1 Introduction

Air emission inventories, particularly greenhouse gas (GHG) emissions, have always been thought of as the primary source of information for international climate change agreements and trading (IPCC 2006; Lieberman et al. 2007). These have been compiled mainly at a national level, but when developed at the subnational level, these datasets can be a precious source of information for policymakers at different administrative levels, in accounting terms, for descriptive analysis and for policy analysis. However, although regional air emission inventories are routinely used, for example, by regional environmental protection agencies to assess the state of the environment, uncertainty is never considered. In some cases, uncertainty coefficients are not even available from the agencies and institutes responsible for the delivery of air emission inventories. However, accounting for uncertainty can often remarkably change an environmental and also, thus, political assessment based on emission inventories.

With respect to the random component of uncertainty obtained from expert judgment, uncertainty analysis deals with random errors based on the inherent variability of a system and the finite sample size of the available data (IPCC 2006). In previous work (Tonin et al. 2016), we addressed the issue of uncertainty in GHG inventories by considering its causes and how to improve the measurements (i.e., data production, data analysis, and data use). We ended up plotting a use-chain model where, when moving from the field of the biophysical environment to the field of social sciences and to decision-making, the analysis of uncertainty expands from (i) quantitative, mainly statistical methodologies aimed at improving the detail of the data to (ii) flexible open approaches aimed at integrating quantitative and qualitative information.

The content of the use-chain model is based on recent literature about uncertainty analysis; an important role in this review is played by the International Workshops on Uncertainty in Greenhouse Gas Inventories, which specifically target the sources of data we make use of. Some previous analyses (Lieberman et al. 2007; White et al. 2011) have highlighted the fact that initially, most workshop contributions, more generally, were more oriented toward data production and referred to the uncertainty related to measurement and to the lack of data. Regarding policy analysis, the experts considered the role of uncertainty in those policy tools that directly depend on emission inventories, such as climate change compliance rules and emission trading schemes. From the most recent workshop, it is worthwhile to note that in terms of data analysis, particularly of integration with other datasets, an increased number of presentations related to the employment of spatial GHG inventories for what concerns economic sectors. Important examples include Charkovska et al. (2018) for the agricultural issue, Halushchak et al. (2015) for fossil fuel extraction, Topylko et al. (2015) for electricity generation, and Charkovska et al. (2015) for industrial activity.

Reducing uncertainty involves a cost in terms of time, human resources, and additional information, so it would be useful to know whether and where the use of such effort is worthwhile (Nocera et al. 2015). Whether it is worthwhile can depend on many different identified criteria, and these need to be clearly presented and justified. The purpose of this paper is to contribute by proposing an initial screening whose aim is to identify situations where the additional work of assessing uncertainty is indeed needed; the paper also suggests basic criteria to justify the choice.

The case study used in this paper moves beyond the traditional contexts (i.e., compliance rules and emission trading) and scale (i.e., national) and shows the possible effects of uncertainty at a subnational level by using tools that are available for local policy action. To screen for uncertainty, we designed a composite approach; an insurance-based method is combined with a hybrid environmental accounting framework to facilitate both descriptive and policy analyses. The descriptive analysis is based on a set of environmental indicators, while the policy analysis is based on a decomposition analytic tool.

Thanks to this initial screening, it is possible to assess rough estimates with and without uncertainty and check their effect on the information provided to policymakers. This will be the basis of understanding whether and to what extent some work on uncertainty (that moves beyond the initial screening) must be undertaken, because estimates including uncertainties will change policy strategy and actions.

The Piedmont region in Italy and its provinces were selected as the case study in which the methodology, the accounting framework, the environmental indicators, and the decomposition analysis will be applied. We chose this case study because the air emission regional inventory is one of the best examples of such inventories in Italy. Their datasets are publicly available, and the uncertainty coefficients have been efficiently compiled by the functionaries in charge of the inventory.

2 Materials and methods

After describing the Piedmont region and its provinces, this section presents the basic accounting module, National Accounting Matrix including Environmental Accounts (NAMEA), used for this analysis. The initial description of the module does not include uncertainty, but in the following subsection, uncertainty is addressed by employing an insurance-based method. Finally, two analytic tools are described: the set of indicators that constitute the environmental themes and the shift-share analysis.

2.1 The Piedmont region and its provinces

The Piedmont region is located in the northwestern part of Italy. In Piedmont, the automotive sector (the FIAT group and activities that surround it) is the dominant industry, followed by the chemical, food, textile, clothing, electronics, and editorial industries. It is important to identify the main locations of all these sectors throughout the region in order to establish a connection with emitted pollutants, starting from the province where Turin, the capital city of the region, lies.

Turin has always been known as the city of cars and has always had a strong connection to that type of industrial production, mainly based on iron compartments. Another important industry in the province is located in the Ivrea area, where a big Italian company (Olivetti) has developed manufacturing focused on electronics and informatics. The cluster of nanotechnology being developed in the province of Turin is a multifaceted enterprise system, from information and communication technologies to biotechnology, aerospace, and audiovisual as well as publishing, banking, and industrial design systems.

Most of the other provinces have industrial profiles, especially Alessandria, Biella, and Verbano-Cusio-Ossola (VCO). In Novara and Cuneo, a variety of activities include agriculture, which also plays an important role in Vercelli and plays an almost exclusive role in Asti. Several industrial sectors have been developed in the Alessandria province, including many specializing in the molding of plastics and chemicals, in heavy engineering, in the confectionery industry, in the production of Borsalino felt hats, and in the processing of precious metals, especially gold. The territory is also an important logistics hub, due to its proximity to the port of Genoa and thanks to its strategic location, which allows transport to the major cities of northern Italy. This allows for the better management of goods arriving and departing from the port, and thus, this is where transport and storage activities are located. With respect to the agricultural sector, there are large cereal plantations, in particular wheat and corn as well as paddy fields.

The Biella province is one of the world centers of textiles and wool and of a general textile machinery sector. The district specializes in the production of fabrics for men and women, of yarns for weaving and knitting, and in all auxiliary processes of the wool textile industry (combing, dyeing, finishing, etc.).

The area of Verbano-Cusio-Ossola boasts a long industrial tradition, both in field extraction (marble and red granite quarries) and manufacturing plants, which settled in that territory due to the good availability of hydropower. There is considerable production of cookware, tableware, and small appliances. In recent years, the energy sector has been characterized by strong production and employment growth, and it recorded an increase in total renewable energy production (e.g., hydroelectric energy, photovoltaic energy).

Because of its position between two of the most important Italian regions, the province of Novara is an important crossroads for commercial traffic along the roads connecting Milan to Turin and Genoa to Switzerland. The Centro Intermodale Merci is the most important center of railway marshalling in northeastern Piedmont, a crucial point for trains incoming from and departing to northern Europe (especially Belgium and the Netherlands). Rice cultivation is a massive presence in the southern part of the province. With the provinces of Vercelli and Lomellina (Lombardy), Novara forms the famous “Triangolo dei chicchi d’oro” (gold grain triangle), which produces about 50% of the total Italian rice yield. The second most widespread crop is corn, while the floriculture sector is also relevant. In the plains, breeding is mainly practiced and cows’ milk is used to produce typical gastronomic products. There are several industrial districts in this area, with companies representing Italian excellence not only in chemistry and metallurgy but also in precision engineering and industrial machinery, textiles, publishing and paper, and food industries.

In Cuneo, the agricultural sector plays an important role (nearly 30% of total business). However, a variety of manufacturing activities are active in the territory, from the textile industry to the confectionery industry and baked goods, to the publishing industry as well as a significant share of companies in machinery and material production.

In Vercelli, the tertiary sector employs the most workers. Agriculture retains significant relevance, in particular for rice production. In the secondary sector, the mechanical industry is characterized by high production specialization in taps and valves. The textile industry also has a strong industrial tradition and exports products most significantly. Some of the major business realities include the leading biomedical and biotech cluster in Piedmont, while chemicals are another major growth area.

Most of the territories of the province of Asti are classified as intermediate rural areas, often hilly, with permanent agriculture and specializing primarily in grape and, thus, in wine production. In this province, we mainly find agro-food systems that, thanks to remarkable brands of wine and food products, favor exports and a notable increase in tourism.

2.2 The hybrid environmental accounts and the insurance-based approach

This section describes the tools employed for the application. First, the NAMEA module allows the connection of economic accounts with environmental data (in our case, air emissions); since emission data present uncertainty, we then introduce an insurance-based method to check how far the estimate can change when uncertainty is taken into account.

2.2.1 Hybrid accounts applied at the local level

The NAMEA (Keuning 1993; de Haan and Keuning 1996; de Boo et al. 1993) is a statistical information system that combines national and environmental accounts in a single matrix; it is based on the input-output approach of Leontief (1970). In NAMEA, the economy is divided into sectors whose contributions to economic and environmental indicators are tracked. This is called a hybrid accounting system because economic aspects are represented in monetary units and environmental aspects in physical units. The fundamental idea of NAMEA is to supplement the conventional national accounting matrix with additional environmental accounts that record the emissions of pollutants. It is also possible to account for environmental themes, such as the greenhouse effect, ozone layer depletion, and acidification.

NAMEA is usually applied at a national level. However, a project (referred to as RAMEA or the Regional Accounting Matrix including Environmental Accounts) was launched to experiment with the application of such hybrid accounts at the regional level. In the RAMEA project, the regional NAMEA-type matrices were prepared for four European regions: Emilia-Romagna (ARPA Emilia-Romagna, lead partner), Noord-Brabant (TELOS, Brabant Centre for Sustainable Development), Małopolska (Mineral and Energy Economy Research Institute of the Polish Academy of Sciences), and South-East England (SEEDA, Environment Agency, Cambridge Econometrics).

Other than RAMEA, some additional experiments of NAMEA-type applications were further tested in Piedmont at regional, provincial, and municipal levels for air emissions and waste (La Notte and Dalmazzone 2012; Dalmazzone and La Notte 2013).

In this paper, we focus on regional and provincial accounts for air emissions in Piedmont. The environmental accounts of the Piedmont NAMEA-type table are compiled with the support of regional emission inventories provided by the Piedmont regional authorities, which follow the EMEP-CorinAIRFootnote 1 inventory. This inventory was compiled since the beginning of the 2000s (specifically for the years 2001, 2005, 2008, and 2010). The estimate procedure has been greatly improved and updated over the years, generating many releases. The regional inventory records data according to the Selected Nomenclature for Air Pollution (SNAP) classification, comprising 11 macro sectors, 75 sectors, and 430 activities for the following pollutants: methane (CH4), carbon monoxide (CO), carbon dioxide (CO2), nitrogen dioxide (N2O), ammonia (NH3), volatile organic compounds (VOC), oxides of nitrogen (NOx), sulfur dioxide (SO2), and particulate matters (PM10 and PM2.5).

The NAMEA-type accounting module allows one to frame together economic data and emissions, and it can be compiled at a local level. The first step is to harmonize the SNAP classification system, which is based on production processes, with the Nomenclature générale des Activités économiques dans les Communautés Européennes (NACE) classification system, which is based on economic sectors. Economic data (local units and number of employees) are drawn from Archivio Statistico delle Imprese Attive (ASIA, statistical inventory of active firms), while air emission data are taken from EMEP-CorinAIR (in tons for all pollutants except CO2, which is in 1000 t). ASIA is the register of active enterprises that integrates national administrative archives with other sectorial registers and with the Italian national STATistical institute (ISTAT) surveys. The aim of ASIA is to provide statistical information every year on the territorial distribution of economic activities and on employment. An ad hoc methodology, in fact, converts the administrative data into statistical information to estimate and validate the characteristics of the identified statistical units.

It is possible to connect air emissions to their generating activity by taking the EMEP 75 sector emissions and allocating them to the corresponding NACE subclasses; that is, SNAP sector-by-sector emissions can be aggregated within the six NACE code classes (the most detailed subclassification available) in order to guarantee as much as possible that all the available information is used in a consistent way. The hybrid account NAMEA, which does not include uncertainty, is discussed in Section 3 of this paper.

2.2.2 The insurance-based approach applied to hybrid local accounts

Approaches proposed to reduce uncertainty can be of different types and be used for different purposes (Tonin et al. 2016). For this application, a methodology based on the mechanism of insurance companies has been chosen. Marland et al. (2014) borrow an approach specifically from life insurance policies, as applied by insurance industries. The approach consists of the addition of a risk charge into their fees to cover the net present value of expected payouts; this represents a sort of insurance for the insurer. The risk charge can be obtained by multiplying the percentage of uncertainty for the price per ton of carbon, so that finally, the price of the released carbon and the risk charge can be added up. The risk charge per ton of carbon (or carbon equivalent) can be determined through this procedure and can be added to the central estimate to complete the estimation of the emissions. This approach follows in the footsteps of the adjustment of emissions to attain some level of confidence in reporting the emissions (Gillenwater et al. 2007; Jonas et al. 2010).

With an initial rough procedure, this logic can be applied in order to screen which pollutants require a deeper analysis; once critical pollutants and critical territorial contexts are pinpointed, more sophisticated approaches can be applied to provide the policymaker with a more robust assessment. However, we must be aware that with this approach, (i) we only address the issue of imprecise estimates and do not consider the issue of inaccuracy and (ii) we do not apply the risk charge concept to accounting emissions at some point in the future.

The risk charge formula applied for the initial screening is taken from the EMEP-CORINAIR methodology (EEA 2013, 2016). This formula is used here to calculate an initial rough estimate of emissions that include uncertainty:

$$ {EUnc}_{ijk}=\left({E}_{ijk}\times \left(1-{UNC}_{ijk}\right)\right)+{E}_{ijk}, $$
(1)

where Eijk is the emission for activity i, pollutant j, and fuel k; UNCijk is the uncertainty coefficient for activity i, pollutant j, and fuel k; and EUncijk is the emission estimate considering uncertainty for activity i, pollutant j, and fuel k. When the uncertainty coefficients are reported as 0%, the estimated value is doubled; when they are reported as 100%, they remain the same. This approach can provide interesting insights to assess the robustness of the inventory datasets. This formula was received from the emission inventory functionaries and was further discussed with them in order to ensure a consistent and coherent application. Of course, more sophisticated formulas could be applied; however, since this application is used for an initial screening, we accepted and applied what is currently used as common practice.

The total uncertainty coefficient for activity i, pollutant j, and fuel k can be calculated through the information available from the emission inventory, which provides uncertainty coefficients for emission factors (EF) and for activity data (AD) at the most refined level (430 activities). It is thus possible to process initial estimates by adding the uncertainty for each activity in terms of the emissions factor and the activity data and then combining them (EEA 2013, 2016). Here we follow uncertainty only proportionally by means of:

$$ {UNC}_{ijk}={UnEF}_{ijk}\times {UnAD}_{ijk}, $$
(2)

where UnEFijk is the uncertainty coefficient assigned to EF for activity i, pollutant j, and fuel k; UnADijk is the uncertainty coefficient assigned to AD for activity i, pollutant j, and fuel k; and UNCijk is the total uncertainty coefficient for activity i, pollutant j, and fuel k. Emission inventories are usually injected into any assessment (from monitoring to policy analysis) as known systems. The use of Eq. (2) breaks this assumption; the emission estimates are not taken as an a priori accepted system. Here we follow the footsteps of other authors. For instance, Jonas and Nilsson (2007) have also questioned the underlying systems perspective by applying uncertainty classes to emission inventories.

In this specific application, the coefficients have been used as follows: the minimum level is the emission estimate as reported in the official inventory, and the maximum level is the outcome of Eqs. (1) and (2) with no range of uncertainty applied (e.g., through the application of the Monte Carlo method, as suggested in the IPCC guidelines). As such, the approach undertaken shows the worst situation that could ever take place; all the inventory data (except those with an uncertainty coefficient equal to 1) underestimate the emissions at the highest value of their uncertainty coefficient. From the worst scenario, the most striking situations can emerge and catch the attention.

The uncertainty coefficients for emission factors and activity data depend on how estimates are measured and processed. Details concerning the meaning of the uncertainty coefficients UnEF and UnAD in the Piedmont inventory are explained in Table 1.

Table 1 Uncertainty coefficients assigned to emission factors and activity data
Full size table

The adopted procedure considers the calculation of uncertainty-modified estimates of the 10 emissions for EF and for AD, for each of the 430 inventory activities. We then aggregate the 8600 records into sectors and households, ensuring that there is consistency with the NACE classification system. The NAMEA-type table is built by entering ASIA data on the economic side and inventory emission data on the environmental side, according to the NACE classification.

2.3 The environmental themes

In order to assess the potential aggregated impact of a number of different air emissions within the same environmental theme, it is possible to combine several air emissions into one indicator by introducing conversion factors. We will consider three environmental themes, each corresponding to a selected indicator (Eurostat 2015).

The first theme concerns the greenhouse effect, the atmospheric-climatic phenomenon caused by emitting CO2, CH4, and N2O, which contribute to global warming. These gases have specific thermal potentials that we can quantify as coefficients to convert them into CO2 equivalents. We use the global warming potentials (GWPs) of gases to calculate the total emissions in a CO2 equivalent—1 for CO2, 21 for CH4, and 310 for N2O (ISTAT 2009; Eurostat 2015). The time period usually used for GWPs is 100 years.

The second theme concerns the phenomenon of acid rain, or the fallout from the atmosphere of acid particles, acid molecules diffused into the atmosphere that are captured and deposited on the ground by precipitation. The issue of acidification (ACID) is mainly caused by gas emissions such as NOx, oxides of sulfur (SOx), and NH3. In order to aggregate the emissions of these pollutants, we multiply each gas for its potential acid equivalent (PAE), which is 1/46 for NOx, 1/32 for SO2, and 1/17 for NH3 (ISTAT 2009Footnote 2).

The third theme concerns high tropospheric ozone concentrations that are harmful to human health and ecosystems. The main atmospheric emissions that contribute to the phenomenon are CH4, NOx, VOCs, and CO. These emissions have their own coefficients to determine the tropospheric ozone formation potential (TOFP). The coefficients to be used are 0.014 for CH4, 1.22 for NOx, 1 for VOCs, and 0.11 for CO (ISTAT 2009; Eurostat 2015). The results are discussed in Section 3.1.

2.4 The shift-share analysis

The shift-share analysis is a tool commonly used in regional science to determine whether a regional economy has competitive advantages over the larger (national) economy. The process considers the change over time of an economic variable (e.g., value added, employment) within economic sectors and then divides that change into various components. In our case study, we consider the regional level as the larger economy and the provincial level as the target, to identify competitive advantages. The shift-share analysis, combined with the hybrid accounts at the regional and provincial levels, allows an explanation of the reasons for different economic performances throughout the territory as well as an identification of a good production mix between specific emissions and economic/social efficiency by decoupling economic development from the increase in emissions. We can in fact apply decomposition analysis in order to investigate the mechanism that affects air emissions, splitting an entity into its components. Changes in some variables are decomposed in variations in its determinants. The methodologies commonly used to decompose emissions trends are index decomposition analyses, input-output structural decomposition analysis, and shift-share analysis. The latter has already been applied to compare regional and national data (Esteban 2000; Mazzanti et al. 2007) and regional–provincial and provincial–municipal data (La Notte and Dalmazzone 2012).

The purpose of this application is to measure the role of the productive structure at the lower hierarchical level considered (in our case, the provincial level) in explaining the emissions efficiency gap between this level and the higher hierarchical level (in our case, the regional level). Shift-share analysis, in fact, decomposes the source of change of the specified “dependent variable” into provincial-specific components (which constitutes the shift) and the portion that follows regional growth trends (which constitutes the share).

We aim to address the question of whether the gap between the considered province and the regional benchmark average depends on environmental-friendly technologies (or the lack of them) in the included economic sectors and/or on a provincial specialization in sectors with higher/lower eco-efficiency.

We first calculate the intensity of the emissions by considering each pollutant with reference to the number of workers employed in each sector. This variable provides insights into the socioenvironmental efficiency of the productive sectors, which is useful in planning a strategy to support environmental innovation at the sector level. We then analyze the relative environmental efficiency of the provincial economic system with respect to the regional average, referring to the GHG pollutants and to the economic sectors included in the hybrid accounts.

The aggregate indicator of emission intensity is represented by “total emissions [E] on number of employees [Empl].” The benchmark is represented by the regional value. This indicator is decomposed as the product of (Es/Empls) × (Empls/Empl), where (Empls/Empl) is the share of sector Empl on the total Empl for all sectors s, with the value of s defined from 1 to h (where h is the number of NACE sectors included in the regional NAMEA).

We define the index of emission intensity as X for the regional average (X = E/Empl); as XPr for the province, \( {X}_{Pr}=\frac{E_{Pr}}{Empl_{Pr}}; \) and as Xs for each sector for both the province, \( {X}_{Pr}^s={E}_{Pr}^s/{Empl}_{Pr}^s \), and for the region, Xs = Es/Empls. We then define the share of total employment as Ps = Empls/Empl for the region and as \( {P}_{Pr}^s={Empl}_{Pr}^s/{Empl}_{Pr} \) for the province

$$ X={\sum}_s{P}^s{X}^s $$
(3)
$$ {X}_{Pr}={\sum}_s{P}_{Pr}^s{X}_{Pr}^s $$
(4)

The shift-share decomposition allows the identification of three effects that explain the gaps, in terms of the aggregate emissions efficiency, between the province and the region.

The first effect (“structural” or industry mix) is given by:

$$ {m}_{Pr}={\sum}_s\left({P}_{Pr}^s-{P}^s\right){X}^s, $$
(5)

where mPr assumes a positive or negative value, respectively, if the region is “specialized” in sectors associated with lower or higher environmental efficiency, given that the gap in total shares is multiplied by the value of X of the regional average (“as if” the province were characterized by the average regional efficiency). The factor mPr assumes lower values if the province is specialized in (on average) more efficient sectors.

The second factor (“differential” or “efficiency”) is given by:

$$ {p}_{Pr}={\sum}_s\left({X}_{Pr}^s-{X}^s\right), $$
(6)

where PPr assumes a positive or negative value, respectively, if the region is less or more efficient in terms of emissions (the shift between provincial and regional efficiency), under the (“as if”) assumption that the share of the number of sector employees were the same for the region and the province.

The third effect (“allocative component”) is given by:

$$ {a}_{Pr}={\sum}_s\left({X}_{Pr}^s-{X}^s\right)\left({P}_{Pr}^s-{P}^s\right). $$
(7)

The aPr factor is positive or negative, respectively, if the province is specialized, relative to the regional benchmark, in sectors characterized by a higher or lower emission intensity.

The outcomes of the shift-share analysis are shown for each province as Supplementary Material and are discussed in Section 3.2.

3 Results

3.1 The accounting module

Table 2 presents the hybrid accounts for the Piedmont region. All economic sectors except the primary sector (i.e., agriculture, forestry, and fishing) are included. In order to compile the economic module for the account and avoid resorting to the economic territory and residence principle,Footnote 3 we use the ASIA database, which does not include the primary sector. By using this database, we ensure consistency with emission inventory data, but we cannot include either the economic or the environmental side with respect to the primary sector. When the uncertainty maximum limit reported in Table 1 is applied to the estimates of Table 2, we obtain Table 3.

Table 2 Hybrid accounts for the Piedmont region (2010)
Full size table
Table 3 Hybrid accounts for the Piedmont region including uncertainty according to the insurance approach (2010)
Full size table

Hybrid accounting tables with and without uncertainty have been compiled for each individual Piedmont province to enable both the assessment of environmental themes and the shift-share analysis.

3.2 The environmental themes

In our application, we first calculate the three environmental themes for the hybrid accounting table without uncertainty; we then calculate the same indicators for the hybrid accounting table with uncertainty (Table 4). Two elements deserve attention: (i) the total amount of pollutants that affect the environmental theme and make one province more environmentally vulnerable than others and (ii) the role played by uncertainty in modifying the estimates. The former is relevant for policymakers, because in environmentally vulnerable provinces, it is important to have reliable and accurate estimates; the latter is relevant because uncertainty plays a major role in processing the final estimates. Each of the two cases justifies a detailed uncertainty assessment.

Table 4 Environmental themes without (w/o) and with (w) uncertainty
Full size table

The immediate way to interpret the results of the environmental themes is to consider the difference between the indicators calculated without uncertainty (w/o Unc) and those calculated with uncertainty (w Unc) in order to check the robustness of the inventory data. The role played by uncertainty can, in fact, affect the strategic analysis in setting environmental priorities.

Torino has the highest value related to GWP; this remains the highest value even when uncertainty is taken into account. However, Fig. 1 clearly shows how uncertainty increases the estimates in all provinces, with peaks in Biella and Novara. This can be explained by the fact that pollutants such as CO2 and CH4 have a high uncertainty in SNAP 03 and SNAP 04, corresponding to the NACE manufacturing activities (in particular in the textile sector) based in Biella, as well as in SNAP 07, which concerns Novara as a crossroads for commercial traffic. Inventory data for Torino, Biella, and Novara are not robust enough to ensure that a correct environmental perspective on GWP is adopted. This screening phase provides a signal that uncertainty should be considered and further processed.

Fig. 1
figure 1

GWP with and without uncertainty in Piedmont provinces (2010)

Full size image

Figure 2 shows the different estimates obtained for acidification when uncertainty is added. Once again, the province of Torino registers the highest estimates, both with and without uncertainty. In provinces such as Novara, Asti, and Biella, the increase of the estimate added by uncertainty is larger than in the other provinces; this can be explained by the fact that the uncertainty in NOx and SO2 emissions is very high in the manufacturing activities in which these three provinces specialize. Inventory data for the four provinces need to be further processed by taking uncertainty into account, using more sophisticated techniques.

Fig. 2
figure 2

Acidification with and without uncertainty in Piedmont provinces (2010)

Full size image

Finally, the estimates for TOFP are presented in Fig. 3. Novara and Biella are once again among the provinces where uncertainty increases the most. In addition to CH4 and NOx, the uncertainty of VOC is especially high for SNAP 03 (i.e., combustion in manufacturing industries), as that of CO is for SNAP 04-06 (i.e., production processes, extraction and distribution of fossil fuels and geothermal energy, solvents, and other product usage); this is explained by the presence of a variety of manufacturing activities (from food transformation to machinery and material production). Considering the importance of Novara as a crossroads for commercial traffic, uncertainty due to pollutants connected to transport activities (SNAP 07) plays a major role. Once again, it would be advisable to consider uncertainty when extracting inventory data for Novara, Biella, and Torino; this initial screening demonstrates that a robustness issue could occur.

Fig. 3
figure 3

Tropospheric ozone formation potential (tons/year) with and without uncertainty in Piedmont provinces (2010)

Full size image

The same message expressed in Table 4 and shown in Figs. 1, 2, and 3 is further illustrated in Fig. 4, where the percentage increase in data with uncertainty (compared to the data without uncertainty) is simultaneously presented for the eight provinces. For the three indicators, Novara and Biella show an increase close to and greater than 0.80, while in the case of Torino, the range is between 0.50 and 0.60.

Fig. 4
figure 4

Estimate increase with and without uncertainty for the three environmental themes in Piedmont provinces (2010)

Full size image

The analysis of estimates with and without uncertainty for environmental themes highlights how the difference between the results from analyzing emission data with and without uncertainty depends to a great extent on the pollutants emitted by specific economic activities. The same typology of pollutant can have a different uncertainty coefficient for different economic activities; thus, it is important to analyze the economic sectors together with the emissions data. In fact, taking into account the prevailing economic activities in a region, a province, or a municipality helps one understand the extent to which uncertainty might impact estimates provided to the final decision maker.

Finally, uncertainty might matter because of the high weight of emissions in a particular part of the territory compared to other parts (as in the case of Torino) or because of the specialization in production activities where the uncertainty coefficient is higher (as is the case mainly in Biella and Novara). Considering the environmental themes, in-depth analyses are be needed for these three provinces, though such analysis might require a further processing of inventory data, for example, by considering medians of the min-max/uncertainty intervals or by applying the Monte Carlo technique (as done in La Notte et al. 2018).

3.3 The shift-share analysis

Environmental themes consider the calculation of uncertainty in absolute values. By using shift-share analysis, we consider relative values by relating emissions to the number of employees in each sector. Here we consider two cases with two different levels of aggregation. Considering the outcomes for the previous section, the provinces of Biella and Torino are presented as showcases.

As a first case, in a previous application (La Notte et al. 2015), different groups of manufacturing activities were agglomerated from the secondary sector, and only transport activity was taken into account from the tertiary sector. The two summary tables provide information about the following:

  • Industry mix: a negative value indicates that in the Biella province, those sectors that employ more workers pollute less; the negative sign occurs for all pollutants except NH3

  • Efficiency (productivity differential): a negative value indicates that in the Biella province, economic activities that employ more workers pollute less than at the regional level; however, this does not happen for CH4, CO, NH3, or PM, meaning that there is room for improvement in eco-efficiency technology

  • Allocative component: a negative value indicates that the Biella province specializes in economic activities that pollute less, the negative sign occurs only for CH4 and PM

It is unusual that the allocative component records a different sign compared to the industry mix and the differential component. This component indicates whether and to what extent the system productively specializes in the fields where it carries out a comparative advantage of efficiency. Despite its efficiency and sector structure, Biella does not productively specialize in what concerns the sectors that mostly emit CO2, COV, N2O, and SO2.

Table 5 shows a different policy message, whether uncertainty is considered or not, especially for what concerns the “efficiency.” Improvement in ecological efficiency is also needed for NOx and SO2. On the other hand, in terms of the allocative component, NOx turns out to be neutral.

Table 5 Shift-share summary table for the province of Biella (sectorial aggregation)
Full size table

The same kind of analysis can be applied for the province of Torino (Table 6). The summary table provides information about:

  • Industry mix: a negative value indicates that in the Torino province, those sectors that employ more workers pollute less; the negative sign only occurs for COV, N2O, and NH3

  • Efficiency: a negative value indicates that in the Torino province, economic activities that employ more workers pollute less than at the regional level; the negative sign occurs for all pollutants

  • Allocative component: a negative value indicates that the Torino province specializes in economic activities that pollute less; the negative sign occurs for CH4, CO, CO2, NOx, and PM10

Table 6 Shift-share summary table for the province of Torino (sectorial aggregation)
Full size table

When taking uncertainty into account, (i) some pollutants that were positive (SO2 in the industry mix) without uncertainty can become neutral with uncertainty and (ii) some pollutants that were neutral without uncertainty can become positive (NH3 in the allocative component) or negative (PM2,5 in the allocative component) with uncertainty.

As the second case, in the current application, we have considered every single economic activity from the secondary and the tertiary sectors to avoid the possibility that selecting only a few economic sectors and aggregating different economic sectors in different ways might affect the final result.

The counter-intuitive case of the allocative component exhibiting a sign which does not follow the logic of combining the signs of both industry mix and efficiency can be explained as follows. Let us consider, for example, the allocative effect of CO in the Province of Torino (Table 8). There are sectors, such as food manufacture (10), other nonmetallic mineral products (23), basic metals (24), and air transport (51), where the structural component (the difference between the provincial level and the regional level) records a negative sign. The emission differential effect in these sectors is very high (negative), and this combination produces a multiplying effect that finally generates a positive outcome. When calculating the industry mix, the negative structural component of sectors 10, 23, and 24 is overcome by the positive number of other sectors, such as transport (49) and electricity (35).

By looking at Tables 7 and 8, we can track differences with and without uncertainty on the numerical values processed but not differences on the sign; what is recorded as negative remains negative, and what is recorded as positive remains positive. A sector decomposition performed in this case study actually confirms that the differences generated by previous applications were indeed the outcome of how the sectors were selected and combined. In order to understand why, in relative values, the difference in estimates with and without uncertainty does not emerge, we must look at another source of information in the Supplementary Material (S1), that is, the economic structure of each province and the relationship of that structure with the economic structure of the region as a whole. In doing this, we see percentages that range from a maximum of 3–4% (with a few exceptions; e.g., Biella records 17% for the textile industry) to an average of 0.0002% for most provinces, across all secondary and tertiary sectors. The fact that there is no remarkable difference among the provinces is exactly what affects the results with respect to relative values.

Table 7 Shift-share summary table for the province of Biella (sectorial disaggregation)
Full size table
Table 8 Shift-share summary table for the province of Torino (sectorial disaggregation)
Full size table

This happens because we have not included the primary sector in the distribution of weights. If we could do so, the distribution in terms of employees and value added would considerably change the structural shape of provinces such as Asti, Alessandria, Cuneo, Novara, and Vercelli. The regional structure, which is heavily affected by the province of Turin, will differ from the provincial structure, where agriculture plays an important role. In principle, a tool such as shift-share analysis could be extremely useful for this kind of application; however, we want to highlight the fact that the way it is applied does indeed matter. In our case study, the outcomes are misleading, because the entire productive structure is not fully represented. When relative values are used, it is thus important not only to choose the tool properly but, most of all, to use it properly. When properly used, the robustness of inventory data in shift-share analysis may play an important role in providing policymakers with the right signals in term of actions to be undertaken. A determination of whether the main cause of emissions (e.g., CO2 emission for the climate agendas of mayors or PM10 and PM2.5 emissions for the health of citizens) lies in the production structure of the provincial economy or lies in the lack of eco-technologies will lead to very different policies.

4 Discussion and conclusions

Taking uncertainty into account could have some effects on policy use, not only when looking at compliance rules and emission trading on the national scale but also when looking at local policy action at regional and provincial levels. But taking uncertainty into account can be time- and resource-consuming, especially for small administrations. This paper shows how to screen whether and in which cases the assessment of uncertainty should be undertaken, because such assessment generates different outcomes for policymaker decisions. Hybrid accounting tables were built considering data drawn from the emission inventory at regional and provincial levels, and uncertainty was added throughout the approach, borrowed from the insurance sector. Once the accounting tables (without and with uncertainty) were available, two sets of policy tools were applied: environmental profiles through environmental indicators, which considered the absolute values of the emissions reported in regional-NAMEA, and the shift-share analysis, which related emissions values to socioeconomic values. The former set of tools showed a remarkable difference between data without uncertainty and data with uncertainty, while the latter tool did not show any difference.

The results obtained with this application allowed a few principles to be generalized. First, the consideration of uncertainty makes a difference in one’s choosing between tools that provide results either in absolute terms (such as the indicators related to the environmental themes) or in relative terms (such as a shift-share analysis). The former tools do indeed record a quantitative difference that can be quite large, while the latter tool might reduce the difference. If not properly used, this tool could also nullify the differences, as happened in our application. Users must be aware of the different peculiarity of each tool and must interpret the outcomes with caution. There is a risk of underestimating the role of uncertainty in data processing and thus of not highlighting the need for an in-depth analysis.

Another important principle to consider is the importance of linking the economic side with the environmental side, since consideration of the size and economic meaning of each sector provides crucial input in understanding how significant it could be to include uncertainty in the estimate of the data; each pollutant can result in a different uncertainty, according to different production processes, and thus can produce a different result with respect to economic activity. Different economic activities can also play diverse roles in specific territorial contexts. The more disaggregated the focus of the analysis is, the greater the worth of the economic and environmental data at the local level is and the more reliable the information.

For example, when we consider the region as a whole, we record the strong footprint of the province of Turin in terms of population and economic activities and thus of the emitted pollutants. Yet the peculiarities of provinces such as Biella, specializing in the textile sector, or Verbania, specializing in the extractive industry and energy, disappear. The possibility of specifically linking certain pollutants to those economic activities arises through analysis at the provincial level. And this linkage becomes much more complicated at the regional level, unless regional estimates result from the aggregation of provincial estimates.

In methodological terms, an insurance-based approach applied as an initial screening tool can be very useful in selecting where and how to concentrate the attention when further data refinements are required in terms of uncertainty. This can be the case particularly when budget and time constraints might affect the possibility of performing a complete analysis for all pollutants and all territorial contexts.

The same insurance-based approach can be refined for activities and pollutants that need to be further analyzed. Adding the full uncertainty amount calculated through the uncertainty coefficient systematically overestimates the impact of uncertainty on current estimates; to avoid this systematic overestimation, it is possible to apply Monte Carlo techniques with appropriate distribution in order to calculate a likely estimate of the emitted pollutants. Once a more sophisticated procedure has developed, the insurance-based approach could be applied at the national level (i) by applying the procedure directly to the national air emission inventory and (ii) by applying the procedure to the regional inventories and then aggregating them. The latter solution can be more difficult and time-consuming, but it would leave open the possibility of accessing local scale details that would otherwise be missed on a national scale.