C). Thus, the storage deficit could be closed if future dams are designed and operated for irrigation and if irrigation efficiency is increased. The Mekong is one of few major basins worldwide where planned dams would turn a current storage deficit into a storage surplus. Finally, it should be noted that variability in Monte Carlo results is much smaller for basins in South Asia than, for example for basins in Africa because of the lower range in reported irrigation efficiencies in South Asia (SI Appendix, Table S1) (48).

Contribution of Current and Future Reservoir Storage to Food Security.

Supplying water from reservoirs to storage-fed irrigation could play an important role for global food security. Yet, the number of people who can be fed from dammed reservoirs is lower than the full biophysical potential from expanding storage-fed irrigation. This difference is because of the storage deficits. We estimate that water stored in current dammed reservoirs can grow food for 345 ± 102 million people. When expanding irrigation to currently rainfed lands and with future dams, this number could increase to 631 ± 145 million or 55% of the biophysical food potential of storage-fed irrigation.

With the equivalent of 68 ± 15 million people, storage-fed irrigation from existing dammed reservoirs on currently irrigated lands contributes most to food production in the Mississippi (Fig. 4A), because of the basin’s large existing storage and the large amount of food produced from its irrigated agriculture (Fig. 4C, green bar). The potential increase in food production from storage-fed irrigation is small and can be met from existing reservoirs (Fig. 4C, hatched part of purple bar).

Fig. 4.

Irrigation from dammed reservoirs could grow food for 631 ± 145 million people, significantly less than the total potential of storage-fed irrigation (1.15 billion). (A) Global distribution of food production of agriculture irrigated from existing dammed reservoirs. (B) Future global distribution of food production after expanding irrigation to currently rainfed land and irrigated from existing and future dammed reservoirs. (C) Current (green) and future (purple) food from storage-fed irrigation and the current and future contribution of reservoirs (hatched), thus highlighting trajectories for storage surpluses/deficits (for 30 basins where irrigation from dammed reservoirs could support most people). (D) Typical trajectories for how differential growth in irrigated agriculture and storage can lead to (un)realized potentials in food production from storage-fed agriculture (SI Appendix, Fig. S1 indicates location of highlighted basins).

In the Ganges-Brahmaputra, storage-fed irrigation on currently irrigated lands could grow food for more people than in the Mississippi (71 ± 6 million) (Fig. 4C, full green bar). However, because there is not enough reservoir storage, storage-fed irrigation currently only produces food for 17 ± 4 million people (Fig. 4C, hatched part of green bar). Future dams could increase food production from storage-fed irrigation (difference between hatched part of green bar and hatched part of red bar in Fig. 4C), yet food production with all future reservoirs is still much less than the biophysical potentials from storage-fed irrigation, shown by the full red bar in Fig. 4C.

These examples highlight that the contribution of storage-fed irrigation to food security depends on current storage infrastructure, the potential to expand irrigation, and the increase in irrigation storage (Fig. 4D). For example, foregone food production because of storage deficits will decrease where storage increases more than future storage-fed irrigation (Fig. 4D, 2). Storage deficits will increase where storage-fed irrigation could grow faster than reservoir storage (Fig. 4D, 3). Where both increase, the change in storage deficits will depend on the respective growth of storage and storage-fed irrigation (Fig. 4D, 4).

Discussion

Momentum is building around a vision for human development that emphasizes climate-resilient, nature-positive, and inclusive foundations, supported by meaningful commitments, standards, and scalable demonstrations (50). Feeding humanity is a fundamental, yet ever more challenging part of achieving this vision. Previous research has shown that this challenge will require concerted actions across sectors and scales to address both food demand and supply (10, 12).

Herein we highlight that unlocking the potentials of rainfed croplands with sustainable irrigation, a supply-side approach that is widely considered necessary for meeting future food demands (11, 12), would have important implications in terms of future water infrastructure. Our results highlight that sustainable water resources sufficient to feed additional 1.15 billion people can only be used if water storage is provided.

Irrigated agriculture fed from existing reservoirs can produce food for 345 ± 102 million people, a number that could nearly double to 631 ± 145 million with future dammed reservoirs. Thus, even if all identified future dams are built despite all associated socioeconomic and environmental challenges, and part of their storage is made available for irrigation, reservoir-based storage would only meet 60% of the biophysical potential for storage-fed agriculture. Our findings show that water storage for irrigation is and will continue to be a major driver of agricultural economic water scarcity (26, 29) and a potential hindrance to leverage sustainable blue water for food security.

Beyond Dammed Reservoirs to Supply Irrigation.

Current and future dammed reservoirs can be important for food security, but future food systems will require integrated solutions beyond building more dams (51). This is not only because of the major socioeconomic and environmental impacts of large dams and associated conveyance infrastructure (30, 41, 49, 52), but also because the potential for dammed reservoirs will not be sufficient to close future storage deficits. In many basins, blue water resources could not be fully used for storage-fed agriculture even if all potential dams were developed (Fig. 3B SI Appendix, Fig. S5A). For example, the storage of the many potential dams in, for example the Ganges-Brahmaputra or in several Sub-Saharan Africa basins, would not suffice to meet the storage-fed agriculture potentials even though those basins are global hotspots of potential dam development (Fig. 4C). In many developed countries, the absence of potential dam sites reflects a limited technical potential compounded with major societal challenges for new dammed reservoirs (30). Even if large dams could provide all storage needed for storage-fed irrigation, irrigation supplied by large dams might be neither sustainable nor economically feasible, even if the stored water fulfils quantitative sustainability criteria.

These findings point to the need to explore alternative storage options and use water allocations for irrigation more efficiently (53, 54). Large dams are probably the most common (31), but not the only scalable solution to support storage-fed irrigation. In fact, small decentralized reservoirs (32), improved agricultural practices (34), and nature-based solutions for improved infiltration and underground storage (55) provide multiple benefits and make more water available for crop production. On larger scales, managed aquifer recharge can be used to store significant amounts of water and provide multiple benefits, for example for irrigation and flood control (33). In basins with existing irrigation storage, maintenance of catchments and reservoirs, and thus reducing the amount of storage lost to sedimentation, is crucial to ensure that existing infrastructure can contribute to future storage-fed irrigation in the long-term (56).

Managing Water Demand to Reduce Water Storage Deficits.

In the past more storage often led to inefficient water use and increased hydrologic extremes (57, 58), rather than fueling an increase in crop productivity (35, 36), which highlights that any increase in irrigation, and associated storage, needs to be embedded in demand management to avoid rebound effects (59). For example, demand for storage depends on biophysical water demands of crops but also on how much of the releases from storage are available to crops on the field.

Increasing the efficiency of irrigation systems will not only increase how much crops can be grown per unit of water but will also minimize the need for water storage. However, many developing regions, where storage-fed irrigation could support most people and where the storage deficit is greatest, are also those with the lowest irrigation efficiencies (SI Appendix, Fig. S3). This highlights the need for joint investments in both storage and irrigation efficiency, embedded in sustainable and hydrologically informed policies (54).

Water storage and conveyance for irrigation will need to be considered on multiple scales. We performed our analysis on a basin scale, if political realities and conveyance infrastructure enabled conveyance of blue water to reservoirs and from reservoirs to irrigated areas within a subbasins. This is not always the case. For example, we found that potential dammed reservoirs could close the storage deficit in the Mekong (Fig. 4C). In reality, reservoirs in the upper Mekong reduce water availability when water is needed most for downstream irrigation because reservoirs are operated to maximize hydropower generation (60). Analyses on smaller scales might highlight potentials for harvesting of local water resources and managing soil moisture (12, 32), and thus local “soft paths” to store more water without more infrastructure (61). Our finding that storage deficits increase in some basins while storage surpluses increase in others indicates the need to study opportunities for regional changes in crop mixes and intensified cropping cycles to maximize the use of storage in basins with a storage surplus (62).

Future Challenges and Opportunities to Plan across Scales and Domains.

Our hydrologic analysis on a basin scale results in estimates for how much water would need to be stored to make use of the full potential of sustainable blue water for agriculture. Our estimates of current storage deficits for agriculture are likely conservative, as unmapped small dams and nature-based solutions already supplement water storage from large dammed reservoirs (31, 32). Our estimates of how much future dams can contribute to future irrigation are likely optimistic for five reasons. First, to estimate an upper limit for reservoir storage, we assume that potential hydropower dams are developed and that part of their storage can be allocated to irrigation. Yet, many dams of the 3,700 identified dam projects might never be developed because of socioeconomic and ecologic challenges. Second, if built, future dams might not contribute to irrigation. Third, domestic water demand, flood control, and balancing renewable energy grids will create additional demand for storage that would compete with irrigation (note these latter two uncertainties are addressed in our Monte Carlo analysis, which assumes that the allocation dam storage to irrigation can be very low, or zero) (Methods). Fourth, more detailed analysis of water needs for people and freshwater ecosystems might lead to reduced estimates of how much blue water can be withdrawn in a sustainable manner and how much water is lost from irrigation from reservoir storage because of deep percolation and evaporation. Last, our analysis assumes that water can be transported within river subbasins. Current irrigation dams are presumably connected to water conveyance infrastructure to supply inflows to dams and to transport stored water to irrigated croplands. For future dams and on a global scale, data are not sufficient to judge the techno-economic feasibility of future conveyance projects (39), another prerequisite to fully harness dammed reservoirs for irrigation. Indeed, water scarcity has already created a significant pipeline of water conveyance projects on the same or larger scales than what we assume in our analysis (49, 52). However, the storage gap might be even larger because conveyance between future dammed reservoirs and future irrigated croplands is not feasible (SI Appendix, Fig. S8 exemplifies the role of conveyance for parts of the Ganges and Tigris basins).

It could be argued that the identified future dam sites, mapped for their hydropower potential alone (44), can be augmented by additional dedicated irrigation dams. However, the development of future dams is limited by the availability of suitable sites, widespread opposition to major dam development (30), and economic challenges for developing irrigation infrastructure (37). Thus, our estimation of those future dams’ contribution to future irrigation should be considered an estimate of the technical potential for irrigation from dammed reservoirs, and not a real-world policy alternative.

From a hydrologic perspective, global warming will shift water supply and demand because of increasing temperatures and changing precipitation patterns (3, 27). This uncertainty is compounded by changes in crop mixes, increasing demand for biofuels (63), and changes in plant metabolism from higher CO₂ concentrations (3, 27), which will alter water demands and the amounts of calories from irrigated agriculture. Future population growth, changing per capita food demands and food trade, creates both uncertainty and opportunities for actively managing how much food will be needed to feed a growing population. While all of these aspects need to be part of a future research agenda, we herein focused on providing a baseline potential to understand the role of water storage to food security under current climate conditions, with existing crop mixes, and assuming that water can be allocated from dammed reservoirs with similar rates as today and redistributed on basin-scales to maximize food production. Future climate conditions, sustainable groundwater storage, multiyear water storage to balance interannual variability, and opportunities to modify crop mixes to reduce water storage needs provide a future agenda for research on infrastructure in the water–food nexus (see also SI Appendix, Supplemental Discussion).

Irrigation from stored water resources could play a crucial role to grow more food within the limits of sustainable land and water resources. However, provision of water storage and associated costs and externalities adds additional social, economic, and ecologic challenges for future food systems. Among all supply and demand side options to increase food and water security, building more dams should be the last resort, given the associated socioenvironmental consequences, but it might not be avoidable in all settings. Our global mapping of potentials for storage-fed irrigation supports research into locally appropriate storage solutions, highlights limits and socioeconomic and ecological trade-offs of a sustainable expansion of irrigation (43, 64), and advances an integrated vision of the role of future water infrastructure for water, energy and food systems.

Methods

We first determined sustainable blue water resources, Q, water demands for current irrigation, D, and future irrigation demands from expanding irrigation onto rainfed croplands, D_F, from global data with a with a 10-km resolution.

Let i denote a grid cell and m denote a month. Then Q(i, m) denotes the sustainable blue water in available cell and month. Q(i, m) is calculated as Q(i, m) = Q_tot(i, m) − Q_Env(i, m) − Q_I+D (i, m), where Q_tot denotes total blue water obtained from the Composite Runoff v1.0 data (65), Q_Env denotes the environmental flows derived using the variable monthly flow (VMF) method (66), and Q_I+D denotes industrial and domestic water demands (66).

Q(i, m) accounts for surface and shallow subsurface runoff, and thus the water that would eventually form river discharge that can be stored in surface reservoirs. Q(i, m) thus accounts for the fact that some water is stored in soils and shallow renewable groundwater from where it could be abstracted for instantaneous use: that is, for use in the month when it accrues. Remaining blue water would then either runoff to streams or feed streams via baseflow in month m or in a later month. Thus, the contribution of subsurface storage is considered, as long as there is no significant multiyear hold-over storage (65). As in other studies (67), we do not distinguish between abstractions from surface water and renewable groundwater because abstractions from sustainable groundwater would reduce baseflow to surface water bodies and it would make no quantitative difference if this water is withdrawn from sustainable groundwater or surface water (67).

We also do not account for multiyear storage in deep groundwater, as irrigation from deep groundwater is predominantly exceeding sustainable rates (25). Yet, we acknowledge that sustainably managed deep aquifers, possibly artificially recharged via managed aquifer recharge (33), can be a relevant alternative to storing water in surface reservoirs (SI Appendix, Supplemental Discussion).

The current monthly irrigation demand D(i, m) is defined as the volume of water needed for meeting the water demands of current crops on irrigated land [as delineated in the MIRCA2000 dataset (68)]. D(i, m) is calculated based on the evaporative demand of 126 crops (nearly 100% of all crops grown) listed in MIRCA2000. Based on MIRCA2000 rainfed lands, Rosa et al. (21, 26) also determined areas with current green water scarcity (insufficient annual precipitation to meet crop water demands) but no blue water scarcity (sufficient blue water to implement irrigation without depleting sustainable water resources).

In those areas, irrigation can be expanded to boost crop production on currently water-limited croplands without depleting environmental flows. For those pixels, we determined future irrigation water demand, D_F(i, m). Previous studies provide more details on the crop model (69) and validation with MODIS data and other global modeling studies (70). We assume that future irrigated agriculture will use the current crop mix and cropping practices. Because of the huge associated uncertainty, we do not consider that implementing irrigation might encourage switching to more water-intensive crops and more frequent cropping, which would feedback on water demands. That means we modeled irrigation of the same crop mix that is currently grown with rainfed practices in the same pixel. It should also be noted that all of our calculations are within quantitative limits of water availability; that is, all of our scenarios are feasible without depleting environmental flows or conflicting with domestic and industrial water uses.

It should be noted that Rosa et al. (26) analyzed the potential for a sustainable expansion of irrigation on a monthly scale. Yet, it is not known if there is the potential need to store water to reach that potential because of interannual mismatches between Q(i, m) and D(i, m). Those unknown needs for water storage, which would feed a sustainable irrigation expansion, constitute the research gap we address with our analyses.

Irrigation storage is often provided by infrastructure that provides water to much larger areas than a single 10 × 10-km pixel through water conveyance on basin or interbasin scales. We thus analyzed storage demands on the scale of hydrologic basins. We obtained outlines of major hydrologic basins from the Hydrosheds dataset (71). As a level of aggregation, we selected Order 4 (according to the Pfaffstetter scheme used by Hydrosheds), resulting in 1,331 basins with a median area of 58,000 km². This scale corresponds to river basins or major subbasins of large river basins (SI Appendix, Fig. S4), a scale on which we assume water conveyance is feasible. Note that this size matches the scale of current major irrigation schemes, for example in the Californian Central Valley, but is still small compared to many planned conveyance projects (52).

Let B denote a given hydrologic basin, then $Q\left(B,m\right)={{\displaystyle \sum }}_{i\in B}Q\left(i,m\right)$, $D\left(B,m\right)={{\displaystyle \sum }}_{i\in B}D\left(i,m\right)$ and $D_F\left(B,m\right)={{\displaystyle \sum }}_{i\in B}D\left(i,m\right)$: that is, all components of the water balance are calculated as sum over all cells in a hydrologic basin.

We then calculate the total water demand for sustainable irrigation in a basin B and for a month m from the irrigation demand on currently irrigated land and currently rainfed lands with a potential for sustainable irrigation as:

$$D_{\mathit{tot}}\left(B,m\right)= D\left(B,m\right)+ D_F\left(B,m\right).$$

Then, we define the mismatch between the sustainable water resources availability and the demand of irrigation on currently irrigated lands as

$$\mathrm{\text{Δ}}\left(B,m\right)= Q\left(B,m\right)- \text{D}\left(B,m\right).$$

We calculate the monthly water surplus, Δ⁺, as

$${\mathrm{\text{Δ}}}^{+}\left(B,m\right)=\text{max}(0,\mathrm{\text{Δ}}\left(B,m\right))$$

and, respectively, the water deficit, Δ⁻, as

$${\mathrm{\text{Δ}}}^{-}\left(B,m\right)=\min (0,\mathrm{\text{Δ}}\left(B,m\right)),$$

with annual values of

$${\mathrm{\text{Δ}}}^{+}\left(B\right)={\sum }_{m=1}^{12}{\mathrm{\text{Δ}}}^{+}\left(B,m\right)$$

and

$${\mathrm{\text{Δ}}}^{-}\left(B\right)={\sum }_{m=1}^{12}{\mathrm{\text{Δ}}}^{-}\left(B,m\right).$$

The above calculations can be repeated for future expansion of irrigation on rainfed land, substituting D(B, m) with D_tot(B, m). The resulting indicators of total future water deficit and surplus are then ${\mathrm{\text{Δ}}}_{\text{tot}}^{+}\left(B,m\right), {\mathrm{\text{Δ}}}_{\text{tot}}^{+}\left(B,m\right)$ on a monthly scale and ${\mathrm{\text{Δ}}}_{\text{tot}}^{+}\left(B\right)$ and ${\mathrm{\text{Δ}}}_{\text{tot}}^{-}\left(B\right)$ on an annual scale.

The water surplus, Δ⁺(B), can be stored to supplement irrigation in months with a water deficit. The potential for this storage-fed irrigation, SFI, is calculated as

$$\mathit{SFI}\left(B\right)=\text{min}(\left|{\mathrm{\text{Δ}}}^{-}\left(B\right)\right|,{\mathrm{\text{Δ}}}^{+}\left(B\right)).$$

Thus, if the absolute water deficit in a basin over all months is smaller than the sustainable water surplus, only the water needed to meet that supply needs to be stored. If the water deficit is larger than the surplus of sustainable water, then that only surplus can be stored, but storage-fed irrigation will be constrained by the availability of sustainable water resources.

This calculation can be repeated accounting for the increased water demands from expanding irrigation to rainfed lands, so that the total potential for future storage-fed irrigation (SFI_tot) read as

$$\mathrm{SF}{I}_{\mathit{tot}}\left(B\right)=\text{min}(\left|{\mathrm{\text{Δ}}}_{\text{tot}}^{-}\left(B\right)\right|,{\mathrm{\text{Δ}}}_{\mathit{tot}}^{+}\left(B\right)).$$

The calculation of these indicators is shown in SI Appendix, Box S1 for the Ganges Basin.

It should be noted that this aggregation and the herein presented analysis can be scaled to smaller (e.g., subbasins, countries) or larger (e.g., regional or continental) scales. It should also be noted that we produce figures and report numbers based on further aggregating results from Pfaffstetter Order 4 basins to main river basins (e.g., the whole Mississippi or the whole Ganges-Brahmaputra) for better geographic referencing. Yet, all indicators for Pfaffstetter Order 4 basins are available from the repository associated with this paper.

Water Availability from Dammed Reservoirs.

The amount of water from a dammed reservoir R that reaches crops on the field, V_irri(R), can be estimated as

$$V_{\mathit{irri}}\left(R\right)={ϵ}_{\mathit{irri}}\left(R\right)*f_{\mathit{irri}}(R)*S(R),$$

where ${ϵ}_{\mathit{irri}}$ is the irrigation efficiency, f_irri is the part of total storage that can be allocated for irrigation, and S is the total storage volume of the dammed reservoir. For current conditions, we consider only dams from Lehner et al. (47) that list irrigation storage as purpose. For a future dammed reservoir, Grill et al. (72) proposed a linear regression to estimate storage volumes

$$S_f\left(R\right)=a*I (R),$$

where S_f is the storage volume of a future dams (in million cubic meters) and I is the installed capacity (in MW), tabulated for each dam site (44). Grill et al. (72) proposed a value of a = 3.19 based on an analysis of more than 200 dams. Thus, S(R) will be substituted by S_f(R) for future dammed reservoirs.

Then the total water availability from reservoirs for basin B is

$$V_{\mathit{irri}}\left(B\right)= {\sum }_{R\hspace{0.17em}\mathit{located}\hspace{0.17em}\mathrm{in}\hspace{0.17em}B}{V}_{\mathit{irri}}\left(R\right)$$

(i.e., the sum over all dammed reservoirs in a hydrologic basin B). Note that this can be calculated for current dammed reservoirs, for current and future dammed reservoirs (i.e., future total storage), and for future dammed reservoirs only (incremental storage). Finally, the storage deficit G(B) is calculated as difference between SFI potential and water from dams available on the field

$$G\left(B\right)=V_{\mathit{irri}}\left(B\right)-\mathit{SFI}(B).$$

Each of the factors required to calculate V_irri(R) for a dammed reservoir is subject to major uncertainty and such is V_irri(B) and our estimates of the storage deficit. To estimate the uncertainty in the storage deficit, we performed a Monte Carlo analysis for each current and future dammed reservoir. For each dammed reservoir R, we calculated V_irri(R) over 100,000 runs. In each run, we randomly drew values for ${ϵ}_{\mathit{irri}}, f_{\mathit{irri}}$, and $a$ ($a$ only for future dams) according to parameter-specific probability distributions we estimated from previous studies and datasets (SI Appendix, Supplemental Methods).

The mean value of V_irri(R) over 100,000 runs can be estimated as

$$\overline{V_{\mathit{irri}}\left(R\right)}= \frac{{{\displaystyle \sum }}_{n=1}^{{10}^5}{V}_{\mathit{irri}, n}\left(R\right)}{{10}^5},$$

with a variance of

$$\mathit{VAR}\left(V_{\mathit{irri}}\left(R\right)\right)=\frac{{{\displaystyle \sum }}_{n=1}^{{10}^5}|V_{\mathit{irri}, n}\left(R\right)-\overline{V_{\mathit{irri}}\left(R\right)}{|}^2 }{{10}^5}$$

and an SD of

$$\mathrm{SD}\left(V_{\mathit{irri}}\left(R\right)\right)=\sqrt{\mathit{VAR}\left(V_{\mathit{irri}}\left(R\right)\right)}.$$

On a river basin level, we then estimate the average value of V_irri(B) as

$$\overline{V_{\mathit{irri}}\left(B\right)}= {\sum }_{D\hspace{0.17em}\mathit{located}\hspace{0.17em}\mathrm{in}\hspace{0.17em}B}\overline{V_{\mathit{irri}}\left(R\right)},$$

with a SD calculated from the sum of dam-level variance

$$\mathrm{SD}\left(V_{\mathit{irri}}\left(R\right)\right)=\sqrt{{\sum }_{R\hspace{0.17em}\mathit{located}\hspace{0.17em}\mathrm{in}\hspace{0.17em}B}\mathrm{VAR}\left(V_{\mathit{irri}}\left(R\right)\right)}.$$

Contribution of Stored Water to Food Security.

Similar to previous studies (26), we assume that crop yields and the associated calorie production scale linearly with the fraction of crop water needs that are met. Thus, the full yields of the pixel-specific crop mix and the associated calorie production are attained if all crop water demands are met. If half of the annual crop water demands in a pixel are met from storage-fed irrigation, we attribute half of the maximum attainable calories to storage-fed irrigation. If reservoirs supply 10% of the full crop water demand in a basin, 10% of total yields are attributed to reservoir storage. Because of different crop mixes in different pixels, and thus different water demands and calorie contents, a unit of stored water will yield different nutritional values in different locations. To estimate how many people can be supported by the produced calories, we assume an average 3,343 vegetal kcal per capita per person per day, accounting for direct human consumption, feed for livestock production, and food waste (21, 26).

To estimate the contribution of existing and potential reservoirs to agriculture, we assumed that all water from dams is first allocated to existing irrigated agriculture until the potential for storage-fed irrigation on currently irrigated land was met. Only after that is the remainder of V_irri allocated to irrigation of currently rainfed croplands. From that analysis, and assuming the above-described relationship between water availability and calories, we calculated how much food could be grown from water provided by current and prospective dammed reservoirs (Fig. 4). Based on the estimated uncertainty in V_irri and V_irri future, we also estimated upper/lower bounds for how many people can be supported by dams.

Data, Materials, and Software Availability

Scripts to perform the analysis (all in R language), required input data, and resulting output datasets and figures are publicly available from https://zenodo.org/record/5932693#.Yfgww-rMIQ8 (73).