Global unsustainable virtual water flows in agricultural trade

Recent studies have highlighted the reliance of global food production on unsustainable irrigation practices, which deplete freshwater stocks and environmental flows, and consequently impair aquatic ecosystems. Unsustainable irrigation is driven by domestic and international demand for agricultural products. Research on the environmental consequences of trade has often concentrated on the global displacement of pollution and land use, while the effect of trade on water sustainability and the drying of over-depleted watercourses has seldom been recognized and quantified. Here we evaluate unsustainable irrigation water consumption (UWC) associated with global crop production and determine the share of UWC embedded in international trade. We find that, while about 52% of global irrigation is unsustainable, 15% of it is virtually exported, with an average 18% increase between year 2000 and 2015. About 60% of global virtual transfers of UWC are driven by exports of cotton, sugar cane, fruits, and vegetables. One third of UWC in Mexico, Spain, Turkmenistan, South Africa, Morocco, and Australia is associated with demand from the export markets. The globalization of water through trade contributes to running rivers dry, an environmental externality commonly overlooked by trade policies. By identifying the producing and consuming countries that are responsible for unsustainable irrigation embedded in virtual water trade, this study highlights trade links in which policies are needed to achieve sustainable water and food security goals in the coming decades.


Introduction
In the last decade, many studies have shown that some of the world's major agricultural baskets rely on unsustainable water use for irrigation (Gleick and Palaniappan 2010, Konikow 2011, Scanlon et al 2012, Kummu et al 2016, Mekonnen and Hoekstra 2016. Irrigation practices are classified as unsustainable when their water consumption exceeds local renewable water availability. In these conditions, irrigation uses water that should be allocated to environmental flows and therefore contributes to environmental degradation and groundwater depletion (Rosa et al 2018a). About 40% of global irrigation water use is at the expenses of environmental flow requirements (Jägermeyr et al 2017) with detrimental effects on aquatic habitats, riparian biodiversity, and ecosystem services (Dudgeon et al 2006, Richter et al 2012, Palmer and Ruhi 2019. Moreover, about 20% of irrigation water use worldwide is from non-renewable groundwater abstractions (Wada et al 2012). Indeed, irrigation water withdrawals may also deplete freshwater stocks (both surface water bodies and aquifers) when their abstraction rates exceed those of natural recharge (Wada et al 2010, Famiglietti 2014, AghaKouchak et al 2015, Richey et al 2015, Rodell et al 2018, Wang et al 2018. The reliance of food production on unsustainable irrigation threatens local and global water and food security (Aeschbach-Hertig and , Turner et al 2019. The problem is worsened when the nexus between irrigation-dependent agricultural production and food consumption occurs through distant interconnections resulting from the globalization of food Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. and water resources through trade or international investments Chapagain 2008, D'Odorico et al 2018). About 20%-24% of water resources embedded in food production-or 'virtual water'are internationally traded (D'Odorico et al 2014). While the virtual water trade associated with agricultural commodities (Konar et al 2011, Hoekstra andMekonnen 2012), and the regions of unsustainable irrigation have been extensively investigated and mapped (Rosa et al 2018a), the globalized dimension of unsustainable irrigation, the loss of environmental flows, and the associated unsustainable virtual water trade are poorly studied (D'Odorico et al 2019).
Research on virtual water has often focused on quantitative analyses of water flows, with no consideration of the environmental impacts of virtual water transfers and of how trade affects the sustainability of irrigation practices (Gawel and Bernsen 2013). Virtual water transfers have been related to groundwater depletion in the United States (Marston et al 2015) and globally (Dalin et al 2017). As important as these studies are, the extent to which virtual water trade contributes to the loss of environmental flows, remains poorly understood. It is not clear to what extent unsustainable surface water consumption for irrigation contributes to the desiccation of rivers and the loss of environmental flows to sustain agricultural production for the export market. Because surface water accounts for about 60% of total consumptive irrigation water use , there is a direct association between agricultural production and patterns of streamflow depletion (Richter 2014) that was not considered in previous analyses of unsustainable irrigation and virtual water trade (Marston et al 2015, Dalin et al 2017. Indeed, many rivers around the world are so strongly depleted that minimum flow requirements to sustain aquatic habitat are not met and in some cases the flow does not even reach the ocean anymore (Richter 2014, Jägermeyr et al 2017. Here, we provide a comprehensive global analysis of unsustainable irrigation water consumption for crop production, accounting for the depletion of both freshwater stocks (including surface and ground water bodies) and environmental flows. We quantify the associated virtual water flows through trade (or 'unsustainable virtual water trade') in year 2000 and 2015 through a global spatially-distributed biophysical analysis of irrigation water consumption considering 130 primary crops or 26 crop classes (Portman et al 2010). We quantify the amount of irrigation water that is unsustainably consumed by comparing irrigation water consumption to local renewable freshwater availability, calculated through a process-based water balance model accounting for environmental flows. We then determine the unsustainable water footprint of crop production and trade (FAOSTAT; http://fao. org/faostat/en/#data) to quantify the unsustainable irrigation water consumption embodied in the international trade of agricultural commodities.
This analysis sheds light on the water unsustainability of crop trade. It identifies the 'culprits' of unsustainable irrigation water consumption in terms of both production regions and consumer countries, and determines the associated virtual water flows. These results can assist in the development of consumption based decision-making tools aiming at meeting Sustainable Development Goals by ensuring sustainable use of water resources to reduce the number of people suffering from water scarcity (Vanham et al 2018).

Assessment of unsustainable irrigation water consumption
We quantified unsustainable water consumption from irrigation worldwide (at 5×5 arcminute resolution) in years 2000 and 2015 for 26 crop classes, based on the MIRCA2000 dataset . For year 2000 (the reference year for the MIRCA2000 global agricultural datasets), we used a global process-based crop water model to assess irrigation water requirements. This model has been extensively used to assess irrigation water requirements (Rosa et al 2018a(Rosa et al , 2018b. The model calculates spatially explicit cropspecific irrigation water requirements (mm yr −1 ) using a daily soil water balance during each crop's growing season. To assess irrigation water consumption, we then multiplied crop-specific irrigation water requirements by the irrigated harvested area of that crop in the year 2000 .
Because there are no up-to-date global datasets of crop-specific irrigated harvested area, we estimated the change in irrigation water consumption between year 2000 and 2015 as proportional to the change in countryspecific irrigation water withdrawals (from FAO's AQUASTAT) and crop-specific change in production (from FAO's FAOSTAT). Specifically, national cropspecific estimates of volumes of irrigation water consumption for year 2000 were scaled to year 2015 based on national agricultural water withdrawal and crop-specific production data (see supplementary materials is available online at stacks.iop.org/ERL/14/114001/ mmedia). To quantify the sustainability of irrigation water consumption in year 2015, we assumed that the variation in irrigation water consumption between years 2000 and 2015 is proportional to country-specific sustainable and unsustainable irrigation expansion potentials. For example, if a country increases irrigation water consumption but has no potential to do so sustainably, all the additional irrigation volumes are assumed to be consumed unsustainably (at the expenses of environmental flows and surface-and ground-water stocks). Country specific values of the sustainability and unsustainability of irrigation expansion were taken from Rosa et al (2018a). Crop-specific production in years 2000 and 2015 were taken from FAO's FAOSTAT database (see section: Assessment of unsustainable virtual water trade). Agricultural water withdrawals in years 2000 and 2015 were taken from FAO's AQUASTAT database.
To determine total blue water consumption (WC) in each grid cell, crop-specific irrigation water consumption values were summed with municipal and industrial water consumption estimates (for the 1996-2005 period) (Hoekstra and Mekonnen 2012). By combining total blue water consumption and renewable blue water availability (WA) we assessed unsustainable irrigation practices. We identified areas of unsustainable irrigation water consumption as those where local renewable blue water resources are less than local total water blue consumption (WC > WA). This methodology to evaluate water sustainability has been extensively validated in studies aiming at analyzing the influence of energy and agricultural production on water resources (Rosa et al 2018a, 2018b, Rosa and D'Odorico 2019. Renewable blue water availability (30 × 30 arcminute resolution, or ∼50 km at the Equator) was assessed following Mekonnen and Hoekstra (2016) and was calculated as the difference between blue water flows generated in that grid cell and environmental flow requirements. Renewable blue water availability accounts for surface-and ground-water volumes that are replenished through the annual hydrological cycle (Rosa et al 2018a(Rosa et al , 2018b. This methodology explicitly links irrigation water consumption to unsustainable irrigation practices. Long term (circa year 2000) blue water flows were assessed from local runoff estimates (Fekete et al 2002) and were calculated using the upstream-downstream routing 'flow accumulation' function in ArcGIS ® . Total blue water consumption (WC) at a 5×5 arcminute resolution was aggregated to the 30×30 arcminute resolution of the global available water (WA) dataset. Following previous global analyses we assumed that 80% of annual blue water flows should be allocated for environmental flows preservation (i.e. remain unavailable to human consumption) (Richter et al 2012, Mekonnen and Hoekstra 2016, Flörke et al 2018. We considered 26 crop classes or 130 primary crops (or nearly 100% of global crop production) (wheat, maize, rice, barley, barley, rye, millet, sorghum, soybeans, sunflower, potatoes, cassava, sugar cane, sugar beets, oil palm, rapeseed, groundnut, cassava, groundnuts, pulses, citrus, date palm, grapes, cotton, cocoa, coffee, other perennials, fodder grasses, other annual) based on the MIRCA2000 dataset (Portmann et al 2010) (supplementary materials). Crop classes 'other annual' and 'other perennials' are labeled as 'Fruits & Vegetables' in figures 2 and 3. 'Sugar Crops' include 'sugar beet' and 'sugar cane'; 'Other Grains' considers aggregated values of 'barley', 'rye', 'millet', and 'sorghum'. Water consumption from industrial and domestic sectors were taken from Hoekstra and Mekonnen (2012) and were assumed to be constant between year 2000 and 2015. Possible inaccuracies in water consumption estimates for domestic and industrial uses are difficult to evaluate with the available data, however, they are expected to have limited impacts on our results. In fact, water consumption from industrial and domestic sectors accounts for just ∼7% of total water consumption. Moreover, previous global studies of irrigation water consumption provided estimates that range from 847 to 1180 km 3 (Siebert et al 2010, Hoekstra and Mekonnen 2012). This range of uncertainty, by far, exceeds total water consumption from the domestic and industrial sectors (80 km 3 ; Hoekstra and Mekonnen 2012).

Assessment of unsustainable virtual water trade
We used international trade matrices and national production data (FAOSTAT) (http://fao.org/faostat/ en/#data) for 302 food commodities to assess the trade T of a food commodity x from a country y to z in year n (2000 and 2015) noted as T (y, z, x, n). Because international trade and national crop production are dynamic and vary year by year, for international trade and national production data we used a five-year average around years 2000 and 2015 to smooth out this variability. For trade matrix data we used FAO's import data, because import reporting is more reliable than export reporting owing to custom reports at the port of entry. Import data are expected to be more accurate, because customs have an incentive to collect data for tax purposes. Exports data are generally poorer, as very few countries tax exports. Moreover, the variation between import and export data of the FAO's dataset has a relatively small effect on the quantification of virtual water flows. Because FAO-STAT's trade data are at the country and annual scales, our analysis is performed at annual and country levels.
We aggregated international trade matrices and national production data for 302 food commodities x into their 130 primary crops c dividing the quantity of produced or traded commodity R(x) by its primary product extraction rate E x : x c x E x was taken from FAO's technical conversion factors for agricultural commodities (http://fao.org/ fileadmin/templates/ess/documents/methodology/ tcf.pdf) (supplementary materials), and R(x) is the quantity of produced or traded commodity x, and c is the set of commodities based on the same primary crop. Primary product extraction rate (E x ) is the fraction of the processed product obtained from the processing of the primary product. For example, FAOSTAT's commodities 'sunflower seed', 'sunflower oil', and 'sunflower cake' are based on the primary crop 'sunflower' and they have an extraction rate E equal to 1, 0.47, 0.49, respectively. To assess production and trade of the primary crop 'sunflower' we divided each commodity by its extraction rate and summed the results to obtain aggregated international trade matrices and national production data for sunflower in year n. The international trade matrices of the 130 primary crops where then aggregated into the 26 MIRCA2000 crop classes (M) to obtain trade fluxes T from a country y to z in year n of crop class M, T (y, z, M, n). The same procedure was followed for FAO-STAT's production data to obtain production data P of country y in year n of crop class M, P(y, M, n). Cropspecific primary product extraction rates are kept constant among countries.
We then assessed unsustainable irrigation water consumption intensity (UWCI) of each MIRCA2000 crop class (M) in each country (y) and year (n), as follows: where UWC is the unsustainable irrigation water consumption of crops in class M, country y, and year n (expressed in m 3 of water). UWC has been assessed by aggregating irrigation blue water consumption (WC) at the country level. P is the aggregated production (expressed in tons) of crops belonging to class M in country y and year n. We then used UWCI to convert the trade fluxes T (expressed in tons) into unsustainable virtual water flows (expressed in m 3 ): where the trade T of a food commodity x from a country y to z in year n (2000 or 2015) is noted as T (y, z, x, n).

Results
The unsustainability of irrigation We find that 52% (569 km 3 ) of global irrigation practices are unsustainable because they deplete freshwater stocks and/or environmental flows (figure 1). About 70% of the global unsustainable water consumption for irrigation (hereunder UWC) is contributed by India (28%), China (16%), Pakistan (13%), and the United States (12%) alone ( figure 2(a)). In many countries a big share of irrigation water consumption is unsustainable (supplementary figure 1) as in the case of India (54% of national irrigation water consumption or 157 km 3 yr −1 ), China (66% or 91 km 3 yr −1 ), Pakistan (61% or 71 km 3 yr −1 ), and the United States (62% or 69 km 3 yr −1 ). The impact of agriculture on UWC strongly varies with crop type and geographic location with wheat, maize, rice, cotton, and fruits and vegetables collectively contributing to 73% (or 417 km 3 yr −1 ) of global UWC (figure 3). While in India and Pakistan, wheat is the major contributor to UWC (32% and 38%, respectively), in China most of the UWC is from rice (33%) and in the United States from maize (29%) ( figure 2(b)).
In year 2015, the United States, India, Pakistan, Mexico, and Spain account for two thirds of UWC embodied in food trade. The United States is the largest exporter, with 22% (19.7 km 3 ) of global unsustainable virtual water transfers, followed by India (19%), Pakistan (14%), Mexico (7%), and Spain (5%). China is the largest importer of UWC-based crops, followed by the United States, Turkey, Mexico and Japan ( figure 4).
India, the United States, Pakistan, Spain, Turkmenistan, Egypt, Uzbekistan, and Australia consistently act as net exporters of UWC-based crops (figure 5), while Canada, United Kingdom, France, Germany, Italy, China, Turkey, Russia, and Indonesia act as net Crops contributing the most to international UWC trade are cotton (33% or 29 km 3 ), rice (15%), fruits and vegetables (12%), citrus (8%), wheat (7%), maize (6%), sugar cane (6%), and soybeans (4%). In the 2000-2015 period, UWC embedded in traded rice, wheat, citrus, cotton, and soybean has increased by 130%, 66%, 125%, 15%, 31%, respectively. importers of UWC-based crops. In the 2000-2015 period, Iran, Peru, Libya, Algeria, and Ethiopia have switched from being net exporters to net importers of unsustainable virtual water. Other countries such as Mexico, Tunisia, and Mozambique have become net exporters. Figure 5 also shows recent history of international food trade with the increasing presence of China as major importer, while it also shows the rise of India and Pakistan as major exporters of non-sustainable agricultural commodities.
In the United States 29% of the UWC is due to crops for export markets ( figure 2(a)). India and Pakistan are the second and third largest exporters of UWC-based crop production, even though they keep 90% and 82% of UWC for domestic consumption, respectively. China is the world largest net importer of UWC-based crops (19% of the global virtual trade of UWC). Moreover, China keeps 97% (89 km 3 ) of its UWC-based crops for domestic consumption.
In year 2015, 42% (8.2 km 3 ) of the Unites States' unsustainable virtual water trade was embedded in cotton export mainly to China, Mexico, Canada, Japan, and Turkey (table 1 and figure 5). Maize and soybeans accounted for 17% and 11% of the United States' unsustainable virtual water exports to China, Mexico, and Japan (table 1). India exported unsustainably produced cotton and rice to China and Bangladesh. Mexico is a major exporter of unsustainably produced citrus and fruits and vegetables to the United States. UWC embedded in rice production accounts for 70% of Pakistan's unsustainable virtual water exports, mainly to China, Afghanistan, and Kenya. Uzbekistan and Turkmenistan export unsustainably produced cotton to China and Turkey. Spain and Morocco are major exporters of fruits and vegetables to European countries.

Discussion
irtual water trade is fundamental to achieve food security in water scarce regions of the world, however, it establishes a disconnection between consumers and the water resources they rely on. This ultimately leads to a loss of ecosystem stewardship (D'Odorico et al 2019) and the enhancement of environmental degradation associated with the drying of rivers and loss of aquatic habitat (Soligno et al 2017(Soligno et al , 2018. Research on the environmental impacts of trade and trade policies (Peters et al 2011, Zhang et al 2017 suggests that production is expected to shift to regions of the world with weaker environmental regulations (Dean et al 2009). In the case of agricultural production and trade, however, the focus has often been on environmental pollution, land use change, and labor rights, while the environmental impacts of unsustainable irrigation and their displacement through trade have remained poorly understood and have just started to be recognized and quantified.
Our results shed light on crops, country, and trade relationships that rely on unsustainable irrigation practices in production and consumption. More than 30% of unsustainable irrigation practices of Mexico, Spain, Turkmenistan, South Africa, Morocco, and Australia are embedded in food exports, while, in India, China, Iran, and Saudi Arabia 90% of unsustainable irrigation volumes are embodied in domestic food consumption. We also find that 60% (53 km 3 ) of global unsustainable virtual water trade is driven by cash crops (cotton, sugar cane, fruits and vegetables).
In particular, cotton alone is responsible for 33% of UWC embedded in international crop trade. These findings show important trade-offs between the economic benefits and the environmental consequences of unsustainable irrigation practices.
Not surprisingly, the fact that only 30% of the increase in virtual water trade (in the 2000-2015 period) is contributed by unsustainable irrigation, confirms previous studies that quantified the increasing reliance of international markets on cropland expansion and rain-fed agriculture, including soybean production in Brazil and Argentina, and oil palm production in Indonesia and Malaysia (Aldaya et al 2010. This study improves our previous assessment of unsustainable irrigation water consumption (Rosa et al 2018a), where we found that in year 2000 about 40% (336 km 3 ) of global irrigation was unsustainable, based on 16 major crops that account for 70% of global crop production. Here, we considered 130 primary crops (or nearly 100% of global crop production) and found that 51% (525 km 3 ) of global irrigation volumes are unsustainable. Moreover, here we also provide crop-specific and country-specific analyses of unsustainable irrigation and evaluate extent to which it is contributed by international trade.

Uncertainty, limitations, and assumptions
The complexity of a global analysis often requires the adoption of suitable assumptions. We used a wellestablished methodology to assess irrigation water consumption based on existing maps of irrigated areas for the year 2000 . However, it is important to note that the estimation of irrigated areas would change significantly using different input data and statistics (Meier et al 2018). Because, to our knowledge, there are not global datasets providing crop-specific irrigated harvested areas after year 2000, we used country-scale statistics to assess irrigation water consumption in year 2015. While this is a limit of global studies aiming at an assessment of irrigation water consumption, our results are in good agreement with recent country-specific changes in irrigation water consumption as available for Australia (−1. Our assessment is based on temporal averages and does not account for inter-annual variability in river discharge and crop water requirements. We performed a sensitivity analysis of irrigation water consumption with respect to changes in climate conditions between 2000 and 2015. We run our crop water model with year 2015 climate forcing, while keeping the same spatial extent of irrigated area as in the MIRCA2000 dataset. We find that there is little sensitivity of irrigation water consumption between the two years (1025 km 3 for year 2000 versus 1035 km 3 for year 2015) (see supplementary table 4 for crop-specific values).
Because fodder grasses (Alfalfa, clover, and grasses) are not present in FAOSTAT's trade data, we excluded them from our unsustainable virtual water trade analysis. It is important to notice that fodder grasses are mainly used for domestic consumption as feed to livestock and therefore they are not commonly traded among countries. Nevertheless, we find that fodder grasses account for 7% of global unsustainable irrigation water consumption ( figure 2(b)).
Our results show little sensitivity to different environmental flow thresholds, as previously highlighted also by Mekonnen and Hoekstra (2016). With the current assumption that 80% of blue water flows should be allocated to environmental flows preservation, we find that, in year 2000, 51% (525 km 3 ) of global irrigation water consumption is unsustainable. When we assume that environmental flows account for 90% and 60% of blue water flows, unsustainable irrigation water consumption in the same year becomes 536 km 3 and 523 km 3 , respectively (see supplementary table 5 for crop-specific results).
As already stressed by Mekonnen and Hoekstra (2016), this low sensitivity to the threshold used to define environmental flows is due to the huge spatiotemporal mismatch between water consumption and availability.

Conclusions
Policymakers and major corporations are broadening the scope of their actions to meet the increasing consumer demand for sustainable commodities and improve corporate social responsibility. For example, there has been a recent commitment to purchase or produce deforestation-free products (Carlson et al 2018, Curtis et al 2018. In an increasingly water scarce world, governments could take specific actions targeting unsustainable irrigation practices by penalizing the associated imports. By identifying the producing and consuming countries that are responsible for unsustainable virtual water trade, this study highlights trade links in which policies are needed to achieve sustainable water and food security goals in the coming decades. Future studies should examine socio-economic implications, such as the feasibility to reduce unsustainable virtual water trade through the adoption of adequate policies. the data; D D C and M C R carried out the simulation to assess irrigation water requirements; C T processed crop trade and production data; and L R wrote the paper with inputs from all authors.