SCALE: Software for CALculating Emergy based on life cycle inventories

Ecological Modelling 248 (2013) 80–91

Contents lists available at SciVerse ScienceDirect

Ecological Modelling journal homepage: www.elsevier.com/locate/ecolmodel

SCALE: Software for CALculating Emergy based on life cycle inventories Antonino Marvuglia a,∗ , Enrico Benetto a , Gordon Rios b , Benedetto Rugani a a b

Public Research Centre Henri Tudor (CRPHT), Resource Centre for Environmental Technologies (CRTE), 66 rue de Luxembourg, L-4002 Esch-sur-Alzette, Luxembourg Cork Constraint Computation Centre, University College Cork (UCC), Western Gateway Building, Cork, Ireland

a r t i c l e

i n f o

Article history: Received 24 June 2012 Received in revised form 11 September 2012 Accepted 13 September 2012 Available online 9 November 2012 Keywords: Emergy calculator OpenSource Graph search Life cycle assessment Life cycle inventory Ecoinvent

a b s t r a c t Emergy analysis is an environmental accounting approach which evaluates the memory of the geobiosphere exergy (environmental work) supporting economic systems, e.g. to make a product or service available via the use of natural resources. This paper presents the software SCALE (Software for CALculating Emergy based on life cycle inventories), which allows a rigorous and replicable calculation of the emergy associated to any product or service. The software exploits an ad hoc algorithm, fulfilling the emergy algebra rules, which requires as input the matrix describing a given system of interconnected processes, containing the amount of natural resources used by the output product and its linked technological processes. Starting from this matrix the algorithm performs a graph search across the corresponding network and records the output emergy associated with each of the explored paths linking the input nodes of the network (i.e. the emergy sources) to the output nodes (i.e. the studied products). The algorithm is specifically devised to be used with data coming (in matrix form) from any available life cycle inventory (LCI) database used in life cycle assessment (LCA) studies. Two case studies are presented to illustrate the features of the software: the first one deals with heat generation from grape marc pellets, the second with drinking water production. The use of large LCI databases for emergy calculation using SCALE could contribute to extend the application of emergy systems evaluation in the practice of sustainability science. To our knowledge, this is the first software tool which allows exact calculation of the emergy of products using LCI databases, and therefore it achieves an operational integration between the well-known and standardized LCI methodology and emergy evaluation. © 2012 Elsevier B.V. All rights reserved.

1. Introduction Life cycle assessment (LCA) is a standardized methodology used to quantify the environmental impacts of products across their whole life cycle (ISO, 2006). Nowadays LCA is one of the most accepted and used tools for the environmental assessment of products and services (Curran, 2006; European Commission, 2010). Like most of the currently existing lifecycle-based impact assessment methods and related indicators, LCA takes an anthropocentric perspective, since the value of resources is related to their scarcity or to their usefulness in economic production systems.1

∗ Corresponding author. Tel.: +352 425 991 4652; fax: +352 425 991 555. E-mail address: [email protected] (A. Marvuglia). 1 Within this paper the term “system” will occur several times. It will refer sometimes to the system meant as a set of interlinked processes (either natural or anthropic) and sometimes in a mathematical meaning, as the system of mathematical (energy and mass balance) equations describing a system meant in the first acceptation. Therefore, in order to avoid misinterpretations, hereafter we will use the word “system” in the first meaning and we will explicitly specify when we refer to the mathematical system of equations. 0304-3800/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.ecolmodel.2012.09.013

Therefore, from the LCA point of view the resources to be protected are the ones requested by economic production activities (via market mechanisms) and whose stocks are scarce. Renewable resources and ecosystem services are often excluded from LCA, despite the fact that their support to human activities and their increasing depletion have been recognized (Hassan et al., 2005). Only recently some researchers attempted to formalize a consistent inclusion of ecosystem services in LCA (Zhang et al., 2010a,b) defining an ecologically based LCA approach (Eco-LCA). Stemming from the work of Odum (1988, 1996), a new paradigm to quantify the value of resources from a nature-centered (or donororiented) point of view has emerged in the scientific community. This approach is based on the concept of emergy (spelled with an “m”). In its traditional definition emergy quantifies the amount of available energy (usually expressed as solar energy and measured in solar emjoules (sej), or its multiples) previously directly and indirectly required to generate a product and/or to support a system and its level of organization (Ridolfi and Bastianoni, 2008). A direct link has been claimed between the concept of emergy and the thermodynamic concept of exergy (Sciubba and Ulgiati, 2005; Bastianoni et al., 2007), although not universally accepted by

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

the exergy community (Sciubba, 2010). As stated by Raugei (2011) “emergy may ultimately be interpreted for all practical intents and purposes as the ‘memory’ of the total exergy that was previously spent to make a product or service available to the end user”. As a potential consequence, “for human-dominated systems [. . .] more emergy assigned to a process’ yield should be interpreted” as “appropriation of more environmental work to produce the used resources and/or more work potentially required2 to replace them” (Raugei, 2011). This definition appears to lend itself very well to establish a link between a formal explanation of emergy and a possible, more practical, interpretation of its “donor side value” (Dong et al., 2008). LCA is based on realistic representations of anthropic production systems, described by models like the Ecoinvent database, which is conventionally used in LCA software. These kinds of models are very detailed, i.e. include thousands of processes connected together by mass and energy flows. The need to integrate the evaluation made with LCA, which ignores ecosystems good and services, and emergy evaluation (EME), which takes them into account, has emerged in different studies (Pizzigallo et al., 2008; Dong and Wang, 2009; Duan et al., 2011) but the attempts done so far have had the character of a synoptic assessment where the two approaches have been used as complementary tools, rather than the form of a structural integration. The question of using LCI models to calculate the emergy associated to man-made goods therefore arises (Ness et al., 2007; Rugani et al., 2011; Rugani and Benetto, 2012) and poses the technical problem of applying the emergy algebra rules to very complicated systems. The aim of this paper is to provide a tool for calculating emergy using life cycle inventory networks. More specifically, a software interface (Software for CALculating Emergy based on life cycle inventories – SCALE) using an ad hoc developed novel algorithm is presented. The software is able to quantify the sej value of every kind of life cycle product and process through a rigorous implementation of the emergy algebra rules.

2. Materials and methods The rationale behind the concept of emergy is the consideration that all the different forms of energy can be sorted under a universal hierarchy of energy transformations and measured with the common metric of the solar emjoule (Odum, 1996; Ridolfi and Bastianoni, 2008; Brown and Cohen, 2008), which then becomes a yardstick by which all energy and material inputs and outputs can be compared with each other. To convert material and energy items in sej, EME uses a conversion factor called transformity (given in sej/j or multiples), or specific emergy (given in sej/g or multiples) – more in general unit emergy value (hereafter UEV) – which is the amount of emergy required to make one unit of a given product or service (Odum, 1996). The methodological framework for EME calculation is represented by the set of algebraic rules known as “emergy algebra”. These rules are completely different from those applied in LCA (Odum, 1996; Brown and Herendeen, 1996) and have been questioned by some authors (e.g. Sciubba, 2010). The rules, listed hereafter, are not discussed further in this paper and used as such: (1) all source emergy to a process is assigned to the process output; (2) co-products from a multi-output process have the total emergy assigned to each pathway;

2

By Nature (ndr).

81

(3) when a pathway splits, the emergy is assigned to each leg of the split based on its percentage of the total energy flow on the pathway; (4) emergy cannot be counted twice within a system: (a) emergy in feedbacks cannot be double counted, and (b) co-products, when reunited, cannot be added yielding a sum greater than the emergy source from which they were derived. Thus, when adding emergy of inflows or outflows that are co-products, only the largest one should be considered. Fig. 1 clarifies the meaning of these rules: an energy systems diagram drawn from a conservation perspective, typical of LCA is compared with an emergy systems diagram drawn from a memorization perspective. Numerical examples are provided to show the emergy calculation in the case of (a) splits & co-products, and (b) feedbacks. For the sake of clarity, co-products are “product items showing different physico-chemical characteristics, but which can only be produced jointly” (Sciubba and Ulgiati, 2005). They are assigned the same emergy (and different transformity) since each of them cannot be produced without providing the whole amount of emergy required for the process. On the other hand, splits are “originating flows showing the same physico-chemical characteristics” (Sciubba and Ulgiati, 2005). Therefore, the emergy of splits will be different, proportionally assigned on the basis of their respective quantities, while their transformity is the same. A better understanding of the fundamental differences between energy and emergy can be gained from (Brown and Herendeen, 1996), where parallel quantitative analyses of several simple systems are performed. The problem of flows allocation in EME is strictly related to the quality of the outputs. When the quality of two outputs is the same (e.g. electricity that can be used both to re-charge the battery of an electric vehicle and to run a hair dryer), EME considers this as a split, while co-products represent flows of different quality (e.g. soybean oil and glycerine obtained in the production of biodiesel from soybean). A comprehensive description of the similarities and differences between the EME and LCA methodological frameworks is presented in (Rugani and Benetto, 2012), where the authors also propose the utilization of the matrix-based computational approach used in LCA (Heijungs and Suh, 2002) for a more consistent quantification of UEVs. This approach is, however, still at its infancy and therefore is not adopted in the present paper. Some attempts have already been made to mathematically illustrate emergy and its properties in a more structured way and to propose emergy computational approaches fully or partially fulfilling the emergy algebra rules. From a conceptual perspective, the two most mathematically structured frameworks for emergy calculations proposed in the literature are the one proposed by Giannantoni (2006), who described an approach based on nonlinear differential equations and on a variant of the functional derivatives concept and the one proposed by Bastianoni et al. (2011), who presented an approach based on the application of set theory. A further original and pioneering attempt, where the calculation steps of emergy algebra are structured into a formal and scientifically consistent framework, has been done by Tiruta-Barna and Benetto (submitted for publication). The authors demonstrate the emergy algebra rules starting from emergy’s definition and using dynamic modeling applied to a simple network of processes. Interestingly enough, they observe that rule #4 has three different formulations depending on the relative magnitude of two characteristic time scales: one derived from the network’s dynamics and one of observation (i.e. the integration time). They further infer that the conventional form of rule #4, usually found in the emergy literature, is valid only for the observation time, which is much higher than the network’s characteristic time. These conceptual approaches do not, however, deal with the actual implementation

82

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

F

F

100

10

FEEDBACK X1 7

3

X1

X 10 S 400

SPLIT

Transformation process

CO-PRODUCTS

3

S 400 X2

SPLIT

50

Process B

Process A

2

Y Y 20

353

X2

55

380 unit of measure: JOULE (J) unit of measure: JOULE (J)

F

F

100

100

FEEDBACK X1 350

240+60

X1

X 500 S 400

Transformation process

SPLIT CO-PRODUCTS

150

S 400 X2

SPLIT

400+60

Process B

Process A Y

Y 500

500

160+40 X2

unit of measure: Solar emJOULE (seJ) transformity S = 1 transformity F = 10 transformity X1 = transformity X2 = 50 → = 350/7 and 150/3 transformity Y = 25 → = 500/20 (A)

unit of measure: Solar emJOULE (seJ) transformity S = 1 transformity F = 10 transformity X1 = transformity X2 = 100 → = (240+60)/3 and (160+40)/2 transformity Y = 9.2 → = (400+60)/50 (B)

Fig. 1. Example of energy vs. emergy calculation with (a) splits & co-products, and (b) feedbacks. Adapted from Bastianoni (1996).

of emergy algebra rules to (large) system’s networks. To this respect, Li et al. (2010) propose a pre-conditioning of the system of equations, which seems however to be unpractical and not suited for large systems, which are characterized by vast networks of interconnected processes. A similar operational methodology is shared by Le Corre and Truffet (2012), who adopt a matrix approach and describe a “path-oriented” method which is further applied to a small graph. The authors assume the emergy paths to be known, anticipating that the computation of emergy paths (that we treat in the present paper) is the subject of a companion paper. They also claim that the only difficulty of their algorithm is the “factorization” of the set of emergy paths in the system’s network. However, this factorization implies a preliminary analysis of the whole network (just as in the case of prior pre-conditioning in Li et al., 2010) before starting the calculations, which again is time consuming and impractical for very large networks. Furthermore, the algorithm used by Le Corre and Truffet (2012) needs the relations between the arcs to be stored in an array. Even though the authors do not perform a memory requirement and computational cost analysis, this operation is normally computationally expensive and requires extensive memory usage. In Patterson (2012) formal matrix algebra methods (the minimum eigenvalue method) have been applied for determining transformities in complex ecological or economic networks, similarly to the approach followed by Collins and Odum (2000). More precisely, the author previously introduced the concept of quality equivalent methodology (QEM) (Patterson, 1983), which has been considered to be a matrix algebra version of Odum’s emergy methodology (Collins and Odum, 2000; Hau and Bakshi, 2004). In Patterson (2012) the author describes a matrix algebra method for determining emergy-based efficiencies of processes in a system, from matrix energy flow data. It is based on

solving, by using singular value decomposition (SVD), a system of simultaneous equations that describe the flow of energy (via interconnected processes) in a complex system. The solution of these simultaneous equations results in the determination of a series of quality coefficients (basically the transformities or specific emergies depending on the fact the data are expressed in energy or mass terms) for each energy form in the system of equations. Although the method proposed by Patterson (2012) is interesting, it just solves the over-determined system of equations obtained from the “supply and use” matrix describing a generic system. This approach is similar to the one used in Marvuglia et al. (2010, 2012), where SVD was used to compute a total least squares (TLS) solution of the over-determined system of equations obtained applying the matrix method to the solution of the inventory problem. The method proposed by Patterson (2012) is undoubtedly prone to automation and scalability, and the structure of the matrices used by the author is the same as the ones we use in SCALE3 ; however, it does not comply with the emergy algebra rules, being the observance of these rules one of the key differences between emergy analysis and QEM. Furthermore, Patterson (2012) calculates the specific solar emergies (as he calls the above mentioned quality coefficients) for each of the processes considered in the system, whereas with SCALE it is possible to calculate the emergy of a specific product, chosen in the selected functional unit (see online supplementary material of the paper). In recent years, complementarities between EME and LCA have been emphasized within the emergy community (e.g. Ness et al.,

3 Except the fact that, opposite to our case, in Patterson (2012) rows describe processes and columns commodity flows.

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

2007). As recognized by Rugani and Benetto (2012), EME could benefit from the use of existing life cycle inventory (LCI) databases, which account for hundreds of environmental interventions in thousands of common industrial processes, such as the Ecoinvent database (Weidema et al., 2011). On the other hand, through the LCI databases, emergy might bring into LCA a complementary concept to assess impacts from ecosystem services’ use. 2.1. Toward the integration of emergy and LCA In order to develop an operational tool for allowing emergy calculation using LCI matrices we formulated the problem in a matrix-based structure which comes directly from the LCI framework, as described above.4 In particular, our background mathematical structure is the matrix method for the solution of the so-called inventory problem (Heijungs and Suh, 2002). The application of emergy algebra to the inventory problem could be realized through the application of one of the algebra rules (namely the rule #4 described above) to the scaled technology matrix5 obtained after solving the system of linear equations describing the technosphere.6 It is worth noticing here that the matrix approach is only applied to the relationship between production systems and natural resources. In other words, the input emergy sources (related to the natural resources) are calculated adopting the baseline approached conventionally used in emergy evaluation and not following the approach proposed in Rugani and Benetto (2012). This application presents two major challenges: (1) allocation of resources in LCI shall be avoided since it is in contrast with algebra rule #2; i.e. all the multi-output processes shall be included without allocation (Rugani et al., 2011); (2) specific rules for co-products, splits and feedbacks make allocation at each node of the network depend on the surrounding nodes and links. This means that rule #4 must be applied at each internal node of the network depending on the surrounding nodes and links. Such a procedure is feasible, and it has been shown already, when carried out manually, for simple systems, e.g. for a 11-by-11 matrix (Li et al., 2010). However, the application of the solution method proposed in this latter paper is unfeasible in cases involving large systems. Benetto et al. (2011a) carried out an automatic implementation in Matlab of the track summing algorithm for a system involving a technology matrix of dimension 7-by-7 including two multi-output processes. In this case study it was pointed out that there was a limitation of the method’s non-scalability to large matrices due to high computational costs. In that case the issue of emergy allocation was prevented by manually modifying the unit process data of the multi-output processes in the technology matrix. Since, in order to compute the emergy values, only the resources consumed for a given production system are relevant, only this part of the environmental interventions matrix7 was used, ruling out the part containing the pollutant emissions. More specifically, a separate dataset was kept for all the co-products of multi-output processes but the full amount of input inventory data (flows of commodities and natural resources) was assigned to all the co-products. As will be explained in the following section, the problem was finally solved in the application presented in this paper by applying graph search theory and related algorithms.

4 For a short, non-exhaustive, presentation of the mathematical background behind the matrix method we forward the reader to online supplementary material of this paper. 5 The technology matrix is the matrix whose generic element aij shows inflows or outflows of commodity i of process j. 6 For details see Heijungs and Suh (2002). 7 The environmental interventions matrix is the matrix containing the pollutant emissions and the natural resources extractions related to the whole system analyzed.

83

3. Calculation Because of the computational hurdles outlined above, so far an automatic implementation of emergy algebra has not been feasible for large systems (Rugani et al., 2011; Li et al., 2010), such as the whole technosphere matrix contained in the database Ecoinvent (Weidema et al., 2011), which comprises over 4000 interconnected processes. The software SCALE is, to the extent of our knowledge, the first tool allowing this kind of up-scaling. SCALE is an operational answer which provides solution to two problems. The first one is assigning the total burden of commodity flows and resource consumption to all the co-products delivered by each of the multi-output processes contained in the life cycle inventory. The second is avoiding double counting of co-products when they reunite. The Ecoinvent database (version 2.2) contains 227 multi-output processes. Let us consider, as an example, one of these processes in the agricultural production sector, namely the process of barley production, called “Barley extensive, at farm” (Fig. 2). This process delivers barley straw and barley grains at the same time. The inventoried values contained in the database refer to already allocated processes. This means that these resources and commodity flows have already been apportioned to each co-product according to certain rules (which can be deducted for each multi-output process by reading the Ecoinvent documentation), obviously infringing emergy algebra rule #2. In order to comply with rule #2 it is necessary to assign the full amount of the inventoried resources and commodity flows to each and every co-product. To solve this problem SCALE calls an ad hoc devised script which allows the extraction of text and numerical data directly from the .xml files downloaded from the Ecoinvent database. The script allows an automatic application of this procedure even to very large LCI databases (see Section 1 of the supplementary material). For the solution of the second problem SCALE uses a specifically implemented variant (based on graph search theory) of the track summing algorithm (Tennenbaum, 1988). The track summing method follows each path in the network of emergy values associated with the energy and mass flows from the sources to the output and divides the total emergy input to a process by the corresponding available energy output to obtain the transformity of the process. This method was first developed by Tennenbaum (1988) and it was presented in Odum (1996). Formally, for each independent source Sk , the emergy Emk,i related to the generic ith node of the graph is calculated by considering every simple path pj (not crossing the same node twice) from the source Sk to the node i: Emk,i =

m 

indep

Emk,i,j

(1)

j=1 indep

where m is the number of paths and Emk,i,j are the independent emergy values which propagate from Sk to the node i, through the path pj . indep

In order to calculate Emk,i,j the above mentioned emergy algebra rules have to be applied, filtering out the double counted emergy values. This algorithm essentially follows three steps: (1) It reads the flows associated with a given system, in its matrix form. This implies that the user does not need to draw the network describing the system. (2) It performs a graph search and memorizes the paths going from each input node (emergy sources) until the output node, provided that: (a) the path does not have a feedback, i.e. it hits a node in the same path which was visited previously; (b) the

84

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

Fig. 2. Multi-output processes contained in the Ecoinvent (version 2.2) database, grouped according to the different categories and example of allocation of the inventory for a generic process in the LCA (a) and EME (b) procedures.

emergy value along this path never goes below a fixed threshold (henceforth called minflow). The choice of a suitable threshold can significantly speed up calculation without increasing the information loss too much. (3) As a last step, the algorithm finally records the output emergy associated with each of the fully explored paths. In practice, the whole network representing the studied system is explored as a graph (where each process represents a node of the system and the nodes are interconnected through the energy and mass flows they exchange). The different nodes are visited following the depth-first search strategy (Russel and Norvig, 2009), which was implemented in C++. In the depth-first search (DFS) the search starts at the root8 and explores as far as possible along each branch before backtracking, i.e. before going back to the last explored node from which it was possible to visit a new branch of the tree. The algorithm tracks emergy values for each independent emergy source, branching off into subsequent paths as they are encountered and maintaining the set of previously visited elements to prevent cycles. Finally it sums up the results independently. In a previous attempt, a variant of the algorithm, using a multi-threaded breadth-first search (BFS) strategy (Russel and Norvig, 2009), had been explored. BFS begins at the root node and explores all the neighboring nodes. Then for each of those nearest nodes, it explores their unexplored neighbor nodes, and so on, until it finds the goal.9 The BFS of the emergy calculation algorithm exploits concurrency, i.e. it allows parallel processing of different paths at simultaneously, thus allowing a shorter running time. It was implemented in the programming environment Scala (Odersky et al., 2010), which has a powerful concurrency model and is thus particularly suited for exploiting parallelization (on multi-core processors). However, although the running time is lower than for its DFS version, the memory requirement for the BFS instantiation of the algorithm is quite high and does not justify a desktop uti-

8

In our case the algorithm is launched as many times as there are emergy sources and each time the node representing the emergy source taken into account represents the root. 9 In our case the goal is the output node, corresponding to the chosen functional unit, i.e. the product whose lifecycle network is being assessed and at which the emergy required will be calculated.

lization, like the one currently envisioned for SCALE. In the case of an upgrade of the software to a cloud computing application, the multi-threaded BFS can be run over 100’s or 1000’s of nodes and will be able to compute solutions to very large problems. The Scala-based solution can work in fact on a distributed platform (cloud is implemented on a cluster of machines and can provide even thousands of virtual computing nodes). The different search strategy followed by the two algorithms is illustrated in Fig. 3. An important element of our computational strategy is the use of a threshold, which must be exceeded for the emergy value to be transferred from a node to the following one in the search. In practice, if this value is lower than a specified minflow parameter called ˛ (multiplied by each initial input value) the algorithm halts the branching of new paths. This limits the order of complexity for space requirements (i.e. for the storage of the number of paths generated by any one source) to O(1/˛) and to O(K/˛) for a problem with K separate input flows. The pseudo-code of the algorithm is provided in Section 2 of the supplementary material. This algorithm has already been applied to a simplified version of the production system for flat glass (Marvuglia et al., 2011). It yielded exactly the same results obtained for the same case study using the software Emsim, which is an emergy simulator that does not allow a direct link to automatic calculation routines, since it requires the system’s diagram to be drawn by the operator (Valyi and Ortega, 2004). In this paper we up-scaled our application dealing with two different case studies, whose matrix of equations was already available from previous LCA studies: (a) a system for the production of heat from grape marc pellets (Jury et al., 2011) and (b) a system for the production of drinking water (Benetto et al., 2011b). They entail technology matrices of size 2052-by-2052 and 2154by-2154, respectively. 3.1. The software interface The software SCALE has a practical and easy-to-use graphical user interface (GUI), which was programmed in C++, using a specific open source library called Qt, which is available with the cross-platform C++ integrated development environment named Qt Creator (Nokia, 2012). Qt is a cross-platform application framework (a library) that is widely used for developing application

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

S A C S

A C

S B

D C

A C

G

D C

A C

G

C

D

D C

G

S B

A C

G

S B

D C

A C

G

B D

C

G

Depth-First Search (DFS)

G

reference graph

S

S

Time progression direction Nodes already visited Node being visited

A

B

C

B D

S

S B

85

A C

C

D C

A

B

G

S B

D C

A C

G

S B

D C

A C

G

S B

D C

A C

G

S B

D C

A

C G

B

D C

G

Breadth-First Search (DFS) Fig. 3. Order in which the nodes are explored in the DFS and BFS algorithms.

software with a graphical user interface (GUI) and also used for developing non-GUI programs such as command-line tools and consoles for servers. An example of application developed using Qt is the music streaming service Spotify (Spotify, 2012). Each calculation is executed in a thread (QThread). It is a dynamic application and that allows one to follow the progression of calculations. The calculation engine of the software contains a matrix computation part and a graph search part. The software GUI contains two main tabs: the “Complete calculation” tab and the “Emergy Calculation” tab. A screenshot of the GUI, showing the “Complete calculation” tab is shown in Fig. 4. Three main menu windows are available in the menu bar (marked with the red box in Fig. 4). (For interpretation of the references to color in this text, the reader is referred to the web version of the article.) The “Home” window (yellow box in Fig. 4) allows: 1. Selection of the workspace (green box in Fig. 4). The program stores the calculation files in a folder chosen by the user. If a folder is not selected, the other modules of the software remain idle. The content of the workspace is shown in the part of the interface marked with the light blue box. The workspace can be opened with the button at the bottom of this box. 2. Import of a matrix coming from a LCI database,10 matrix algebra operations (“Matrix Calculation” part of the window) and emergy accounting (“Emergy Calculation” part of the window). These three parts are shown within the orange box in Fig. 4. When a user uploads a .csv file containing the matrix11 describing the system, the algorithm reads its content and splits it into two parts: the technology matrix and the environmental intervention matrix, which contains the raw materials consumption. After selecting a matrix to be imported in the workspace, the user is asked to specify a functional unit vector. The user can start typing the beginning of the name of the process in the matrix which delivers the desired commodity (i.e. the product being studied and for which a functional unit has therefore to be specified). An automatic

10 The matrix the user imports is a matrix containing both the technology and the environmental intervention parts. However, only the part of the environmental interventions matrix related to raw materials consumption will be imported for making the emergy calculations (Rugani and Benetto, 2012). 11 The matrix comes from the “un-allocation” procedure carried out using the algorithm described in Section 2 of the supplementary material.

name-completion function will suggest the name of the possible processes contained in the matrix whose names are compatible with the string typed by the user. Then the user can select the desired product and input a desired functional unit (e.g. 1 MW of energy or 1 kg of the selected product, etc.). Once the matrix has been imported and a functional unit has been chosen, the “Matrix Calculation” module of the software solves the corresponding system of linear equations and scales the two parts of the matrix using the calculated solution vector (the vector of scaling factors) according to the standard matrix method for the solution of the inventory problem, well described in (Heijungs and Suh, 2002; see also Section 1 in the supplementary material). When the matrix calculation has been completed, the results can be saved on the workspace to be used later for further investigations or simply to be used later, only to run the emergy calculation part. The emergy calculator requires three inputs named , <flow multiplier> and . The is a text file (normally a .txt file). It contains three columns. Each row represents a branch of a path: the first entry of the row represents the name of the start node of this branch (e.g. NODE A); the second entry represents the name of the end node of the branch (e.g. NODE B) and the last entry contains a scalar ranging between 0.0 and 1.0, which represents the ratio between the emergy value going from NODE A to NODE B and the total energy value going out from NODE A. A warning message pops up if the user introduces graph files containing values >1 in the third column. The <flow multiplier> is the above described threshold (the minflow). The third input, the , is another text file containing as many equalities as the total number of emergy values entering the input nodes (i.e. the nodes of the graph which “receive” an input emergy value which will be distributed across the all graph). The equalities are separated by a space. If a node receives more than one emergy input, all its input shall be reported in the file and the emergy calculator sums them up before running the graph search algorithm. An example is shown in Fig. 5, where the node A takes an input of 1000 emergy units (sej or Msej, i.e. mega solar emjoules, etc.) from whatever natural resource and an input of 2000 emergy units from another natural resource; the node B takes an input of 3000 and another of 4000 emergy units from two different resources, and so forth. In this case the input is represented by the text file: A = 1000 A = 2000 B = 3000 B = 4000 C = 10,000 C = 10,000.

86

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

Fig. 4. Screenshot of SCALE’s GUI (“Complete Calculation” tab).

S5

10000

S4

10000

S1

1000

C A B

2000

S2

3000 4000

D

S3 Fig. 5. Example of a network with node receiving multiple emergy input from the emergy sources (S1,S2,. . .,S5).

Finally, a “New Calculation” button at the bottom of the area of the GUI marked with the yellow box in Fig. 4 allows the user to reset all the selected variables (orange box in Fig. 4) and start a new calculation. Clicking on the “Emergy Calculation” tab, a second window of the software opens up (Fig. 6), which allows the utilization of the emergy calculator, skipping the previous steps, in the case the user has previously accomplished the matrix calculation and saved the result in a folder. The required inputs and “Input file” (i.e. the file) can be uploaded using the proper buttons; the threshold (i.e. the <flow multiplier>) can be directly typed by the user. Needless to say that the reliability of the results obtained from SCALE depends on the reliability of the emergy values assigned to the sources. In this context, it is worth recalling that, as any other computational framework, EME is characterized by three types of uncertainty: parameter, scenario and model uncertainty (Ingwersen, 2010). On the other hand, as remarked by Ingwersen (2010), “describing the uncertainty in parameters, scenarios and models requires significant effort and must draw from previous applications of various models and across various scenarios”. It is

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

87

Fig. 6. Screenshot of SCALE’s GUI (“Emergy Calculation” tab).

also important to remind here that, being emergy strictly dependent on the path followed from the source to the final product, the level of accuracy with which the entire system (from geo-biosphere to the technological system delivering a final product) is described plays a crucial role in the calculation. This influences then the accuracy of the results, even upstream the application of the emergy rules. The objective of SCALE is hence to allow a calculation of the emergy of products for large systems, coherently with the emergy algebra rules and in a fast and automatic way. The characterization of the uncertainty of the UEVs used to compute the emergy of the inputs sources is left to the user. The level of detail with which the studied systems are described depends on one side from the user (for the description of the foreground system) and on the other side from the environmental database from which the system’s matrix is extracted (for the description of the background system).12 About the accuracy of the data, as noticed by Campbell and Ohrt (2009), as a rule of thumb, emergy analysts aim to achieve esti-

12 For the definition of foreground and background systems see supplementary material.

mates that are within 10–15% of the actual value of the variable used in the analysis. However, the authors themselves notice that, even though numerous errors have been found and corrected in the emergy analyses conducted over the last two decades, the results and the final conclusions of an emergy analysis are rarely changed by subsequent corrections. This reinforces the position we assume in this paper, which does not intend to deal neither with data uncertainty, nor with model uncertainty, but only with a rigorous and automatic application of the emergy algebra to large networks.

4. Results 4.1. Production of pellets from grape marc and combustion to generate heat The functional unit chosen is 1 MJ of heat obtained from the combustion of pellets produced from fresh grape marc, as detailed in Jury et al. (2011). The size of the technology matrix describing the system (i.e. including foreground and background system) is 2052-by-2052.

88

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

Fig. 7. Representation of the emergy values (and “lost” emergy values) related to the case study of heat production from grape marc pellets.

The emergy sources have been grouped under the 8 categories shown in Table 1. Summing up the emergy contribution of each source, we obtained 1.96 × 104 sej as the total emergy input of the system. These emergy inputs subsequently re-distributed along each path of the system’s network. SCALE calculates the emergy associated with each stretch of the graph (from node to node) representing the system (following the emergy algebra rules) and it obtains the total emergy associated with the final node of the graph, i.e. the emergy of the chosen functional unit. The results obtained are summarized in Figs. 7 and 8 and extensively discussed in Section 5.

4.2. Production of drinking water The functional unit is 1 m3 of potable water, as detailed in Benetto et al. (2011b). The size of the technology matrix describing the system is 2154-by-2154. Also in this case the emergy sources have been grouped under the same 8 categories used in the case study #1. The values of the respective emergy inputs divided by category, as well as the total emergy input of the system (3.56 × 107 Msej) are shown in Table 2. The results obtained with the software SCALE for this case study are resumed in Figs. 9 and 10. The same observations formulated for the previous case study (see Section 5) hold true also in this case.

Table 1 Emergy inputs divided by category for the case study of heat production from grape marc pellets.

Fig. 8. Representation of the output emergy value, running time and total percentage emergy “lost” related to the case study of heat production from grape marc pellets.

5. Discussion Some preliminary insights can be gained by the observation of the graphs shown in Section 4. The colors of the curves plotted in Fig. 7 are the same as the colors of the y axes they refer to. On the

Table 2 Emergy inputs divided by category for the case study n.2 (potable water from water treatment plant).

Emergy sources

Emergy value (Msej)

Emergy sources

Emergy value (Msej)

Renewable, atmospheric-gaseous res. Renewable, energy resources Renewable, land resources Renewable, water resources Non-renewable, fossil resources Non-renewable, metal ores Non-renewable, mineral resources Non-renewable, nuclear energy resources

0.00E+00 5.96E+01 2.63E+02 1.20E+02 1.16E+04 1.41E+03 6.12E+03 7.27E+00

Renewable, atmospheric-gaseous res. Renewable, energy resources Renewable, land resources Renewable, water resources Non-renewable, fossil resources Non-renewable, metal ores Non-renewable, mineral resources Non-renewable, nuclear energy resources

0.00E+00 3.96E+04 4.17E+04 3.12E+04 8.36E+05 7.65E+05 3.39E+07 4.94E+03

Total emergy input

1.96E+04

Total emergy input

3.56E+07

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

89

Fig. 9. Representation of the emergy values (and “lost” emergy values) related to the case study of drinking water production.

x axis the negative of the Log10 of the surveillance threshold which we set up as minflow is plotted. The blue curve with triangle markers shows the amount of emergy “lost” (i.e. not accounted for) as a function of the threshold’s value. Obviously, by decreasing the threshold one increases the probability that nodes able to exceed the fixed threshold (i.e. which transfer to the following node an amount of emergy higher than the fixed threshold) do exist. As a consequence, the number of fully explored paths increases and correspondingly the amount of emergy not accounted for decreases. One could also argue that, the further one decreases the threshold13 the more the complexity (i.e. the higher the ramification of the explored graph) and this increases the probability to encounter paths that constitute feedbacks (i.e. loops). Consequently, the total flow that one excludes from the accounting, due to the occurrence of a loop, has to increase (as shown by the green curve). (For interpretation of the references to color in this text, the reader is referred to the web version of the article.) The y axes on the left hand side of the graph have to be used as follows: when using the black curve with square markers, on the y axis (left side) one can read the number of complete paths that are followed by the algorithm from one of the sources to the final node; when using the black curve with triangle filled markers, on the y axis (left side) one can read the number of loop violations, i.e. the number of times a path would cross itself due to the occurrence of a loop. In this case the algorithm stops the calculation and re-starts from the next path in the search queue. In agreement with what was said before, the graph shows that the number of completed paths and the number of paths affected by loops increases as the minflow threshold decreases. An analogous reasoning also holds true for the results of the drinking water case study, summarized in Fig. 9. Referring still to the first case study (grape marc pellets), in Fig. 8 the green curve refers to the running time of the algorithm, the red curve refers to the output emergy from the system (i.e. the emergy associated with the functional unit) and the blue curve refers to

13 Decreasing the threshold implies exploring more and more paths. In fact, the number of the nodes exceeding the value of the threshold will increase if the threshold decreases.

the percentage of emergy (Em), with respect to the total emergy input, which is “lost” (i.e. not accounted for) because one of the two control conditions described in Section 3 (i.e. a path would form a loop or the emergy flow at a node does not exceed the threshold) is met. (For interpretation of the references to color in this text, the reader is referred to the web version of the article.) Interestingly enough, it is possible to see that there is a value of the threshold, equal to 10−6 Msej, for which the percentage of emergy “lost” is as little as 3% of the total input. On the other hand, when the threshold decreases below this value, the running time increases exponentially and the emergy output changes by a negligible amount (i.e. it passes from 1.90 × 104 Msej for a threshold of 10−6 Msej to 1.93 × 104 Msej for a threshold of 10−8 Msej). 10−6 Msej is thus an optimal value, allowing a running time as little as ca. 3 min 36 s. Concerning the case study presented in Section 4.2 (drinking water), one can similarly observe in Fig. 9 that for a value of the threshold equal to 10−6 Msej the percentage of emergy “lost” is 3.5% of the total input. Again, when the threshold decreases

Fig. 10. Representation of the output emergy value, running time and total percentage emergy “lost” related to case study of drinking water production.

90

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91

below this value, the running time increases exponentially and the emergy output changes by a negligible amount (it passes from 3.47 × 107 Msej for a threshold of 10−6 Msej to 3.48 × 107 Msej for a threshold of 10−8 Msej). Also in this case 10−6 Msej is thus an optimal value, allowing a running time of ca. 10 min 14 s.

6. Conclusions The paper describes a matrix application that is a variant of the track summing algorithm for the calculation of the emergy associated to a man-made product. The algorithm (programmed in C++) has been implemented in a software package endowed with a simple GUI, which is now in its beta version: the software SCALE (Software for CALculating Emergy based on life cycle inventories). The network of interconnected processes comprising the studied system is described as an oriented graph, and the emergy of the chosen functional unit is calculated using the depth first search (DFS) strategy, well known in computer science. SCALE allows the calculation of emergy flows in complex real systems like the entire technosphere described by the LCI database Ecoinvent. Most importantly, it makes this possible in an automatic way that complies with the rules of emergy algebra. In this paper SCALE was applied to two different case studies: (1) production of pellets from grape marc and their combustion to generate heat; (2) production of drinking water through a treatment process representative of European conditions. The technology matrices describing these systems have sizes 2052-by-2052 and 2154-by2154, respectively. An important control parameter of the algorithm running in SCALE is the threshold value called minflow. If the emergy value going from a node of the graph to the subsequent one in the explored path is lower than the fixed threshold (which is equal to a certain percentage of the total emergy input to the system and is generally expressed in Msej) then the algorithm halts the branching of new paths for that input. The value of the threshold should be chosen on a case-by-case basis, but the applications described in this paper show that for systems whose technology matrix contains over 2000 rows (and columns) a value of the threshold representing a very good trade-off between execution time and accuracy of the final result is 10−6 Msej. For this value of minflow the running time of the algorithm was ca. 3 min 36 s for the first case study and ca. 10 min 14 s for the second case study. The development of SCALE presented in this paper goes in the direction of advancing the research and the current debate on the absence of standardization in emergy. Indeed, the latter is determined by a high adaptability and customization of the emergy calculation models, which is rather a characteristic of emergy that may determine low accuracy and representativeness, even though this does not prevent emergy models from being accurate and representative enough if well-structured and implemented (undoubtedly a highly time consuming task). In this connection, the automated calculation process based on LCI database allowed by SCALE can make emergy evaluation of systems easier, quicker and essentially feasible, thus contributing to increase reproducibility and transparency, make procedures uniform (standardization), and make different assessments comparable to each other. With the application of the software to more and more case studies it will be possible to create a sort of inventory of recommended values for the threshold, so that a good practice code will be established, with the identification of some criteria to guide the user in the choice of the parameters. The final version of SCALE will allow the achievement of reproducible, consistent and transparent calculations of emergy values for thousands of products in the LCI databases. This would also enable to build a large and harmonized database of both globally and regionally scaled

UEVs, which could be eventually standardized and used in classical emergy analyses. Furthermore, the algorithm could be applied on a case-by-case basis to a specific product’s life cycle modeled using conventional LCA software using the matrix approach, like e.g. the OpenSource tool OpenLCA (OpenLCA, 2012), thus allowing an exact calculation of the emergy associated with the studied products and therefore achieving a complete combination of LCA and the emergy perspective for environmental assessment. To this aim, a future development of SCALE could be the enhancement of LCA software with specific functionalities dedicated only to emergy calculation and not requiring specific LCA competences. This would allow users to perform both classical LCAs and carry out emergy accounting using the different protocols incorporated into the revised LCA framework (the LCI, the matrix-based calculation, etc.). An additional feature of SCALE might also be the computation of the percentage contribution to the emergy output of each of the resource categories in which the emergy inputs can be classified, for example according to the rationale followed in Rugani et al. (2011): renewable resources, R, such as water, land and biomass, and non-renewable resources, N, such as fossil fuels, minerals and metals. Acknowledgements The authors would like to acknowledge Prof. Ligia Tiruta-Barna (INSA Toulouse, France) for the invaluable discussions during the research project and the French national research fund (ANR) for funding of part of the SCALE software development. Benedetto Rugani gratefully acknowledges the National Research Fund of Luxembourg, which cofounded this works together with the European Commission under the Marie Curie Actions (FP7 – COFUND). Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.ecolmodel.2012.09.013. References Bastianoni, S., 1996. Auto organizzazione, emergy e transformity. In: Dejak, et, al. (Eds.), Chimica Fisica per le scienze Ambientali. ETAS LIBRI, Milan, Italy. Bastianoni, S., Facchini, A., Susani, L., Tiezzi, E., 2007. Emergy as a function of exergy. Energy 32 (7), 1158–1162. Bastianoni, S., Morandi, F., Flaminio, T., Pulselli, R.M., Tiezzi, E., 2011. Emergy and emergy algebra explained by means of ingenuous set theory. Ecological Modelling 222, 2903–2907. Benetto, E., Tiruta-Barna, L., Rugani, B., 2011a. Toward consistent emergy calculations with LCI databases: implementing the emergy algebra. In: Proceedings of the SETAC Europe 21st Annual Meeting, Milan, Italy, May 15–19, 2011. Benetto, E., Tiruta-Barna, L., Rugani, B., Baudin, I., 2011b. Evaluating natural resources use for potable water production. In: Proceedings of the Life Cycle Management (LCM 2011), Berlin, Germany, August 28–31, 2011. Brown, M.T., Cohen, M.J., 2008. Emergy and network analysis. In: Jörgensen, S.E., Fath, B. (Eds.), Encyclopedia of Ecology. Elsevier, Amsterdam, Netherlands. Brown, M.T., Herendeen, R.A., 1996. Embodied energy analysis and EMERGY analysis: a comparative view. Ecological Economics 19, 219–235. Campbell, D.E., Ohrt, A., 2009. Environmental Accounting Using Emergy: Evaluation of Minnesota, USEPA Project Report, EPA/600/R-09/002, 138 pp. Collins, D., Odum, H.T., 2000. Calculating Transformities with an Eigenvector Method. In: Brown, et al. (Eds.), Emergy Synthesis 1: Theory and Applications of the Emergy Methodology. Center for Environmental Policy, University of Florida, Gainesville, USA, pp. 265–280. Curran, M.A., 2006. Life Cycle Assessment: Principles and Practice. EPA/600/R06/060, U.S. Environmental Protection Agency, National Risk Management Research Laboratory, Cincinnati, OH, p. 80. Dong, X., Ulgiati, S., Yan, M., Zhang, X., Gao, W., 2008. Energy and eMergy evaluation of bioethanol production from wheat in Henan Province, China. Energy Policy 36, 3882–3892. Dong, L., Wang, R., 2009. Hybrid Emergy-LCA (HEML) based metabolic evaluation of urban residential areas: the case of Beijing, China. Ecological Complexity 6, 484–493.

A. Marvuglia et al. / Ecological Modelling 248 (2013) 80–91 Duan, N., Liu, X.D., Dai, J., Lin, C., Xia, X.H., Gao, R.Y., Wang, Y., Chen, S.Q., Yang, J., Qi, J., 2011. Evaluating the environmental impacts of an urban wetland park based on emergy accounting and life cycle assessment: a case study in Beijing. Ecological Modelling 222, 351–359. European Commission, Joint Research Centre, Institute for Environment and Sustainability, 2010. International Reference Life Cycle Data System (ILCD) Handbook – General guide for Life Cycle Assessment – Detailed Guidance, 1st ed. EUR 24708 EN, Publications Office of the European Union, Luxembourg, p. 417. Giannantoni, C., 2006. Mathematics for generative processes: living and non-living systems. Journal of Computational and Applied Mathematics 189, 324–340. Hassan, R., Scholes, R., Neville, A. (Eds.), 2005. Ecosystems and Human Wellbeing: Current State and Trends, vol. 1. Island Press, Washington, DC, USA (available online: http://www.maweb.org/en/index.aspx). Hau, J.L., Bakshi, B.R., 2004. Promise and problems of emergy analysis. Ecological Modelling 178, 215–225. Heijungs, R., Suh, S., 2002. The Computational Structure of Life Cycle Assessment. Kluwer Academic Publishers, Dordrecht, Netherlands. Ingwersen, W.W., 2010. Uncertainty characterization for emergy values. Ecological Modelling 221, 445–452. ISO 14040, 2006. Environmental Management: Life Cycle Assessment: Principles and Framework. International Organisation for Standardisation, Geneva, Switzerland. Jury, C., Kneip, G., Huck, V., Marvuglia, A., Benetto, E., 2011. Environmental impacts of pellets production from vinery residue – a site dependent result? In: Proceedings of the Life Cycle Management (LCM 2011), Berlin, Germany, August 28–31, 2011. Le Corre, O., Truffet, L., 2012. Exact computation of emergy based on a mathematical reinterpretation of the rules of emergy algebra. Ecological Modelling 230, 101–113. Li, L., Lu, H., Campbell, D.E., Ren, H., 2010. Emergy algebra: improving matrix methods for calculating transformities. Ecological Modelling 221, 411–422. Marvuglia, A., Benetto, E., Rugani, B., Rios, G., 2011. A scalable implementation of the backtracking algorithm for emergy calculation with life cycle inventory databases. In: Proceedings of the 25th International Conference on Environmental Informatics (EnviroInfo 2011), vol. 2, Joint Research Centre (JRC), Ispra, Italy, October 5–7, 2011, pp. 755–764. Marvuglia, A., Cellura, M., Pucci, M., 2012. A generalization of the orthogonal regression technique for life cycle inventory. International Journal of Agricultural and Environmental Information Systems 3 (1), 51–71. Marvuglia, A., Cellura, M., Heijungs, R., 2010. Toward a solution of allocation in life cycle inventories: the use of least-squares techniques. International Journal of Life Cycle Assessment 15 (9), 1020–1040. Ness, B., Urbel-Piirsalu, E., Anderberg, S., Olsson, L., 2007. Categorising tools for sustainability assessment. Ecological Economics 60, 498–508. Odersky, M., Spoon, L., Venners, B., 2010. Programming in Scala – A Comprehensive Step-By-Step Guide, 2nd ed. Artima Developer, Walnut Creek, CA, USA, 852 pp. Odum, H.T., 1988. Self-organization, transformity and information. Science 242, 1132–1139. Odum, T.H., 1996. Environmental Accounting – Emergy and Environmental Decision Making. John Wiley & Sons, New York, USA.

91

Patterson, M.G., 1983. Estimation of the quality of energy sources and uses. Energy Policy 11, 346–359. Patterson, M.G., 2012. Are all processes equally efficient from an emergy perspective? Analysis of ecological and economic networks using matrix algebra methods. Ecological Modelling 226, 77–91. Pizzigallo, A.C.I., Granai, C., Borsa, S., 2008. The joint use of LCA and emergy evaluation for the analysis of two Italian wine farms. Journal of Environment Management 86, 396–406. Raugei, M., 2011. Emergy indicators applied to human economic systems – a word of caution. Ecological Modelling 222, 3821–3822. Ridolfi, R., Bastianoni, S., 2008. Emergy. Encyclopedia of Ecology, 1218–1228. Rugani, B., Benetto, E., 2012. Improvements to emergy evaluations by using life cycle assessment. Environmental Science and Technology 46 (9), 4701–4712. Rugani, B., Huijbregts, M.A.J., Mutel, C., Bastianoni, S., Hellweg, S., 2011. Solar energy demand (SED) of commodity life cycles. Environmental Science and Technology 45, 5426–5433. Russel, S.J., Norvig, P., 2009. Artificial Intelligence: A Modern Approach, 3rd ed. Prentice Hall, Englewood Cliffs, NJ, USA. Sciubba, E., 2010. On the second-law inconsistency of emergy analysis. Energy 35, 3696–3706. Sciubba, E., Ulgiati, S., 2005. Emergy and exergy analyses: complementary methods or irreducible ideological options? Energy 30, 1953–1988. Tennenbaum, S., 1988. Network energy expenditures for subsystem production. M.S. Thesis, Environmental Engineering Sciences, University of Florida, Gainesville, USA, 132 pp. Tiruta-Barna, L., Benetto, E., submitted for publication. A conceptual framework and interpretation of emergy algebra. Ecological Engineering. Unpublished results. Valyi, R., Ortega, E., 2004. Emergy simulator, an open source simulation platform dedicated to systems ecology and emergy studies. In: Ortega, E., Ulgiati, S. (Eds.), Proceedings of IV Biennial International Workshop “Advances in Energy Studies”. Unicamp, Campinas, SP, Brazil, June 16–19, 2004, pp. 349– 360. Weidema BP, et al., 2011. Overview and methodology. Data quality guideline for the Ecoinvent database version 3. Ecoinvent Report 1 (v3), The Ecoinvent Centre, St. Gallen. Zhang, Y., Singh, S., Baral, A., Bakshi, B.R., 2010a. Accounting for ecosystem services in life cycle assessment. Part I. A critical review. Environmental Science and Technology 44, 2232–2242. Zhang, Y., Baral, A., Bakshi, B.R., 2010b. Accounting for ecosystem services in life cycle assessment. Part II. Toward an ecologically-based LCA. Environmental Science and Technology 44, 2624–2631.

Web references Nokia, 2012. http://qt.nokia.com/ (last accessed April 2012). Spotify, 2012. http://www.spotify.com/int/ (last accessed April 2012). OpenLCA, 2012. http://www.openlca.org/index.html (last accessed April 2012).