Main

Agriculture is the largest water user worldwide, both in terms of blue and green water1,2,3,4. Blue water refers to water in rivers, lakes, wetlands and aquifers. Green water is the soil water held in the unsaturated zone, originating from precipitation and eventually evaporating through and from plants and soils5,6,7. Irrigated agriculture receives blue water (from irrigation) and green water (from precipitation), whereas rain-fed agriculture receives only green water. Other sectors such as domestic water supply, industrial and power plants use only blue water resources. In international reporting and statistics, such as in Aquastat of the Food and Agriculture Organization of the United Nations (FAO)8, blue water is the main component of interest9.

Freshwater is a limited resource and its use by different sectors contributes to water scarcity in many places around the world9. Assessments of recent sectoral water use at high spatial detail, including by crops, are essential to evaluate and monitor global to local water stress. They are also essential for informed water management decisions and agricultural practices. Billions of people live under water stress conditions3,10, characterized by intense competition for water between different sectors, reduced ecosystem services for humans and contributing to a fast decline in aquatic biodiversity during the last decade11,12. Due to a growing global population and changes in lifestyles including dietary changes, global crop production has steadily increased over the past decades. Monitoring agricultural water use is crucial to address these challenges and ensure sustainable water management.

Few spatially distributed global crop water use modelling assessments, which provide their results open access, have been conducted recently. Mekonnen and Hoekstra1 provided an influential study on the global blue and green water use of different crops representative for the year 2000 (multi-year average of 1996–2005), in a spatially distributed way. Since then, more recent spatially distributed global assessments have been conducted, such as the work by Chiarelli et al.13 and Mialyk et al.14,15, but detailed assessments for the current decade are lacking. Our analysis provides a detailed assessment for the year 2020, by using the most recent SPAM (Spatial Production Allocation Model) crop harvested area and production data (10 × 10 km grid-cell resolution) (called SPAM2020 data), released by the International Food Policy Research Institute (IFPRI) in 2024 (ref. 16).

We compute blue and green water consumption (WC) for the 46 SPAM crops and major crop groupings for the year 2020 (Extended Data Table 1) in high spatial resolution (5 arcminutes—approximately 10 km at the equator). Our global gridded model, which we call CropGBWater, is provided as open-access Pythton script and runs with a daily time step. We use only open-access input data. We run the model with climate data from 2018 to 2022, representing the year 2020 as an average. We used this multi-year average approach to level out climatological fluctuations, similar to Mekonnen and Hoekstra1. We aggregate daily blue and green WC results per crop to annual and monthly amounts, which we provide open access. More particularly, we provide spatially distributed crop WC results in mm and in m3, with the distinction between irrigated blue (WCbl), irrigated green (WCgn,irr) and rainfed green (WCgn,rf) water.

We run the model again with SPAM data for the years 2000 (called SPAM2000 data) and 2010 (SPAM2010 data) and 5-year daily climate data from 1998 to 2002 and 2008 to 2012, respectively, representing the year 2000 and 2010 as an average, respectively. We do this to evaluate how global crop WC amounts have changed in 20 years. Our study thereby provides insights into changes in the globally largest water user, up to the current decade. The only other study that assesses evolution in global WC in high-spatial resolution, up to the year 2019 (based on SPAM2010 data), is Mialyk et al.14,15.

We advance on previous studies such as Chiarelli et al.13 and Mialyk et al.14,15 in different ways:

  • We make an assessment for the year 2020, based upon the most recent SPAM2020 data. These SPAM data are the most detailed spatially distributed crop production data for the year 2020 available. They are based on FAO statistics and country (sub)national statistics, making them aligned with official statistics.

  • We conducted our core modelling at a much finer spatial resolution than Mialyk et al.14,15. Our core modelling of evapotranspiration (in mm) is done at a spatial resolution of 5 arcminutes (0.0833°). Mialyk et al.14,15 conducted their core modelling at a spatial resolution of 0.5° (30 arcminutes), a much coarser resolution. Our core modelling thus contains 36 grid cells for each grid cell of Mialyk et al.14,15.

  • Our model does not simulate crop growth, in contrast to Mialyk et al.14,15, but it balances the quality of output with minimizing data input and computational resources. This ensures robust yet relatively fast results without unnecessary complexity for users. We thus developed a relatively simple yet robust model, thereby providing a potential powerful open-access tool for capacity building in the global south.

  • We use only open-access input data and the open-access software Python. Chiarelli et al.13, for example, developed their WATNEEDS script in MATLAB, a software package that is not open access and therefore much less accessible for many stakeholders

  • Our model can be used by any stakeholder with basic modelling skills, including in the global south, due to the use of open-access software and data

  • We used an improved green–blue water distinction. In Chiarelli et al.13, for example, the distinction between green and blue water use was achieved by employing two separate water balance settings: first, under rain-fed conditions, categorizing all water consumption as green, and then under rainfall with irrigation conditions, assuming its total water consumption encompasses both green and blue components. In our study, we improve the green–blue water distinction by tracking the stocks and flows of soil water within the root zone on a daily time step, systematically differentiating them as ‘green’ or ‘blue’, according to Chukalla et al.17 and Hoekstra18.

Results

Crop water consumption in 2020

Global annual blue WC (WCbl) amounts to 1,080 km3, increasing to 1,228 km3 when the flooding phase of paddy rice is considered (Fig. 1 and Table 1). Global green WC (WCgn) amounts to 5,588 km3, resulting in a total crop blue plus green WC (WCbl+gn) of 6,668 km3 or 6,817 km3 when paddy flooding is included.

Fig. 1: Total annual crop blue and green WC for 2020.
figure 1

a, Spatial distribution of blue WC in 103 m3 (map includes paddy WC). b, Blue WC for main crops and per continent in km3. c, Spatial distribution of green WC in 103 m3. d, Green WC for main crops and per continent in km3. Maps in a and c created with ArcGis using country/continent borders from GADM (https://gadm.org/data.html).

Full size image
Table 1 Crop blue and green WC amounts for main crops and continents for 2020 and related SPAM statistics on production and harvested area
Full size table

Both blue and green WC show a heterogeneous spatial distribution across the globe (Fig. 1, Table 1 and Supplementary Data 1 and 2). Asia accounts for the highest total blue (782 km3 or 918 km3 including paddy) and green WC (2,606 km3) amounts. In decreasing order of total WCbl and WCgn separately, the Americas account for 135 km3 (140 km3 including paddy) and 1,321 km3, respectively, whereas Africa accounts for 112 km3 (119 km3 including paddy) and 1,087 km3, respectively. Europe (excluding Russia) accounts for 39 km3 (39 km3 including paddy) and 355 km3, Oceania for 7 km3 and 75 km3 and Russia (including its European part) for 6 km3 and 144 km3, respectively. For WCbl+gn, the amounts are 3,388 km3 (3,524 km3 including paddy) for Asia, 1,455 km3 (1,460 km3 including paddy) for the Americas, 1,199 km3 (1,206 km3 including paddy) for Africa, 394 km3 for Europe, 150 km3 for Russia and 82 km3 for Oceania.

The countries with the highest total WCbl (without paddy), in decreasing order, include China (244 km3), India (234 km3), Pakistan (98 km3), the USA (68 km3), Egypt (50 km3), Iran (39 km3), Mexico (23 km3) and Indonesia (19 km3) (Supplementary Data 2).

The five crops with the highest global WCbl are in decreasing order rice (328 km3 or 477 km3 including paddy), wheat (212 km3), maize (96 km3), sugar cane (96 km3) and cotton (70 km3) (Fig. 1, Table 1 and Supplementary Data 1). Combined they account for 74% of the total WCbl (or 77% when including paddy). For WCgn, these are maize (764 km3), rice (707 km3), wheat (553 km3), soybean (505 km3) and oil palm (302 km3). The largest amounts for WCbl+gn are rice (1,035 km3 or 1,183 km3 including paddy), maize (860 km3), wheat (765 km3), soybeans (521 km3) and oil palm (303 km3).

In many locations, one of the five main crops with the highest global WCbl is the dominant blue water user (Fig. 2). Rice is the dominant water user in many regions of Asia, including China, Japan, South East Asia and India. Also in certain parts of the Americas and Africa, rice is the dominant water user. Wheat is the dominant water user in large parts of northeast China, northern India and Pakistan, Central Asia and the Middle East, along the Nile River and in parts of the USA and Mexico. Maize is the dominant water user in much of the American Midwest, parts of Mexico and southern Brazil, the Mediterranean and parts of northeast China. Sugar cane is the dominant water user in parts of the Caribbean and Brazil and in large parts of Pakistan and India. Cotton is the dominant water user in parts of the southern USA, large parts of Central Asia and in the northeast of Africa in the Nile basin.

Fig. 2: Grid cell information on the blue WC.
figure 2

a, Dominant crop for WCbl within a grid cell. For five main crops (wheat, sugar cane, maize, rice, cotton), we identify the crop with the largest blue WC amount within a grid cell (restriction to grid cells where blue WC ≥ 1,000 m3). b, Seasonality of WCbl within a grid cell, shown by month with highest WC amount. The month with the largest blue WC amount within a grid cell is identified (restriction to grid cells where blue WC ≥ 1,000 m3, exclusion of paddy phase of rice). The inset graph shows global monthly WCbl amounts for all 46 crops in km3. Maps in a and b created with ArcGis using country/continent borders from GADM (https://gadm.org/data.html).

Full size image

Monthly WCbl amounts vary considerably, depending on the month of the year and the location (Fig. 2). Global monthly amounts increase steadily from January (33 km3) up to a peak in July (166 km3) and decrease steadily up to December (48 km3). The highest monthly amounts in the Northern Hemisphere are generally during the late spring and summer months of that hemisphere (May–August). In Asia, especially on the Indian subcontinent, this period is sooner (March–April), before the onset of the monsoon. Along the Equator, monthly highs are during different periods of the year. In the Southern Hemisphere, such as the south of Brazil and Argentina, monthly highs are during the months of December and January.

Crop water consumption change from 2000 to 2020

We compare results for SPAM2020 with the results for SPAM2010 and SPAM2000, representing the years 2010 and 2000, respectively. We find that WCbl+gn in 2020 has increased by 9% from 6,270 km3 in 2010 to 6,817 km3 (including paddy) (Fig. 3a and Supplementary Data 1). Blue WC has increased from 1,043 km3 to 1,080 km3 and WCpaddy from 138 to 149 km3, resulting in a total blue WC increase of 4%. Green WC has increased by 10% from 5,090 km3 (WCgn,irr 804 km3 plus WCgn,rf 4,727 km3). Total WC amounts have thus increased substantially during the period 2010–2020.

Fig. 3: Comparison of total blue and green WC in km3 for SPAM2020, SPAM2010 and SPAM2000.
figure 3

a, WC by different components (WCbl, WCpaddy, WCgn,irr and WCgn,rf); SPAM2000 excludes ‘other crops’. b, Total harvested area (million ha or Mha) with differentiation between irrigated and rain-fed area. c, Crop WCbl+gn for SPAM 2010 and SPAM2020; R2 correlation coefficient; red line is the 1:1 gradient. d, Crop WCbl+gn for SPAM 2000 and SPAM2020. All data in Supplementary Data 1.

Source data

Full size image

For SPAM2000 we compute a WCbl+gn of 4,897 km3 (Fig. 3a). This amount, however, is not directly comparable with the SPAM2020 and SPAM2010 total amounts, because in SPAM2000 the crop ‘othe–Other crops’ contains no data. The total amount is thus an underestimation. Blue WC is 932 km3, WCpaddy 123 km3 and green WC 3,842 km3 (WCgn,irr 727 km3 plus WCgn,rf 3,115 km3).

The total harvested area increased from 1,184 million hectares (Mha) in the year 2000 (of which 165 Mha ‘other crops’) to 1,301 Mha in 2010 (+10%) and to 1,407 Mha (+19%) in 2020 (Fig. 3b). Irrigated harvested area increased from 278 in 2010 to 292 Mha in 2020 wheras rain-fed area increased from 1,023 Mha to 1,115 Mha. The expansion of harvested area is an important driver for the increased WC amounts we compute.

A comparison for individual crop WCbl+gn (in km3) shows high correlation values between SPAM2020 and the results for SPAM2010 (correlation R2 = 0.9824; Fig. 3c) and SPAM2000 (correlation R2 = 0.9509; Fig. 3d). This shows a high degree of consistency between the three modelled datasets. WCgn+bl crop amounts tend to be higher for SPAM2020 compared to SPAM2010 and SPAM2000. Examples are rice (1,035 versus 996 and 963 km3 without paddy or 1,183 versus 1,134 and 1,087 km3 with paddy), maize (860 versus 704 and 593 km3), soybeans (521 versus 443 and 325 km3), sugar cane (292 versus 282 and 238 km3) and oilpalm (303 versus 179 km3). Regarding WCbl, higher amounts are observed for 24 out of 42 SPAM2010 crops. This includes most crops with high WCbl values. Rice, wheat, sugarcane, maize and cotton account for more than two-thirds of the total WCbl amount for all three years.

Maize shows a global increase in WCbl+gn of 45% from 2000 to 2020 (harvested area increase +44%; Supplementary Data 1), with large increases in specific countries (Fig. 4a). This includes the USA (+19%), China (+74%), Brazil (+70%), India (+44%), Nigeria (+17%), Argentina (+139%), Indonesia (+27%), Tanzania (+393%), Angola (+322%) and Ethiopia (+51%). Also the WCbl of maize has gradually increased from 63,293 Mm3 in 2000 to 82,792 Mm3 in 2010 and 95,758 Mm3 in 2020. Increased production of maize during the past decades is referred to as the ‘maize boom’ and has taken place around the globe19. Large increases in maize production and related WCbl+gn are driven by an increasing consumption of maize for food (specifically in Africa), for feed (specifically in the Americas, Europe and Asia) and biofuels20. Global trade in maize has risen significantly20.

Fig. 4: Comparison of WCbl+gn in Mm3 for SPAM2020, SPAM2010 and SPAM2000 specific crops, with indication of main countries.
figure 4

a, Maize. b, Soybeans. c, Rice (with paddy). d, Sugarcane. e, Groundnuts. f, Cassava. REST, remaining countries; RSA, Republic of South Africa; DRC, Democratic Republic of the Congo.

Source data

Full size image

Soybeans show a global increase in WCbl+gn of 60% from 2000 to 2020 (Fig. 4b), with large increases in the USA (+24%), Brazil (+161%), Argentina (+67%), India (+67%), Paraguay (+152%), Canada (+94%), Bolivia (+133%), Uruguay (+8,984%) and the Republic of South Africa (RSA) (+524%). Harvested area increased 63% (Supplementary Data 1). Similar to maize, increased food, biofuel and especially animal feed requirements around the world have largely driven the increase in soybean production, expansion in harvested land and global trade21,22.

Rice shows an overall global increase in WCbl+gn of 9% (Fig. 4c), with the two largest producers India and China showing only smaller increases (6% and 2%, respectively). Larger increases are observed in Thailand (+11%), Bangladesh and Vietnam (each +13%), Myanmar (+19%), Pakistan (+25%), the Philippines (+24%) and Nigeria (+112%). Harvested area increased 7%. Also the WCblue increases consistently, from 295,638 Mm3 (419,085 Mm3 with paddy) in 2000, to 303,474 Mm3 (441,418 Mm3 with paddy) in 2010 and 328,229 Mm3 (476,849 Mm3 with paddy) in 2020.

Sugarcane and groundnuts show large global increases in WCbl+gn of 23% (Fig. 4d) (increase harvested area 37%) and 27% (Fig. 4e) (increase harvested area 36%), respectively. We compute specifically large increases in Brazil (+72%) and Thailand (+58%) for sugarcane and Sudan (+194%), Nigeria (+121%), Myanmar (+141%) and Chad (+79%) for groundnuts.

A very high relative global increase in WCbl+gn is observed for cassava (+82%; Fig. 4e). Due to its importance as a food security crop in western, central and eastern Africa23, large increases occur in this region, such as Nigeria (+244%), the Democratic Republic of the Congo (+157%), Uganda (+158%), Tanzania (+51%), Angola (+132%) and Ghana (+53%). Global harvested area increased 76% (Supplementary Data 1).Traditionally a famine reserve and a subsistence crop, the status of cassava is now evolving fast as a cash crop and as raw material in the production of starch (and starch based products), energy (bio-ethanol) and livestock feed in the major producing countries23.

These (spatial) changes in WC have an impact on water management in river basins. We relate the total blue WC for SPAM2010 and SPAM2020 to renewable (blue) water availability in major river basins of the world (Fig. 5). For many river basins, the relation between WCbl and water availability increases for SPAM2020 compared to SPAM2010, indicating additional pressure on water resources. These include the Indus (from 106 to 130%), the Ganges (111 to 121%), the Yellow (48 to 51%), the Colorado in the USA (38 to 46%), the Murray in Australia (10 to 26%), the Nile (22 to 24%), the Yangtze (19 to 23%), the Mekong (8 to 15%), the Niger (7 to 14%), the Danube (5.5 To 5.8%) and the Zambezi (0.8 to 0.9%) river basins. In some river basins no change is computed, such as in the Congo River basin (0.2 to 0.2%). In a few river basins, we compute a reduction, such as in the Mississippi (23.8 to 23.6%) or the Orange River basin in southern Africa (10 to 7%). For all river basins combined, we compute an increase from 9 to 11%.

Fig. 5: Blue WC as percentage to renewable water availability for main river basins.
figure 5

a, SPAM2020. b, SPAM2010. For selected basins, percentage values are given in bars (upper bar for SPAM2010, lower bar for SPAM 2020). Red amounts (larger than 100) indicate that environmental flows are violated. Maps in a and b created with ArcGis using country/continent borders from GADM (https://gadm.org/data.html).

Source data

Full size image

Discussion

We here provide a spatially explicit crop blue and green WC assessment for the year 2020, comparing it to the years 2010 and 2000. We find that total WCbl+gn has increased by 9% from 2010 to 2020. For many crops, WC in terms of m3 have increased. Total harvested area has also increased, by 19% since 2000, primarily due to an increase in rain-fed areas whereas the increase in irrigated area is smaller. The expansion of harvested area is an important driver for the increased WC amounts. This expansion has also led to deforestation, especially in the global south21,22,24,25. Additionally, intensification on existing lands, due to factors such as increased nutrient application or multiple cropping practices, will have an important influence. This highlights the growing demand for water resources in agriculture, emphasizing the need for sustainable water-management practices to support both rain-fed and irrigated agriculture under increasing pressure from global food demand.

Increasing blue and green WC for crops puts additional pressure on available blue and green water resources. The increases in 20 years’ time we compute are substantial, especially as previous global to local water stress assessments have shown that water stress was already abundant in the past3,6,10,26,27. We compute substantial WC increases for crops such as maize, soybeans, groundnuts, sugarcane and cassava, driven by increased food, feed and biofuel demands19,20,21,22,23. This increased WC puts additional pressure on already water-stressed river basins such as the Indus, Ganges or Yellow river basins, which provide food security for many hundreds of millions of their inhabitants.

Our crop WC assessment for the decade of the 2020s provides essential information for updating global to local water-stress assessments, for informing general water management and agricultural strategies and policies. Our assessment shows that due to increases in crop WC, competition for limited water resources between different sectors will only increase. Competition with other sectors such as municipal water use, industrial water use, energy water use28,29 and water for mining is set to grow with increasing water demands, depending on geographical setting, time of the year and regional water-management practices. Furthermore, water for ecosystems and their services needs to be accounted for11,12,30.

Water is essential for food security, yet increasing WC for crops puts food security at risk. Solutions to deal with this water–food nexus include producing more food with less water (‘more crop per drop’)31,32,33, the cultivation of less water demanding crops in water-stressed river basins32,34,35, the shift to more plant-based diets among affluent populations36,37 and the reduction of food losses and waste along the supply chain38,39. Also essential is integrated water management, including solutions in other water demand sectors, within a wider nexus approach addressing trade-offs and synergies40,41,42.

Our model does not simulate crop growth, but it balances the quality of output with minimizing data input and computational resources. This ensures robust results without unnecessary complexity for users. Our model can thus be used by any stakeholder, using open-access input data in the open access software Python. This is especially important in the global south, where stakeholders often lack the funds to purchase input data or specific software. This approach is consistent with the mission of Consortium of International Agricultural Research Centers (CGIAR). Our model can be used by any stakeholder with basic modelling skills. Our analysis here offers high spatial resolution of the core modelling, the use of the most recent SPAM2020 data, the use of open-access data and software, the balance between simplicity and robustness and the provision of all our results and software open access.

We provide the Python script of our model CropGBWater (supplementary data 10 in ref. 43) and all monthly and annual gridded outputs (in mm and m3) for all 46 crops open access (supplementary data 4–7 for the year 2020 in ref. 43). Additionally, we provide annual gridded outputs (in mm and m3) for all 42 SPAM2010 crops (supplementary data 8 in ref. 43) and 20 SPAM2000 crops (supplementary data 9 in ref. 43) open access.

Comparison with other studies and limitations

A detailed discussion on the comparison of our results with other studies and a discussion on limitations can be found in the Supplementary Information. In the Supplementary Information, we compare our results with the studies of Mekonnen and Hoekstra1, Mialyk et al.14,15 and Chiarelli et al.13, including for selected crops on the grid level. We find:

  • A high correlation between the results of Mialyk et al. and our study for WCbl+gn for individual crops for the year 2010, whereas for WCbl, on average, the crops in our study show higher amounts, leading to a much higher total WCbl for all crops combined (1,043 km3 compared to 768 km3 in the year 2010). This is in line with the discussion of their results by Mialyk et al., who stated that their total and individual crop WC are consistently smaller compared to four global studies for the year 2000.

  • A high correlation in individual crop WC amounts for SPAM2000 between our study and Chiarelli et al.13 and Mekonnen and Hoekstra1. Our CropGBWater model results in, on average, slightly lower individual crop total WC amounts but, in general, with slightly higher blue WC amounts, due to the new blue–green water partitioning module in CropGBWater.

Differences in modelling results are the result of multiple reasons, including (but not limited to):

  • The improved green–blue water distinction used in our study (Extended Data Figs. 1 and 2) and in Mialyk et al.14,15 but not in Mekonnen and Hoekstra1 or Chiarelli et al.13 do not.

  • Different input data, such as crop distribution and harvested area, climate data or soil data (Extended Data Table 2).

  • The use of different crop calendars. Using a fixed crop calendar for different periods (decades) is not fully correct as crop varieties, farmers adaptation and change in the rainfall patterns could influence farmers’ decisions on when to plant and harvest. Any crop calendar choice is therefore an approximation of reality. Different crop calendar databases44,45,46,47,48,49,50,51,52 exist, with each database having its own strengths, limitations and uncertainties. Extended Data Table 3 shows the data sources we used. We add an additional SPAM2020 run for rice with a different crop calendar, that is, of the RiceAtlas49 (grid results added as supplementary data 11 in ref. 43). We show that the total WC for SPAM2020 rice modelled with the crop calendar of the rice atlas amounts to 852 km3, of which 339 km3 WCbl (Supplementary Fig. 10). The total amount of our study—1,035 km3—is thus 18% higher than the amount calculated with the rice atlas calendar. The WCbl amount is quite similar, that is, 328 km3 (Table 1). This shows that the choice of crop calendar can make quite some difference with respect to the final WC amounts. Using recent spatially detailed crop calendar databases is therefore recommendable, such as the RiceAtlas49 or the recent MIRCA-OS database of Kebede et al.53.

  • The inclusion of capillary rise (Mialyk et al.14,15) or not (our study) also influences final results. We chose not to include this factor to reduce data requirements of our model.

  • We assume full irrigation, that is, that the irrigation is sufficient to meet the irrigation requirement, an assumption made in many global studies1,13. However, in reality, farmers may supply irrigation water below the required amount, in particular, in those regions where water is scarce. The assumption of full irrigation may lead to an overestimation of the blue WC.

In Supplementary Information Section 2, we further discuss certain limitations inherent to any modelling study. Model results can also be compared with remote sensing products54, that is, for total crop WC amounts, but we did not conduct such an analysis.

Methods

Model CropGBWater overview

We developed a Python-based global gridded crop green and blue WC assessment tool, which we call CropGBWater, building upon the earlier established work by Mekonnen and Hoekstra1. The latter model has not been published and was therefore never freely available for the scientific community. Operating on a daily timescale, CropGBWater dynamically simulates root-zone water balance and related fluxes, such as actual crop WC under suboptimal water availability, by incorporating methodologies and assumptions outlined by Allen, Pereira, Raes and Smith55.

The tool tracks daily water balance fluxes, including incoming (precipitation, irrigation, capillary rise) and outgoing (surface runoff, percolation, evapotranspiration) water fluxes, and soil moisture stocks to determine the actual green and blue WC using the shadow water balance method described in Chukalla et al.17 and Hoekstra18. Applicable at global scale, we applied the model with input data at a 5-arcminute resolution (0.0833° or roughly 9.2 km around the Equator) (Extended Data Table 1). As climate input data, we used daily values for the 5-year period 2018–2022, with a resolution of 0.1° (Extended Data Table 1). The model runs at 5 arcminutes. Outputs are generated at 5-arcminute resolution, providing results on both an annual and monthly basis for green and blue WC. Supplementary data 16 in ref. 43 provides for SPAM2020 the Climate-SPAM-grid that links climate grid cells (resolution 0.1°) to SPAM grid cells (resolution 0.0833°). Volumetric WC amounts are obtained from modelled mm amounts by means of SPAM harvested area. An overview of the framework is given in Extended Data Fig. 1. An overview of root-zone soil water stock and fluxes and its green and blue fraction is provided in Extended Data Fig. 2.

The calculation at each grid involves five key steps:

  • Step 1 establishes initial green and blue soil moisture levels for time t − 1 and determines the transpiration reduction factor (Ks).

  • In Step 2, actual evapotranspiration (ETa) and other root-zone water fluxes for time t, including runoff and deep percolation, are computed. In addition, irrigation requirement (IR) is calculated and crop water requirement (CWR) are determined.

  • Step 3 determines the actual green and blue ETa at time t using the composition ratio of green and blue water at time t − 1.

  • In Step 4, root-zone water stocks, including their green and blue components, are updated, along with the transpiration reduction factor for the current time increment, and this process iterates through Steps 1 to 4 for successive timesteps.

  • In the final Step 5, monthly and annual crop actual green and blue WC, CWR and IR are calculated from daily values.

Model CropGBWater different steps

Step 1: establish the initial green and blue soil moisture status for time t − 1 and compute the transpiration reduction factor (Ks)

Assume the initial soil moisture storage (St−1) and its green (Sg t−1) and blue (Sb t−1) components

Assumption: at t = 0, the soil moisture content (S)(t − 1) = Sg is at field capacity (FC) and is fully green, that is, S(t − 1) = Sg = Smax = RAW and Sb = 0. Thus, the ratio of Sg/S = 1, Sb/S = 0 and soil moisture depletion Dr(t − 1) = 0 as the soil moisture is near field capacity (FC). Normally sowing time is at the onset of rainfall and thus soil moisture level in the root zone near FC.

Establish the green moisture fraction (Cg) and the blue moisture fraction (Cb):

$${C}_\mathrm{g}\left(t-1\right)={\left(\frac{{S}_\mathrm{g}}{S}\right)}_{t-1}$$
$${C}_\mathrm{b}\left(t-1\right)={\left(\frac{{S}_\mathrm{b}}{S}\right)}_{t-1}$$

Determine the Ks(t − 1): read the corresponding transpiration reduction factor (Ks(t − 1) for the soil moisture content vs Ks curve).

Step 2: calculate the actual evapotranspiration (ETa) and other root-zone water fluxes (such as runoff and deep percolation) for time t.

Irrigation: irrigation requirements are established by assessing root-zone depletion. When the root-zone depletion reaches or exceeds the level of readily available soil moisture (RAW), irrigation becomes necessary and the irrigation amount matches the level of depletion, as indicated below.

$$\mathrm{IR}(t)=\left\{\begin{array}{ccc}{D}_\mathrm{r}(t-1)&{{\mathrm{if}}}&{D}_\mathrm{r}(t-1)\ge\mathrm{RAW}\\ 0 & & \mathrm{otherwise}\end{array}\right.$$

Surface runoff, according to AquaCrop reference manual56:

$$\mathrm{RO}\left(t\right)=\frac{{\left(P-0.05\times254\left[\frac{100}{\mathrm{CN}}-1\right]\right)}^{2}}{P+0.95\times254\left(\frac{100}{\mathrm{CN}}-1\right)}(t)$$

where RO is surface runoff in mm, P is precipitation in mm and CN is curve number.

Deep percolation56:

$$\mathrm{DP}(t)=\tau \left({\theta }_{\mathrm{SAT}}-{\theta }_{\mathrm{FC}}\right)\times\frac{\mathrm{e}^{{\theta }_{t}-{\theta }_{\mathrm{FC}}}-1}{\mathrm{e}^{{\theta }_{\mathrm{SAT}}-{\theta }_{\mathrm{FC}}}-1}1,000z$$
$$0\le \tau =0.0866{{K}_{\mathrm{sat}}}^{0.35}\le 1$$

where τ is drainage characteristics, Ksat is saturated hydraulic conductivity in mm per day, θSAT and θFC are moisture content at saturation and field capacity, θt is moisture level for day t.

The DP is, however, not calculated in this study. The drainage (deep percolation) calculation is dependent on the saturated hydraulic conductivity (Ksat) as a crucial input, which cannot be obtained in global soil datasets such as from the SOILGRID database.

Capillary rise:

CR(t) = 0 (is not considered in this study)

Crop water requirement:

$${{\rm{CWR}}({{t}})=\mathrm{Kc}\times\mathrm{ET}}_{0}(t)$$

Actual evapotranspiration:

$${\mathrm{ET}}_\mathrm{a}\left(t\right)={{K}_\mathrm{s}\left({\rm{t}}-1\right)\times{K}_\mathrm{c}\times\mathrm{ET}}_{0}\left(t\right)$$
$$\mathrm{Or},\,{\mathrm{ET}}_\mathrm{a}\left(t\right)={K}_\mathrm{s}({{t}}-1)\times\mathrm{CWR}(t)$$

Step 3: determine the actual green and blue ETa at time t using the composition ratio of green and blue water at time t − 1.

$${\mathrm{Green}}\;{\mathrm{ET}}_\mathrm{a}\left(t\right)={\mathrm{ET}}_\mathrm{a}(t){\left(\frac{{S}_{\mathrm{g}}}{S}\right)}_{t-1}$$
$${\mathrm{Blue}}\;{\mathrm{ET}}_\mathrm{a}\left(t\right)={\mathrm{ET}}_\mathrm{a}(t){\left(\frac{{S}_{\mathrm{b}}}{S}\right)}_{t-1}$$

Step 4: update the green and blue water stocks at time t and repeat step 1 to step 4 for

subsequent timesteps.

The green, blue and total soil moisture storage in the crop root zone:

$${S}_\mathrm{g}\left(t\right)={S}_\mathrm{g}\left(t-1\right)+P\left(t\right)-\mathrm{RO}\left(t\right)\times{\left(\frac{P}{P+I}\right)}_{t}-\left(\mathrm{E{T}}_\mathrm{a}+\mathrm{DP}\right)\left(t\right)\times{\left(\frac{{S}_\mathrm{g}}{S}\right)}_{t-1}$$
$$\begin{array}{l}{S}_\mathrm{b}\left(t\right)={S}_\mathrm{b}\left(t-1\right)+I\left(t\right)+\mathrm{CR}\left(t\right)-\mathrm{RO}\left(t\right)\times{\left(\frac{I}{P+I}\right)}_{t}\\\qquad\qquad-\left(\mathrm{E{T}}_\mathrm{a}+\mathrm{DP}\right)\left(t\right)\times{\left(\frac{{S}_\mathrm{b}}{S}\right)}_{t-1}\end{array}$$
$$S\left(t\right)=({S}_\mathrm{g}+{S}_\mathrm{b})\left(t\right)$$
$$S\left(t\right)=S\left(t-1\right)+P\left(t\right)+I\left(t\right)+\mathrm{CR}\left(t\right)-\mathrm{RO}\left(t\right)-{\mathrm{ET}}_\mathrm{a}\left(t\right)-\mathrm{DP}\left(t\right)$$

Depletion at the end of the day:

$${D}_\mathrm{r}\left(t\right)=S\left(t\right)-S\left(t-1\right)$$

or

$${D}_\mathrm{r}(t)={D}_\mathrm{r}(t-1)-P(t)-I(t)-\mathrm{CR}(t)+\mathrm{RO}(t)+\mathrm{E{T}}_\mathrm{a}(t)+\mathrm{DP}(t)$$

Step 5: monthly and seasonal/annual crop actual green and blue WC aggregated from daily values:

$$\mathrm{Monthly}/{\mathrm{seasonal}}\;{\mathrm{green}}\;{\mathrm{ET}}_{a}({\mathrm{Green}}{\mathrm{ET}}_{\mathrm{a,s}})=\mathop{\sum}\limits_{t=1}^{n}{\mathrm{Green}}{\mathrm{ET}}_{\mathrm{a}}({{t}}),$$
$${\mathrm{Monthly}}/{\mathrm{seasonal}}\;{\mathrm{blue}}\;{\mathrm{ET}}_{\mathrm{a}}({\mathrm{Blue}}{\mathrm{ET}}_{\mathrm{a,s}})=\mathop{\sum}\limits_{t=1}^{n}{\mathrm{Blue}}{\mathrm{ET}}_{\mathrm{a}}({{t}})$$

where n is number of days in the period (a month, season or year). When a crop grows in two calendar years, the values of similar months are aggregated.

$${\mathrm{Seasonal}}\;{\mathrm{water}}\;{\mathrm{consumption}},\,{\mathrm{ET}}_{\mathrm{a,s}}=\mathop{\sum}\limits_{t=\mathrm{SOS}}^{\mathrm{EOS}}{\mathrm{ET}}_{\mathrm{a}}({{t}})$$
$${\mathrm{Or}},{\mathrm{ET}}_{\mathrm{a,s}}={\mathrm{Green}}{\mathrm{ET}}_{\mathrm{a,s}}+{\mathrm{Blue}}{\mathrm{ET}}_{\mathrm{a,s}}$$

where SOS is start of the season and EOS is end of the crop season.

Input data

An overview of all input data is provided in Extended Data Tables 1 and 2. SPAM data for the year 2020 (ref. 16) (called SPAM2020), differentiating between 46 crops, were used. SPAM2010 differentiates between 42 crops and SPAM2000 between 21 crops. The crop ‘21 othe–Other crops’ in SPAM2000 does not contain data, which makes it not usable for the modelling. Fodder crops, included in Mialyk et al.14,15 or Mekonnen and Hoekstra1, are not included in SPAM. Otherwise, all crops in SPAM are included in Mekonnen and Hoekstra and vice versa1. Latter authors differentiate between more individual crops.

Crop coefficient Kc and length of growth stages were used from Allen, Pereira, Raes and Smith55. Soil data (total available water capacity and curve number (CN)) were taken from Fischer et al.57 and Jaafar, Ahmad and El Beyrouthy58, respectively. Climate data (reference evaporation (ETo) and precipitation) were obtained from FAO59 and Boogaard et al.60, respectively.

The crop calendar we used to estimate WC draws on reliable sources, specifically the SAGE46,47 and MIRCA2000 (ref. 44) datasets, which provide global data on crop phenology. For some crops and specific countries, additional regional sources45,48 were incorporated to enhance accuracy. Extended Data Table 3 provides an overview on used sources. This multi-source approach helps ensure that the crop calendar aligns as closely as possible with observed growing periods worldwide. We thereby account for double-cropping.

For rice, we added an additional blue WC calculation for the paddy flooding phase, as the water applied during this phase is substantial and evaporates throughout the growing season1,61. We used a 25-cm bund to store water and imitate flooding conditions.

We compare our results representative for the year 2020 with the results representative for the years 2010 and 2000.

The direct modelling input data and data pre-processing scripts for SPAM2020 are included in supplementary data 12–16 in ref. 43.

WC related to renewable water availability for major river basins

We quantify the relation of blue WC to renewable blue water availability in major river basins. We define renewable water availability as natural renewable water minus environmental flows (EFs):

$${\mathrm{renewable}}\;({\mathrm{blue}})\;{\mathrm{water}}\;{\mathrm{availability}}={\mathrm{natural}}\;{\mathrm{renewable}}\;{\mathrm{water}}\;{-}\;{\mathrm{EF}}$$

Natural renewable water in high spatial resolution (0.1° or 11.1 km at the equator) was taken from Vanham et al.62, computed with the hydrological model LISFLOOD63. The model works at a daily time step for the period 1980–2018 and generates natural water availability as the sum of renewable surface and groundwater. We use the geodataset on river (sub)basins of the Global Runoff Data Centre64 to aggregate grid natural renewable water amounts to the basin level.

EFs are required to maintain ecosystem integrity in streams, rivers, wetlands, riparian zones and estuaries. To quantify EFs, we used the presumptive standard for EFs by Richter et al.65, widely used in water-management studies10,29,66,67,68. This methodology defines 80% of the natural flow as EF. The remaining 20% is considered as water available for human use, in this paper defined as renewable water availability. This presumptive standard is supported by empirical studies showing that flow alterations within 20% support native fish species and flow alteration beyond this level strongly affects biodiversity and ecosystem structure and function69. The global renewable water availability amounts to 12,025 km3 annually.

We did not conduct a full water stress assessment, for which all water demand stakeholders (such as agriculture, municipal water use, mining and industrial water use) are required. The reason is that we focus specifically on WC for crops.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.