Coupling GIS and LCA for biodiversity assessments of land use

For this study, we developed cradle-to-gate inventory models of ethanol production from corn (Zea mays) and sugar beet (Beta vulgaris) feedstocks. Corn-based ethanol can be produced from grain alone or grain and stover, which is the cellulosic residue of the stalks and leaves. To demonstrate the use of GIS for spatial, nonlinear crop production modeling, we described all process inventories in terms of total rather than average or marginal ethanol output and used multiple reference flows of increasing sizes. The reference flows are 10%, 25%, 50%, 75%, and 100% of 2.76 MMT of ethanol, which is the ethanol contained in 3.5 × 10⁹ l of reformulated gasoline containing 5.7% ethanol by volume, the estimated demand for gasoline in 2010 in California (California Biomass Collaborative 2006; Williams et al. 2007). We convert these ethanol production targets into metric tons of required annual crop harvest for each crop (Table 1). We refer to each target output–feedstock combination as a scenario since our crop production models are purely hypothetical and not meant to be predictions. All process models were obtained from the GaBi 4 database, except for the GIS-based crop production model (described in Section 2.3) and a dummy process in GaBi 4 which receives the GIS model results. The growing of biofuel crops is the only process in our product system for which we model and assess land use impacts. If one models the locations of new ethanol conversion plants (Voivontas et al. 2001), it would be possible to include those land use impacts as well, but they would generally be much smaller. There are likely to be processes upstream and downstream of biofuel crop production that have land use impacts. While these are outside the scope of this study, the ultimate aim would be to use GIS-based inventory modeling for all relevant processes of the product life cycle.

Our study area consists of the four southernmost counties of the San Joaquin Valley of California (Fig. 1). The San Joaquin Valley of California used to be semi-arid grassland before vast parts of it were converted to irrigated cropland in the nineteenth and twentieth centuries. These counties (Fresno, Tulare, Kings, and Kern) are among the most productive agricultural lands in the nation (US Department of Agriculture 2002). Today, roughly half of the land area is in farms with primary agricultural revenue generated from wine, dairy, fruit, tomatoes, and nuts (Umbach 2002). Irrigation water evaporates in the relatively dry climate, leaving salts on the surface. Some parts of the San Joaquin Valley are already unusable for crop production due to the raised salinity of the topsoil. However, small patches of native grassland, shrubland, and wetland remain. Native wildlife species use both pristine and cultivated land as habitat. The San Joaquin Valley is the region with the largest number of threatened or endangered species in the mainland United States (US Fish and Wildlife Service 1998). Native species are threatened mostly by habitat loss from conversion to cropland and urban development.

GIS and LCA are fundamentally different and complementary. The former is specifically designed to organize and analyze spatial data, while the latter inventories and analyzes data on product systems and typically does not operate on geospatial information. Although LCA’s lack of spatial differentiation is frequently criticized, there is no consensus on how it should be integrated. A decade ago, it was suggested that GIS should be integrated with LCA (Bengtsson et al. 1998), but implementation has been slow. Azapagic et al. (2007) described a framework for linking LCA, GIS, and environmental models, although GIS would merely be used to provide locational information on pollution sources as input for other models and to display the spatial distribution of predicted impacts. Burke et al. (2008) used GIS to map biotopes before and after a single mining operation and to generate characterization factors. Results were appended to the LCIA findings, but GIS data were not apparently passed to the LCA model for assessment. To assess desertification potential, Núñez et al. (2010) used GIS to calculate a score as an elementary flow along with ecoregion-specific characterization factors. Combining these values with the area of the process and the area of the ecoregion determined the impact.

The procedures developed for and applied in this work are shown in the flowchart in Fig. 2. In the LCA system, all conventional cradle-to-gate processes and flows of an agriculture-based end-product are shown (Kim and Dale 2005a). To incorporate land use, a crop production model is inserted (center of the LCA box in Fig. 2), which receives its intermediate inputs from an upstream cradle-to-gate LCI model and feeds its crop outputs to a downstream gate-to-gate LCI model of ethanol conversion. In the goal and scope phase, scenarios are defined with target demand for ethanol production, which are transformed into crop demand based on conversion factors (Table 1). A GIS land use model calculates the amount and location of crop land needed to meet the target (Section 2.3). A land use map is thus generated for each scenario, which is then converted into a map of habitats (Section 2.4). In principle, GIS could be used to model the land use of all processes with significant land requirements, such as resource extraction, transportation infrastructure, and industrial sites. GIS-based inventory models can be regarded as a spatially explicit component within a traditional life cycle inventory model. Such LCA–GIS coupling facilitates spatially explicit, nonlinear input–output modeling of intermediate and elementary flows. In the present study, the coupling is limited to calculating elementary flows of habitat-type areas as functions of fuel crop type and production level. The habitat-type areas and all other inventory data are used to generate a crop production process in the LCA model (Fig. 2). Inputs such as irrigation water, fertilizer, pesticides, and fossil fuel and outputs such as greenhouse gases and nitrates are based on averages in our study. They likely will vary with locations and could also be modeled explicitly in GIS, given that the relevant data are available. In the impact assessment phase, the habitat-type areas are used to calculate indicator results for potential biodiversity impacts (Geyer et al., submitted).

Crop scenario models can be extremely sophisticated to capture all relevant influences such as land productivity, climate, farmer behavior, market forces, land use policies, infrastructure, and economic incentives. As our purpose in this study was to demonstrate the coupling of LCA and GIS in order to assess impacts on biodiversity, we opted to develop a relatively modest scenario model. This model also neglects land use associated with upstream and downstream activities such as resource extraction and siting of biorefineries. There is nothing in this framework, however, that would preclude use of a more sophisticated scenario model or the inclusion of ancillary land use changes if appropriate data were available.

Our crop production model maps each annual fuel crop production target onto a set of land parcels while accounting for land productivity, basic production economics, and different land use policies. For each crop type and production target in Table 1, the GIS model returns a list of land parcels that, if used for fuel crop production, would collectively achieve the required annual output. To achieve this, we roughly follow the GIS-based modeling approach suggested by Bryan et al. (2008) that integrates soil productivity for each biofuel crop type with the potential revenues and costs of production. The resulting spatially explicit land use scenarios are intended to be feasible and plausible, but not meant as predictions of actual land use changes due to given fuel crop demand.

For this demonstration, corn and sugar beets were selected as exemplary fuel crops because the San Joaquin Valley is suitable for growing them, the technologies for converting feedstocks to ethanol are known, and the necessary spatial data were available. The crops differ in the conversion technology as corn grain generates starch-based ethanol, corn stover produces cellulosic ethanol, and sugar beets can be converted to a sugar-based ethanol. The GIS modeling, implemented in ArcGIS 9.2 (ESRI 2009), consisted of the following four steps.

The first step was mapping the production potential for each feedstock (Noon and Daly 1996; Voivontas et al. 2001; Price et al. 2004; Hughes and Mackes 2006; Bryan et al. 2008; Hellmann and Verburg 2010). The production potential for corn and sugar beets was based on estimates by soil types and expressed as spatially varying crop yields (Soil Conservation Service 1994). On suitable soils, the corn grain yield ranges from 4,400 to 9,400 kg ha⁻¹ year⁻¹. In the first set of corn-based ethanol scenarios, we assumed that the corn stover would be left in the field as residue to conserve topsoil under best management practices. In additional scenarios, we assumed that 30% of the corn stover would also be used for cellulosic ethanol, with the remaining 70% left in the field. Sugar beet yields range from 45,000 to 78,000 kg ha⁻¹ year⁻¹ on soils that are suitable, which differ from those suitable for corn (Fig. 1). Commercial biofuel crops cannot be grown in certain areas because of land management or prior commitments of land use. Therefore, potential yields on lands owned by public agencies or conservation groups and on existing urban development were set to zero based on GIS layers of land ownership and land use (Lovett et al. 2009). These unavailable lands are depicted in black in Fig. 1. Using all the spatial data from above, 1-ha grid cell maps of production yields in kilograms per hectare per year were derived for corn (with and without stover) and sugar beets.

The second step utilized economic data for these crops specific to the San Joaquin Valley or similar regions to derive a map of potential economic net returns (University of California Davis 2004). Establishment, production, and some harvest costs are fixed per unit area, while other harvest costs and gross revenue are related to volume of harvest. Maps of potential net return (gross revenues from the crop yield minus fixed and variable production costs) in dollars per hectare per year were produced for each crop type by subtracting fixed and variable costs from potential revenue for each 1-ha cell. Potential net return (dollars per year) and crop yield (MT per year) for each land parcel in the study area were calculated by summing over all 1-ha cells in each parcel. Parcel size, ranging from 5 to 15,000 ha, was also computed by the GIS software.

The third step employed a simple ranking and selection algorithm to generate parcel sets that efficiently produce the annual target level of the crop for each scenario (Table 1). It is necessary to make assumptions about agricultural markets and about individual landowner decisions when developing crop production scenarios (Sheehan et al. 2003; Santelmann et al. 2004). Our approach was to generate a list of all parcels ranked by their potential net return per hectare under current conditions, i.e., with the parcel having the highest return on top and the one with the lowest return on the bottom. The parcel with the highest net return was selected, its potential output was added to the cumulative total output, and its area added to the cumulative area. Parcels were then selected in descending order of returns until the cumulative output met the target level. The result of this procedure is an economically plausible set of parcels which collectively meet the annual target level.

The final step was to calculate the types and amounts of (intermediate) inputs, such as irrigation water, fertilizers, pesticides, and use of agricultural machinery required for each land use scenario (set of selected parcels). There was no spatially differentiated data on input requirements available for our case study. Average values are used instead, which are independent of parcel location and therefore only functions of parcel area.

“Habitat” is the area or places inhabited by a species where its life history requirements for food, cover, and reproduction are met. Biologists often categorize sets of places into a finite set of mutually exclusive habitat types based on their shared attributes of vegetative cover and structure as they relate to those life history requirements of wildlife species (Mayer and Laudenslayer 1988). Each of these habitat types can be characterized by its degree of suitability for each species in a wildlife–habitat relationships database. A habitat type is typically suitable for many different species, and many species will have more than one suitable habitat type. Each species thus has its own potential geographical distribution depending on which habitat types it finds suitable for its needs. Combining geographic information on habitat distribution with knowledge of species–habitat relationships provides a means of mapping potential species distributions, with approaches ranging from relatively simple relational models to more complex models that account for species dispersal and population dynamics (Guisan et al. 2006). At a minimum, we suggest basing characterization models for biodiversity indicators on inventory results that contain pertinent habitat information. More specifically, we suggest that inventory models should contain elementary input flows that represent habitat types associated with land cover and use. In California, experts have defined a set of mutually exclusive and collectively exhaustive standard habitat types, including eight for agricultural landscapes, as the basis for modeling species’ geographical distributions (Mayer and Laudenslayer 1988). Nearly half of the terrestrial vertebrate species in California utilize the state’s agricultural lands, particularly those that are more mobile such as birds and bats (Brosi et al. 2006). For most of these species, agricultural land is more valuable for feeding and cover than for reproduction. The continuing importance of the San Joaquin Valley as a major flyway for migratory birds illustrates this point. Even some threatened or endangered species can persist with modest set-asides in the working landscape. For instance, the San Joaquin kit fox can survive in farmland if its prey base is maintained and there are sufficient patches of untilled land for denning sites (US Fish and Wildlife Service 1998). Corn is a member of the habitat-type irrigated grain (IGR) crops, whereas sugar beets are irrigated row and field (IRF) crops. Thus, these feedstocks not only differ in their location and conversion processing but also in the habitat they provide for wildlife.

We compiled a baseline habitat map from multiple sources. The Multisource Land Cover layer (California Department of Forestry and Fire Protection 2002) was the best available geospatial data for native or naturalized habitats but had lumped all farmland into a single map class. Therefore, we merged this information with maps of specific crop types of the study area (California Department of Water Resources 2010) and reclassified the crops to their corresponding habitat type. The resulting baseline layer comprised 29 habitat types, including eight agricultural habitats. For each scenario, baseline habitat types in the selected parcels were replaced with the habitat type associated with the fuel crop of the scenario in the GIS habitat model (Fig. 2) to generate a new wildlife habitat-type layer. The area of each of the 29 habitat types was then derived for the entire study area. For each scenario, the sum of the 29 habitat areas equals the total study area, which reconfirms that the 29 habitat types are mutually exclusive and collectively exhaustive.

In addition to all other input and output flows, the GIS-based crop production model thus also generates 29 elementary input flows, the habitat composition, containing habitat-type areas in hectares. The inventory model characterizes each considered amount of annual fuel crop output with a habitat composition, which can be used for attributional analysis. For consequential analysis, the inventory model relates land use changes (in this case to meet increasing fuel crop production targets) to the resulting changes in habitat composition. It should be noted that inventory modeling based on total inputs and outputs is a digression from standard LCA, which uses average process inputs and outputs for attributional analysis and marginal changes in inputs and outputs for consequential analysis. Average and marginal data can be readily derived from total input and output data, but not vice versa. Part 2 of this paper series (Geyer et al., submitted) demonstrates how the elementary input flows of the habitat composition can be used to calculate potential biodiversity impacts.