Issues associated with LCA methodology fell into two categories. The first centers around the formulation of questions for analysis and boundary issues, including models to address these issues. The second set focuses on technologies assessed, mechanisms of impact and data representation.
3.2.1 Methodology I: methods: scope, system boundaries and model integration
Goal and scope setting is the first step in performing a LCA and thus influences the subsequent phases (Tillman
2000). It falls to the trained practitioner to enforce the validity of goal and scope setting for the required analyses. For example, the level and quality of data collected must be sufficient in order to fulfill the original aim of the study (Singh et al.
2010). The system boundaries are set during this stage, and these are key to any analysis (Bird et al.
2011). The choice of system boundary in comparative studies may lead to different conclusions and decisions about which products to promote, as in the classic debate surrounding nappies/diapers (Bond
2005), and have an influence on rankings through variation in scope or the impact that is assessed (Suh et al.
2004). Bio-based (poly)lactic acid, for example, out-performed conventional polymers on energy consumption and GHGs; however, clear ranking was lost when ecosystem quality metrics were assessed (Yates and Barlow
2013).
Allocation methods (the way in which impacts are assigned to co-products) give quantitatively different results, which can also lead to incomparability (Wang et al.
2011). An assessment of UK wheat ethanol with regional N
2O parameters reported GHG emissions (midpoint values) of 51, 64 and 68 g CO
2eg/MJ for energy allocation based on energy, economics and substitution, respectively (Yan and Boies
2013). This affects both comparison and compliance, as policies mandate different allocation types: by substitution on the RTFO, RFS2, LCFS and RED for electricity co-production, by energy content in RED and as fall back for RFS2 and LCFS and by economic value as a fall back for the RTFO. Thus, the same pathway ranks differently across policies. Direct GHG emissions for bioelectricity from rapeseed using allocation choices from primary renewable policies improved on the reference case by 60 % (sub peas) or 21 % (sub soy) with substitution, 33 % for energy allocation and 16 % for economic allocation (Wardenaar et al.
2012). Such incompatibility can occur even with methodologies compliant with industry or regulatory standards for biofuels GHG reporting (Whittaker et al.
2011).
System boundaries can become inconsistent because of allocation decisions or because the scope or the goals of analysis changes. For example, methods for reducing the overall farm environmental footprint will not necessarily result in a low carbon crop (both of which are assessed with valid LCAs) and vice versa, and the differing purposes lead to different boundaries, which are often conflated (Whittaker et al.
2013). The farm-based approach considers the overall environmental footprint of a farm site, whereas a crop-based analysis focuses on crop- and product-specific aspects that go beyond the farm gate. Agricultural assessments are particularly sensitive to scope changes in response to purpose (Roy et al.
2009), which has major implications for bioenergy policy decisions. For example, the RED was designed to promote the production of biofuels from ‘non-food’ biomass feedstocks and does this by specifying in the GHG accounting methodology that cereal residues are not allocated upstream emissions from cultivation. This immediately places these feedstocks at an advantage but is an example where the intention and approach of the analysis has been heavily influenced by policy (Whittaker
2014). Perhaps more significantly, the detail of goal and scope is seldom conserved when analyses are merged into the policy process, because data, for example GHG values, are taken from a range of studies to inform policy or regulation or to be the basis of financial incentives, such as feed-in tariffs. Often, several studies, which may have been originally produced for a different purpose, are taken to inform policy (McManus and Taylor
2015).
This tendency to aggregate the GHG estimates for comparative decisions is logical but overlooks the complexity and detail specificity associated with GHG analyses for bioenergy systems. For example, Farrell et al. (
2006) conducted a meta-analysis of early corn ethanol LCAs and found that, even for direct effects (attributional analysis), the GHG and net energy results varied substantially. Studies with the least favourable results for corn ethanol yielded a carbon-intensity of nearly 120 g CO
2eq/MJ, which is approximately 30 % higher than gasoline. However, those studies failed to allocate any biorefinery impacts to co-products and used outdated data for key industrial processes. For rapeseed biodiesel in Europe, a careful analysis found calculated emissions ranging from 5 to 170 g CO
2eq/MJ (Malça and Freire
2011), arising primarily from differences in modelling soil emissions and land use, co-product allocation and high uncertainty associated with some emissions parameters.
Currently, GHG assessments dominate the biofuels sustainability debate; for example, more than half of 53 sampled LCA studies of lignocellulosic biofuels published between 2005 and 2011 were GHG focused (Borrion et al.
2012). Although bioenergy is a global commodity, local values and drivers may differ, and effects like GHG emissions are not always observed near their cause. Location-specific metrics are expanding, though, particularly as water captures more attention and the relevance of other resource competition emerges in the discourse. However, issues other than GHG tend to be specific to the location of production or processes. For example, Scown et al. (
2011) demonstrated the location specificity of water consumption/withdrawal and pollution in biofuel production systems. Integrating such impacts is hampered since LCA is, traditionally, not a tool that examines local impacts and thus has crucial gaps. For example, water, particulate emissions and wider impacts on ecosystem services are often not well modelled and may lack spatial or biophysical data. Gaps in data availability and data quality, as addressed elsewhere, are generally not highlighted in the final result, which are often distilled down to a single number (e.g. N
2O emissions from soil under the IPCC guidelines (De Klein et al.
2006)).
Analyses with complex boundaries often result in multiple models in the final analysis. Ecosystem, market and social dynamics contribute to impact processes. Structural changes, such as infrastructure and technology changes and other progress along the innovation trajectory and adoption rates and patterns, contribute to scenarios for impact and are also beyond LCA’s scope. All of these introduce the need to rely on and integrate increasingly complex and speculative modelling and shape the development of assessment tools (Wicke et al.
2015) and lead to efforts to develop general quality criteria for modelling and recommend best-practice LCIA models for particular impacts from the plethora available (Hauschild et al.
2013). Because advances in elucidating or representing the mechanisms that underpin models occur in many different academic disciplines (see, e.g. Arbault et al.
2014), integrating the sets of data or models to give a picture of the life cycle impacts is challenging, though essential for LCA to contribute policy setting or analysis (Wicke et al.
2015). In connecting models to describe various portions of the overall value chain and potential consequences, the overall complexity and level of detail in the analysis can mask areas where boundaries conflict, even for relatively narrow studies. Ensuring that the full set of models integrate harmoniously and preserve reasonable levels of transparency, rigour and robustness is an ongoing challenge that worsens as use of custom-built tools proliferate (see, for example, CA-GREET
2014).
3.2.2 Methodology 2: technical use: mechanisms, technologies and representations of data
System sustainability is often estimated by summing the impacts of value chain components and attributional LCAs (see, e.g. Bento and Klotz
2014). In order to assess the sustainability of biofuel production, it is necessary to consider each biofuel from each feedstock, according to its own merits, and alongside specified sustainability criteria (Royal Society
2008). There is, however, a lack of the necessary codified sustainability criteria and a recognition of data needs to support them (Hecht et al.
2008), and most progression in LCA-based reporting methods has focused on developing a single method (predominantly based around GHG measurement (McManus and Taylor
2015)) to assess all biofuels in a similar manner. The success of this is mixed, especially with the increasing importance of social and economic sustainability parameters, and with mismatches between regulatory mechanisms in methodological approaches such as allocation and default values for fuels and systems.
To be included in an assessment, mechanisms for interactions among portions of the physical system or value chain are also needed. The fundamental interactions are bio/geophysical, such as in soil carbon accumulation and mobility, or in nutrient and water cycles, and give rise to feedbacks at a variety of spatial and time scales (Bagley et al.
2014). These mechanisms draw on scientifically active areas where knowledge is evolving rapidly with both high and low uncertainties (see, e.g. Balvanera et al.
2014; Greene et al.
2015) and provide an idea of the necessary scale of the system boundary needed to capture relevant impacts. Sustainability assessments of biofuel systems show very strong sensitivity to soil emissions, particularly nitrogen. Using different N
2O emission methods gave GHG values (midpoint) differing by 25 % when using IPCC Tier 1 methods or a UK-specific model within the same assessment of UK wheat ethanol (Yan and Boies
2013). A meta-analysis of LCA studies on European rapeseed biodiesel found that GHG intensities correlated directly with how soil emissions were modelled (Malça and Freire
2011). Some of these mechanisms differ substantially between first generation and advanced bioenergy systems. For example, perennial crops can have belowground carbon allocations more than four times higher than a traditional corn-soy rotation, and belowground biomass increased by 400–750 %, measured for miscanthus, switchgrass and native prairie during establishment ( Anderson-Teixeira et al.
2013). The potential for perennial bioenergy crops to increase soil carbon under some conditions has contributed directly to their preferential ranking in over first generation pathways in policy instruments.
Multi-sector analyses or so-called consequential factors introduce additional complex mechanisms for interactions among the value chain and potential consequences. One route is through mechanisms that do not directly include the primary product. For example, Scown et al. (
2014) showed that the net GHG and water impacts of utilizing lignin for heat and power at cellulosic biorefineries or exporting lignin for co-firing with coal varied greatly depending on long-term trends in power plant retirement and new construction. If the export of additional biopower to the grid encourages early retirement of aging coal-fired power plants, the GHG footprint of ethanol in their scenario was predicted to be approximately 50 % lower than if biopower exports encourage deferred construction of new natural gas combined cycle (NGCC) power plants.
Interaction mechanisms can also be through market mediation that emerges within or across borders and/or sectors, which may rely on speculative global market dynamics. Though such consequential approaches started with general product life cycles (Weidema
1993; Zamagni et al.
2012), attempts to assess the impact of global commodity market-driven land use dynamics in relation to biofuels have made it a common concept to guide policy in attempting to avoid unintended consequences in the form of indirect land use change (McManus and Taylor). The drawback is that consensus around the approach has not yet emerged (Marvuglia et al.
2013; Rosegrant and Msangi
2014; Schmidt et al.
2015); thus incorporating such market-mediated impacts for consequential analysis increases the variability dramatically ( Vázquez-Rowe et al.
2013). For example, estimates of ILUC GHG impacts ranged from about 3 to above 220 g CO
2eq/MJ rapeseed biodiesel and ~5 to ~100 g CO
2eq/MJ for bioethanol from maize over the last 5 years, even in current analyses when ranges for such values have tightened nor are the authors of that analysis optimistic about such uncertainty decreasing soon (Ahlgren and Di Lucia
2014). Thus far, ILUC has received the bulk of attention among consequential LCA of biofuels. But market-mediated mechanisms are also being explored for other properties, such as indirect fuel use (Rajagopal et al.
2011) and rebound effects (Vivanco and van der Voet
2014; Smeets et al.
2014). Likewise, social impacts, crucial for sustainability assessments, arise through cross-cutting mechanisms (see, e.g. Benoît et al.
2010). About 20 % of annual LCA publications touch on or address social factors (McManus and Taylor
2015) and guidelines and methodologies are developing broadly because of the labyrinthine relationships among groups and the range of potential impacts (Wu et al.
2014). Many of these are reflected qualitatively in international sustainability standards (see, for example, FAO’s Compilation of Bioenergy Sustainability Initiatives
2011). In nearly all these mechanisms, the science or state of knowledge is changing rapidly. As new insights emerge, accounting for impacts in the LCA that depend on them is lagging.
The absence of spatiotemporal components, which underlie most mechanisms in LCA, is of particular concern for bioenergy. Time is most commonly incorporated with set-year analyses (e.g. N years in the future) and linear annualization. While the former is a viable, if limited, approach, the latter is problematic for agriculture. Simple annual crop rotations that make up the bulk of agriculture (corn, soy, wheat, etc.) are well represented this way, but, for perennial crops, where yields develop over time and management practices may vary from year to year, simple annualization is not always reflective of reality. The difference between establishment and production periods for such crops are key to emissions estimates; simulations of nitrogen loss from 2-year-old switchgrass were 360–410 % higher than for mature stands at the same fertilization levels, depending on harvest number, resulting in a 15-year average of 20–30 % that of cotton under the same conditions (Sarkar et al.
2011). Likewise, the temporal mismatch between conversion or harvest and carbon uptake in forest-based resources introduces decadal and longer time frames, and concerns over the potential carbon debt caused between harvesting and re-establishing timber stands have become an important issue for climate and bioenergy policy (Lamers and Junginger
2013).
Location-dependent issues for LCA are not limited to siting information and land use changes; they are dominated by the limitations in inventories from location dependence in input data and the location dependence of impact metrics water or biodiversity (Seager et al.
2009) as well as trade, market mediated and social impacts, all addressed in other sections. Spatial characteristics of the feedstock production system are often assessed extraneously in economic terms. Scown et al. (
2013) found that, in their scenario that incorporated corn stover, wheat straw and Miscanthus in the USA, only 80 % of available biomass was geographically concentrated enough to warrant utilization for biofuel production; the remaining biomass was too dispersed, resulting in prohibitively high transportation costs. While highly sensitive to supply chain characteristics, LCA is still evolving to incorporate the intersection between developing infrastructure and logistics, such as transportation modelling (Strogen and Horvath
2013; Strogen and Zilberman
2014). The spatial aspects of resource management will also begin to contribute in bioenergy LCA, as in critical resource assessment (Sonnemann et al.
2015), potentially to pivotal effect, because there are more critical resources than just land in biofuels, and land is a critical resource in more than biofuels. Generally, these issues introduce reliance on multi-model, multi-scale systems. Data integration, uncertainty integration, error propagation, certainty and multi-parameter output representation all become important and are difficult to convey succinctly.
3.2.3 Best practice and optimization—farming
Agriculture and farming in particular deserve special mention as competition for land increases. Financial incentives awarded to the farmer to encourage planting of bioenergy crops create an interaction among LCA, policy and farming practices (Glendining et al.
2009; Natural England
2013), as does the global commodities feed/food market. The effectiveness of such incentives to serve climate policy goals depends on the accuracy of calculated avoided life cycle GHG emissions and the system boundaries considered. Farmers are key to the provision of empirical data; increasing the certainty of the assessments for that portion of the life cycle. However, because agricultural practices differ by region, even over relatively small distances, and management decisions have large impacts in bioenergy (Davis et al.
2013), transferability of data or existing analysis from a region where much is known to one where little is known (a common technique in other sectors) is questionable.
LCA for agricultural production is well-established (Roy et al.
2009), but LCA for policy planning in bioenergy incorporates the potential implementation of large-scale biomass production, which lacks certainty. Improvements in agricultural yields have been substantial over the past decades (e.g. US corn yields have roughly doubled since 1976, USDA data). Projections of productivity changes over time are speculative but important for long-term planning. For example, carbon payback times decrease 30–50 % when crop yields reach the 90th percentile in global yield, representing crop productivity increases and/or substantial management changes across global averages (Gibbs et al.
2008), in effect identifying the benefits of addressing the global yield gap. There are potential benefits in non-GHG impacts also. For example, in advanced bioenergy landscapes, it is possible for very small trade-offs in the economic balance to have a large favourable impact on biodiversity under particular land conservation regimes (Evans et al.
2015). These represent challenging aspects for cross-sector agricultural LCAs to include. Other complicating factors that are of increasing importance include conservation mechanisms which impact biodiversity (see, e.g. Evans et al.
2015), precision agriculture where practices vary between or even with fields (Weekley et al.
2012), alternate cropping systems (Perrin et al.
2014) and land sparing practices (e.g. Cohn et al.
2014) which are often multifunctional. Multifunctional landscapes provide many sustainability benefits (Perfecto and Vandermeer
2010) but expand the system boundaries increasing data requirements (Rossing et al.
2007) and uncertainty (Jung et al.
2013).
Risk, policy uncertainty, nascent markets, infrastructure and supply chains threaten adoption, which makes projection of roll-out rate or the extent of production challenging. This increases reliance on representing statistical projections in data and uncertainty and shifting policy estimates of expansion (Rajagopal and Plevin
2013), including how to reflect the ranges in the results, and whether knowledge of the demographics can contribute to analyses or types of analyses that are more useful from a policy perspective (Plevin
2010).
3.2.4 Robust high quality data and gaps
Data availability, transparency, curation and sharing are key to the long-term success of LCA both generally and for comparisons used in bioenergy planning. Reliability of the results from LCA studies strongly depends on the extent to which data quality requirements are met, where common problems include lack of transparency and data variation and gaps. For lignocellulosic ethanol, for example, data inconsistencies contribute to conflicts in published LCA results of GHG emissions (e.g. Wiloso et al.
2012), ranging from −1.25 to 0.84 kg CO
2eq/km travelled when using E100 (Borrion et al.
2012). There are specific cases where there is minimal data, for example, enzyme production (Spatari et al.
2010), or where data is outdated, for example, pesticide production (Audsley et al.
2009). Primary data sets, despite their overall strength, also sometimes have errors (see, e.g. Cooper et al.
2012). Data set compositions vary regionally, and sometimes sectorally, in the completeness of their data on which GHGs are included (Ansems and Ligthart
2002). Reusability of data is highly dependent on sufficient data documentation, and standards are emerging (see, e.g. Sonnemann et al.
2013) to address this, for example, in the area of LCAs used for Environmental Product Declaration (EPDs) (Ingwersen et al.
2012).
While transferability is a common approach to supply missing data, using the value from another region for a missing parameter, the approach frequently fails for bioenergy. Data for non-global environmental impact categories frequently vary by region, often for categories to which agricultural assessments are highly sensitive (e.g. ecotoxicity). Some impact categories, such as water quality and use impacts in bioenergy, can be local, regional or both (Dale et al.
2010) and are further complicated by the importance of the local water context (e.g. drought to excess). Water consumption, for example, may range from 5 to 2138 L per litre of ethanol depending on regional irrigation practices in the USA (Chiu et al.
2009)). Region-specific data are also often not available (for example land use and biodiversity (Dale et al.
2010)). Industrial data for emerging processes and technologies is frequently limited due to commercial confidentiality and also as the data for such systems and an understanding of their impact is nascent. New scientific understanding of various impacts, such as the impact of air pollutants on human health (see, e.g. Hajat et al.
2015; Novák et al.
2014), or newer issues such as ‘black carbon’ (see, e.g. Cai and Wang
2014; Otto et al.
2014), also introduce data or method limitations.
Under some conditions, data sets/databases used in a LCA reflect external value judgments, of which users may or may not be aware or take time to discover. For example, there are levels of detail within and between datasets, some with wider boundaries than others. There are also value judgments embedded in ‘off the shelf’ impact assessment methods. Some are more explicit than others, for example EcoIndicator 99 and ReCiPe both show three options ranging from an ‘absolutely only proven cause and effect’ to a precautionary approach. Pragmatically, these databases and software packages are frequently the most accessible solution for the practitioner since primary data is either unavailable or too time consuming to gather (Hetherington et al.
2014). This is reflected in the literature, where uptake of these databases is rapid; studies citing the widely used ECOINVENT database launched in 2000 (Frischknecht et al.
2004) had reached 2240 by mid-2014 (scopus search, 18th July 2014). However, these proprietary databases and software tools can limit transparency, independent reproducibility and transferability as the underlying data cannot be shared in publications. The high-cost setting up and maintaining the quality of database impedes open access (Hellweg and Canals
2014). A number of open data sets and standardization efforts seek to address this, in particular the European Platform on LCA’s European reference Life Cycle Database and the International Reference Life Cycle Data system common data format standard, as well as and the USDA LCA Commons and NREL US Life Cycle Inventory Database.
1
Because a wide range of stakeholders use the results, indications of data quality are increasingly important. Qualitative indicators can be used along with their quantitative counterparts to address data transparency issues such as data source and to address the confidence and uncertainty of data (see for example, Ansems and Ligthart
2002) under particular circumstances, such as in the electricity sector (see, e.g. Garraín et al.
2015). Qualitative indicators help to contextualize the relevance of the data so that policymakers are able to make more informed decisions about the circumstances to which the data apply, thus making policy judgements based on the available data more reliable. Such indicators are sometimes contained within commercial software packages (for example SimaPro—Pre Consultants (
2014) and GaBi - PE International (
2014)). While these packages can aid very quick and effective calculations, it is also very easy for general users to employ them and default databases without an understanding of the uncertainty or quality of the data contained within them.