Accounting for the water impacts of ethanol production

Biofuels account for 1–2% of global transportation fuel and their share is projected to continue rising, with potentially serious consequences for water resources. However, current literature does not present sufficient spatial resolution to characterize this localized effect. We used a coupled agro-climatic and life cycle assessment model to estimate the water resource impacts of bioenergy expansion scenarios at a county-level resolution. The study focused on the case of California, with its range of agroecological conditions, water scarcity, and aggressive alternative fuel incentive policies. Life cycle water consumption for ethanol production in California is up to 1000 times that of gasoline due to a cultivation phase that consumes over 99% of life cycle water use for agricultural biofuels. This consumption varies by up to 60% among different feedstocks and by over 350% across regions in California. Rigorous policy analysis requires spatially resolved modeling of water resource impacts and careful consideration of the various metrics that might act to constrain technology and policy options.


Introduction
Faced with erratic petroleum prices, security concerns, and climate change, governments across the globe have implemented policies aimed at increasing the volume of alternative fuels in the energy mix. These policies have helped lead to a six-fold increase in biofuel production over the past decade. Biofuels currently represent 1-2% of the global transport fuel and are projected to continue expanding rapidly (REN21 2008, IEA 2009.
The timeline laid out in the current version of the 2007 United States Energy Independence and Security Act (EISA) targets 9 billion gallons of renewable fuels for 2008 and 36 billion gal (or about 20% of total vehicle fuel consumption) 4 Author to whom any correspondence should be addressed. by 2022 (US Congress 2007). In a similar vein, the European Commission (EC) recently agreed to a proposal to increase its binding volumetric target to 10% renewables in the transport fuel mix and 20% of all energy by 2020 (UNCTAD 2008, European Commission 2009. As production of biofuel continues to expand, especially given that this expansion is largely driven by policy, the environmental and social implications of these fuels have come under scrutiny, especially their life cycle greenhouse gas (GHG) emissions. Recognizing that emissions vary greatly depending upon feedstock and production methods (Farrell et al 2006) and that at least some biofuels have considerable 'indirect' emissions (for example, from remote land use change mediated through the international food market) (Hertel et al 2010, Searchinger et al 2008 many governments have begun promulgating policies using life cycle assessment (LCA) to Much less attention is being paid to water resources, but biomass energy carriers are usually orders of magnitude more water-intensive than conventional energy sources (King and Webber 2008, Chiu et al 2009, Gerbens-Leenes et al 2009, Wu et al 2009, Service 2009). This concern could be, as a report from the Institute for Agriculture and Trade Policy pointed out, the 'Achilles heel' of biofuel production (Keeney and Muller 2006).

Water scarcity
Population growth and dietary changes are projected to drive a 70-90% increase in demand for water worldwide in the next 50 years (Molden 2007), even as new supplies are dwindling. Concerns about environmental and social impacts have curtailed new reservoir construction while existing reservoirs are being lost to siltation and groundwater lost to overdraft at a rate of about 60-70 km 3 yr −1 (Serageldin 2001), or about half the capacity of Egypt's Aswan High Dam reservoir. By 2025, 1.8 billion people are predicted to be living in absolute water scarcity and 2/3 of all people to be experiencing some water stress (UNEP 2007).

Water and biofuels
Water is consumed all along the biofuel supply chain. Figure 1 shows the major uses of water in both the agricultural and industrial phases of biofuel production. Some of the water applied as irrigation runs off or infiltrates and is later available for other productive uses. Furthermore, some of the water consumed in agricultural production is so-called 'green water'-naturally available soil moisture. To avoid confusion, and to provide a comprehensive accounting of water use, this analysis uses a common definition of water consumption; water is considered consumed if it is removed from any potential further use for the remainder of one hydrologic cycle.
Two types of consumption are considered in this analysis: • Evapotranspiration (ET)-combination of evaporation and transpiration.
• Industrial/biorefinery consumptions-water consumed in industrial processes through uses such as cooling and incorporation into finished products.
Many studies of the water resource impacts of biofuel production only consider uses at the biorefinery (NRC 2007, Keeney andMuller 2006). While industrial consumption is important, especially for local water resources, this method does not account for agricultural water consumption and therefore does not fully characterize the life cycle effect of biofuels on water resources.
Some recent studies (King and Webber 2008, Chiu et al 2009, Service 2009, Wu et al 2009 include agricultural consumption in their analyses, but only account for water that is applied to fields through irrigation. Considering only irrigation water implies that the 80% of global agriculture, and 85% of US corn production, that is exclusively rainfed consumes no water. While irrigation water is a vital resource, rainwater is also of value, and if not devoted to biofuel feedstock production could be allocated to other crops, to environmental services, or to reservoir and/or groundwater recharge (Molden 2007).
In studies considering all evapotranspiration (Gerbens-Leenes et al 2009), lack of spatial resolution leads to an illusion of uniformity in what is actually a very heterogeneous system. Owing to differences in crop physiology, management, energy yield, and climate, the amount of water required to produce a gallon of biofuel varies spatially. Furthermore, because water availability also varies over space and time, the implications of consuming a given volume of water will not be uniform. This is an important difference between water resource impacts and GHG emissions, which have essentially uniform and widespread effect wherever and whenever they occur.
In this paper we develop a quantitative framework for evaluating the water resource effects of biofuel expansion for use in LCA and policy analysis. We take as a case study the state of California, which has been a leader in the development of LCA-based fuel policies and which, through its wide variety of agricultural systems and climate types, allows us to draw robust and broadly applicable conclusions. California is also facing severe water supply and allocation challenges. We propose a list of quantitative metrics to enable rigorous analysis of the water impacts of bioenergy production and policy for a variety of concerns and contexts.

Experimental methods
We used the Penman-Monteith model, a well-established crop evapotranspiration model that uses plant physiology and climate data to calculate water consumption on a daily timestep (Allen et al 1998). These calculations were performed at a county-level resolution in order to capture the spatial heterogeneity of water resource requirement for bioenergy production in California.
The calculated ET was then incorporated into a life cycle assessment of biofuel water consumption.

Evapotranspiration modeling
Crop water consumption is estimated in many studies by calculating ET using the Penman-Monteith model developed by the UN Food and Agriculture Organization (Allen et al 1998, Chapagain and Hoekstra 2004, Hoekstra and Chapagain 2007, Gerbens-Leenes et al 2009. As shown in equation (1), Penman-Monteith estimates ET as the product of a reference crop evapotranspiration (ET o ) and a crop coefficient (K c ). ET o characterizes climate effects and is calculated from temperature, solar radiation, wind speed, and relative humidity. K c accounts for the effect of characteristics such as crop height, surface coverage, and albedo, that distinguish a crop from the reference surface (Allen et al 1998).
where ET c is total evapotranspiration (mm day −1 ) from crop c; K c is a physiological crop constant that varies from 0 to 1; and ET o is reference crop evapotranspiration (mm day −1 ), representing ET from a grass surface of 0.12 m height that is calculated as follows (equation (2)).
We applied the Penman-Monteith model, to calculate crop embedded water, using the Consumptive Use Program (CUP)-the Penman-Monteith model parameterized and refined for California by the California Department of Water Resources. This model has been validated using nine years of data from the instrumentation network of the California Irrigation Management Information System (Orang et al 2005). CUP uses measured monthly solar radiation, maximum, minimum, and average temperature, dew point, and wind speed to compute ET o for climatic regions within California. K c values are adapted from values published by the United Nations Food and Agriculture Organization (Doorenbos andPruitt 1984, Allen et al 1998) for analogous crop types. We did not apply a stress coefficient (K s ). In other words, we estimated ET for standard conditions, which assumes crops achieve full production for the given climate (Allen et al 1998). Under water stress, both ET and yield will be reduced, such that the ET/yield ratio reported here is not expected to be highly sensitive to water stress above the wilting point.

Production scenarios
Increased ethanol production could be met in a variety of ways. In the US, common energy crops include corn, other grains, and sugar beets. Possible second-generation feedstocks include agricultural waste, municipal solid waste, and dedicated biomass energy crops such as switchgrass and miscanthus.
County-level agricultural production and per-acre productivity data used in this study are drawn from the USDA National Agricultural Statistics Service's (NASS) agricultural census. NASS county-level production data are also used for weighting of statewide averages. All counties in California were used in the study, though individual feedstocks were only considered in those counties where they (or comparable crops in the case of biomass feedstocks) are currently grown.
We assume that field crops will be displaced for feedstock production-specifically corn, wheat, rice, sorghum, barley, oats, cotton, beans, and fodder crops. These are low value crops, and are therefore more likely to be replaced by biofuel feedstocks than are higher value crops such as fruits and vegetables.
Data on refining processes and outputs are drawn from the EBAMM model (Farrell et al 2006), the GREET model (Wang 2009), and the NREL model biorefinery (Aden et al 2002).

Case-study feedstocks.
We modeled four feedstock types: corn, sugar beet, a low-yield biomass energy crop (LYB) similar to the low-input high-diversity grasslands investigated by Tilman et al (2006) and a high-yield biomass crop (HYB) representing dedicated energy crops such as miscanthus.
Corn grain is the primary feedstock for biofuel production in the United States. In California, plantings rose almost 30% between 2006(USDA 2008 with the majority of production going to supply the state's livestock operations. California produces among the highest sugar beet yields globally; only 25 000 acres were dedicated to sugar beet cultivation in the state in 2009 but this figure could expand to exceed the 1970 high of 300 000 acres (USDA 2008).
No sufficiently broad field tests of perennial grasses such as miscanthus (Miscanthus giganteus) and switchgrass (Panicum virgatum) to date provide reliable data on their water productivity. Furthermore, existing crop evapotranspiration models are calibrated for current crop systems and have not yet been applied to most biomass crops. As a result, this analysis uses two hypothetical biomass crop feedstocks-one low-yield and one high-yield-using outside data to project biomass yield, water consumption, and ethanol productivity.
The low-yield biomass (LYB) crop is modeled here on grassy fodder crops (hay and haylage) currently grown in the state. Similar to lignocellulosic feedstock crops, fodders have been bred and cultivated to maximize total plant biomass rather than one specific plant product as is the goal with most crops.
The average productivity of these crops is approximately 8.2 dry tonnes of biomass per hectare annually-similar to the yields anticipated from low-input high-diversity grasslands (Tilman et al 2006).
High-yield biomass (HYB) crops in this analysis are modeled as producing 20 dry tonnes of biomass per hectare on average annually after Williams (2006) with the relative yields in various California regions modeled on common biomass crops currently grown in the state. This is comparable to yields predicted for energy crops such as miscanthus (Hastings et al 2009). The water consumption dynamics of these hypothetical HYB crops are modeled using biomass crop water use efficiency values developed by Berndes (2002).

Case-study exclusions.
We did not analyze the embedded water in agricultural residues used as a feedstock because there was no established method or empirical constraints for allocating embedded water among products and coproducts. Sugarcane, while a major feedstock for ethanol production globally, is not grown widely in California and so is not analyzed here. Biodiesel is also not considered because soy, the primary agricultural feedstock for biodiesel in the US, is grown in such small quantities in California that its area is not reported in the USDA agricultural census.
We do not take into account indirect effects of feedstock production on water resources in which water consumption is altered far from the production site by market-mediated land use change (Searchinger et al 2008, Hertel et al 2010. A standard method for their quantification has not been developed and is the subject of much debate. Also, although pollution can reduce the amount of available water, we did not evaluate it here.

Results
In all the feedstock cases studied, the agricultural production phase represented more than 99% of life cycle water consumption on average with biorefineries consuming less than 1% of total ethanol embedded water. Our analysis shows a clear difference in fuel embedded water among the feedstocks modeled. Table 1 shows the average embedded water in ethanol from each of the feedstock crops weighted by county feedstock production (equation (3)). The statewide average embedded water in ethanol from the feedstocks studied varied by more than 50% due to plant physiological properties and where individual feedstocks are grown. The production-weighted averages presented in table 1 are derived from a dataset exhibiting substantial county-level heterogeneity in both yield and crop ET. Figure 2 illustrates the breadth of values seen for these crop characteristics.  The feedstocks studied varied in both their water consumption and their yield per hectare. This result leads to both spatial and feedstock-related variation in water consumption per MJ of fuel produced.
The variation stems from climatic factors such as temperature, wind, and relative humidity within the study area as well as from physiological differences among the feedstock crops studied. Water consumption ranged from less than 1 acre foot yr −1 for cultivation of low-yield biomass in Modoc County to more than 4 acre feet yr −1 for cultivation of sugar beets in Imperial County. Ethanol yield ranged even more dramatically, from less than 1000 L ha −1 for low-yield biomass in Sierra County to almost 14 000 L ha −1 from high-yield biomass in Imperial County.
This variation in yield and ET among counties creates patterns in the geographic variation of water consumption. Figures 3(a) and (b) focus on the low-yield biomass feedstock because it is grown in every county in the state. Figure 3(a) presents the water consumed per liter of fuel from each of these places, while figure 3(b) shows the amount consumed per hectare cultivated. The contrasting patterns of these two maps stem from the fact that while more water is consumed in cultivation of this crop in the southern reaches of the state, those areas are also more productive per unit consumption. Imperial County in Southern California was found to have both the highest water consumption of any county in the state and the greatest yield. If regional average field crops were displaced by biofuel feedstocks, there would be a decrease in total water demand statewide, since some of those displacements would occur on land previously occupied by very water-intensive crops such as rice and alfalfa. If these heavily irrigated crops were preferentially displaced, the water savings would be significant. If, however, the bioenergy targets were met through expansion of irrigated agriculture, the net increase in water demand would be substantial.

Discussion
The amount of water required to produce ethanol from purpose-grown feedstocks in California was found to range from under 500 L of water per liter of fuel to over 3500. In comparison, production of a comparable (by energy content) volume of gasoline requires from 2.1 L of water for conventional petroleum crude to almost 14 L for fuel from tar sands (King andWebber 2008, Gleick 1994).
It would be overly simplistic, however, to compare these figures directly-concluding that petroleum fuels are over 100 times more water-efficient than is ethanol. While industrial processing of biofuels and petroleum fuels require comparable amounts of water, biofuels also require water for the cultivation of feedstocks, whereas the feedstock for petroleum refining is pumped from the earth. Although over 99% of the water 'embedded' in an agricultural biofuel is used in feedstock cultivation, this consumption is much more spatially diffuse than the refining of the feedstock into biofuel. Feedstocks, grown over a large area, are concentrated in one locale to be refined into biofuel. As a result, in some cases effects on the local resource base may be larger from the industrial phase of the biofuel life cycle than from the agricultural phase.
Furthermore, a new refinery, whether for biomass or petroleum refining, represents a new demand on water resources. In contrast, feedstock production may result in an increase or a decrease in water demand depending on what, if anything, is replaced. In cases where the energy crops displace more water-intensive cultivation, this shift could lead to a net decrease in local water consumption.
Location of water consumption is also an important consideration. For example, cultivation of rainfed crops in tropical regions where water is plentiful may not result in water scarcity although it may consume a great deal of water. In contrast, use of scarce groundwater for refining petroleum in arid regions where it is produced might put excessive strain on that resource base even though it may be comparatively 'efficient' in simple terms of water consumption per unit energy. This is an important difference between analyses of life cycle water consumption, and those of GHG emissions; use of water has very different implications in different contexts, whereas GHG emissions have essentially the same effect wherever they are produced.

Functional units for analysis
Life cycle assessment (LCA) requires that impacts be normalized and reported in terms of a common 'functional unit'. Current literature on bioenergy water consumption implicitly assumes that impacts on water resources can be normalized and compared through the use of a single metric. Metrics considered have included gal H 2 O applied per mile traveled (King and Webber 2008), L H 2 O applied per L ethanol (Chiu et al 2009) (Wu et al 2009). One of the important messages of our research is that in moving the energy system to use water more responsibly, different measures of consumption will be important in different contexts and that there is no one consistently superior functional unit for analysis.
While the 'water footprint' of a biofuel can be calculated in terms of L H 2 O consumed per MJ, in assessing the effects of a production system we may at times be concerned with ascertaining the volume of water consumed per hectare used to grow feedstock. These two metrics can lead to very different pictures of the production system as is illustrated by figures 3(a) and (b). If our interest lies in optimizing the former, figure 3(a) would indicate that we should focus production in the southern reaches of California. If, however, we are interested in the equity of resource distribution, or in minimizing the energy-intensive pumping of water across great distances, figure 3(b) would indicate that production in the northern reaches of the state is the more viable option.
The dichotomy illustrated in figures 3(a) and (b) reveals the importance for water resource analysis of choosing the proper functional unit for analyzing the issue at hand. Table 2 presents many of the metrics that might be of value and concern in evaluating the water resource implications of biofuel production. It lays out the analytical utility of each metric, as well as the context and scale in which each might be used as a basis for optimization and the assumptions embedded in doing so.
A critical issue relating to these metrics is whether the sustainability of water resource management hinges on the metric of L H 2 O consumed per unit product (in this case MJ of ethanol) or on L H 2 O applied per unit product. The latter metric's focus only on irrigation application implies that rainfed crops consume no water, which is clearly not the case. Rainwater is a valuable resource that could be dedicated to other productive use, to environmental services, or to aquifer recharge if not consumed in biofuel feedstock cultivation (Molden 2007). On the other extreme, some researchers argue that evapotranspiration should be the primary or exclusive metric of concern, since irrigation water can run off or infiltrate to rejoin the exploitable resource base. While this may be true on a macroscale, in many localities water that is applied inefficiently is lost to productive use, causing irrigation demand to be a more important concern than total ET.
No unit is de facto best for analysis and optimization of water resources in the bioenergy system. Instead, local nuance must be accounted for if policies are to adequately address this important implication of bioenergy expansion.

Policy implications
Biofuel policies aimed at reducing life cycle GHG emissions are a step in the right direction, but many policy makers are beginning to recognize that a fuel with a life cycle GHG footprint better than that of its petroleum analog is not necessarily environmentally benign (Verdonk et al 2007, Schlegel and Kaphengst 2007, Hecht et al 2009.
In some cases, optimizing a biofuel production system with regard to greenhouse gas emissions could increase the strain on water resources. For example, the developing understanding of market-mediated or 'indirect' land use change (Searchinger et al 2008, Fargione et al 2008 may cause production of biofuel feedstocks to trend away from currently cultivated land. This could mean extensification of agriculture, potentially bringing new strain on irrigation water resources. Furthermore, the imperative to increase yields in order to minimize cropland required for biofuel could cause growers to irrigate cellulosic crops such as miscanthus, which are currently grown in rainfed conditions.
Water resource implications of bioenergy policies should be considered in the rule-making process to ensure that these policies do not drive changes that will put undue stress on water supply. We recommend the following actions and further research to analyze and elaborate them into practical form: • Implement a water accounting system. Our analysis shows the feasibility of calculating the water embedded in biofuels from different feedstocks grown in various regions. • Establish water intensity regulations for low-carbon fuels.
Calculated or reported water consumption and application for biofuel production should be incorporated into regulatory frameworks with incentives for implementing best management practices. • Regulate siting and design of biorefineries. While water consumption by biorefineries is a relatively small portion of total biofuel embedded water, it may have a large local effect. Careful siting and design of biorefineries will minimize conflicts between different water uses.