Background & Summary

Agriculture requires around 90% of the global freshwater use and accounts for nearly 70% of the global water withdrawals, as widely reported in the literature1. The livestock sector is estimated to cover between 30–40%2,3 of this use to provide us 16% of our global food supply through meat, eggs, and dairy products4.

In the agri-food sector, the concept of water footprint (‘WF’) is widely used to indicate a volumetric measure of the direct and indirect water consumed to obtain a unit of agricultural product, considering its entire supply chain5. It is a multidimensional indicator, geographically and temporally defined. When accounting for the production of plant-based foods, the WF is strictly associated with the cultivation of the primary crops directly consumed by humans. On the other hand, the WF associated with animal-source foods is not only related to the drinking and service water for the animal itself but also to the indirect consumption associated with the primary crops fed to livestock along their entire life cycle6.

Three factors, i.e., efficiency, production system, and livestock diet, influence the WF of animal products the most. These three factors are strictly interconnected and they influence one another. Efficiency usually refers to the ratio between inputs and outputs, i.e. the amount of feed needed by each animal species to obtain a unit of animal product. The type of production system depends on the scale, purpose, and nature of the farming enterprise, which translates into extensive and intensive systems and several kinds of intermediate systems between these two (semi-intensive, semi-extensive systems). Lastly, the livestock diet refers to a balanced ration which represents the amount of feed that supplies the proper type and proportions of nutrients needed for an animal to perform a specific purpose while staying healthy and ensuring its well-being. Despite efficiency and production systems being strictly dependent on livestock species and capital investments, respectively, both are greatly influenced by the animal diet adopted by farmers to feed their livestock.

The international scientific literature is unanimous in setting that the feed production phase is responsible for almost the entire water demand associated with this sector (~98%)3. Despite being widely recognized, few investigations were done on the specific WF not only of primary crops but also of their by- and co-products7, and, to the best of our knowledge, no specific studies on the WF of crop products when used as feed, i.e. taking into account only the mass fraction effectively ingested by livestock. Several studies investigated the water use associated with animal production and animal-source food consumption in different conditions and scenarios, for specific livestock species8,9, most of the time on a local or regional scale10,11,12,13, while just a few focused on a global comprehensive quantification2,3,14,15. Furthermore, it should be noted that rarely do global studies evaluate the water use and/or water footprint separately in its two consumptive components of green and blue water. Mekonnen and Hoekstra3 and Heinke et al.2 reported the WF of animal products in its components, obtained by modeling the water used to produce the feed crops fed to animals and using agricultural crop yields adapted from the national statistics from FAO4. FAO yields data, however, do not account for differences in agricultural water management such as the use of irrigation, which is usually responsible for substantially higher yields compared to rainfed production of some crops (e.g., soybean, sugarcane, and rapeseed)7.

Here we address the gaps found in the literature by assessing the WF for 140 animal feed products produced in 262 countries. Specifically, the WATNEEDS16 model was used to get outputs on the green and blue water requirements of primary crops from which the feed products are derived from. The model was run on specific areas, based on maps of spatially distributed crop-specific rainfed and irrigated harvested areas, and using climatic data referring to the period 2015–2019. The model outputs were combined with crop- and system-specific yields (the distinction being made between yields in rainfed and irrigated conditions), feed use fractions, and conversion factors to obtain the green and blue water footprints in the two different agricultural systems. Specifically, we estimated the green water footprint in rainfed areas and the green and blue water footprint in irrigated areas.

Methods

The methodology is outlined in every step and divided into subsections to provide a detailed description of the sources used, the assumptions made, the model employed, and the resulting database of water footprint records.

The analysis was performed on a global scale for 262 countries and 140 feed products, based on climate data averaged between 2015 and 2019. The results aim to be representative of the current use of water associated with the adoption of a unit of measure of the different feed products in livestock diets according to their origin.

Feed products and primary crops

A list of 140 feed products of plant origin was selected from more than 200 feeds and 100 mineral sources reported in the Feed Tables17 (Table 1). All of the feeds of animal origin (animal and dairy products), mineral products, amino acids, fodder feeds from alfalfa, and feeds from the bakery industry were excluded. For all the feed products fed to livestock in the form of by-products or crop residues and not as a whole crop, it was defined the primary crop from which these feeds are derived from (e.g., brewers grains derived from barley, wheat gluten feed from wheat) (Table 1). Thus, when different co-products are obtained from one crop (e.g. straw, oilseed cakes, molasses, pulps), an allocation method was used to earmark the different shares of water use for each crop product5,185. The approach suggested in the Water Footprint Assessment Manual5 assigns the share of water use attributable to each by-product by referring to the agricultural yield and the water demand of its primary crop. The most suitable environmental allocation method for this kind of study is widely discussed. The literature reports three methods according to mass, energy content, or economic value as the most acknowledged. Despite the Water Footprint Assessment Manual5 recommending the economic allocation, more recent studies suggest considering the metabolizable energy (ME) content when dealing with co-products intended for livestock feed18. Furthermore, the dependency of economic allocation on price fluctuations is often controversial compared to mass and energy allocation, which are stable overtime18 (see increasing oilseed prices after 2020 due to the effect of Covid19 pandemic in Supplementary Information). The ME value indicates the fraction of energy an animal can metabolize for each feed product and depends on the animal species19. However, since data on prices and metabolizable energy are not available for all the 140 feed products included in the analysis, the energy and economic allocation was implemented for oilseed and their co-products, while mass allocation was used for the main analysis. Oilseeds are mainly processed into cake and oil co-products; while cakes represent the highest mass fraction among the co-products, oilseed oils have both higher market values and higher metabolizable energy values. Indeed, the difference between adopting mass, energy, or economic allocation methods is more relevant than for the other feed products.

Table 1 Main feed products included in the analysis (full list of 140 feed products in Supplementary).
Full size table

The main analysis uses mass factors estimated as a combination of the relative mass MF, accounting for the feed material mass as a fraction of the total mass of the crop, and the feed use fraction (FUF), accounting for the fraction of the yield that is effectively ingested and used as feed by the animals. Dry matter content (DM) was taken into account for dried products as well as moisture content for fresh products.

The energy and economic allocations were done using ME (values for ruminants and monogastric) and price values, respectively, and the extraction rate of cake meal during the oil processing.

Water footprint and the WATNEEDS model

Calculations on the volume of water associated with the production of each feed product and the consequent WF were performed using the WATNEEDS model16, an agro-hydrological model that is crop-specific, spatially distributed, and physically based. The model solves the vertical soil water balance at a 5 arc-min resolution to return a spatially distributed crop-specific monthly analysis of green (GW) and blue water (BW) requirements. The GW represents the water volume supplied directly by precipitation and then absorbed as soil moisture by roots, while BW is the volume provided by irrigation5. With this model, it is possible to estimate the actual evapotranspiration (ET) and calculate the crop water requirements (CWR), and the irrigation requirements (BW) based on soil and crop-specific data, and climate inputs. The initial release of the WATNEEDS model (26 crops from the MIRCA2000 dataset20) was adapted to simulate the 42 crops included in the SPAM2010 updated dataset21. Further details on model inputs data sources and model components are available in Chiarelli et al.16.

The WF (green water footprint, GWF, and blue water footprint, BWF) was calculated for each feed product i, for each country c, and for each agricultural system s = {rainfed, irrigated, rainfed with low agricultural inputs, rainfed with high agricultural inputs} following the approach reported by Hoekstra et al. in the Water Footprint Assessment Manual5:

$${{GWF}}_{i,c,s}\left[\frac{{m}^{3}}{{ton}}\right]=\frac{{{GW}}_{i,c,s}\,[{mm}]\times 10}{{{yield}}_{i,c,s}\,\left[\frac{{ton}}{{ha}}\right]\times {DM}\left[ \% \right]}\times {{AF}}_{i,c,m}[ \% ]$$
$${{BWF}}_{i,c,s}\left[\frac{{m}^{3}}{{ton}}\right]=\frac{{{BW}}_{i,c,s}\,[{mm}]\times 10}{{{yield}}_{i,c,s}\,[\frac{{ton}}{{ha}}]\times {DM}[ \% ]}\times {{AF}}_{i,c,m}[ \% ]$$

where AFm is the allocation factor (whether it is m = mass, energy content, or economic value), DM is the dry matter content of each primary crop (conversion from fresh to dry matter yield), while the monthly gridded outputs of GW and BW requirements obtained with climate and soil-gridded data were annually aggregated by taking harvested area-weighted averages to obtain mean country values for the period 2015–2019 (mm yr−1). Similarly, gridded data on crop-specific yields for the different agricultural systems were averaged on the corresponding crop-specific national harvested areas (ton ha−1). While GWF was estimated for all the agricultural systems, the BWF was only for the case of irrigated agriculture.

The allocation factor was estimated for each feed product, country, and for the different methods (mass, energy, economic), as follows:

$${{AF}{\rm{\_}}{mass}}_{i,c}=\frac{{{MF}}_{i,c}\times {{FUF}}_{i}}{{M}_{{pc}}}$$
$${{AF}{\rm{\_}}{energy}}_{i,c}=\frac{{{ME}}_{i,a}\times {{MF}}_{i,c}}{{\sum }_{1}^{n}{({ME}}_{i,a}\times {{MF}}_{i,c})}$$
$${{AF}{\rm{\_}}{economic}}_{i,c}=\frac{{p}_{i}\times {{MF}}_{i,c}}{{\sum }_{1}^{n}({p}_{{\rm{i}}}\times {{MF}}_{i,c})}$$

where MF is the mass fraction of each feed product, M is the mass of the primary crop pc (equal to the sum of the MF of all co-products of a crop), FUF is the feed use fraction (equal to 1 for all feed products excluding straws and crop residues), ME is the metabolizable energy depending on the animal species a = {ruminants, monogastric}, and p is the reported international FOB price.

Data sources

The list of feed products was retrieved from the INRAE-CIRAD-AFZ Feed Tables17, developed by the collaboration of the Institut National de la Recherche Agronomique (INRAE), the Centre de Coopération Internationale en Recherche Agronomique pour le Développement (CIRAD), and the Association Française de Zootechnie (AFZ).

For the selected years, the model inputs include potential reference evapotranspiration (ETo) (University of East Anglia’s Climate Research Unit Time Series version 4.01 dataset22) and daily precipitation (Climate Hazards Group InfraRed Precipitation with Station v. 2.0 CHIRPS dataset23; National Oceanic and Atmospheric Administration’s Climate Prediction Center Global Unified Gauge-Based Analysis of Daily Precipitation CPC dataset24). Spatially distributed (5 arc minute resolution) irrigated/rainfed crop-specific harvested areas and yields were retrieved from IFPRI’s Spatial Production Allocation Model (SPAM) for the year 201021. Additionally, areas and yields from the rainfed high- and low-input farming systems were included in the analysis21. Soil type information was taken from Batjes25, based on soil data from the FAO Harmonized World Soil Database. Crop coefficients and growing stages were taken from Allen et al.26 and adapted from Siebert and Döll27.

The mass fraction values for the commodities are crop- and country-specific when data was available otherwise national average was used; data were taken from multiple sources, including mainly FAO28, Feedipedia29, and other sources30,31. The FUF of each product was taken from the latest version of the GLEAM model (v. 3.0) developed by FAO30. The DM content was taken from the Feed Tables17.

Concerning the other allocation methods, the ME content for ruminants and monogastric species (data on adult and growing pigs were averaged) were taken from the Feed Tables17, while FOB prices of oilseeds and their co-products were retrieved from reports of the USDA32,33.

Data Records

The water footprint records for the selected 140 feed products and 262 countries for the 2015–2019 average period are available in Figshare as an Excel file (https://doi.org/10.6084/m9.figshare.27324984)34.

The file contains a Summary of data included in the sheets. The results are presented for each agricultural system s and allocation method m.

The sheets are named as “ xWF_s_m”:

  • x refers to the green (G), blue (B), or total (T, equal to G + W) water;

  • s refers to the agricultural system rainfed (R), rainfed low-input (R_LOW), and rainfed high-input (R_HIGH), or irrigated (I);

  • m refers to the mass (M), energy (EnR, EnM for ruminants and monogastric, respectively), and economic (Eco) allocation method.

The Data sheet contains information on the selected feed products, the primary crops, the crop groups, the dry matter content, and the allocation factors of each product. The variables are reported in Table 2.

Table 2 Overview of variables included in the dataset on the national water footprints of livestock feeds.
Full size table

Figures 1–3 and Table 3 show the GWF and BWF in rainfed and irrigated conditions for 47 out of 140 selected feed products among the most consumed as animal feed and for 40 out of 262 countries where the cropland area is higher than 5 million hectares and covering 89% of the global cropland area.

Fig. 1
figure 1

Green water footprint in rainfed conditions (GWF_R) for the main agricultural producer countries and the most consumed feed products. Results for all the feed products and countries in the full dataset.

Full size image
Fig. 2
figure 2

Green water footprint in irrigated conditions (GWF_I) for the main agricultural producer countries and the most consumed feed products. Results for all the feed products and countries in the full dataset.

Full size image
Fig. 3
figure 3

Blue water footprint in irrigated conditions (BWF_I) for the main agricultural producer countries and the most consumed feed products. Results for all the feed products and countries in the full dataset.

Full size image
Table 3 Global average of GWF and BWF, in rainfed and irrigated conditions (GW_R, GW_I, and BW_I, respectively).
Full size table

Technical Validation

The use of the WATNEEDS model was tested and validated by comparing the outputs around the year 2000 with the ones estimated by Siebert and Döll27 and obtaining a discrepancy lower than 3% for the cumulative results, while a crop-by-crop pixel-by-pixel comparison showed a difference lower than 20% for 90% of the harvested area between the two datasets. The results of the technical validation and sensitivity analysis for the year 2000 are reported and described in Chiarelli et al.16. Another comparison was made on the crop water use (CWU) obtained with the WATNEEDS model for 15 selected primary crops for the year 2016 with the results of Mialyk et al.35 averaged on 2010–2019. The results showed that both rainfed and irrigated CWU were generally smaller than our estimates, as explained in the “Comparison of crop water use” of Mialyk et al.35. Furthermore, the outputs of the same authors were compared with the updated results of the WATNEEDS model on the new rainfed and irrigated areas and the new list of crops from SPAM201021 (see Data Sources) averaged on 2015–2019 for 33 primary crops, used in this and other analyses36. The comparison confirmed the generally smaller estimates for Mialyk et al., as it was reported for the year 2016. The large regional differences are reported in Supplementary Information for the 6-most-used livestock feed crops (maize, wheat, rice, soybean, rapeseed, sunflower seed) for rainfed agricultural systems.

Regarding the estimated WF records, we provide a comparison with two other global studies: one providing annual data from 1990 to 201535, and another reporting an average around the year 20007. The comparison was conducted for 15 primary crops, as the WFs of all by-products, co-products, and crop residues used as livestock feeds are derived from the respective primary crops through the application of one of the presented allocation methods. Furthermore, we provide the comparison for BWF and GWF both in terms of WF per unit of product (in m3/ton) for both production- and harvested area-weighted averages, and in terms of global production WF (in km3) for the selected 15 crops. The unit WFs are averaged on the harvested area reported by FAO4 for 2015–19 and splitted in irrigated and rainfed areas proportionally to data from SPAM2010, while production was obtained by multiplying the same areas for the irrigated and rainfed yields21. Global production WF was obtained by multiplying the country-estimated BW and GW (in mm) by the corresponding country- and crop-specific harvested area, for rainfed and irrigated production systems separately, and by a factor of 10−8, then summing up all the national production WFs to obtain the global value in cubic kilometers (km3). Note that while Mialyk et al. report a total value for GWF (GWF_t) and they distinguish between BWF from irrigation (BWF_I) and from capillary rise (BWF_t – BWF_I), on the other hand, Mekonnen and Hoekstra estimate just the BWF for irrigation (BWF_I) but they distinguish between GWF from rainfed (GWF_R) and irrigated (GWF_I) production systems, as we did. The comparison shows how our production WFs are generally larger than those estimated by Mialyk et al., as already discussed for the CWU, but also larger than those reported by Mekonnen and Hoekstra. However, looking at the variations in the total harvested area from 1996–2005 to 2015–20194, we obtained a good agreement for GWFs of production. Maize, rice, cassava, soybean, sunflower seed, rapeseed, groundnuts, and sugarcane show a net increase, while barley, millet, potatoes, and sugar beet have a net decrease in total GWF, in line with the reported variations in the harvested area (Table 4). On the contrary, the total estimates of production BWFs turn out to be generally smaller despite an increase in total irrigated area. Concerning unit WFs comparison, better agreement is obtained for production-weighted averages WFs, in particular for GWF. Table 5 shows how the difference in GWF and total WF between studies almost always varies in the range of ± 30%, which is an expected and accepted interval according to a study on sensitivity and uncertainty in crop water footprint accounting37. The same authors underline how BWF is usually more sensitive and with higher uncertainty than GWF.

Table 4 Comparison of production WF for 15 main crops to other global studies7,35 in cubic kilometers (km3).
Full size table
Table 5 Comparison of unit WF for 15 main crops to other global studies7,35 in cubic meters per tonne (m3/ton).
Full size table

Usage Notes

Globally, the highest contribution to the WF of products is given by rainwater, i.e. GW component, being the only water available for rainfed agriculture (around 80% of the harvested area) and covering on average 60% of the CWR in irrigated agriculture. Overall, higher GWF and BWF are obtained for primary crops rather than by- and co-products since the water use is entirely allocated to the crop product whichever allocation method is adopted. Higher values of GWF are estimated in rainfed conditions due to lower agricultural yields reached without supplying irrigation to the plants. The highest values of GWF are estimated for cottonseed and cocoa products, followed by some pulses (e.g., cowpea) and millet, which are crops characterized by low crop yield (around 1 ton/ha on average) and high CWR. On the other hand, when moving to BWF pulses stand out as BW-demanding crops, including beans, faba beans, lentils, lupins, vetches, and peas.

Outputs highlight how the best practice results to be that of using agricultural by-products as animal feeds to save water resources while keeping agricultural leftovers in the food chain and reallocating primary crop use to direct human food consumption. However, the extent to which this practice could be adopted in the livestock sector remains closely linked to conditions such as the different animal species, the nutritional needs in the various animal life stages, the production systems, and the local or regional availability of agricultural by-products. While the use of somekind of agricultural by-products is already widespread in extensive and semi-extensive systems often due to their economic convenience, wider room for growth could be found in the industrial systems where by-products could replace some high-resource-demanding concentrate feeds38. In this context, monogastric species such as pigs are ideally the most suited species due to their omnivorous diet.

Potential applications

The dataset of water footprint is intended to be utilized for multiple purposes. An estimate of the water use associated with the production and consumption of animal-source food is useful to make both producers and consumers aware of what they produce and what they eat, of the environmental impacts of the sector, and of the current and future issues that such trends may induce in several regions, e.g., situations of water scarcity. On the other hand, an estimate of the water footprint of different kinds of animal feeds might be helpful to guide farmers in choosing the less water-demanding diet for their livestock while ensuring sustainability standards for their production. Furthermore, the dataset gives insights into the existing differences in terms of water productivity and water use efficiency among feed products but also among countries. The outputs, in fact, might be used also for virtual water assessments when feed products are traded among countries such as for soybeans.

This approach can be seen as a method to tackle the problem at its roots by preventing, or at least reducing, inefficiencies and the waste of water resources in the sector when possible. However, the dataset should always be consulted together with the nutritional profiles of each feed product in order to check the most suitable and efficient strategies to ensure the health and well-being of animals, people, and the environment.

Uncertainties and limitations

The results are affected by several uncertainties. Addressing these uncertainties would necessitate more simulations with different crop models and input datasets. However, this level of analysis exceeds the aim of our study, which is limited to estimating the WFs of livestock feeds. As such, we do not attempt to quantify the uncertainties in this section. Instead, we provide a brief overview of their key sources and offer suggestions for minimizing them in future studies. The main source of uncertainty arises from the quality and resolution of input data. The quantification of the CWR and the two separate requirements of GW and BW with the WATNEEDS model is sensitive to several factors (initial condition of soil moisture, Kc values, planting dates and crop calendars, multiple growing seasons, input of climatic and soil data, and others) to which are added further uncertainties when accounting for the WFs of crops. While sensitivity analysis and potential uncertainties and limitations on the use of the WATNEEDS model are reported in Chiarelli et al. and in the comparison of crop water use reported and discussed by Mialyk et al., other additional factors affect the next steps of the analysis to account for the WFs values. Two additional factors mainly influence the final result of WF:

  • data on estimated rainfed and irrigated crop yields, here taken from SPAM2010; while rainfed and irrigated harvested area from SPAM2010 was scaled to match the total average harvested area in 2015–19 from FAO, crop yields were not scaled and harmonized to the average study period;

  • the adopted method of allocation for the by- and co-products and crop residues.