Supply versus use designs of environmental extensions in input–output analysis: Conceptual and empirical implications for the case of energy

Abstract Input–output analysis is one of the central methodological pillars of industrial ecology. However, the literature that discusses different structures of environmental extensions (EEs), that is, the scope of physical flows and their attribution to sectors in the monetary input–output table (MIOT), remains fragmented. This article investigates the conceptual and empirical implications of applying two different but frequently used designs of EEs, using the case of energy accounting, where one represents energy supply while the other energy use in the economy. We derive both extensions from an official energy supply–use dataset and apply them to the same single‐region input–output (SRIO) model of Austria, thereby isolating the effect that stems from the decision for the extension design. We also crosscheck the SRIO results with energy footprints from the global multi‐regional input–output (GMRIO) dataset EXIOBASE. Our results show that the ranking of footprints of final demand categories (e.g., household and export) is sensitive to the extension design and that product‐level results can vary by several orders of magnitude. The GMRIO‐based comparison further reveals that for a few countries the supply‐extension result can be twice the size of the use‐extension footprint (e.g., Australia and Norway). We propose a graph approach to provide a generalized framework to disclosing the design of EEs. We discuss the conceptual differences between the two extension designs by applying analogies to hybrid life‐cycle assessment and conclude that our findings are relevant for monitoring of energy efficiency and emission reduction targets and corporate footprint accounting.


SEEA: terms and definitions
The System of Environmental and Economic Accounting (SEEA), which was adopted as an international standard by the United Nations Statistical Commission, is a central reference point for the integration of physical and monetary accounts. The framework is based on the definition of three key flows: Flows from the environment into the economy are natural inputs, flows within the economy are products 1 , and flows from the economy to the environment are residuals.
Natural inputs encompasses all energy flows that are removed and captured from the environment by economic entities. This includes mineral and energy resources (e.g., crude oil, natural gas, coal, peat, and uranium) and inputs from renewable energy sources (e.g., solar, wind, hydro, geothermal and biomass). Products are energy carriers that are produced i.e. generated by an economic entity. This comprises fuels, electricity and heat that is generated and sold to third parties by an economic entity.
Energy products include electricity and heat from the combustion of biomass and solid waste. Some energy products may be used for non-energy purposes. For example, naphtha is used in the manufacture of plastic. A distinction must be made between primary and secondary energy products. Primary energy products are produced directly from natural inputs; they differ only by the amount of energy lost during production (e.g., natural gas evaporates during extraction). Secondary energy products are the result of a transformation of primary, or other secondary, energy products (e.g., crude oil into petroleum products, or fuel oil into electricity). Residuals comprise a number of energy flows. Most focus is on energy losses, S2-2 which includes losses through flaring and venting of natural gas and losses during production and transformation processes. Energy losses during distribution may arise from leakages of liquid fuels, loss of heat during transport of steam, and losses during gas distribution, electricity transmission and pipeline transport. Figure S2-1 displays the energy conversion chain, discerning the following energy flows and processes.
Primary energy industries, i.e. producers, capture natural inputs from the environment (R) and incinerate solid wastes (W) to produce primary energy products (V P ). Part of the input returns back to the environment as residuals in the form of transformation losses (L) and primary energy industries' own consumption for heating, pumping, traction, and lighting (O). Secondary energy producers use primary or other secondary energy products as a transformation input (U t ) to produce secondary energy products (V s ). Again, a fraction of the input leaves the process as a residual flow and is lost to the environment (L + O). Additionally, primary producers' own consumption of secondary energy products as well as secondary producers' own consumption of primary energy products (U e ) is also accounted on the input side. The total in-and out-flows of the energy industries (transformation processes) are balanced: R + W + U = L + O + V. Energy markets for primary and secondary energy products receive inputs from domestic (V = V p + V S ) and foreign producers (IM). Markets distribute energy products to non-energy industries (u in ), households (y h ), domestic energy industries (U = U t + U e ) and foreign industries (EX). Markets add energy to stocks for future use (S add ) and withdraw (S with ) energy from last year's production respectively (∆S = S add − S with ). Again, the accounting of all inputs to and outputs of the distribution processes (i.e. flows to and from markets) must lead to a balanced system: V + IM + S with = EX + U + u in + y h + S add .

S2-4
The economic value of energy flows Physical flows of natural inputs and residuals are not captured by monetary transactions (UN et al., 2014). All monetary transactions are interactions between economic entities. A monetary transaction is one in which an economic entity makes/receives a payment, incurs a liability or receives an asset stated in monetary units (EC, IMF, OECD, UN, & WB, 2009). Flows from the environment, the dilution of pollutants or the disposal of wastes are considered as "free gifts of nature" (Duchin, 2010;Leontief, 1970). Natural inputs and residuals are boundary flows and therefore represent an "extension" of the monetary IO table and the scope of the System of National Accounts (SNA). Products (primary and secondary), on the other hand, are flows that result from production processes within the economy. They  Pauliuk and colleagues (2015) illustrate how supply-use tables (SUTs) can be completely described as a bipartite directed graph. Graphs are an established mathematical concept and frequently deployed for describing relations between elements (Diestel, 2017). A graph model consists of edges or arrows that represent flows of objects and vertices (or nodes) that represent the processes in the system. Graphs that have two disjoint sets of nodes and that only have directed edges that connect a node in one set to a node in the other are called bipartite directed graphs. A SUT is a tabular representation of a bipartite directed graph where industries and markets (two disjoint sets of nodes) connect via flows of commodities (directed edges). Industries represent transformation nodes and markets distribution nodes. The former transform input commodities into output commodities, whereas the latter do not transform inputs, but simply transfer the intermediate output of one or more transformation nodes to the transformation nodes that use this intermediate exchange as an input (Pauliuk et al., 2015). In the following, we apply a simplified representation of bipartite directed graphs using boxes (representing nodes) and arrows (flows), a notation that is common in material flows analysis (Bertram et al., 2017;Mao, Dong, & Graedel, 2008).

Graph visualisation of EE-MIOTs
The next section shows how monetary flows in a single-region MIOT can be described as a bipartite directed graph, which is then followed by an extended graph visualization of the supply-extended and the use-extended SRIO model. The descriptions hereafter supplement the description of the two extension designs from the main text (see materials and method section). Figure S2-2 presents two variants of the single-region MIOT structure. The upper part shows an aggregated version where energy industries (primary and secondary) and non-energy industries (e.g., manufacturing and services), as well as energy markets and non-energy markets, are lumped together. Vector x stands for the gross production of industries. This is the amount of commodities supplied to the (domestic) market. Gross output x is used either as an intermediate input for production (Z) or for S2-5 final consumption (y). y comprises consumption of households, government and non-profit organizations serving households, as well as gross fixed capital formation and changes in inventories and, in a SRIO framework, exports. Industries' payments to primary i.e. factor inputs (or value-added in production) are represented as v. This includes compensation for employees and capital (interest rates), taxes and profits. In the SRIO model of the present study, monetary transactions representing imports (im), either competitive or non-competitive, are endogenized in the MIOT. This means imported commodities are treated as if they were produced within the economy, hence imports are included as part of the gross production vector (x) of the monetary IO table. Industries (v + im + Z = x) and markets (x = Z + y) are balanced and the expenditures on the consumption side (y) must equal the payments for factors and imports on the production side (im + v = y).
The lower part Figure   Based on the foregoing, the next figure shows a more disaggregated graph-based representation of the two extension-designs form the perspective of the MIOT. Figure S2-3 distinguishes between IO industries producing energy commodities and industries producing non-energy commodities. Note that here withdrawal (S_with) and addition (S_add) to stocks are explicitly represented as solid arrows in the corresponding extension design. Figure S2-3 depicts the supply-type and Figure S2-3 the use-type extension.

Energy footprints for 1999, 2007 and 2014
Comparing the final demand footprints and the associated variations for the years 1999, 2007 and 2014 reveals a relatively stable pattern over time. Figure S2-5 shows that the footprints of households and other final demand were always larger when applying a supply-extended IO model and the ones of S2-10 export demand consequently always smaller. In other words, the energy footprints of export demand were always larger when applying a use-extended IO model. Across all years, the strongest convergence between the two model results, viewed in both relative and absolute terms, is found for the footprints of households. The household footprints of 2007 were most similar showing an absolute difference of only 48 PJ or 7%, when taking the hypothetical mean of the two results as a benchmark. In absolute terms, again export footprints were most dissimilar. However, the observation that the ranking of the final demand footprints is sensitive to the extension design, does not apply to the results of 1999. For this year, both IO models rank the categories in the same order.

Energy footprints of households and exports disaggregated by final products
This section gives a brief overview of the differences between the energy footprints of household ( Figure   S2-6) and export demand ( Figure S2-7) disaggregated by products. The waterfall-charts shows the top-12 product footprint differences sorted in descending order from left to right, where the products with the largest absolute/gross difference, for example for the household footprint this is real estate services (68) with a difference of 23 PJ, is on the far left. Please note that all products not explicitly presented in this figure (hence aggregated in NEC) have an absolute difference that is below the smallest product difference shown here. The waterfall chart reveals how the differences on the product level accumulate and cancel each other out in the aggregated household footprint. The dark bars indicate that the supply-extended product footprint is larger than the use-extended product footprint whereas a lighter hatched bar stands for the opposite. The arrow stands for the total difference between the footprints.

S2-12
Disaggregating the footprints by final products reveals that only a small number of products makes up the largest part. In the case of households, for both model results, the same six products account for at least 75% of the total energy footprint. This is refined petroleum products with approx.

Concordance table between energy products and IO industries.
Because official Energy Accounts do not provide an allocation of energy supply, energy loss and nonenergy use to IO industries, the following adjustments were made. Firstly, a concordance table was constructed (Table S2-1) that links the supply of energy products to IO i.e. NACE industries using the statistical classification of products by activity (CPA) (Statistics Austria, 2004). Secondly, according to the information contained in the more detailed raw data set, only four manufacturing industries (producing petrochemicals, rubber, glass and basic metal products) are using energy products for nonenergy purposes. Lacking detailed information, total non-energy use was therefore allocated to these four industries assuming a constant ratio between final energy consumption and non-energy use.
Thirdly, transformation and transportation losses were allocated to the IO industries producing the energy products.

Comparison of energy footprints calculated with EXIOBASE
EXIOBASE is a global multi-region input-output database that includes 44 countries and 5 aggregated rest of the world regions (see country codes in SI.1) for the years 1995 to 2016. The industry-by-industry table differentiates between 163 sectors (Tukker, Giljum, & Wood, 2018;Wood et al., 2015). The environmental extension includes various energy flows, for example losses or direct consumption of households, for 74 energy carriers.

S2-14
As can be seen in Figure S2-8, according to the energy supply-use extensions of EXIOBASE, in 2014 total natural inputs (R) into the global economy amounted to 573.4 EJ (Exa Joule). Global energy use of industries (U) -which includes transformation and transportation losses, energy-industries own use and non-energy use -was 434.8 EJ (75.1% of global natural inputs). Direct energy consumption for final use (Y) -which comprises demand of households, governments, non-profit organisations and changes in inventorieswas 137 EJ (24.9% of global natural inputs). Losses and energy-industries own use amounted to 179.9 EJ and non-energy use to 34.6 EJ in 2014. supply-extended and all use-extended energy footprints must add up to the same total of 573.4 EJ.