Diet change—a solution to reduce water use?

We used food supply information from the FAO (United Nations Food and Agriculture Organization) food balance sheets (FAO 2013). To limit the effects of possible year-to-year variations, we used averages over the most recent three data years (2007–2009). The food supply figures provided were totals that included domestic production and net imports, adjusted for any changes in stocks (FAO 2001). For each country, we aggregated food products into 13 groups, of which eight were used in the diet adjustment calculations. These were vegetal product groups of cereals, fruits and vegetables, oil, oilseeds and roots and animal-based groups of eggs, meat and milk. Additionally, the food groups of fish, beverages, spices, stimulants and sugar were included in the calculations but were not used for the diet adjustment for various reasons (see section 2.2 Diet scenarios). The group composition can be found online (supplement table S1, available at stacks.iop.org/ERL/9/074016/mmedia). The data were available for 176 countries and for a global population of 6.6 billion people (97% of the total population). Harvest and post-harvest losses were derived for each country and food group from FAOSTAT (FAO 2013), while waste percentages for processing, distribution and consumption were derived from Gustavsson et al (2011) and Kummu et al (2012).

In the first phase of the diet adjustment, the 'Recommended Diet' (RD), we changed the food intake to conform to each country-specific average dietary energy requirement ADER (FAO 2012b) and to the WHO recommendation of macronutrient intake (WHO 2003), as described in table 1. In addition to the dietary recommendations, some constraints in diet change were considered in the diet adjustment. As many of the world's wild fisheries are already overexploited (FAO 2005), we constrained the consumption of fish to its current level. Whilst there is potential to increase aquaculture production (Tidwell and Allan 2001), the related environmental effects are complex (Bostock et al 2010), and the required water footprint data are not available; thus its assessment was left for further studies. For health reasons, additional alcoholic beverages or sugar were not allowed to increase the insufficient dietary energy intake in the diet adjustment. Moreover, per capita consumption of spices and stimulants was not changed, as we believe that these are not used to fulfil nutritional needs but rather are used for cultural and taste reasons.

Starting from the RD, we reduced the contribution of animal products to the diet in four steps (table 2), while still accounting for the dietary guidelines and for other limitations that were described before. In scenario A50, we limited the protein intake from all animal products to 50% and from meat to 16.7%. Consequently, animal protein consumption did not change much in countries where it was already low compared to the total protein intake, while animal protein was considerably replaced by protein from other food groups in countries where animal products were a major staple food. The limits for protein intake from animal products were then gradually decreased to zero in scenarios A25–A0 (table 2).

To preserve a good amino acid composition and to avoid a nutritionally unsatisfactory diet, the protein removed from the meat group was substituted by protein from oilseeds and root crops. In contrast, the optimisation algorithm was allowed to freely compensate for animal protein reduction in the milk and egg groups. The optimisation code is provided in the Supplement for further experimentation.

We used quadratic programming to calculate the changes in diet for each step. Although quadratic programming (Nocedal and Wright 2006) is an optimisation method, the goal was not to optimise the water consumption of a diet but to find a diet that would fulfil two criteria: i) meet the dietary objectives of each scenario (table 2) and ii) minimise the change in the diet, i.e., aim to retain the typical diet for each country. The original diet was assigned as the optimisation objective, and constraints were used to enforce the requirements of each scenario. Quadratic programming provided a virtual cost for any deviations from the objective function; therefore, the result closely followed the traditional, culturally acceptable diet while satisfying the nutritional constraints.

We estimated the consumptive water use of the diet in the different scenarios by multiplying the food supply (i.e., including food intake and losses) with water footprints of the crops (Mekonnen and Hoekstra 2011b) and with animal products (Mekonnen and Hoekstra 2010). Wherever possible, we used product-and country-specific values for the footprints, but if country-specific values were unavailable, global averages were used instead. The blue water consumed for pasture and grass irrigation was derived from the Global Crop Water Model GCWM (Siebert and Döll 2010) because this water use was not considered in the water footprints of animal feed calculated by (Mekonnen and Hoekstra 2010). We assigned the volume of water used for pasture and grass crop irrigation to the blue water footprints of beef and milk. Our approach does not take into account that it might be contradictory to the objectives of sustainable use of resources to constrain production of livestock commodities when it is based on the grazing of extensive pastureland, which is otherwise difficult or impossible to exploit by agriculture to produce human food.

We accounted for global food trade by aggregating exported blue and green water, which was calculated for each foodstuff and each exporting country, into a common pool. Imports of blue and green water were then assigned to specific countries in relation to imported and produced food group quantities, as similarly done by Kummu et al (2012). Food group exports and imports were derived from FAO commodity balance sheets (FAO 2013). Therefore, our methodology does not account for trade relations between specific countries, which also change over time. While it is already challenging to trace current and historical flows of green and blue water embodied in processed products and in livestock commodities between specific countries, import and export relationships would certainly be completely reorganized under large-scale diet changes. Modelling these potential changes in trade flows was out of the scope of this study. We find that the aggregated imports and exports better highlight the intrinsic water footprint effect of a national diet, without emphasising the volatile trade matrix between countries.

We found a large variability in national diets, which are in distinct agreement with the quite broadly defined WHO dietary recommendations. The largest differences to the guidelines were in vegetable consumption, but there were appreciable deviations in all of the other diet quality related aspects as well. Nevertheless, the world average diet was well balanced at 11% of the energy intake from protein (recommendation 10–15%) and 27% from fat (15–30%), and the total energy intake was 2446 kcal/cap/d (population weighed average ADER is 2357 kcal/cap/d) (table 3). Of the total protein intake, on average, one-third originated from animal-based foodstuffs (figure 1), which represented 17% of the total energy intake. Even the global average of fruit and vegetable consumption of 443 g/cap/d was higher than the recommended minimum of 400 g/cap/d. The baseline water footprints for the global food consumption were 2350 l/cap/d for green water and 388 l/cap/d for blue water, respectively (table 3). We found a large variation between the countries in the baseline dietary status in total food intake, as well as in the composition of the diet, as shown below.

Zoom In Zoom Out Reset image size

Adjusting diets to the recommendations (i.e., RD scenario) reduced global dietary energy consumption by 3.6%, which showed that the world food supply, even after losses, is more than adequate from the energy intake perspective. However, the distribution of food is far from even. Food intake decreased in 102 countries, with a total population of 4.3 billion because of the adjustment, whilst it increased in 74 countries with a total population of 2.3 billion (table 3). The average reduction of energy intake in the countries with a decreasing food intake was 199 kcal/cap/d, while in the countries with an increasing food intake, the required addition was 121 kcal/cap/d. The percentage of energy from protein increased globally from 11 to 12% (recommendation 10–15%), and the contribution of fat to the energy intake decreased from 27 to 25% (recommendation 15–30%). Very different adjustments were, however, required for individual countries (figures 2(B) and (C)).

Zoom In Zoom Out Reset image size

The changes in global water footprints under the RD scenario (figure 3(A), 4(A) and (B)) followed the dietary adjustments: the global blue water footprint declined by 1%, and the green water footprint declined by 2% (table 4), with large differences between the countries (figures 4(A) and (B)). In terms of blue water, a slight increase was calculated for large parts of Africa, South and Southeast Asia and Western South America. The largest increases were obviously found for countries suffering from undernourishment. The largest decreases in blue water use were calculated for Kazakhstan, Portugal and Iran. In terms of green water, the changes were mostly rather modest. The notable increases were concentrated in Africa—especially toward the south—and in Bolivia, due to its need for an increased food supply. The largest decreases were observed in Niger and in Burkina Faso, where the shift from cereals towards fruits and vegetables were accountable for the change. Also, many countries around the Mediterranean, as well as Kazakhstan, showed a notable decrease in green water use (figure 4(B)).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In scenarios A50–A12.5, the reduction in blue water consumption was rather modest (−4% to −9%, table 4). Scenario A0, in which all animal products were removed from the diet, had a more appreciable effect: a 14% reduction (figures 5 and 6). The green water savings were larger than those for the blue water, e.g., reaching a reduction of 15% in A12.5 and 21% in A0 (figures 5 and 6, table 4).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The differences in the contribution of animal products to the protein supply (figure 1) caused large differences in the change of water footprints between the regions (figure 5). In Southeast Asia and South Asia, as well as in Sub-Saharan Africa, the water savings by diet change were quite limited, and some parts of Asia even showed some increases when stricter limits were imposed. These areas already have low animal protein content in their original diet. In contrast, both blue and green water footprints decreased clearly across the scenarios in North America, Australia and Oceania. In these regions, the biggest decrement was from RD to A50, as the original diets were rich in animal protein, which was well over the limit of A50. In Latin America, Western Europe and Eastern Asia, mainly the green water footprint was reduced.

The effects of shifting away from animal-based diets on water footprints were much more notable at the country level than at the global scale (figures 3(B) and (C), figures 4(C)–(F)). In 140 countries, the blue water footprint was smaller in A0 than in RD (figures 3(C), 4(E)), while it increased in 36 countries. In 166 countries, the green water footprints decreased between RD and A0 (figures 3(C), 4(F)), with increasing footprints in 10 countries.

The results allow the identification of two factors explaining the low reduction—or even the increase—of blue water footprints when animal protein in the diet was reduced. First, the high protein content of animal products requires a large amount of compensating vegetal products. The second explanation is the relatively high proportion of irrigated agricultural land used for the production of cereal crops (61%), while only 7.4% of the irrigated land is used for fodder and as pasture (FAO 2012a). While oilseeds and roots have a smaller blue water footprint per gram of protein than animal products (apart from fish), cereals—the largest protein source in most countries—have a higher one. The global averages for water footprints per gram of dietary protein are presented in supplement tables S2 and S3.

Vanham et al (2013) conclude that a healthy European diet conforming to regional recommendations could save 3–30% in total water consumed for food production in Europe, while a lacto-ovo vegetarian diet would result in a savings of 27–41%, compared to current eating habits. Our results for Europe, without East European countries, are within their ranges; RD, with only dietary recommendations applied, decreased total water consumption by 13%, whilst A0 decreased the total water footprint by 30% compared to the Original Diet. It should be noted that healthy diet compositions by Vanham et al (2013) are designed with more attention to health effects and foodstuff availability than could be achieved by our minimal change methodology.

Springer and Duchin (2014) proposed that a dietary energy supply of 3000 kcal/cap/d, and limiting protein from animal products to 20%, would be sustainable goals in developed countries. Ercin and Hoekstra (2014) also determined diet change to be an important component of a sustainable future. The scenarios in these two papers combine a number of factors, and while the results point at the same direction as those of this study, the effect of dietary change alone cannot be accurately compared.

Renault and Wallender (2000) compare the American diet to diets with reduced animal product content. They found that a diet with a 50% reduction in animal products would reduce total water use by 37%. Our A25, with a slightly higher animal product reduction, provided a savings of 25% for the US. With regard to a lacto-ovo vegetarian diet, Renault and Wallender (2000) found a 52% reduction of water use. This is higher than the 35% water footprint reduction of our A0, even though we removed all of the animal products except fish.

Our results can also be compared with other suggested actions to increase food availability or to reduce the environmental burden of food production. By adjusting diets to the recommended energy intake and composition, total water footprints would be reduced from 2738 l/cap/day to 2682 l/cap/day, while a diet without meat, milk and eggs would reduce the global mean water footprint further to 2149 l/cap/day. The use of current volumes of water with a lower dietary water footprint would allow the production of more food to feed an additional 136 million people (RD scenario) or even 1.8 billion people (A0 scenario). This would, however, require that similar improvements of efficiencies are possible for other resources used in food production, as well the avoidance of improvements in water use and in the increased food supply, which would cause an additional environmental burden with regard to other resources (e.g., crop nutrients, land).

Kummu et al (2012) found that halving losses and waste within the food supply chain (Kummu et al 2012) could reduce cropland area by 11% and blue water use for food production by 12%, or it could allow the production of food for one billion extra people. Our scenario, A25, represents a case where the protein intake from animal-based foodstuff is decreased by 47% (see table 4). This would lead to a reduction of green water use by 11%, comparable with the impacts of halving food losses (Kummu et al 2012). As green water availability depends on the hydrological cycle, its use is, in the end, a measure of cropland use. In terms of blue water, savings similar to those of halving food losses would require total abstinence from animal products (i.e., scenario A0, 14% decrease). Further, Foley et al (2011) estimated that shifting away from animal-based foods could add up to 49% to the global food supply without expanding croplands. This is a maximum savings potential that assumes full abstinence from animal products and no qualitative diet optimisation. Foley et al (2011) also noted that some amount of animal-based foods in diets provide benefits, and they suggest moderating their use instead of abandoning them. Our results can be viewed as quantifying some of the impacts of the moderation suggested by Foley et al (2011), while also taking dietary requirements into account.

Differences in resource savings between our study and the others presented are due, at least in part, to the fact that our analysis was based on diets conforming to the WHO nutritional recommendations. Not only did we scale the energy intake to a locally recommended value, but we also shifted the balance between food item groups to represent a qualitatively acceptable diet. This improvement had a cost in terms of water resources, as without it, water savings in scenario A0 would have been 4 percentage points higher for green and 1 percentage point higher for blue water. Moreover, our approach was based on existing dietary habits and assumed the smallest changes possible to reach the scenario criteria (table 2). We also took food losses and waste into account in our calculations. Further, we included the irrigation (i.e., blue water use) of pasture and grass feed, which are not taken into account in other studies based on the same water footprint data (Mekonnen and Hoekstra 2010). Without this additional water use, blue water reduction would be negligible in A50 and A25, and even full abstinence from animal protein would only save 7% blue water. Therefore, our approach resulted in an increased understanding of the practical potential for worldwide water savings by dietary choices.

The relative importance of water savings with diet change strongly depends on the available water resources of a specific country. To quantify this, we compared the water use change of scenarios A25 and A0 (in relation to the RD scenario) to available green-blue water (GBW) resources (Kummu et al 2014) at the country level. The largest relative savings from diet change were achieved on water scarce areas in the Middle East (up to 371% in A25 and up to 559% in A0), Africa (34% and 79%), Central and East Asia (40% and 74%) and in some Latin American countries (17% and 25%) (figures 7(A) and (B)).

Zoom In Zoom Out Reset image size

We further assessed the potential of diet change to alleviate GBW scarcity by calculating its impacts on the frequency of GBW scarcity, calculated by (Kummu et al 2014), who based their calculations on the method developed by Gerten et al (2011). Although our RD diet differs somewhat from the reference diet by Kummu et al (2014), we assumed that the relative changes in water use could be applied to their GBW scarcity results. It should also be noted that the approach by Kummu et al (2014) assumes a scenario where all food is produced locally, whereas in our approach, food trade is accounted for. We present the results in percentage points, representing the decrease in the percentage of years with water scarcity (figures 7(C) and (D)). The impact is greatest in Africa, although even there, diet change alone is not sufficient to completely eliminate the GBW scarcity. In total, the A25 scenario removed the GBW scarcity from 67 million people and reduced the frequency from 376 million people, whilst for the A0 scenario, these were 168 million and 637 million, respectively.

The savings in water consumption come from several factors. The total dietary energy consumption was kept constant between the scenarios A50–A0, but the nutritional composition was allowed to change within the range of WHO recommendations (see table 1).

This resulted in the protein and fat content of the food intake decreasing (11% and 36%, respectively, for A0 vs RD) when the use of animal products was restricted. This shifts the consumption towards carbohydrates that, in practice, provide the rest of the food energy. To quantify the impact of reduced fat and proteins, we generated the diets of the A50–A0 scenarios as well, which included constant protein and fat content but was still within the optimisation constraints. This was not possible for all of the countries; however, by pushing the optimisation as far as was practical without manual intervention, a global increase of 7% in green water use was observed in A0. With blue water, the difference was minimal: under +1%.

It can thus be said that the water savings we observed were partly realised because animal products were replaced from the energy perspective, and protein content was not preserved. Nevertheless, the scenarios follow the dietary recommendations used; and while for some countries, the calculated food composition may not be completely practical, the results offer a meaningful insight into the potential of decreasing water consumption in food production without sacrificing nutritional needs.

In this analysis, we concentrated on the impacts of different diets on water use, but there are many factors that affect the practicality of replacing the products of one food group with those of another. For example, soil and fertiliser requirements may differ from one group to another. Greenhouse gas emissions of different foods is another factor to be considered, although according to studies (Pradhan et al 2013, Smith et al 2013), a shift towards diets with less animal products would be beneficial also from this point of view.

We acknowledge that there may be overwhelming obstacles in changing the food production chains, as well as convincing consumers to suddenly and drastically change their eating habits. This is why the existing consumption patterns within each food item group were kept intact when shifting the intake between them. Moreover, this article merely concentrates on estimating the potential to reduce water use by changing diets, without assessing the social or technical aspects of it.

We assumed throughout that agricultural practices would remain unchanged, which means that water footprints of different food item groups per mass unit would remain similar, even though total production amounts might change. This is not necessarily true. For example, if an increasing amount of a specific agricultural product is needed, the production may spread to less favourable areas that are not able to sustain the crop or livestock with rainwater, which would increase blue water consumption. It is important to notice that the increase in blue water consumption in many countries in our scenarios is not due to this factor, as it is currently unaccounted for—it is simply the result of a shift between the food groups. With regard to animal-based food, we assumed that the ratio of different types of production systems stays constant throughout the scenarios. As water footprints of grazing livestock are often larger than those of animals kept in an industrial setting (Mekonnen and Hoekstra 2012), there is the potential for resource optimisation in future studies.

If water and cropland savings are sought, the practical steps should be considered carefully. Permanent meadows and pastures that mainly provide feed for grazing livestock account for almost 70% of the world's agricultural land (FAO 2013), but they are not necessarily readily converted for crop production or other uses. Therefore, constraining animal protein production in these areas may not produce concrete resource savings and might even increase the pressure on crop production and thus resources. From this perspective, only livestock fed using directly human-edible crops or feed competing with them for agricultural land should be targeted. On the other hand, the total area used for agricultural production is also important. As pristine natural ecosystems are still being phased out by agriculture (DeFries et al 2010), the efficiency of production in terms of cropland area should be improved (Godfray et al 2010, Tilman et al 2001).

Further, the effect of diet changes on global trade patterns might have unexpected environmental and economic consequences. Foodstuffs replacing meat protein may not be practical to produce in the current meat exporting countries. This restructuring of the global food system would deserve serious attention not only from a natural resource and nutritional perspective, but also from the perspectives of employment, the security of supply and self-sufficiency.

Limiting animal-based food content often helps to decrease agricultural sprawl, but as we have shown, the amount of resources required to produce an acceptable diet are not the same everywhere. This, together with the varying availability of those resources, makes the building blocks for sustainable food composition—whether domestically produced or imported—inherently local. A step forward would be to assess how food production could be optimised in each country, or even globally, from an available resources point of view.