Science of the Total Environment 435–436 (2012) 230–243

Contents lists available at SciVerse ScienceDirect

Science of the Total Environment

j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenv

LCA data quality: Sensitivity and uncertainty analysis

M. Guo, R.J. Murphy ⁎Department of Life Sciences, Imperial College of Science and Technology and Medicine, London SW7 2AZ, UK

H I G H L I G H T S

► LCA data quality issues were investigated by using biopolymer case studies.► Dynamic characterization models with varying time horizons are proposed for LCAs.► A method proposed to integrate statistical methods into LCA for uncertainty analysis.► Uncertainty and sensitivity analysis enabled assigning confidence to LCA outcomes.

⁎ Corresponding author. Tel.: +44 2075945389; fax:E-mail addresses: miao.guo06@imperial.ac.uk (M. G

(R.J. Murphy).

0048-9697/$ – see front matter © 2012 Elsevier B.V. Alldoi:10.1016/j.scitotenv.2012.07.006

a b s t r a c t

a r t i c l e i n f o

Article history:Received 13 March 2012Received in revised form 27 June 2012Accepted 1 July 2012Available online 31 July 2012

Keywords:Life cycle assessmentData qualityUncertainty analysisSensitivity analysisTime horizonBiopolymer

Life cycle assessment (LCA) data quality issueswere investigated by using case studies on products from starch–polyvinyl alcohol based biopolymers and petrochemical alternatives. The time horizon chosen for the character-ization models was shown to be an important sensitive parameter for the environmental profiles of all the poly-mers. In the globalwarming potential and the toxicity potential categories the comparison between biopolymersand petrochemical counterparts altered as the time horizon extended from 20 years to infinite time. These casestudies demonstrated that the use of a single time horizon provide only one perspective on the LCA outcomeswhich could introduce an inadvertent bias into LCA outcomes especially in toxicity impact categories and thusdynamic LCA characterization models with varying time horizons are recommended as a measure of the robust-ness for LCAs especially comparative assessments. This study also presents an approach to integrate statisticalmethods into LCA models for analyzing uncertainty in industrial and computer-simulated datasets. We calibrat-ed probabilities for the LCA outcomes for biopolymer products arising from uncertainty in the inventory andfrom data variation characteristics this has enabled assigning confidence to the LCIA outcomes in specific impactcategories for the biopolymer vs. petrochemical polymer comparisons undertaken. Uncertainty combined withthe sensitivity analysis carried out in this studyhas led to a transparent increase in confidence in the LCAfindings.We conclude that LCAs lacking explicit interpretation of the degree of uncertainty and sensitivities are of limitedvalue as robust evidence for decision making or comparative assertions.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

Life cycle assessment (LCA) as outlined by International StandardsOrganization (ISO) (2006) has been widely applied as a decision sup-port tool to identify the important environmental factors in product sys-tems and to evaluate and compare the environmental profiles of diverseproducts e.g. biofuels and biomaterials. In LCA practice data quality is-sues have been broadly discussed since the 1990s (USEPA, 1995), butrobustness of the modeled results is not commonly addressed in theLCAs. Here, we focus on two issues that have emerged from literaturereview and our experience with LCAs, namely 1) a lack of temporalinformation, and 2) a lack of uncertainty analysis in LCAs.

Often, sensitivity analysis is carried out in LCAs to test systemboundaries (Kim and Dale, 2009; Cherubini and Ulgiati, 2010),

+44 2075842056.uo), r.murphy@imperial.ac.uk

rights reserved.

allocation approaches (Kim and Dale, 2002; Gnansounou et al., 2008;Luo et al., 2009; Morais et al., 2010), parameter values (Estermann etal., 2000; Laser et al., 2009) and characterization methods (Dreyer etal., 2003). Temporal effects on both the life cycle inventory (LCI) andlife cycle impact assessment (LCIA) results are rarely concerned e.g.landfill emission under different time horizons (Finnveden, 1999) andthe time-dependency of characterization models. This work employeda case-study of starch-based biopolymer foam materials to explore thetime horizon effects on characterized results for global warming poten-tials (GWPs), ozone depletion potentials (ODPs) and human andeco-toxicity potentials using the midpoint level (problem-orientated)approach.

The GWP is a simplified index introduced by the IntergovernmentalPanel on Climate Change (IPCC) to evaluate the overall potential climatechange response to greenhouse gases (GHGs) (Forster et al., 2007). Inthe IPCC model, direct GWP values of GHGs are given under differenttime-horizons (20, 100, 500 years). Long-lifetime compounds tend tocontribute more to total GWP in a ‘forward-looking’ perspective e.g.

231M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

CClF3, whereas compounds with short-lifetimes become less importantafter removal or depletion with time e.g. CH4. Among three time hori-zons above, the 100-year level is widely used as a ‘default’ time horizonfor expressing GWPs, for example it is the time horizon of GWP speci-fied in PAS 2050 (BSI, 2011).

The concept of ODP developed by the World Meteorological Organi-zation has been used to evaluate the effects of compounds on strato-spheric ozone (Wuebbles et al., 1998; Calzonid et al., 2000). The ODPvalues of major depleting substances (halogenated compounds) listedin the Montreal Protocol (UNEP, 1999) was used as a foundation forthe ODP characterization models (WMO, 2007). These ODP values arebased on the assumption of steady state with constant emission inde-pendent of the time horizon (WMO, 2007). Although most of the ODPsubstances are long-lived compounds, their lifetime varies substantial-ly between 1 and 3000 years. Thus attempts were made to estimatetime-dependent ODP e.g. the semi-empirical approach proposed bySolomon and Albritton (1992). This semi-empirical model covered a5 to 500 year time-scale where the ODP value of compounds whichhave shorter lifetimes than the reference gas (CFC-11) decrease withincreasing integrated time. However, the general ODP model with in-finite time has been the most widely used as the baseline approach(Goedkoop and Spriensma, 2001; Guinée et al., 2001).

Toxicity related impact categories also raise time-dependency is-sues. Generally, the methods most commonly used for calculation oftoxicity potential are steady-state models with infinity as the time ho-rizon (Huijbregts et al., 2000; Guinée et al., 2001). However, Huijbregts(2000) stated that there are substantial drawbacks to suchmodels. Forinstance, shorter-term toxic impacts may be underestimated due tothe domination of persistent substances which actually take effectsafter long-time exposure. Therefore, toxicity potentials under specifictime horizons (20, 100 and 500 year horizons) were introduced in dy-namic characterization models where organic substances and metalshowed greater time-dependency effects due to their longer residencetime in comparison with other inorganic substances (Huijbregts,2000). Another important issue in toxicity impact categories is uncer-tainty in modeled outcomes — the various models differ substantiallyin terms of scope, modeling principles and, most importantly, can failto arrive at consistent characterization factors. To harmonizemodelingapproaches and characterization factors, a life cycle initiative waslaunched in 2002 by United Nations Environment Program and the So-ciety of Environmental Toxicology and Chemistry and great effortsweremade to identify the influential parameters and sources of differ-ences in toxicity-related models (Hauschild et al., 2008). Based on arange of existing models (e.g. Impact 2002, USES–LCA), a scientificconsensus model USEtox was developed where infinite time is usedas sole time-horizon (Rosenbaum et al., 2008).

Most of the midpoint characterization models reviewed above havebeen incorporated into the CML 2001 method library (Guinée et al.,2001). However, dynamic endpoint models (termed as a ‘damage orien-tated approach’ defined at the level of protection area) (Finnveden et al.,2009) have also been developed, such as the time-dependent Eco-indicators 99 with three perspectives where time horizons 100 year and100,000 year are integrated(Goedkoop and Spriensma, 2001). Althoughthe application of various time horizons in LCIAs has been recommendedfor robustness analysis in addition to the usual infinite time (with the ex-ception of GWP100) (Goedkoop and Spriensma, 2001; Guinée et al.,2001; European Commission, 2011) static models have been very widelyadopted as the sole approach inmany LCA studies. LCAs derived only fromstatic models run the risk of bias in providing evidence for policy or com-parative assertion as it has been demonstrated already in previous LCAsthat variable time-horizons produce significant influences on GWP scores(Basset-Mens et al., 2009; Kendall et al., 2009; Levasseur et al., 2010).

Uncertainty analysis is not commonly performed in LCAs (Huijbregtset al., 2001; Björklund, 2002; Ross et al., 2002) although great effortshave been made on the classification, definition, and sources of uncer-tainty as well as methodological aspects for expressing uncertainty. At

LCI level, publicly available LCA databases, e.g. Eco-profiles databasefor plastic developed by Plastic Europe (Boustead, 2005) only provideaverage inventory data with no uncertainty information. To tackle thisissue, a SETAC-Europe LCA working group was established (Bretz, 1998)and a framework for modeling uncertainties in LCI was developed,where data uncertaintieswere classified into lack of data anddata inaccu-racy. A pedigreematrix approachwas recommended for the estimationofdata inaccuracy (Huijbregts et al., 2001). Sugiyama et al. (2005) sug-gested applying the Pedigree matrix approach to literature-based databut to use statistical methods to quantify the uncertainties in industrialinventories. Unlike the Pedigree matrix approach, which has been intro-duced into LCA database such as Eco-invent (PRéConsultants, 2004), sta-tistical methods have only been used to a limited extent in LCAs tocharacterize the data quality (Capello et al., 2005; Sugiyama et al., 2005;Ciroth and Srocka, 2008; Mullins et al., 2011; Seabra et al., 2011). In theLCIA phase, a range of available approaches for uncertainty analysishave been reviewed by Björklund (2002). Amongst them, Monte Carlosimulation was the most commonly recommended (Contadini et al.,2002; Ciroth et al., 2004; Hung and Ma, 2009). Monte Carlo simula-tions have now been built into commercial LCA software e.g. SimaPro(PRéConsultants, 2004) but are still only applied in few LCA studies(Miller et al., 2006; Basset-Mens et al., 2009; Spatari et al., 2010).

The current research used LCA case studies of biopolymer foams toinvestigate the effects of various time horizons on characterized LCIAprofiles and to present an uncertainty analysis approach integratingstatistical methods into the LCA model.

2. Methodology

2.1. LCA case study

Themainmaterial studied was a wheat-based foam (WBF) provid-ed by Green Light Products Ltd. The major components of WBF arewheat flour derived from a specific variety of winter wheat and poly-vinyl alcohol (PVOH). A cradle-to-grave approach was adopted forthe LCA case-studies of various WBF-based products and potentialproducts. The wheat farming, flour milling, manufacturing of WBFproducts, distribution, WBF product use and end-of-life were includedin the LCA system boundary. The system boundary definition has beengiven by Guo (Guo et al., 2011a). As summarized in Table 1, compari-son of threeWBF concept products with petrochemical polymer coun-terparts are included in the case studies — a coolbox, a display board,and a refractory lining former for concrete formwork.

In the LCA model, the economic allocation approach was adoptedfor most of the stages where multiple-products occurred. The rationalfor choosing this allocation approach is that those co-products in-volved have clear market values but unknown/uncertain substitutes(avoided products) in their second life-cycle. An ‘avoided burdens’ ap-proach was applied in the cases where energy related co-products orclosed-loop recycling occurred. A carbon counting approach followingthe carbon stoichiometry was used to ‘track’ the carbon flows duringthe life cycle of theWBF based products (Guo et al., 2011a). Sensitivityanalysis on the allocation method was not undertaken in the currentstudy, but can be explored in further research. The LCA inventorywas developed by using site-specific primary data collected from in-dustrial sources and results from computer simulations and laboratoryexperiments supplemented with secondary data from publicly avail-able sources and the Ecoinvent database (v2.0) (Guo et al., 2011a).The primary datasets were developed in collaboration with HeygatesLtd, Green Light Product Ltd, Brunel University, Hydropac Ltd andother manufacturers to represent the latest industrial processes forwheat farming and milling and for WBF production and usage. Thesimulation model Denitrification–Decomposition (DNDC) was used toestimate the field emissions from agricultural land duringwheat farming(six fields modeled) (Guo et al., 2011b). The inventories for the anaero-bic digestion (AD) of WBF-based products were derived from a UK

Table 1Functional unit and specifications for case study products (Guo et al., 2011b).

Case studies Functional unit WBF EPS or PE

Case study 1 coolbox Single 8.5 l capacity corrugated box insulated with WBF or LDPEfoam to maintain a temperature below 5 °C for 24 h for thetransport of temperature-sensitive contentsService life — single use, estimated up to 48 h

Density: 25 kg/m3

Thickness: 26 mmWeight: 213.4 g

Virgin LDPEDensity: 35 kg/m3

Thickness: 20 mmWeight: 241.9 g

Case study 2 display board Single display board used for indoor advertisement applicationwith 2 m2 surface area, 10 mm thicknessService life — 3-months

Density: 25 kg/m3

Weight: 0.5 kgHDPE with 20% recycled contentDensity: 85 kg/m3

Weight: 1.7 kgCase study 3 refractory lining former A dome-shape refractory lining former with 2.4 m3 volume

Service life — single useDensity: 70 kg/ m3

Weight: 168 kgVirgin EPSDensity: 25 kg/ m3

Weight: 60 kg

232 M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

commercial AD plant and novel laboratory experimental results (Guoet al., 2011a). The inventories for otherWBF disposal options includinghome composting, industrial composting and landfill were developedusing meta-analysis (Guo, 2010). The EU average data for petrochem-ical product (LDPE/HDPE/EPS) production processes (EUMEPS, 2001;Boustead, 2005) were used in the LCA model to represent the currenttechnology for the PE/EPS pellet production, transformation of PE/PSpellets to LDPE/EPS foams, and extrusion of HDPE. The emissions atuse phase are considered as zero due to the relatively short life-cycleof products modeled. The end-of-life scenarios (recycling, landfilland incineration) for petrochemical products were mainly based onthe Ecoinvent database (v2.0).

2.2. LCA data quality analysis

2.2.1. Sensitivity analysisLCA was performed using SimaPro software (v7.1.8). The default

time horizon recommended in the International Reference Life CycleData System (ILCD) Handbook was adopted in this study (EuropeanCommission, 2011). Thewidely-used CML 2 baseline 2000 (v2.04) char-acterization model was used as the default, ‘static’ LCIA method wherethe time horizons were 100-year for GWP and infinite time for theODP and toxicity potential. Dynamic characterizationmodels incorporat-ed into the CML 2001 method library (v2.0.4) including GWP (20-year,100-year, 500-year), ODP (5-year, 10-year, 15-year, 20-year, 25-year,30-year, 40-year, infinite), and toxicity (20-year, 100-year, 500-year, infi-nite) were applied in sensitivity analysis to investigate the effects of dif-ferent time horizons on the LCIA outcomes. The sensitivity analysisresults therefore show the effects of time-dependency onGHGs, ODP sub-stances and toxic compounds as reflected in the characterized results forGWP, ODP, and human and eco-toxicity potentials of the polymer life cy-cles (biopolymers vs. petrochemicals).

The scenario sensitivity analysis method suggested by Björklund(2002) was applied in our work. This method involves calculating dif-ferent scenarios, to analyze the influence of input parameters on eitherLCIA output results or rankings. In this study, a reversal of the rank orderof theWBF vs. the petrochemical alternatives due to using the dynamiccharacterization model rather than the static default was taken as theindicator that the LCA result was sensitive to time horizon factors. Fora single product system (no comparison with a petrochemical polymeralternative), an arbitrary level of a 10% change in the characterized LCIAprofiles was chosen as the threshold above which the influence of timehorizon was considered to be significant.

2.2.2. Uncertainty analysis

2.2.2.1. Uncertainty analysis of LCA inventory. At LCI level, the uncertain-ty introduced into inventory due to the cumulative effects of input un-certainty and variability of inventory data was quantified by usingeither statistical methods or expert judgment-based approach.

For the industry-based inventory withmultiplemeasurements or asimulated inventory containing variability parameters, the statistical

analysis was performed in MATLAB (The MathWorks, Inc.). Maximumlikelihood estimation (MLE) was used to assess the characteristicparameters of each hypothetical distribution. Two non-parametricgoodness-of-fit (GOF) tests – Chi-square and Kolmogorov–Smirnov(K–S test) tests – were used to compare the probability density func-tion (PDF) of each hypothesized distributionwith the frequency distri-bution of the observed or simulated datasets and to test the nullhypothesis H0 that the observed frequency distribution is consistentwith hypothesized distribution.

The Chi square test statistic is given as

χ2 ¼Xn

i¼1

Oi−Eið Þ2=Ei: ð1Þ

Where n is number of bins; Oi is the observed frequency or numberof counts in bin i, Ei is the expected frequency of the hypothesized dis-tribution in bin i. Based on the degrees of freedom df, defined as n−1,and the χ2 value, the critical values of the Chi-square distribution isgiven to indicate the probability p for the occurrence of a given χ2

value if the null hypothesis is true. If pb (significance level, ¼ 0:05),then the null hypothesis H0 is rejected.

The K–S test statistic is defined as

dmax ¼ supx

F xð Þ−G xð Þj j: ð2Þ

Where the data is grouped into bins; F(x) is cumulative empiricalfrequency and G(x) is the cumulative density function for the hypoth-esized distribution; sup is the supermum of set of data |F(x)−G(x)|for each bin.

According to the total number of data, and the number of binsn, acritical value at significance level α (α=0.05 was applied) is givenfor the K–S test. If dmax is greater than critical value, then null hypoth-esis H0 is rejected.

Both test statistics χ2 and dmax were used as measures of how farthe observed or simulated data deviated from the hypothesized dis-tribution. Thus, when the null hypothesis was rejected, the rank ofχ2 and dmax still indicates the best representative distribution of theobserved and simulated data — the larger the disagreement betweenthe observed and the expected frequency, the greater were the χ2

and dmax value obtained. In most cases, the best distributions identi-fied by the two statistical tests were in agreement. When discrepancyoccurred, the distribution identified by the Chi-square test was usedfor large sample size datasets and K–S test distribution used forsmall datasets. This is to avoid type II errors (null hypothesis is ac-cepted although is actually false) which can be introduced when ap-plying the Chi-square test on small sample sizes where the K–S testis more powerful (Lilliefors, 1967).

For industrial or literature data only represented by a single value,an expert judgment-based approach – the Pedigree matrix – was ap-plied to estimate data uncertainty caused by temporal, geographical,or technological gaps. The Pedigreematrix approach originally developed

233M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

by Weidema and Wesnæs (1996), was further modified and adoptedin other studies (Kennedy et al., 1997; Huijbregts et al., 2001;Björklund, 2002) and has been introduced into the Eco-invent data-base (Frischknecht et al., 2007). This approach transforms the dataquality indicators to probability distributions by representing thedata quality indicator value by a ‘default’ lognormal distribution. Spe-cifically, uncertainty in inventory data is characterized by six characteris-tics (reliability, completeness, temporal geographical and technologicalcorrelation and sample size). Each characteristic is divided to five levelswith a score (1 to 5), and an uncertainty factor in terms of contributionto the square of the geometric standard deviation (SD) is given to eachscore of the six characteristics (Frischknecht et al., 2007).

Another issue is the estimation of the uncertainties introduced intothe LCI by using empirical models, for instance, the GHGs emissionsfrom diesel combustion calculated using the IPCC Tier 1 approach(2006). To estimate the combined uncertainty for the IPCC-derivedGHGs inventory, Monte Carlo simulation recommended in the IPCC2006Guidelines (2006)was adopted in current study. The uncertaintiesin emission factors (EF) were determined according to uncertaintyranges given in the IPCC Guidelines (2006) — it was assumed that PDFof EF is uniform and that the range corresponds to the 95% confidenceinterval. RiskAMPMonte Carlo Add-in Library version 2.97 (Profession-al Edition, Structured Data, LLC) statistical analysis softwarewas used toperform theMonte Carlo simulationwith 5000 iterations at significancelevel α=0.05. The generated sets of IPCC-simulated data were furtheranalyzed by the MLE and GOF statistical methods.

2.2.2.2. Uncertainty analysis of LCIA. At the LCIA level, Monte Carlo sim-ulation was applied to estimate the uncertainties of LCIA results intro-duced by the statistical variability or temporal, geographical ortechnological gaps in the LCI data. Based on the uncertainty of LCIdata expressed as a probability distribution, the Monte Carlo functionbuilt in SimaPro 7.0 software was run with 1000 iterations at a signif-icance level α=0.05.

3. Results and discussion

3.1. Time horizon

The time-dependency of GWP, ODP, and toxicity potential wereexamined in the current study to assess the sensitivities of LCA out-comes to the different time horizons used. The comparison results be-tween WBF and petrochemical polymers in three case studies arepresented in Figs. 1–5.

3.1.1. Time horizon of GWPAs shown in Fig. 1, the extension of the time horizon from 20 years

to 500 years not only reduced the GWP profiles of all WBF life cyclesbut also changed their relative ranking (ranking between variousend-of-life scenarios) and their comparison with petrochemical poly-mer alternatives. These changes are due to differing emission profilesfor the short-lived gas CH4. The GWP of WBF with landfill andcomposting decreased by 40–50% and 30–37% respectively, whereasonly a 13–20% decline was observed in the AD scenario when ex-panding the time horizon from20 to 500 years. ThusWBFwith landfillshifted from being inferior in GWP terms to being a superior system toWBFwith AD due to this time horizon perspective. Similar trendswerefound with the petrochemical polymers: their GWP scores decreasedwith the extending time horizon due to the importance of short-lived emissions, especially CH4 and N2O to the GWP profile. EPS andPE with landfill were more sensitive to time horizon than those withincineration and recycling end-of-life scenarios. Overall, the time hori-zon had a greater influence on WBFs due to their higher CH4 emissionprofiles over their life cycles than the petrochemical polymers. In otherwords, the advantages of WBFs over conventional petrochemical

polymers in the GWP category increased as the time horizon for theGWP model extended.

3.1.2. Time horizon of ODPAs given in Fig. 2, generally, the ODP profiles of petrochemical

polymers were sensitive to the time horizon but this was not so forthe WBFs. The relatively stable ODP profiles of WBF products withtime horizon (increase less than 10%) were mainly driven by the bal-anced ODP profiles for transportation/field operation and PVOH pro-duction processes. With extension of the time horizon, ODP impacts ofthe short-lived gas CBrClF2 emitted from transporting natural gas (naturalgas is the feedstock for PVOH production) declined but this was balancedby increased ODP impacts from the long-lived emission CBrF3 releasedfromcrudeoil production (feedstock for diesel required for transportationand field operations involved in wheat farming). Therefore, the rank be-tween the WBF various end-of-life scenarios remained stable. The ODPscore of HDPE display board with incineration increased 60–70% withthe extended time horizons due to a decline in ODP savings gained froman avoided product, natural-gas-dependent electricity where short livedgas CBrClF2 played an important role. The decrease in ODP scores of EPSproducts with the extended time horizon was due to the short lived gasCBrClF2 emitted from transporting natural gas, which was the dominantenergy source for EPS transformation. These can explain the comparisonbetween WBF and HDPE/EPS — the advantages of HDPE incinerationoverWBFdecreasedwith the extended timehorizon,whereas the inferiorODP score of EPS product to WBF with AD at the shorter time horizonwere reversed under the infinite time horizon. No significant influenceof time horizon on the coolbox comparisonwas observed—WBF coolboxshowed significant environmental advantages over LDPE alternativeacross all time horizons.

Overall, with the extension of the time horizons on ODP, the com-parison of WBF and PE remained relatively stable. The constructioncase study was sensitive to extension of time horizon and the envi-ronmental advantages of WBF over EPS diminished as time horizonextended.

3.1.3. Human toxicity and eco-toxicity time horizonsAll the toxicity impacts generally increased with the extended time

horizon, especially on terrestrial and marine eco-toxicity, where a sub-stantial rise in environmental burdens was observed (Figs. 4, 5). This isdue to heavy metals which are modeled as the long-term emissions pro-ducing higher toxicity impacts over the longer time horizons (Guinée etal., 2001).

Contributional analysis (not shown here) showed that the toxicityprofiles of WBF products were driven by the production of infrastruc-turematerials involved in PVOH andWBF production aswell as by elec-tricity generation and transmission but not driven by the end-of-life.Thus, the relative rank ofWBF life cycle with different waste treatmentsremained stable.

The impacts of the petrochemical product life cycles with landfill orincineration on human toxicity and aquatic eco-toxicity were dominatedby emissions ofmetallic ions (e.g. vanadium, nickel) linked to thesewastedisposal processes. These environmental impact scores increased withthe extended time horizons. This effect was significant enough to changethe ranking between the petrochemical and WBF products. Generallywith increasing time horizon WBF gained more environmental advan-tages over PE and EPS with incineration and landfill end-of-life scenarios.However, the petrochemicals with recycling end-of-life remained equiv-alent or environmentally superior toWBFs on human toxicity and aquaticeco-toxicity across all time horizons. This can be explained by the fact thathuman toxic and aquatic eco-toxic impacts of petrochemical productswith recycling depended on the balance between environmental creditscaused by ‘avoided’ virgin PE or EPS products and environmental burdenscaused by polymermanufacturing or transformation and so kept relative-ly stable across the extended time horizons. For all petrochemical poly-mers (with recycling incineration and landfill), terrestrial eco-toxicity

Fig. 1. Influence of time horizons on characterized GWP profiles (method: CML 2001). A: Case study 1 (unit per coolbox); B: case study 2 (unit per display board); C: case study 3(unit per refractory lining former).

234 M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

Fig. 2. Influence of time horizons on characterized ODP profiles (method: CML 2001). A: Case study 1 (unit per coolbox); B: case study 2 (unit per display board); C: case study 3(unit per refractory lining former).

235M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

Fig. 3. Influence of time horizons on characterized profiles on human toxicity impact categories (method: CML 2001). A: Case study 1 (unit per coolbox); B: case study 2 (unit perdisplay board); C: case study 3 (unit per refractory lining former).

236 M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

Fig. 4. Influence of time horizons on characterized profiles on terrestrial eco-toxicity impact categories (method: CML 2001). A: Case study 1 (unit per coolbox); B: case study 2(unit per display board); C: case study 3 (unit per refractory lining former).

237M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

Fig. 5. Influence of time horizons on characterized profiles on aquatic eco-toxicity impact categories (method: CML 2001). A, B: Case study 1 (unit per coolbox); C, D: case study 2(unit per display board); E, F: case study 3 (unit per refractory lining former).

238 M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

scores were mainly driven by emissions from fuel combustion or electri-cal energy transmission particularly the high energy demands by LDPEand EPS transformation process (transforming from LDPE/PS pellets toexpanded foams) and extra energy required for recycling. Therefore,with the extended time horizon the terrestrial eco-toxic score of HDPEdisplay board with recycling, LDPE and EPS products (with recycling, in-cineration and landfill) significantly increased which strengthened theenvironmental advantages of WBF products over petrochemical coun-terparts.

The overall toxicity impacts of petrochemical polymers were moresensitive to the extension of the time horizon than WBF. Thus withincrease in time horizon, WBFs either strengthened their advantagesover petrochemical polymers ormoved towards being superior or equiv-alent systems to petrochemical products on the toxicity impact catego-ries. LCA comparison results appeared more sensitive to time horizon intoxic impact categories than GWP and ODP. These findings indicate thatfor LCA studies, especially comparative LCAs on materials containinglong-lifetime toxic compounds, dynamic characterization models with

Table 3GOF results for key AD data.

Best-fitted distribution SDb

Chi-square testa K–S testa

Energy recoveryc Lognormal Lognormal 0.334Energy consumptionc GEV Lognormal 6.064%

a Number of bin=10; significance level α=0.05.b SD of observed samples.c Unit for energy recovery is kWh electricity/per m3 biogas; unit for energy con-

sumption is percentage of electricity generated.

239M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

varying time horizons should be adopted in order to provide the bestinformation for policy makers to ensure that policies can be formulatedin full recognition of the effects of different time-frames on the LCA out-comes (e.g. medium-term vs. long-term).

3.2. Uncertainty analysis

3.2.1. Uncertainty analysis of LCI inventoryThe DNDC-simulation results for N2O emissions from a single wheat

field are presented as a detailed example of the application of statisticalmethods at the LCI level. Although the null hypotheses were rejected,the rank of χ2 and dmax indicated the best-fit hypothesized distributionfor the DNDC-simulated N2O emissions was a beta distribution (Table 2).

In the DNDC generated N2O emission, most of the best-fitted distri-butions identified by the two non-parametric methodswere consistent.When discrepancy occurred, the results obtained from Chi-square testtook priority for uncertainty calibration of computer-simulated inven-tory as the sample size was 5000. However, when analyzing industrialinventorywithmultiplemeasurements, best-fitted distributions identi-fied by the K–S test were considered as more representative due to thesmall sample size. For instance, when concerning the data variability inthe AD of WBF products, the energy recovery efficiency and energy con-sumptionwere themajor concerns. Thus daily data collected from a com-mercial AD plant over a three-month period (Jan–March 2009) wereanalyzed to derive statistical dispersions and best-fitted distributionsidentified by K–S test was used in current LCA study (Table 3).

3.2.2. Uncertainty analysis of LCIA results

3.2.2.1. Uncertainty in LCIA profiles of WBF. Based on the probabilitydistribution of the computed category indicator results (e.g. the histo-gram given in Fig. 6A), the uncertainty ranges for the characterizedLCIA profiles of WBF products were derived (e.g. Fig. 6B). Here, theWBF coolbox life cycle with all the components disposed via AD isgiven as a detailed example. The 95% confidence interval given inTable 4 indicated that in 95% of the cases the characterized LCIA re-sults for WBF coolbox life cycle would fall within the range (U, V).The coefficient of variation (CV) is the normalized indicator of disper-sion in the category indicator results. The CV (Table 4) and the errorbars in Fig. 6B suggest that a large degree of uncertainty is introducedinto the toxicity potential impact scores of the WBF coolbox (withAD), especially its aquatic eco-toxicity results, due to the large uncer-tainties in the major toxic drivers i.e. heavy metal emissions. Con-versely, scores for abiotic depletion potential, GWP100, acidificationpotential and eutrophication potential for the WBF coolbox (withAD scenario) showed low variance (Fig. 6B). Similar results werealso observed in other WBF case studies with diverse end-of-life sce-narios — the higher confidence in LCA results on GWP100, abiotic de-pletion, acidification and eutrophication potentials indicated that the

Table 2Example: GOF results for DNDC-simulated N2O field emissions from a single wheatfield.

Hypothesized distributions Chi-square testb K–S testb

Statistic χ2 H0 Statistic dmax H0

Normal 500.384 Reject 0.073 RejectLognormal 420.060 Reject 0.058 RejectUniform 1020.097 Reject 0.248 RejectTriangle 802.023 Reject 0.187 RejectWeibull 507.572 Reject 0.076 RejectRaylaigh 1358.673 Reject 0.323 RejectBeta 418.099 Reject 0.056 RejectGEVa 423.496 Reject 0.067 RejectGamma 420.275 Reject 0.063 Reject

a GEV = generalized extreme value.b Number of bin = 50; significance level α = 0.05.

environmental profiles of WBF products well represented the realityon these impact categories, and also provided a better basis for theLCA comparisons between WBF and petrochemical products.

3.2.2.2. Uncertainty in LCIA comparisons. The Monte Carlo simulationwas run with 1000 iterations at the 95% confidence level to estimateuncertainties in the LCA comparisons between WBF products andpetrochemical counterparts. As illustrated in Fig. 7 where the coolboxis given as a detailed example, uncertainty analysis of the LCIA re-vealed that there was over 85% probability that the WBF option deliv-ered better cradle-to-grave LCIA results than the LDPE product inmost impact categories. As identical end-of-life scenarios (recycling)were modeled for the cardboard components, the results indicatedthat it was quite certain that the WBF with the AD scenario in generalwould be an environmentally superior choice in this thermal packag-ing application than the LDPE even with 100% recycling of the LDPE.However, specifically within the acidification potential and eutrophi-cation potential categories, there is much less certainty and only arather low probability (0–17%) that the WBF with AD scenario incurslower impacts than LDPE with recycling. For fresh water eco-toxicitypotentials the uncertainty analysis also reveals that no clear state-ment can be given about which polymer would offer the more envi-ronmentally friendly choice in this impact category.

Examples of uncertainty analysis results are presented in Table 5,where WBF products with composting and AD scenarios are comparedwith petrochemical polymerswith recycling and landfill scenarios. Gen-erally, therewas an over 85% probability thatWBFwas environmentallysuperior to LDPE and HDPE (exception was the HDPE with recyclingscenario) in most impact categories. In the construction case study,the advantages ofWBF over EPS on POCPwas certain (100% probability)and there were high probabilities that WBF delivered better scores onaquatic eco-toxicity potential and abiotic depletion than EPS with land-fill. However, in other impact categories the uncertainty analysis indi-cated considerable variation with the different end-of-life scenarios.Across all the case studies, WBF consistently showed high probabilitiesof incurring higher impacts on acidification potential and eutrophica-tion potential than the petrochemical polymers.

4. Conclusions

According to the criteria used for our analysis time horizon was dem-onstrated to be a sensitive parameter for the LCIA profiles ofWBF and alsofor the LCIA comparisons between WBF and petrochemical polymers. Inthe GWP and toxicity potentials impact categories WBF either strength-ened its advantages over petrochemical polymers or shifted towardsbeing a superior or equivalent system to these products as the time hori-zon extended from 20 to 500 years or infinite time. These case studiesprovide scientific evidence to support the time horizons recommendedin the ILCD (European Commission, 2011) and also demonstrate that dy-namic LCA characterization models that can represent varying time hori-zons should be applied in LCAs. However, the 100-year time horizon isgenerally used in LCA practice as a single time horizon for GWP and

Fig. 6.Uncertainties analysis forWBF coolbox with AD scenario (unit: per coolbox). A: Probability distribution of characterized GWP100 profiles (number of bin=50); B: Uncertainties forcharacterized LCIA profiles (the error bars represent the uncertainty range in terms of the ratio of the 2.5th and 97.5th percentile U, V to mean value).

240 M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

Table 4Uncertainties for characterized LCIA profiles of WBF coolbox with AD scenario (unit: per coolbox).

Impact category Mean SD CV Ua (2.50%) Va (97.50%)

Abiotic depletion (kg Sb eq) 5.86E−03 1.85E−03 31.50% 3.21E−03 1.04E−02Acidification (kg SO2eq) 7.46E−03 1.73E−03 23.20% 4.66E−03 1.16E−02Eutrophication (kg PO4

3−eq) 1.62E−03 4.35E−04 26.80% 1.03E−03 2.55E−03GWP100 (kg CO2 eq) 8.03E−01 2.43E−01 30.30% 4.06E−01 1.35E+00ODP (kg CFC-11 eq) 8.14E−08 3.10E−08 38.00% 3.97E−08 1.50E−07Human toxicity (kg 1,4-dB eq) 2.45E−01 9.87E−02 40.30% 1.16E−01 4.92E−01Fresh water eco-toxicity (kg 1,4-dB eq) 6.12E−02 3.29E−02 53.80% 2.67E−02 1.55E−01Marine eco-toxicity (kg 1,4-dB eq) 8.84E+01 4.30E+01 48.70% 3.70E+01 1.90E+02Terrestrial eco-toxicity (kg 1,4-dB eq) 2.30E−03 8.88E−04 38.60% 1.14E−03 4.72E−03POCP (kg C2H4) 2.49E−04 1.03E−04 41.30% 1.09E−04 5.13E−04

a U: 2.5th percentile V 97.5th percentile.

241M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

infinite time is the commonly applied single time horizon for ODP andtoxicity potentials impact categories. Based on our analysis and publishedwork cited elsewhere in this paper it is clear that use of a single time ho-rizon can introduce inadvertent bias into the LCA outcomes. While it isdesirable to adopt a default time horizon e.g. 100 years for GWP and in-finity for other impact categories, there is a clear need to consider theimplications of alternative timehorizons for LCA outcomes and interpre-tation. We recommend that in general LCA studies, time horizons vary-ing from 20-years to infinite-time for impact categories like GWP, ODPand toxicity potentials should be examined in the LCIA phase as a mea-sure of robustness for LCAs, especially comparative LCAs, in order to de-liver unbiased information for policy makers. Even for specific LCAstudies providing information for policy-making with targeted time ho-rizon (e.g. next 100 year), the different time perspective should be stillexamined to provide transparent LCIA results with full recognition ofthe effects of various time horizons.

The use of standard statistical methods needs to be encouraged inLCAs. This study presented an approach to integrate statistical meth-ods into LCA research for analyzing uncertainty in industrial andcomputer-simulated datasets. In the case studies, we calibrated prob-ability of the LCA outcomes for WBF products arising from uncertaintyin the inventory and data variation characteristics which has beenvaluable in assigning confidence to the specific LCIA outcomes (WBFvs. petrochemical polymers). In particular this approach enabled usto identify for particular products the impact categories in which

-20%-30%-40%-50%-60%-70%-80%-90%

Abiotic depletion

Acidification

Eutrophication

Global warming (GWP100)

Ozone layer depletion (ODP)

Human toxicity

Fresh water aquatic ecotox.

Marine aquatic ecotoxicity

Terrestrial ecotoxicity

Photochemical oxidation

Fig. 7.Monte-Carlo simulation results of characterized LCIA comparison betweenWBF and LDcoolbox with AD of WBF plus cardboard recycling scenario).

benefit forWBF over petrochemical alternatives was clear and, equallyimportantly, other categories where this was much less so or was ab-sent. Uncertainty combined with the sensitivity analysis carried outin this study has led to a transparent increase in confidence in theLCAfindings. At the same time, it suggests that LCAs lacking explicit in-terpretation of the degree of uncertainty and sensitivities are of limit-ed value as robust evidence for policy or comparative assertions.

Finally, several key LCA data quality issues have emerged from thisresearch,where further improvements are needed to facilitate the deliv-ery of robust conclusions and unbiased information: 1) themethodolog-ical rigidity of characterization models particularly the rigid timehorizon effects; 2) lack of uncertainty analysis and interpretation in LCIreporting and LCIA outcomes; 3) integration of statistical methods intoLCAs. Thus, further exploration of methodological aspects in data qualityanalysis for LCI and LCA studies is needed in order to enable clear dem-onstration of how aspects of representativeness, robustness and confi-dence in inventory data and LCIA are handled within a given LCA study.

Acknowledgments

This study is based on research financially supported by the UKDepartment of Trade and Industry through the Technology Programme.We would like to thank all the participants in the consortium projectled by Green Light Products Ltd and Brunel University.

A < B A >= B

100%90%80%70%60%50%40%30%20%10%0%-10%

PE coolbox (unit: per coolbox; A = LDPE coolbox with 100% recycling scenario; B =WBF

Table 5Uncertainty analysis for LCIA comparisons of WBF vs. petrochemical products over their life cycle (indicated by probability).

Notes:

= Over 50% probability that WBF delivers lower impacts than petrochemical polymer.= Over50% probability that WBF delivers higher impact than petrochemical polymer.

242 M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

243M. Guo, R.J. Murphy / Science of the Total Environment 435–436 (2012) 230–243

References

Basset-Mens C, Kelliher FM, Ledgard S, Cox N. Uncertainty of global warming potentialfor milk production on a New Zealand farm and implications for decision making.Int J Life Cycle Assess 2009;14:630–8.

Björklund A. Survey of approaches to improve reliability in LCA. Int J Life Cycle Assess2002;7:64–72.

Boustead I. Eco-profiles of the European Plastics Industry. Pastics Europe: Associationof Plastics Manufacturers; 2005.

Bretz R. SETAC LCAWorkgroup: data availability and data quality. Int J Life Cycle Assess1998;3:121–3.

BSI. PAS 2050: 2011 Specification for the assessment of the life cycle greenhouse gasemissions of goods and services; 2011.

Calzonid J, Casperseni N, Dercasc N, Gaillarde G, Gosseh G, Hanegraafb M, et al. Bioenergyfor Europe: Which ones fit best?- A comparative analysis for the Community; 2000.

Capello C, Hellweg S, Badertscher B, Hungerbuhler K. Life-cycle inventory of waste solventdistillation: Statistical analysis of empirical data. Environ Sci Technol 2005;39:5885–92.

Cherubini F, Ulgiati S. Crop residues as raw materials for biorefinery systems - A LCAcase study. Appl Energy 2010;87:47–57.

Ciroth A, Srocka M. How to obtain a precise and representative estimate for parametersin LCA - A case study for the functional unit. Int J Life Cycle Assess 2008;13:265–77.

Ciroth A, Fleischer G, Steinbach J. Uncertainty calculation in life cycle assessments - Acombined model of simulation and approximation. Int J Life Cycle Assess 2004;9:216–26.

Contadini JF, Moore RM, Mokhtarian PL. Life cycle assessment of fuel cell vehicles - Amethodology example of input data treatment for future technologies. Int J LifeCycle Assess 2002;7:73–84.

Dreyer L, Niemann A, Hauschild M. Comparison of Three Different LCIA Methods:EDIP97, CML2001 and Eco-indicator 99. Int J Life Cycle Assess 2003;8:191–200.

Estermann R, Schwarzwälder B, Gysin B. Life cycle assessment of Mater-Bi and EPSloose fills Study prepared by COMPOSTO for Novamont, Novara, Italy; 2000.

EUMEPS. Life Cycle Assessment of Industrial Use of Expanded Polystyrene Packaging InEurope. Brussels: European Manufacturers of Expanded Polystyrene; 2001.

European Commission-Joint Research Centre - Institute for Environment andSustainability: International Reference Life Cycle Data System (ILCD). Handbook-Recommendations for Life Cycle Impact Assessment in the European context. 1stedition. Luxemburg: Publications Office of the European Union; 2011. November2011. EUR 24571 EN.

Finnveden G. Methodological aspects of life cycle assessment of integrated solid wastemanagement systems. Resour Conserv Recycl 1999;26:173–87.

Finnveden G, Hauschild MZ, Ekvall T, Guinee J, Heijungs R, Hellweg S, et al. Recentdevelopments in Life Cycle Assessment. J Environ Manage 2009;91:1-21.

Forster P, Ramaswamy V, Artaxo P, Berntsen T, Betts R, Fahey DW, et al. Changes in Atmo-spheric Constituents and in Radiative Forcing. In: Solomon S, Qin D, Manning M,Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL, editors. Climate Change 2007:The Physical Science Basis. Contribution of Working Group I to the Fourth Assess-ment Report of the Intergovernmental Panel on Climate Change. Cambridge, UnitedKingdom and New York, NY, USA: Cambridge University Press; 2007.

Frischknecht R, Jungbluth N, Althaus H-J, Doka G, Heck T, Hellweg S, et al. Overviewand Methodology. ecoinvent report No.1. Dubendorf: Swiss Center for Life CycleInventories; 2007.

Gnansounou E, Dauriat A, Panichelli L, Villegas J. Energy and greenhouse gas balancesof biofuels: biases induced by LCAmodelling choices. J Sci Ind Res 2008;67:885–97.

Goedkoop M, Spriensma R. The Eco-indicator 99 A damage oriented method for LifeCycle Impact Assessment Methodology Report; 2001.

Guinée JB, GorreeM, Heijungs R, Huppes G, Kleijn R, Koning A, et al. Life Cycle AssessmentAn Operational Guide to the ISO Standards Final report (part 1, 2, 3). Ministry ofHousing, Spatial Planning and the Evironment & Center of Environmental Science-Leiden University; 2001.

Guo M. Life Cycle Assessment (LCA) of Light-weight Eco-composites. Ph.D. Dissertation,Imperial College London, 2010.

Guo M, Trzcinski AP, Stuckey DC, Murphy RJ. Anaerobic digestion of starch–polyvinylalcohol biopolymer packaging: Biodegradability and environmental impact assessment.Bioresour Technol 2011a;102:11137–46.

Guo M, Li C, Bell JNB, Murphy RJ. Influence of Agro-Ecosystem Modeling Approach onthe Greenhouse Gas Profiles of Wheat-Derived Biopolymer Products. Environ SciTechnol 2011b;46:320–30.

Hauschild MZ, Huijbregts M, Jolliet O, Macleod M, Margni M, van de Meent D, et al.Building a Model Based on Scientific Consensus for Life Cycle Impact Assessmentof Chemicals: The Search for Harmony and Parsimony. Environ Sci Technol2008;42:7032–7.

Huijbregts MAJ. Priority Assessment of Toxic Substances in the frame of LCA–Time horizondependency in toxicity potentials calcuated with the multi-media fate, exposure and

effects model USES-LCA. University of Amsterdam, Amsterdam, The netherlands:Institute for Biodiversity and Ecosystem Dynamics; 2000.

Huijbregts MAJ, Thissen U, Guinee JB, Jager T, Kalf D, van de Meent D, et al. Priorityassessment of toxic substances in life cycle assessment. Part I: Calculation of toxicitypotentials for 181 substances with the nested multi-media fate, exposure and effectsmodel USES-LCA. Chemosphere 2000;41:541–73.

Huijbregts MAJ, Norris G, Bretz R, Ciroth A, Maurice B, von Bahr B, et al. Framework formodelling data uncertainty in life cycle inventories. Int J Life Cycle Assess 2001;6:127–32.

Hung ML, Ma HW. Quantifying system uncertainty of life cycle assessment based onMonte Carlo simulation. Int J Life Cycle Assess 2009;14:19–27.

IPCC. 2006 IPCC Guidelines for National Greenhouse Gas Inventories National Green-house Gas Inventories Programme; 2006.

ISO. Environmental management —Life cycle assessment —Principles and framework.BS EN ISO 14040. London: British Standards Institution; 2006.

Kendall A, Chang B, Sharpe B. Accounting for time-dependent effects in biofuel lifecycle greenhouse gas emissions calculations. Environ Sci Technol 2009;43:7142–7.

Kennedy D, Montgomery D, Rollier D, Keats J. Data Quality. Int J Life Cycle Assess1997;2:229–39.

Kim S, Dale BE. Allocation procedure in ethanol production system from corn grain - I.System expansion. Int J Life Cycle Assess 2002;7:237–43.

Kim S, Dale BE. Regional variations in greenhouse gas emissions of biobased productsin the United States-corn-based ethanol and soybean oil. Int J Life Cycle Assess2009;14:540–6.

Laser M, Larson E, Dale B, Wang M, Greene N, Lynd LR. Comparative analysis of efficiency,environmental impact, and process economics for mature biomass refining scenarios.Biofuels Bioprod Biorefin 2009;3:247–70.

Levasseur A, Lesage P, Margni M, Deschenes L, Samson R. Considering Time in LCA:Dynamic LCA and Its Application to Global Warming Impact Assessments. EnvironSci Technol 2010;44:3169–74.

Lilliefors HW. On the Kolmogorov-Smirnov Test for Normality with Mean and VarianceUnknown. J Am Stat Assoc 1967;62:399–402.

Luo L, van der Voet E, Huppes G, de Haes HAU. Allocation issues in LCA methodology: acase study of corn stover-based fuel ethanol. Int J Life Cycle Assess 2009;14:529–39.

Miller SA, Landis AE, Theis TL. Use of Monte Carlo analysis to characterize nitrogenfluxes in agroecosystems. Environ Sci Technol 2006;40:2324–32.

Morais S, Martins AA, Mata TM. Comparison of allocation approaches in soybean biodiesellife cycle assessment. J Energy Inst 2010;83:48–55.

Mullins KA, Griffin WM, Matthews HS. Policy Implications of Uncertainty in ModeledLife-Cycle Greenhouse Gas Emissions of Biofuels. Environ Sci Technol 2011;45:132–8.

PRéConsultants. Simapro Database Manual; 2004.Rosenbaum R, Bachmann T, Gold L, Huijbregts M, Jolliet O, Juraske R, et al. USEtox—the

UNEP-SETAC toxicity model: recommended characterisation factors for humantoxicity and freshwater ecotoxicity in life cycle impact assessment. Int J Life CycleAssess 2008;13:532–46.

Ross S, Evans D, Webber M. How LCA studies deal with uncertainty. Int J Life Cycle Assess2002;7:47–52.

Seabra JEA, Macedo IC, Chum HL, Faroni CE, Sarto CA. Life cycle assessment of Braziliansugarcane products: GHG emissions and energy use. Biofuels Bioprod Biorefin2011;5:519–32.

Solomon S, Albritton DL. Time-Dependent Ozone Depletion Potentials for Short-Termand Long-Term Forecasts. Nature 1992;357:33–7.

Spatari S, Bagley DM, MacLean HL. Life cycle evaluation of emerging lignocellulosic ethanolconversion technologies. Bioresour Technol 2010;101:654–67.

Sugiyama H, Fukushima Y, Hirao M, Hellweg S, Hungerbuhler K. Using standard statisticsto consider uncertainty in industry-based life cycle inventory databases. Int J LifeCycle Assess 2005;10:399–405.

UNEP. TheMontreal Protocol on Substances that Deplete the Ozone Layer. United NationsEnvironment Programme; 1999.

USEPA. Guidelines for Assessing the Quality of Life Cycle Inventory Analysis. Washington,DC: United States Environmental Protection Agency; 1995.

Weidema BP, WesnæsMS. Data quality management for life cycle inventories–an exampleof using data quality indicators. J Cleaner Prod 1996;4:167–74.

WMO. Scientific Assessment Of Ozone Depletion: 2006. Global Ozone Research andMonitoring Project—Report No. 50. Geneva, Swizerland: World MeteorologicalOrganization; 2007.

Wuebbles DJ, Jain AK, Patten KO, Connell PS. Evaluation of ozone depletion potentials forchlorobromomethane (CH2ClBr) and 1-bromo-propane (CH2BrCH2CH3). AtmosEnviron 1998;32:107–13.