Eco-efficiency of electric vehicles in the United States: A life cycle assessment based principal component analysis

Journal of Cleaner Production 212 (2019) 515e526

Contents lists available at ScienceDirect

Journal of Cleaner Production journal homepage: www.elsevier.com/locate/jclepro

Eco-efficiency of electric vehicles in the United States: A life cycle assessment based principal component analysis Nuri C. Onat a, *, Murat Kucukvar b, Shiva Afshar c a

Qatar Transportation and Traffic Safety Center, College of Engineering, Qatar University, Doha, Qatar Department of Mechanical and Industrial Engineering, College of Engineering, Qatar University, Doha, Qatar c Department of Industrial Engineering, College of Engineering and Natural Sciences, Istanbul Sehir University, Istanbul, Turkey b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 8 August 2018 Received in revised form 14 November 2018 Accepted 5 December 2018 Available online 6 December 2018

This research presents an integrated sustainability assessment framework applied to electric vehicle technologies in the United States of America. Two methods; principal component analysis and life cycle assessment are jointly used to present a novel integrated framework for eco-efficiency analysis of battery electric vehicles for each state in the USA. Three electricity production scenarios; 1) marginal electricity mix; 2) average electricity mix; and 3) 100% solar energy are investigated. Three environmental (water withdrawal, energy consumption and carbon emission) and one economic indicator as life cycle costing are merged to obtain the eco-efficiency scores of each state. The scenarios are compared by applying ANOVA and Tukey/HSD test regarding their environmental and economic values. Then, a comparison is done based on the eco-efficiency values of states in each scenario, separately. The results showed that the maximum eco-efficiency scores are obtained in three states such as Indiana, Texas and New Mexico based on marginal electricity scenario, average electricity mix scenario and solar energy scenario, respectively. The findings also revealed that 100% solar charging scenario is the most environmentally friendly option because of the environmental impacts in terms of water, energy and carbon footprints. The researchers concluded that the proposed integrated framework for eco-efficiency of electric vehicle technologies has a strong application potential for policy making in sustainability performance assessment where multiple sustainability indicators' are aimed to be integrated into the decision making process, especially to deal with the multi-collinearity associated with environmental life cycle impact data. © 2018 Published by Elsevier Ltd.

Keywords: Life cycle assessment Principal component analysis Eco efficiency Electric vehicles Carbon-energy-water footprints

1. Introduction Over the past decades, environmental problems such as climate change, water pollution and depletion of natural resources have become major global concerns. An increasing energy consumption in transportation networks is considered one of the main causes of today’ major environmental problems (Ercan et al., 2016). These concerns are even more highlighted in developed countries such as United States of America (USA) due to a vastly growing transportation sector (Alirezaei et al., 2017; Ercan et al., 2017). The U.S. transportation sector consumes approximately 30% of the total energy used in the country and responsible for 92% of this petroleum-based energy demand The amount of fuel required to

* Corresponding author. E-mail address: [email protected] (N.C. Onat). https://doi.org/10.1016/j.jclepro.2018.12.058 0959-6526/© 2018 Published by Elsevier Ltd.

meet the growing transportation demand accounts for nearly 70% of the total oil consumption in U.S., and around 65% of this amount is used by personal vehicles in the USA (EIA, 2018). This significant fuel consumption makes the transportation sector the second largest carbon emitter after the power generation and supply sector. To this end, alternative technologies such as electric vehicles (EVs) have been considered as sustainable solutions worldwide (Onat, 2015a). Especially, lower tailpipe emissions and energy consumption of EVs compared to internal combustion vehicles (ICVs) make EVs more sustainable options, depending on the source of electricity generation (Hawkins et al., 2013). 1.1. The state-of-the art: sustainability assessment of electric vehicles According to literature review studies (Hawkins et al., 2012; € f et al., 2014; Onat et al., 2018), Global Warming Potential Nordelo

516

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

(GWP) and energy consumption are the top two environmental impact categories studied for electric vehicles. On the other hand, the social and economic impacts of electric vehicles are not sufficiently studied. Only a couple of studies addressed the social and economic impacts. Life cycle assessment is the mainly utilized approach for assessing the sustainability impacts of electric vehicles. Life cycle assessment (LCA) is a widely used systematic approach to quantify the environmental impacts of products, e et al., 2002; Guine e et al., 2011; processes, or services (Guine Onat et al., 2017a). The LCA technique is widely used method because of its ability for analyzing environmental impacts of transportation activities for all life cycle phases such as raw material extraction and processing, production, distribution, use and end-of-life (Heijungs et al., 2010; Kucukvar et al., 2014a,b; Onat et al., 2014a,b; Tatari et al., 2015). In the literature, LCA has been applied to assess the sustainability impacts (environmental, economic, and social) of buildings (Nuri C. Onat et al., 2014a), electricity generation (Kucukvar et al., 2018, 2017; Shaikh et al., 2017), manufacturing sectors (Kucukvar et al., 2016, 2015), construction sectors (Kucukvar et al., 2014a,b), food supply chains (Kucukvar and Samadi, 2015; Park et al., 2016), electric vehicles (Bartolozzi et al., 2013; Daniel and Rosen, 2002). For example, Samaras and Meisterling (2008) utilized the LCA to measure carbon emissions of plug-in hybrid electric vehicles (Faria et al., 2012). used LCA approach to compare EVs and gasoline vehicles based on their economic and environmental impacts. Onat et al. (2015) used the hybrid LCA to compare the sustainability of multiple electric vehicle types for the 50 states in the USA based on the driving patterns, battery structures and energy preparation scenarios. Onat et al., (2014a,b) assessed environmental impacts of conventional vehicles, hybrid electric vehicles, battery electric vehicles and plug in hybrid electric vehicles in the USA based on 19 indicators considering three different charging scenarios. Onat et al. (2016) presented an integrated approach combining inputoutput based LCA and multicriteria optimization technique to determine the most optimal passenger vehicle distribution for the USA (Onat, 2015b). A wide range of integrated LCA approaches are developed in the literature. To name a few, Life Cycle Sustainability Assessment (LCSA) along with system dynamics modeling (Onat et al., 2016a, 2016b), LCSA þ fuzzy TOPSIS (Onat et al., 2016c), LCA þ agent based modeling (Noori et al., 2016), LCA þ Data Envelopment Analysis along with agent based modeling (Onat et al., 2017b), and multi-region input-output based LCA (Zhao et al., 2016) have been applied to investigate the sustainability impacts of alternative vehicle technologies. In sustainability assessment of these technologies, one of the important issues is to deal with multiple sustainability indicators. The LCA technique is jointly applicable with one of the dimension reduction techniques to assess the environmental impacts of a system including large number of environmental indicators (Egilmez and Park, 2014). Dimension reduction provides reliable results in computing eco-efficiency, which is one of the metrics widely applied in many studies for assessing the sustainability performance when dealing with multiple indicators (Egilmez et al., 2015; Park et al., 2015; Tatari and Kucukvar, 2012). This is mostly because of the fact that eco-efficiency captures both economic and environmental sustainability indicators in the computations (Huppes and Ishikawa, 2005). While all the works focusing on improving the sustainability assessment approaches of electric vehicles answered various important questions, the issue of multi-collinearity and generation of a composite sustainability metric remain as a challenge. Furthermore, in the literature no study found analyzing eco-efficiency of electric vehicles.

1.2. State-of-the art: eco-efficiency analysis Creating composite sustainability indices have been a complicated challenge when considering many environmental and economic indicators with different measuring units. In order to reduce the complexity of computations, weighting models are utilized to reduce the dimension of these variables (Cerutti et al., 2013). The results obtained from these models are influenced by the weights assigned for each indicator. Some linear programing techniques such as Data Envelopment Analysis (DEA) and Principle Component Analysis (PCA) become suitable alternatives because of their independency to the subjective weights (Park et al., 2015). The DEA approach is applicable to measure the environmental impact of a system for multi-attributed data and has the capability to deal with spurious, modal and outlier data (Cook and Zhu, 2007). However, the results obtained from this approach are highly sensitive to the correlation exists between the sustainability indicators. If the indicators are correlated to each other, the PCA becomes more suitable approach due to dealing with correlated indicators and obtaining rigorous results. PCA has been used frequently in the literature as a dimension reduction technique to produce composite sustainability indexes by generating one or a few new indicators in order to simplify the computations (Li et al., 2012). Salvati and Carlucci (2014) used a PCA method as a case study to determine the contribution among 99 indicators and determined their contribution in sustainability index, which is obtained from applying the factor-weighting model. Reisi et al. (2014) obtained a sustainability index for transportation in Melbourne using the PCA approach to combine nine social, environmental and economic rova  and Kolosta (2015) ranked 27 countries in indicators. Bolca Europe by considering their aggregated sustainability development index regarding environmental, social and economic indicators by using the PCA approach. Mascarenhas et al. (2015) used PCA to reduce the number of indicators to compute the sustainability score of Algrave's spatial plan in Pourtogul. Mascarenhas et al. (2015) applied the PCA approach to find a composite sustainability index to assess the performance of ten energy systems for the rural electrification industy in India. Jiang et al. (2018) proposed a composite three-dimensional (economic, environmental, and social) sustainability assessment model using PCA to analyze corporate sustainable performance based on The World Business Council for Sustainable Development (WBCSD) defined eco-efficiencyhttps://www.sciencedirect.com/science/article/pii/ S0959652617300434 - fn1 as: “… the delivery of competitively priced goods and services that satisfy human needs and bring quality of life, while progressively reducing ecological impacts and resource intensity throughout the life-cycle to a level at least in line with the Earth's estimated carrying capacity” (Ehrenfeld, 2005). BASF, one of the leading chemical companies in the world, has used eco-efficiency analysis, based on LCA according to ISO 14040 rules, to assess the economic and environmental impacts of chemicals, processes, and products (Lozano and Lozano, 2018; Saling et al., 2002). On the other hand, one study was found in the literature, which presented an integrated approach combining LCA and PCA. The researchers used the EIO-LCA þ PCA approach to compute the eco-efficiency score in 273 industrial sectors in the USA (Park et al., 2015). 1.3. Research motivation and novelty In the literature, life cycle impacts of electric vehicles are highly studied. These models usually analyzed the environmental and economic impacts in isolation without a proper integration of economic and environmental dimensions of sustainability using a composite metrics such as eco-efficiency or sustainability

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

performance index. While life cycle assessment is considered a systemic method for conducting comprehensive environmental impact analysis, there is no research found for eco-efficiency analysis of emerging electric vehicle technologies. On the other hand, earlier studies showed that there can be significant positive correlations between environmental footprint categories of transportation sectors based on results obtained from life cycle assessment models (Choi et al., 2015; Park et al., 2015). Considering the fact that environmental indicators can be highly correlated to each other, a combination of PCA and LCA will become a novel approach due to dealing with correlated indicators and obtaining composite eco-efficiency metrics, which considers not environmental impacts but also economic value added. Furthermore, eco-efficiency analysis will provide a holistic understanding of how emerging electric vehicle technologies will affect the economic and environmental indicators, simultaneously. Therefore, current method presents the first empirical analysis on eco-efficiency analysis of alternative electric vehicle technologies in the United States of America. To realize this goal, two methods such as PCA and LCA are jointly applied for the eco-efficiency analysis of batter electric vehicles. In this paper, we applied PCA and LCA jointly to rank the sustainability performances of electric vehicles for three different scenarios. These three scenarios are named as; 1) state-based average electricity generation mix scenario considers the average electricity generation in the U.S., 2) state-based marginal electricity mix generation scenario is based on the marginal electricity generation in the U.S. and 3) 100% solar power charging stations scenario just utilize solar energy as the resource of energy for battery charging system. The LCA results for environmental impacts for all these three scenarios are obtained from (Onat et al., 2015). Based on these three scenarios, the sustainability performances of battery electric vehicles (BEVs) are evaluated in the operational phase of their life cycle, which is the most environmental-impact (carbon, energy, water) intensive phase and highly sensitive to regional variations. In this research, eco-efficiency is used as metric to assess the sustainability, which provides a metric that combines economic benefits and environmental impacts. To assess the environmental impacts regarding the selected environmental indicators such as carbon footprint, energy use and water consumption, a novel twophased model of PCA and LCA (the so-called PCA þ DEA framework) is developed. This integration enabled us to rank each state based on their eco-efficiency values for each scenario. To this end, the rest of the paper is organized as follows. The methodology and data description is explained in section 2. The results of life cycle inventory, eco-efficiency, and ANOVA and Tukey/HSD tests are presented in Section 3. Finally, the conclusion, limitation of the work and future work are presented in Section 4. 2. Methods 2.1. LCA and PCA approach Three environmental and one economic indicator are considered to obtain the eco-efficiency as the ratio of economic output to environmental index. In order to obtain a specific value for environmental impact index LCA and PCA approaches are jointly used. The life cycle impacts corresponding to the environmental and economic indicators computed by applying LCA is the input of PCA to make a specific value as the composite environmental impacts (CEI). Fig. 1 shows each step when calculating CEI and ecoefficiency metric. Application of PCA method is explained in section 2.5 in detail. In order to prevent having negative values of CEI a large enough positive number should be added to the output of PCA. To calculate the eco-efficiency, both direct and indirect economic outputs are considered. The total economic activity

517

throughout the life cycle of BEVs (LCC) is an indicator of economic contribution in the terms of economic value added to gross domestic product (GDP) through consumption. The LCC is the nominator and the CEI is the denominator of the eco-efficiency ratio. 2.2. Life cycle assessment of battery electric vehicles The operation phase is the most energy-water-carbon intensive phase as well as spatially more sensitive compared to manufacturing and end-of-life phases (N.C. Onat et al., 2016). Therefore, the manufacturing and end-of-life impacts are not considered. The functional unit of the LCA is per vehicle-miles traveled (VMT). The operation phase impacts are composed of well to tank (WTT) and tank-to-wheel (TTW), which are upstream and direct impacts, respectively. Since there is no direct water consumption and tailpipe emissions in the operation phase of BEVs, TTW carbon emissions and water consumptions are zero for BEVs, regardless of the spatial variations. However, there are energy consumption in both WTT (the amount of energy required to generate electricity) and TTW (the amount of energy consumed during travel of a BEV) phases. Hence, the environmental impacts of BEVs can be calculated as follows: Fc,i ¼ FC x (WTTc,I þ TWWc,i)

(1)

Where, F is the footprint for the impact category c in state i. FC is per mile fuel consumption in kWh. WTT and TWW stand for well to tank and tank-to-wheel phase impacts in impact category c in state i. WTT impacts are calculated based on state-specific energy mixes. TTW energy consumption is equal to direct energy consumption of an average BEV, which is approximately 0.3 kWh. Similarly, life cycle cost impacts are obtained from literature (Noori et al., 2015) for the same vehicle type applying the same assumptions. Fig. 2 shows the boundary of the analysis. The analysis is conducted for three different electricity generation scenarios; 1) state-based average electricity generation mix scenario considers the average electricity generation in the U.S., 2) state-based marginal electricity mix generation scenario is based on the marginal electricity generation in the U.S. and 3) 100% solar power charging stations scenario just utilize solar energy as the resource of energy for battery charging system. In LCA, the electricity generation mixes play very important role when quantifying the upstream environmental impacts of electric vehicles. The representative scenarios are reflecting the best, average, and worst cases. For example, the average electricity generation mix, represent a mid-range (business as usual), the solar scenario represent the best option where the electric vehicles perform the best in terms of environmental impacts. On the other hand, marginal electricity generation mix represents the worst scenario where the electricity required to run electric vehicles is mainly provided by fossil fuels and limiting electric vehicles’ environmental impact reduction potential, even sometimes make them worse options compared to the tradition vehicle types. Considering the rationale behind the selection of the scenarios in this analysis, all other possible impacts would fall into the interval of these three results. For more detail information about how the LCA impacts are calculated, the complete life cycle inventory, and data source, please see Onat et al. (2015) and Noori et al. (2015). 2.3. The application of PCA To compose the environmental sustainability index, one of the linear programming techniques is utilized to make a combination of three environmental impacts. PCA is one of the approaches used for unsupervised (data without any response variable) multiattribute and highly correlated data. PCA is based on a linear

518

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

Fig. 1. Analysis steps to calculate Eco-efficiency.

Fig. 2. System boundary of the LCA.

programming approach, which is widely used for reducing the dimension of multi-attribute data. This approach makes one or several components (principle components) as new variables (Zi ), which are a linear combination of the main indicators, while there is no correlation between the components. Among all the components, only a few first components include the most information and variance from the dataset. Therefore, they are kept as new variables and the remaining are removed from the calculations. The mathematical framework of PCA is shown in given in Eq. (2) (James et al., 2013).

Z1 ¼ at1 ¼ a11 x1 þ a12 x2 þ …a1n xn ; Z2 ¼ at2 ¼ a21 x1 þ a22 x2 þ …a2n xn ; « Zp ¼ atp ¼ ap1 x1 þ ap2 x2 þ …apn xn :

(2)

Where Z1 ;Z2 ;…;Zp are the components and the aij is the coefficient

of xj in ith component. The coefficient of each xj is obtained by dividing its related eigenvector over the square root of its eigenvalue. Each individual component is computed as a linear combination of the variables to cover the most of the information in the dataset with the largest variance and each component is orthogonal to its previous components.

2.3.1. Data normalization The output obtained from LCA technique is a matrix consists of the states of U.S. as the rows and three environmental indicators and economic output as the columns. This matrix is used as the base of the following calculations. Since the data obtained from the LCA has different measuring units, a normalization technique is used to reduce the lopsidedness and the magnitude of environmental and economic output variables by executing a log transformation technique. This normalization will lead to have more

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

2.4. Mathematical framework for eco-efficiency

accurate results of PCA. 2.3.2. Finding the correlation matrix of selected indicators After normalizing data, correlation matrix of three environmental indicators is computed in Eq. (3). The indicators with correlation values close to 1 (or 1) have a strong correlation.

rij ¼

1 n1

n X

Xsi Xsj

519

(3)

For calculating the eco-efficiency as a sustainability metric for the performance of electric vehicles regarding both environmental and economic aspects, the raw eco-efficiency is defined as the ratio of life cycle cost (LCC) to composite environmental impacts (CEI).

Eco  efficiency ¼

Econmic Output ðLCCÞ Composite Environmental Impacts ðCEIÞ

s¼1

(8)

Where rij is the correlation coefficient of the indicator i and j, Xsi and Xsj are the amount of indicator i in state s and Xsj are the amount of indicator j in state s.

In order to make the eco-efficiency score comparable between the states ; the raw eco-efficiency values are rescaled by applying a min-max technique, which is also performed by Park et al. (2015).

2.3.3. Computing eigenvalues and eigenvectors In order to decide about the number of components in the PCA, eigenvalues and eigenvectors are calculated in Eq. (4) and Eq. (5), respectively.

NormalizedðEi Þ ¼

jR  lIj ¼ 0

(4)

Ei  Emin Emax  Emin

(9)

Ei is the raw eco-efficiency value for state i and Emin and Emax are the minimum and maximum values of eco-efficiency among all the states, respectively.

where R is the indicators correlation matrix and l represents the eigenvalues and I is the unit matrix. Number of eigenvalues that obtained by solving Eq. (4) is equal to the number of principal components.

3. Results

  R  lj I Fj ¼ 0

By utilizing the LCA and economic output values of the states for each scenario, a comparison between three scenarios is performed by applying the Analysis of Variance (ANOVA), which is a statistical analysis technique utilized to compare the means of several populations in previous studies (Dominguez et al., 2015; Jothi Basu et al., 2015). ANOVA serves as a statistical tool to estimate the differences in groups. This comparison is essentially a statistical hypothesis testing in which the null hypothesis (H0) is that all population means are equal with confidence of 1- H0. ANOVA considers the proportion of variance between the populations over the variance within the populations and calculates the F-value. For the large amount of F-value, it will be more likely to reject the null hypothesis. If the corresponding p-value of F-value which is extracted from F distribution is less than the a, the null hypothesis is rejected and claim that the means of the populations are not lezequal. The framework of ANOVA is shown in Eq. (8) (Gonza Rodríguez et al., 2012):

(5)

where lj is the eigenvalue of component j and Fj is its eigenvector value (Soler Rovira and Soler Rovira, 2009). 2.3.4. Deciding about the number of components The components which their eigenvalues are grater or equal to 1 and consequently include high variance in dataset are used to calculate PCA values and the remains are omitted, since they do not include a large amount of variability of dataset and do not have any impressive effects in our results. If only the eigenvalue of the first component is equal or greater than one, it is principle component; otherwise, principle component is a linear combination of those Zj in which their eigenvalues are greater or equal to one. (Dong et al., 2015). 2.3.5. Computing the PCA values for each state for three different scenarios After computing all components, we can compute PCA value using Eq. (4) for each state for three different scenarios.

l1 Z1 þ l2 Z2 þ … þ lj ZP PCA value ¼ l1 þ l2 þ … þ lP

(6)

2.3.6. Adding a large enough positive value to PCA values to avoid negative values We added a large enough number to each PCA value to avoid non-positive amounts as our CEI, which is one of the index that help us to assess the sustainability in a system (Adler and Golany, 2001).

CEIi ¼ PCi þ ε

(7)

Where, CEIi is the composite environmental impact score of state i, PCi is the PCA value of state i and ε is a positive constant number and is bigger than the smallest negative PCA value.

3.1. Result of LCA

H0 : M1 ¼ M2 ¼ M3

(10)

H1. At least two scenaios have different averages for one specific indicator.Where, M1 ,M2 and M3 are the average of each environmental indicators in scenario 1, 2 and 3, respectively. In order to determine the difference between the means of each indicator in three scenarios the one way- ANOVA technique is used with 95% confidence interval. The results of ANOVA are provided in Table 5. The results demonstrate that, for all four variables (carbon, water, energy and LCC) the null hypothesises are rejected due to their very small P-values (0.000 < 0.05). These results (See Table 1) shows that the alternative hypothesis (H1) will be true and at least two scenarios do not have equal means for each indicator. Considering the results obtained from ANOVA (null hypothesises are rejected for all four indicators), there are at least two scenarios for each indicators that have unequal averages. Therefore, to determine which scenarios have different means for each indicator, one method is doing t-tests for each two scenarios but this method will increase the type I error (The probability of rejecting a true null hypothesis). There is another method, which is used, very

520

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

Table 1 ANOVA results.

Carbon Water Energy LCC

Between Groups Whitin Groups Between Groups Whithin Groups Between Groups Whithin Groups Between Groups Whithin Groups

DF

Sum. Squ

Mean.Squ

F-Value

P-Value

2 144 2 144 2 144 2 144

1397359 304889 6.21 13.37 277.6 23.7 3.00 7.10

698679 2117 3.11 0.09 138.8 0.16 1.50 0.05

330

0.000

33.45

0.000

843.2

0.000

30.4

0.000

common after observing the rejection of H0 in ANOVA, which called Tukey HSD test. This test defines confidence intervals for each two groups and with regard to their difference of averages determines whether there is any significant difference between their means or not (Lowry, 2008). The purpose of Tukey's HSD test is to determine which groups in the sample differ, while ANOVA investigates whether groups in the sample differs. Hence, Tukey's HSD test clarifies which groups among the sample in specific have significant differences. The results of Tukey test which have been shown in Table 2, represent that for carbon, energy and LCC the differences between each two scenarios are significant because their lower and upper bound values have the same sign (both of them are positive or negative) and zero is not in their confidence interval; on the other word, Mi -Mj are not equal to zero. The results of HSD test implies that minimum distance between the investigated sample groups that must exist before the difference between these groups is to be considered statistically significant. The small amounts of Pvalue (0.000 < 0.05) also illustrate that the null hypothesis is rejected for each two scenarios except water consumption in scenario 2 and 3 since zero is in the interval of their lower and upper bound and consequently, the p-value (0.16) is not small enough to reject H0. (See Table 2). Regarding the results that obtained from the ANOVA and Tukeytests and by considering the averages of three environmental variables, the third scenario has the minimum amount of carbon and energy footprint with significant differences compared to scenario 1 and 2 and the water consumption also has the least value among three scenarios although its difference is not remarkable in comparison with Scenario 2. Considering the descriptive statistics of three environmental indicators which are provided in Table 3, The average of carbon emission of first and second scenarios are 16.98 and 20.47 times while the means of water consumption are 25 and 6.5 times and the averages of energy use are 3.33 and 3.77 times more than Scenario 3, respectively. Consequently, Scenario 3 is the best scenario duo to its extreme lowest environmental impacts in comparison with the first and second scenarios. Scenario 1 has the

lower averages of carbon emission and energy consumption but and almost similar average of water consumption in comparison with scenario 2 (See Table 3). Therefore, the third, first and second scenarios had the lowest environmental impacts, respectively. LCC is another index, which has the important effect on the ecoefficiency scores. The mean LCC values of Scenarios 1, 2, and 3 are, 3.37, 3.02, and 3.15, respectively. Scenario 1 has the highest LCC value, while Scenario 3 have the lowest one (See Table 3).

3.2. Results of principal component analysis The average of correlation coefficients among the indicators for three scenarios are presented in Table 4. There are strong positive correlations among all indicators in Scenario(s) 2 and 3. This means that the more water and fuel are consumed, the more energy is used. In all scenarios except for Scenario 1, all indicators have strong and positive correlations. However, in Scenario 1, the water withdrawal indicator has negative correlations with the amount of energy consumption and carbon footprint. Regarding to the significant correlations between the indicators we used PCA method to compute CEI index. The values of percentage of variance and eigenvalue of the PCA components are shown in Table 5. In order to decide about the number of component to obtain PCA values, it is necessary to select the components which their cumulative percentage of variances cover the most information in the dataset. Therefore, their eigenvalue should be greater or equal to 1. For all three scenarios only the first components have the most percentage of variance (79.9, 89.87 and 97.28) and their eigenvalues are more than one (See Table 5). Therefore, the first component (Z1 ) is used to obtain PCA vales for all three scenarios. The correlation between each indicator and the first component are shown for three scenarios in Table 6. In Scenario 1, there are strong positive correlation between Energy consumption and Carbon emission values and the scores of PCA, while there is a strong

Table 2 Tukey HSD-test results. Mi-Mj

Lower Bound

Upper Bound

P-value

Carbon

Scenario2- 1 Scenario3- 1 Scenario 3-2

40.12 183.82 223.95

18.11 205.83 245.96

62.14 161.80 201.93

0.000 0.000 0.000

Water

Scenario 2-1 Scenario 3-1 Scenario 3-2

0.37 0.48 0.11

0.51 0.63 0.26

0.22 0.34 0.03

0.000 0.000 0.16

Energy

Scenario 2- 1 Scenario 3-1 Scenario 3-2

0.50 2.64 3.13

0.30 2.83 3.33

0.69 2.44 2.94

0.000 0.000 0.000

LCC

Scenario 2-1 Scenario 3-1 Scenario 3-2

0.35 0.21 0.13

0.45 0.32 0.03

0.24 0.11 0.24

0.000 0.000 0.000

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

521

Table 3 Descriptive statistics of variables. Range

Min

Max

Mean

Std deviation

Scenario 1

Water Carbon Energy LCC

2.45 311.18 3.46 0.62

0.08 3.53 1.77 3.02

2.54 314.70 5.22 3.64

0.50 195.32 3.76 3.37

0.52 76.06 0.69 0.16

Scenario2

Water Carbon Energy LCC

0.13 73.49 0.34 0.46

0.08 205.76 4.10 2.84

0.21 279.25 4.44 3.3

0.13 235.45 4.26 3.02

0.04 20.81 0.11 0.12

Scenario 3

Water Carbon Energy LCC

0.01 9.08 0.05 1.60

0.01 7.86 1.11 2.40

0.02 16.94 1.16 4.00

0.02 11.50 1.13 3.15

0.00 1.87 0.01 0.32

Table 4 The correlation coefficient between variables. CC

Water Energy Carbon

Scenario 1

Scenario 2

Scenario 3

Water

Energy

Carbon

Water

Energy

Carbon

Water

Energy

Carbon

1 0.79 0.58

0.79 1 0.90

0.58 0.90 1

1 0.69 0.97

0.69 1 0.83

0.97 0.83 1

1 0.96 0.99

0.96 1 0.97

0.99 0.97 1

Table 5 The eigenvelues and percentage of variances of the components for three scenarios. Scenario 1

Component 1 Component 2 Component 3

Scenario 2

Eigenvalue

Percentage of variance

Eigenvalue

Percentage of variance

Eigenvalue

Percentage of variance

2.40 0.48 0.12

79.89 15.94 4.17

2.68 0.31 0.01

89.50 10.21 0.29

2.98 0.02 0.00

99.38 0.50 0.12

negative correlation between the water consumption values and PCA. This means that by increasing the value of water consumption PCA value is decreasing in this scenario. For second and third scenario, all the correlations are positive and close to one. Therefore, by increasing the values of each indicator PCA value is increasing, consequently. The variables factor maps show the vector of the environmental indicators in three scenarios. Dim 1 and Dim 2 display the percentage of variance of the first and second component in PCA, respectively (See Fig. 3). The negative correlations among water consumption and energy use and carbon footprint due to their opposite directions are observed in the first scenario, where all other indicators in scenario 2 and 3 have the positive correlations. In all three scenarios, the first components (Dim 1) represent the largest percentages of variance. Furthermore, the correlations among the indicators and their related PCA scores are also observable by drawing an orthogonal line from the endpoint of each vector to the Dim 1 axis for each dimension. The greatest correlations among PCA values and environmental indicators belong to third scenario, since in case of the obtained value is close

Table 6 The correlation between the variables and the first components.

Energy Carbon Water

Scenario 3

Scenario 1

Scenario 2

Scenario 3

Component 1

Component 1

Component 1

0.96 0.86 0.85

0.89 0.99 0.95

0.99 0.99 0.99

to 1. For computing the composite environmental impact (CEI), after doing log transformation to reduce the skewness of environmental indicators and economic output, PCA is applied for each scenario. A large enough number (6) is also added to each computed PCA value to avoid having the negative values as the CEI. Then, the ecoefficiency scores as a ratio of life cycle cost over composite sustainability index for three scenarios are computed for all the states. Afterward, the states are ranked considering their descending orders of eco-efficiency scores. The values of CEI and log transformed LCC and eco-efficiency scores of the states are shown in Table 7. The values of eco-efficiency in Table 7, present the raw ecoefficiency values, which are obtained by dividing the LCC values of different states in to the CEI values. To rescale the values of raweco efficiency we used the min-max technique (Eq. (6)) to normalize the raw-eco-efficiency scores and put them into the zero and one-interval. The values of normalized eco-efficiency for three scenarios are shown in Fig. 4. In Scenario 1, ID has the highest amount of eco-efficiency. Since the eco-efficiency has a direct relation with life cycle cost (LCC) value and the inverse relation with CEI, this state has the minimum score of CEI among all the states. The amount of LCC (3.35 cents/ mile) makes it the best state by considering both economic and environmental impacts. VT is the second state, which has the high value of eco-efficiency since it has the second lowest value of CEI. DC has the maximum amount of CEI and this leads to make it the least eco-efficient state (See Fig. 4a). In Scenario 2, TX has the maximum amount of eco-efficiency due to its low value of CEI and large enough amount of LCC.

522

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

Fig. 3. The variables factor map (PCA) for a) Scenario 1, b) Scenario 2 and c) Scenario 3.

Totally, it is concluded that the western and central states have higher amount of eco-efficiency than eastern provinces because of their less environmental impacts and consequently their lower amounts of CEI, which is one of the most important factors that determine the value of eco-efficiency in the states. IL has the highest values of all three environmental indicators and consequently the largest value of CEI. Although this state has the maximum amount of economic output, the large value of CEI makes it the last eco-efficient state. (See Fig. 3b). In Scenario 3, NM has the maximum amount of eco-efficiency duo to its minimum amount of CEI. VT and AZ are the second and third scenarios with high score of eco-efficiency, respectively. While the CEI value of AZ is less than VT, AZ is more eco-efficient since it has the greater value of LCC than VT. IL is the least eco-efficient, which has the maximum value of CEI and its LCC score is not high enough to make a significant change in its low value of eco-efficiency (See Fig. 4c).

technologies was performed with joint approach of DEA and LCA. In this work, we applied PCA as an alternative and more robust method that deal with the multi-collinearity. The results of the bivariate correlation analysis are provided in Table 1. According to the correlation analysis results, a significant positive correlation was found between the eco-efficiency scores of the previous work (Onat et al., 2017a,b) and current study. The correlation analysis is conducted to determine whether there is any meaningful relation between the results obtained from a previous benchmarking study based on DEA approach versus this study (PCA approach). It is found that there are strong positive relations between the results of two approaches in Scenario 2 (0.88) and scenario 3 (0.89) and the moderate positive relations (0.43) between the results of Scenario 1 between two approaches. The proposed PCA based framework was not only statistically replaceable to the previous study but also methodologically more appropriate due to evident significant correlations between the carbon and energy footprint indicators.

3.3. Comparison of eco-efficiency results with previous DEA analysis

4. Conclusion, discussions, and future work

In a previous work, the eco-efficiency of states in the USA calculated by applying DEA technique to find the sustainability index. The DEA efficiency scores are presented in Table S1 in the supplementary information (SI) file in the journals website. The eco-efficiency was the ratio of life cycle cost over that sustainability index. In this study, the correlation in between efficiency scores calculated using DEA and PCA are investigated. The correlation coefficients in between these two dataset (efficiency scores) are shown in Table 8. This section provides comparison of eco-efficiency results with previous work by (Onat et al., 2017a,b), because it is always important to validate the result with literature. In the previous work, eco-efficiency assessment for the electric vehicle

In this paper, transportation focused sustainability performance assessment of electric vehicle technologies was carried out by using a novel LCA þ PCA method. The objectives were, first, to quantify the environmental impacts of different electric vehicle technologies in USA and, second, determine the sustainability performances with eco-efficiency analysis. A hierarchical methodology that consists of LCA and PCA was utilized to quantify the transportationfocused eco-efficiency of each state as a ratio of economic outputs to environmental impacts. Three environmental (CO2 emission, energy use and water consumption) and economic output indicators are considered to benchmark the sustainability performance of BEVs for the states in U.S. The eco-efficiency results are presented for three scenarios such as BEVs charged through state-

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

523

Table 7 CEI, LCC and raw eco-efficiency (EE) scores for three scenarios. State

CEI1

LCC1

EE1

State

CEI2

LCC2

EE2

State

CEI3

LCC3

EE3

ID VT WA OR SD MT NY CA TN ME NH NJ IL AL SC CT MD AZ NC NE AR MN VA PA WI IA NV CO MO KS ND MI GA OK OH KY LA WV WY MS IN TX UT NM MA DE FL RI DC

1.049 1.158 1.778 2.765 3.629 4.893 4.410 5.106 5.477 5.086 5.203 5.702 6.072 5.792 5.828 5.470 6.104 6.065 6.092 6.420 5.970 6.418 6.338 6.383 6.617 6.623 6.401 6.887 6.844 6.806 6.760 6.832 6.501 6.725 7.116 7.056 6.706 7.138 7.112 6.834 7.289 7.037 7.197 7.274 6.773 7.338 7.075 7.595 8.256

1.209 1.122 1.209 1.209 1.273 1.232 1.105 1.173 1.246 1.122 1.122 1.206 1.273 1.208 1.215 1.122 1.232 1.220 1.215 1.273 1.182 1.256 1.237 1.232 1.270 1.256 1.209 1.291 1.276 1.264 1.256 1.264 1.168 1.195 1.258 1.246 1.182 1.258 1.250 1.195 1.258 1.189 1.209 1.207 1.122 1.206 1.154 1.122 1.215

1.152 0.969 0.680 0.437 0.351 0.252 0.251 0.230 0.228 0.221 0.216 0.211 0.210 0.209 0.209 0.205 0.202 0.201 0.199 0.198 0.198 0.196 0.195 0.193 0.192 0.190 0.189 0.187 0.187 0.186 0.186 0.185 0.180 0.178 0.177 0.177 0.176 0.176 0.176 0.175 0.173 0.169 0.168 0.166 0.166 0.164 0.163 0.148 0.147

TX CA ID OR UT WA WY AR KS OK NV SD AZ CO NM FL NY MT ND NE AL GA LA MS NC SC TN WI DC DE MD NJ PA IA CT MA ME NH RI VT VA MN MO KY IN MI OH WV IL

3.447 3.568 3.671 3.671 3.671 3.671 3.671 3.846 3.846 3.846 4.131 4.564 4.550 4.550 4.550 4.429 4.807 5.244 5.975 5.975 6.290 6.290 6.290 6.290 6.290 6.290 6.290 6.474 6.758 6.758 6.758 6.758 6.758 7.630 6.957 6.957 6.957 6.957 6.957 6.957 7.430 7.630 7.764 8.488 8.488 8.488 8.488 8.748 9.132

1.061 1.068 1.067 1.067 1.067 1.067 1.067 1.075 1.075 1.075 1.082 1.116 1.082 1.082 1.082 1.045 1.056 1.067 1.116 1.116 1.139 1.139 1.122 1.122 1.122 1.122 1.122 1.116 1.105 1.105 1.105 1.105 1.105 1.193 1.069 1.069 1.069 1.069 1.069 1.069 1.122 1.116 1.122 1.183 1.174 1.174 1.174 1.174 1.193

0.308 0.299 0.291 0.291 0.291 0.291 0.291 0.279 0.279 0.279 0.262 0.245 0.238 0.238 0.238 0.236 0.220 0.203 0.187 0.187 0.181 0.181 0.178 0.178 0.178 0.178 0.178 0.172 0.163 0.163 0.163 0.163 0.163 0.156 0.154 0.154 0.154 0.154 0.154 0.154 0.151 0.146 0.145 0.139 0.138 0.138 0.138 0.134 0.131

NM VT AZ WY TX NV DE UT CA FL SD CO CT ND KS OK KY WA SC NC NE LA GA ID MT MD MN AR OR DC WI NH ME VA NJ IA MI OH MS RI MO IN TN MA WV AL PA NY IL

2.322 3.341 2.579 3.306 3.962 3.170 4.214 4.128 3.947 4.590 4.822 4.612 5.336 5.329 4.856 4.725 5.436 6.584 5.237 5.329 5.561 5.574 5.546 5.587 6.015 6.354 6.250 6.015 7.420 6.986 6.880 7.608 6.775 7.327 7.203 6.678 7.409 7.513 6.590 6.986 6.909 7.203 6.670 7.894 8.691 8.137 8.971 8.988 10.431

0.875 1.253 0.875 1.030 1.163 0.875 1.163 1.065 0.993 1.131 1.131 1.065 1.224 1.194 1.065 1.030 1.163 1.386 1.099 1.099 1.131 1.131 1.065 1.065 1.131 1.194 1.163 1.099 1.335 1.224 1.194 1.308 1.163 1.253 1.224 1.131 1.253 1.253 1.099 1.163 1.131 1.163 1.065 1.194 1.253 1.131 1.194 1.194 1.224

0.377 0.375 0.339 0.311 0.294 0.276 0.276 0.258 0.252 0.246 0.235 0.231 0.229 0.224 0.219 0.218 0.214 0.211 0.210 0.206 0.203 0.203 0.192 0.191 0.188 0.188 0.186 0.183 0.180 0.175 0.174 0.172 0.172 0.171 0.170 0.169 0.169 0.167 0.167 0.166 0.164 0.161 0.160 0.151 0.144 0.139 0.133 0.133 0.117

specific (1) average electricity generation mix, (2) marginal electricity generation mix, and (3) fully solar charging station. Comparison of LCA of three scenarios by using ANOVA and Tukey-tests show that third scenario has the minimum average of all three environmental indicators and can be introduced as the cleanest scenario. In Scenario 3, because of maximum utilization of solar power, the value of water consumption and energy use has decreased notably in most of the states. Scenario 1 has lower environmental impacts compared to those of Scenario 2. Scenario 1 has the maximum value of economic output and Scenario 3 and Scenario 2 have the second and third highest averages of economic output, respectively. Additionally, because of the high correlation between environmental indicators, the PCA approach is applied to reduce the dimension of three environmental indicators and generate a unique composite environmental impact. Next, eco-efficiency of each state is computed and the states were ranked regarding their increasing value of eco-efficiency. In Scenario 1, ID has the highest amount of eco-efficiency. In Scenario 2, TX has the maximum

amount of eco-efficiency due to its lowest value of CEI. It can be concluded that the western and central states have higher ecoefficiency values compared to eastern states. In Scenario 3, NM has the maximum eco-efficiency score. Since sustainability assessment encompasses multiple indicators, making a selection among alternatives has been an issue for decision-making and policy analysis purposes. Considering the knowledge gaps in the literature as well as the challenges highlighted by the WBCSD, the main contribution of our method is to combine environmental and economic life cycle impacts and deal with multi-collinearity between indicators of sustainability. By adopting such framework, policy makers can get benefit of capturing both economic benefits as well as the environmental impacts in a single, refined, and strong representative metric. Furthermore, the proposed method is applied to alternative electric vehicle technologies for the first time, thus providing a strong foundation for further benchmarking and regional policy analysis towards achieving sustainable transportation in the United States. The proposed method is also applicable for the sustainability

524

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

Fig. 4. Eco-efficiency scores of the a) first scenario b) second scenario c) third scenario.

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526 Table 8 The correlation between current study and the previous work (Onat et al., 2017a,b).

Correlation Coefficient

Scenario 1

Scenario 2

Scenario 3

0.43

0.88

0.89

problems within and beyond transportation sector, where there are correlated sustainability indicators and consequently need a dimension reduction technique. For example, a similar method can be applied for many other industries such as construction, energy and manufacturing where life cycle environmental, economic and social impacts are aimed to be integrated into sustainability performance assessment, especially to deal with the multi-collinearity associated with the life cycle inventory data. Especially, there is a transition from life cycle assessment to life cycle sustainability assessment (Onat et al. 2017a,b). In this framework, environmental LCA, social LCA and life cycle costing (LCC) results are individually analyzed without any common framework integrating and analyzing all these categories at the same time. In the future work, this method can fill this knowledge gap and be used to benchmark the eco-efficiency of alternative vehicle technologies based on their environmental, economic and social indicators, which is termed as triple bottom line of sustainability. Many countries around the world including United Kingdom, Norway, Germany, France, China, and India, are planning to ban diesel or gasoline vehicles within a few decades (CNN, 2017). On the other hand, the sustainability impacts of electric vehicles might greatly vary among these countries as regional and temporal variations play important role on the impacts of electric vehicle technologies. For instance, BEVs are the most favorable options only in 24 states in the U.S. in terms of carbon footprint, while they were not better options in the rest due to the variations in regional and temporal variations (Onat et al., 2015, 2018). When promoting emerging technologies, decision-making as well as optimum allocation of resources is crucial and thus, the assessments should consider multiple dimensions including temporal, spatial variations, as well as different sustainability indicators. Hence, future work should investigate the impacts of alternative vehicle technologies for the world's major economies where there is a strong policy support for these vehicles. While the assessment provided in this study focused on a limited number of indicators, the future work can include more social, economic, and the environmental indicators such as income, tax, safety, import dependence, human health, etc. Such extension would allow the policy makers to benchmark the impacts, thus serve the United Nation's sustainable development goals, and contribute to principles of “doing more with less” and “resource productivity”. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi.org/10.1016/j.jclepro.2018.12.058. References Adler, N., Golany, B., 2001. Evaluation of deregulated airline networks using data envelopment analysis combined with principal component analysis with an application to Western Europe. Eur. J. Oper. Res. 132, 260e273. https://doi.org/ 10.1016/S0377-2217(00)00150-8. Alirezaei, M., Onat, N.C., Tatari, O., Abdel-Aty, M., 2017. The climate change-road safety-economy nexus: a system dynamics approach to understanding complex interdependencies. Systems 5, 6. Bartolozzi, I., Rizzi, F., Frey, M., 2013. Comparison between hydrogen and electric vehicles by life cycle assessment: a case study in Tuscany. Italy. Appl. Energy 101, 103e111. https://doi.org/10.1016/j.apenergy.2012.03.021. rov Bolca a, P., Kolosta, S., 2015. Assessment of sustainable development in the EU 27 using aggregated SD index. Ecol. Indicat. 48, 699e705. https://doi.org/10.1016/

525

J.ECOLIND.2014.09.001. Cerutti, A.K., Beccaro, G.L., Bagliani, M., Donno, D., Bounous, G., 2013. Multifunctional Ecological Footprint Analysis for assessing eco-efficiency: a case study of fruit production systems in Northern Italy. J. Clean. Prod. 40, 108e117. https:// doi.org/10.1016/J.JCLEPRO.2012.09.028. Choi, J., Park, Y.S., Park, J.D., 2015. Development of an aggregate air quality index using a PCA-based method: a case study of the US transportation sector. Am. J. Ind. Bus. Manag. 05, 53e65. https://doi.org/10.4236/ajibm.2015.52007. CNN, 2017. These Countries want to Ban Gas and Diesel Cars [WWW Document]. http://money.cnn.com/2017/09/11/autos/countries-banning-diesel-gas-cars/ index.html. (Accessed 12 December 2017). Cook, W.D., Zhu, J., 2007. Data irregularities and structural complexities in dea. In: Modeling Data Irregularities and Structural Complexities in Data Envelopment Analysis. Springer US, Boston, MA, pp. 1e11. https://doi.org/10.1007/978-0-38771607-7_1. Daniel, J.J., Rosen, M.A., 2002. Exergetic environmental assessment of life cycle emissions for various automobiles and fuels. Exergy An Int. J. 2, 283e294. https://doi.org/10.1016/S1164-0235(02)00076-6. Dominguez, R., Cannella, S., Framinan, J.M., 2015. On returns and network configuration in supply chain dynamics. Transport. Res. Part E Logist. Transp. Rev. 73, 152e167. https://doi.org/10.1016/J.TRE.2014.11.008. €o €k, M., Pi, G., 2015. Sustainability assessment of the natural gas Dong, X., Guo, J., Ho industry in China using principal component analysis. Sustainability 7, 6102e6118. https://doi.org/10.3390/su7056102. Egilmez, G., Kucukvar, M., Park, Y.S., 2015. Supply chain-linked sustainability assessment of the US manufacturing: an ecosystem perspective. Sustain. Prod. Consum. https://doi.org/10.1016/j.spc.2015.10.001. Egilmez, G., Park, Y.S., 2014. Transportation related carbon, energy and water footprint analysis of U.S. manufacturing: an eco-efficiency assessment. Transport. Res. Transport Environ. 32, 143e159. https://doi.org/10.1016/ j.trd.2014.07.001. Ehrenfeld, J.R., 2005. Eco-efficiency: philosophy, theory, and tools. J. Ind. Ecol. 6e8. https://doi.org/10.1162/108819805775248070. EIA, 2018. Transportation energy use [WWW document]. https://www.eia.gov/ energyexplained/?page¼us_energy_transportation. (Accessed 11 July 2017). Ercan, T., Onat, N.C., Tatari, O., 2016. Investigating carbon footprint reduction potential of public transportation in United States: a system dynamics approach. J. Clean. Prod. 133, 1260e1276. https://doi.org/10.1016/j.jclepro.2016.06.051. Ercan, T., Onat, N.C., Tatari, O., Mathias, J.-D., 2017. Public transportation adoption requires a paradigm shift in urban development structure. J. Clean. Prod. 142, 1789e1799. https://doi.org/10.1016/J.JCLEPRO.2016.11.109. Faria, R., Moura, P., Delgado, J., de Almeida, A.T., 2012. A sustainability assessment of electric vehicles as a personal mobility system. Energy Convers. Manag. 61, 19e30. https://doi.org/10.1016/j.enconman.2012.02.023.  2012. Fuzzy data treated as functional lez-Rodríguez, G., Colubi, A., Gil, M.A., Gonza data: a one-way ANOVA test approach. Comput. Stat. Data Anal. 56, 943e955. https://doi.org/10.1016/j.csda.2010.06.013. e, J.B., Heijungs, R., Huppes, G., Kleijn, R., de Koning, A., van Oers, L., Wegener Guine Sleeswijk, A., Suh, S., Udo de Haes, H.A., de Bruijn, H., van Duin, R., e, M., 2002. Life Cycle Assessment. Operational Guide to Huijbregts, M.A.J., Gorre the ISO Standards. I: LCA in Perspective. IIa: Guide. IIb: Operational Annex. III: Scientific Background, The Netherlands: Ministry of … https://doi.org/10.1007/ BF02978784. e, J.B., Heijungs, R., Huppes, G., Zamagni, A., Masoni, P., Buonamici, R., Guine Ekvall, T., Rydberg, T., 2011. Life cycle assessment: past, present, and future y. Environ. Sci. Technol. 45, 90e96. https://doi.org/10.1021/es101316v. Hawkins, T.R., Gausen, O.M., Strømman, A.H., 2012. Environmental impacts of hybrid and electric vehiclesda review. Int. J. Life Cycle Assess. 17, 997e1014. https://doi.org/10.1007/s11367-012-0440-9. Hawkins, T.R., Singh, B., Majeau-Bettez, G., Strømman, A.H., 2013. Comparative environmental life cycle assessment of conventional and electric vehicles. J. Ind. Ecol. 17, 53e64. https://doi.org/10.1111/j.1530-9290.2012.00532.x. e, J.B., 2010. Life cycle assessment and sustainability Heijungs, R., Huppes, G., Guine analysis of products, materials and technologies. Toward a scientific framework for sustainability life cycle analysis. Polym. Degrad. Stabil. 95, 422e428. https:// doi.org/10.1016/j.polymdegradstab.2009.11.010. Huppes, G., Ishikawa, M., 2005. Why eco-efficiency? J. Ind. Ecol. 9, 2e5. https:// doi.org/10.1162/108819805775248052. James, G., Witten, D., Hastie, T., Tibshirani, R., 2013. An Introduction to Statistical Learning, an Introduction to Statistical Learning. https://doi.org/10.1007/978-14614-7138-7. Jiang, Q., Liu, Z., Liu, W., Li, T., Cong, W., Zhang, H., Shi, J., 2018. A principal component analysis based three-dimensional sustainability assessment model to evaluate corporate sustainable performance. J. Clean. Prod. 187, 625e637. https://doi.org/10.1016/J.JCLEPRO.2018.03.255. Jothi Basu, R., Bai, R., Palaniappan, P.K., 2015. A strategic approach to improve sustainability in transportation service procurement. Transport. Res. Part E Logist. Transp. Rev. 74, 152e168. https://doi.org/10.1016/J.TRE.2014.10.015. Kucukvar, M., Cansev, B., Egilmez, G., Onat, N.C., Samadi, H., 2016. Energy-climatemanufacturing nexus: new insights from the regional and global supply chains of manufacturing industries. Appl. Energy 184, 889e904. https://doi.org/ 10.1016/j.apenergy.2016.03.068. Kucukvar, M., Egilmez, G., Onat, N.C., Samadi, H., 2015. A global, scope-based carbon footprint modeling for effective carbon reduction policies: lessons from the Turkish manufacturing. Sustain. Prod. Consum. 1 https://doi.org/10.1016/

526

N.C. Onat et al. / Journal of Cleaner Production 212 (2019) 515e526

j.spc.2015.05.005. Kucukvar, M., Egilmez, G., Tatari, O., 2014a. Evaluating environmental impacts of alternative construction waste management approaches using supply-chainlinked life-cycle analysis. Waste Manag. Res. 32, 500e508. https://doi.org/ 10.1177/0734242X14536457. Kucukvar, M., Haider, M.A., Onat, N.C., 2017. Exploring the material footprints of national electricity production scenarios until 2050: the case for Turkey and UK. Resour. Conserv. Recycl. 125 https://doi.org/10.1016/j.resconrec.2017.06.024. Kucukvar, M., Noori, M., Egilmez, G., Tatari, O., 2014b. Stochastic decision modeling for sustainable pavement designs. Int. J. Life Cycle Assess. 1e15. Kucukvar, M., Onat, N.C., Haider, M.A., 2018. Material dependence of national energy development plans: the case for Turkey and United Kingdom. J. Clean. Prod. 200, 490e500. https://doi.org/10.1016/j.jclepro.2018.07.245. Kucukvar, M., Samadi, H., 2015. Linking national food production to global supply chain impacts for the energy-climate challenge: the cases of the EU-27 and Turkey. J. Clean. Prod. 108, 395e408. https://doi.org/10.1016/ j.jclepro.2015.08.117. Li, T., Zhang, H., Yuan, C., Liu, Z., Fan, C., 2012. A PCA-based method for construction of composite sustainability indicators. Int. J. Life Cycle Assess. 17, 593e603. https://doi.org/10.1007/s11367-012-0394-y. Lowry, R., 2008. One-way analysis of variance for independent samples [WWW document]. http://vassarstats.net/textbook/ch14pt1.html. Lozano, F.J., Lozano, R., 2018. Assessing the potential sustainability benefits of agricultural residues: biomass conversion to syngas for energy generation or to chemicals production. J. Clean. Prod. 172, 4162e4169. https://doi.org/10.1016/ J.JCLEPRO.2017.01.037. Mascarenhas, A., Nunes, L.M., Ramos, T.B., 2015. Selection of sustainability indicators for planning: combining stakeholders' participation and data reduction techniques. J. Clean. Prod. 92, 295e307. https://doi.org/10.1016/ J.JCLEPRO.2015.01.005. Noori, M., Gardner, S., Tatari, O., 2015. Electric vehicle cost, emissions, and water footprint in the United States: development of a regional optimization model. Energy 89, 610e625. https://doi.org/10.1016/j.energy.2015.05.152. Noori, M., Zhao, Y., Onat, N.C., Gardner, S., Tatari, O., 2016. Light-duty electric vehicles to improve the integrity of the electricity grid through Vehicle-to-Grid technology: analysis of regional net revenue and emissions savings. Appl. Energy 168, 146e158. https://doi.org/10.1016/j.apenergy.2016.01.030. €f, A., Messagie, M., Tillman, A.-M., Ljunggren So €derman, M., Van Mierlo, J., Nordelo €derman, M., 2014. Environmental Van Mierlo, J., Messagie, M., Ljunggren So impacts of hybrid, plug-in hybrid, and battery electric vehicles-what can we learn from life cycle assessment? Int. J. Life Cycle Assess. 19, 1866e1890. https:// doi.org/10.1007/s11367-014-0788-0. Onat, N.C., 2015a. A Macro-level Sustainability Assessment Framework for Optimal Distribution of Alternative Passenger Vehicles. University of Central Florida. Onat, N.C., 2015b. Integrated Sustainability Assessment Framework for the U.S. Transportation. University of Central Florida. Onat, N.C., Gumus, S., Kucukvar, M., Tatari, O., 2016a. Application of the TOPSIS and intuitionistic fuzzy set approaches for ranking the life cycle sustainability performance of alternative vehicle technologies. Sustain. Prod. Consum. 6, 12e25. https://doi.org/10.1016/j.spc.2015.12.003. Onat, N.C., Kucukvar, M., Halog, A., Cloutier, S., 2017a. Systems thinking for life cycle sustainability assessment: a review of recent developments, applications, and future perspectives. Sustain. Times Vol. 9, 706 9e706. https://doi.org/10.3390/ SU9050706, 2017. Onat, N.C., Kucukvar, M., Tatari, O., 2018. Well-to-wheel water footprints of conventional versus electric vehicles in the United States: a state-based comparative analysis. J. Clean. Prod. 204, 788e802. https://doi.org/10.1016/ J.JCLEPRO.2018.09.010. Onat, N.C., Kucukvar, M., Tatari, O., 2016b. Uncertainty-embedded dynamic life cycle sustainability assessment framework: an ex-ante perspective on the impacts of

alternative vehicle options. Energy 112. https://doi.org/10.1016/ j.energy.2016.06.129. Onat, N.C., Kucukvar, M., Tatari, O., 2015. Conventional, hybrid, plug-in hybrid or electric vehicles? State-based comparative carbon and energy footprint analysis in the United States. Appl. Energy 150, 36e49. https://doi.org/10.1016/ j.apenergy.2015.04.001. Onat, N.C., Kucukvar, M., Tatari, O., 2014a. Scope-based carbon footprint analysis of U.S. residential and commercial buildings: an inputeoutput hybrid life cycle assessment approach. Build. Environ. 72, 53e62. Onat, N.C., Kucukvar, M., Tatari, O., 2014b. Towards life cycle sustainability assessment of alternative passenger vehicles. Sustain. Times 6. https://doi.org/ 10.3390/su6129305. Onat, N.C., Kucukvar, M., Tatari, O., Egilmez, G., 2016c. Integration of system dynamics approach towards deeping and broadening the life cycle sustainability assessment framework: a case for electric vehicles. Int. J. Life Cycle Assess. 21, 1009e1034. Onat, N.C., Kucukvar, M., Tatari, O., Zheng, Q.P., 2016. Combined application of multi-criteria optimization and life-cycle sustainability assessment for optimal distribution of alternative passenger cars in. U.S. J. Clean. Prod. 112 https:// doi.org/10.1016/j.jclepro.2015.09.021. Onat, N.C., Noori, M., Kucukvar, M., Zhao, Y., Tatari, O., Chester, M., 2017b. Exploring the suitability of electric vehicles in the United States. Energy 121. https:// doi.org/10.1016/j.energy.2017.01.035. Park, Y.S., Egilmez, G., Kucukvar, M., 2016. Emergy and end-point impact assessment of agricultural and food production in the United States: a supply chain-linked Ecologically-based Life Cycle Assessment. Ecol. Indicat. 62, 117e137. https:// doi.org/10.1016/j.ecolind.2015.11.045. Park, Y.S., Egilmez, G., Kucukvar, M., 2015. A novel life cycle-based principal component analysis framework for eco-efficiency analysis: case of the United States manufacturing and transportation nexus. J. Clean. Prod. 92, 327e342. https://doi.org/10.1016/j.jclepro.2014.12.057. Reisi, M., Aye, L., Rajabifard, A., Ngo, T., 2014. Transport sustainability index: Melbourne case study. Ecol. Indicat. 43, 288e296. https://doi.org/10.1016/ J.ECOLIND.2014.03.004. €mer, B., Wittlinger, R., Zombik, W., Schmidt, I., Saling, P., Kicherer, A., Dittrich-Kra Schrott, W., Schmidt, S., 2002. Eco-efficiency analysis by basf: the method. Int. J. Life Cycle Assess. 7, 203e218. https://doi.org/10.1007/BF02978875. Salvati, L., Carlucci, M., 2014. A composite index of sustainable development at the local scale: Italy as a case study. Ecol. Indicat. 43, 162e171. https://doi.org/ 10.1016/J.ECOLIND.2014.02.021. Samaras, C., Meisterling, K., 2008. Life cycle assessment of greenhouse gas emissions from plug-in hybrid vehicles: implications for policy. Environ. Sci. & Technol. 42, 3170e3176. https://doi.org/10.1021/es702178s. Shaikh, M.A., Kucukvar, M., Onat, N.C., Kirkil, G., 2017. A framework for water and carbon footprint analysis of national electricity production scenarios. Energy 139. https://doi.org/10.1016/j.energy.2017.07.124. Soler Rovira, J., Soler Rovira, P., 2009. Assessment of aggregated indicators of sustainability using PCA: the case of apple trade in Spain. In: 6th International Conference on Life Cycle Assessment in the Agri-food Sector. Towards a Sustainable Management of the Food Chain, Zurich, Switzerland. Tatari, O., Kucukvar, M., 2012. Eco-efficiency of construction materials: data envelopment analysis. J. Construct. Eng. Manag. 138, 733e741. https://doi.org/ 10.1061/(ASCE)CO.1943-7862.0000484. Tatari, O., Kucukvar, M., Onat, N.C., 2015. Towards a triple bottom line life cycle sustainability assessment of buildings. In: Science for Sustainable Construction and Manufacturing Workshop Volume I. Position Papers and Findings, p. 226. Zhao, Y., Onat, N.C., Kucukvar, M., Tatari, O., 2016. Carbon and energy footprints of electric delivery trucks: a hybrid multi-regional input-output life cycle assessment. Transport. Res. Transport Environ. 47 https://doi.org/10.1016/ j.trd.2016.05.014.