Mapping and Tackling Energy Poverty Vulnerability at the Local Scale—A Methodology for Sustainable Urban Planning

Abstract

Energy Poverty Vulnerability is a multidimensional phenomenon, which depends both on the energy efficiency of dwellings and on the related socio-economic aspects. Its diagnosis at the intra-urban scale has been addressed by few studies; therefore, its spatial distribution is still difficult to assess, as well as poorly integrated into urban planning processes. The research addresses these gaps by proposing a methodology to assess and map Vulnerability to Energy Poverty (V_ep) at this scale, using open and institutional data that can be integrated into Italian urban planning processes. Based on the state of the art, a set of indicators is selected to construct the Local Index of Vulnerability to Energy Poverty (LIV_ep). The workflow includes an assessment of the applicability of Principal Component Analysis (PCA) and the mapping of the results through Geographic Information System (GIS). The methodology is applied to the municipality of Milan, Italy. PCA identified four Principal Components of V_ep, which were mapped in a GIS environment. The first two components, interpreted as Socio-Economic Vulnerability (PC1) and Energy Vulnerability (PC2), were subsequently combined into LIV_ep, providing a single planning-oriented index. A sensitivity analysis, based on an alternative equal-weighted aggregation of the normalized indicators, showed strong consistency in the overall ranking of census sections, although some differences emerged in identifying the most vulnerable areas. The main contribution of the study concerns a spatial decision-support framework, based on open data and GIS, for building a local-scale knowledge base of V_ep, supporting the spatial targeting of actions to tackle V_ep, and fostering the integration of energy, social, and urban regeneration policies. The proposed methodology is potentially transferable to geographical contexts other than Italy, provided that comparable census and energy-related data are available from institutional sources or through urban planning processes.

1. Introduction

According to the World Cities Report of the United Nations Human Settlements Programme, the number of cities globally doubled between 1975 and 2020, making the urban environment the living context of the majority of the world’s population and urban planning essential to face new urban challenges [1].

It is well known that the attraction of the population by cities is historically attributable to the promise of more socially and economically advantageous living conditions compared to settlements inserted in rural contexts, which have become increasingly marginal [2]. On the other hand, the concentration of human activities in cities has made the latter progressively energy-intensive organisms, in which the increase in climate-changing emissions and local temperature seriously threatens the quality of life and, more generally, urban sustainability [3,4].

This phenomenon, in addition to undermining the competitiveness of urban areas, hinders progress towards the goal of carbon neutrality needed to combat climate change, whose achievement is increasingly advocated by major international organizations [5,6,7].

In Europe, the recent Climate Law, enacted as part of the Green Deal, makes this target legally binding by 2050 [8].

Furthermore, by 2050, the European Commission recommends zero net land consumption to Member States, calling on spatial planning to move more and more towards urban regeneration and densification, rather than encouraging further expansionary dynamics of cities, especially in the widespread form of sprawl [9].

According to Eurostat, in 2022 25.8% of final energy consumption in Europe comes from homes, in particular for heating (63.5%), followed by domestic hot water production (14.9%), lighting (13.9%), cooking food (6.3%), cooling (0.6%) and other uses (0.9%).

In this scenario, increasing the energy efficiency of existing residential buildings brings numerous benefits, not only for the reduction in climate-changing emissions in urban areas, but also for the fight against Energy Poverty (EP).

In fact, the poor energy performance of the current residential building stock is considered one of the three main causes of EP, according to the definition given to the latter by the European Commission, together with the low income of residents and the high cost of energy [10].

EP appears to be dramatically increasing in Europe, first as a result of the general impoverishment caused by the COVID-19 pandemic [11], then due to the Russia-Ukraine war crisis, which has led to turbulence in energy markets, price volatility and energy insecurity. As highlighted in the State of the Energy Union 2022 report, the phenomenon of EP no longer affects only low-income households, “but also those with lower and middle income and potentially others in some Member States as well”, putting at risk the achievement of ambitious European decarbonisation targets [12].

In 2023, 10.6% of the EU population said they were unable to heat their homes adequately. Compared to 2022, this share increased by 1.3 percentage points, in a context of energy crisis and inflation [13]. In addition, this new situation threatens sustainable energy planning in general [14].

Dubois and Meier suggest that it is not possible to adopt a single approach to combating EP in all Member States, as some suffer more than others from the problem of deprivation of energy services, with potentially greater difficulties in financing and effectively implementing policies against EP. However, they highlight that the most effective measure concerns the energy efficiency of homes, which represents the only permanent remedy for the lack of energy services [15].

This logic is consistent with the principle of ‘Energy Efficiency First’, recently promoted by the latest update of the European Energy Efficiency Directive [16].

Over the years, scholars and international design companies have been highlighting how it is possible to accelerate the energy transition of cities by intervening on entire neighborhoods or urban districts, rather than on individual buildings, in order to increase the impact of the actions taken [17,18].

Furthermore, if an effective strategy is to intervene at the district scale, acting primarily on the most problematic urban areas in terms of energy consumption, it is equally important to assess the socio-economic characteristics of residents and their possible interconnections with the poor energy performance of homes. If only the energy efficiency of buildings is taken into account, neglecting the related social aspects, or vice versa, there is a risk of acting incompletely, not adequately addressing all the components that characterize the EP phenomenon [19].

In Italy, according to the National Information System of Energy Performance Certificates (SIAPE), based on data collected from over seven million certificates, more than 70% of residential buildings fall into the worst classes (E, F and G), a percentage that drops to about 52% for non-residential use [20].

EP is estimated to be twice the European average already in 2021, as reported by the Italian National Institute of Statistics (ISTAT) [21].

Against this background, and considering the need to limit land take, the Italian government has long promoted energy efficiency interventions in the existing residential building stock through tax incentives and deductions, such as the so-called ‘Ecobonus’. These measures received a significant boost with Decree-Law No. 34/2020, which introduced the financial instrument known as the ‘Superbonus 110%’. The latter was also indirectly aimed at alleviating EP among Italian households; according to an Italian consulting company, households benefiting from the measure achieved energy bill savings of 30.9% when their dwellings improved by two energy classes [22].

However, if not accompanied by local general and energy planning, these policies tend to result in mostly isolated interventions on individual buildings, more frequently single-family houses, due to the greater technical and managerial obstacles associated with multi-family buildings [23].

Although beneficial, isolated interventions are insufficient to address a problem that can affect large parts of a city and should therefore be regarded as a collective issue, requiring a response from local planning.

It should be emphasized that these measures have been implemented within a national regulatory framework that lacks structural policies on urban regeneration, which provides the broader context within which such actions could be integrated, following an integrated approach that also contributes to limiting further land take [24].

In this direction, territorializing incentives and public investments, i.e., targeting them towards areas with the highest levels of EP, where their greater effectiveness is expected, requires spatial knowledge of the geography of EP. Such knowledge can be developed and made available through municipal urban planning processes.

The latter can also lead to the drafting of urban plans—general and integrated energy plans—that provide a support framework at the local level, promoting the energy efficiency of the residential building stock and the reduction in EP in actions, rules and regulations aimed more generally at promoting urban regeneration.

In fact, local authorities play a central role in the fight against EP, thanks to their greater knowledge of territorial needs and the ability to implement targeted interventions integrated with national policies [25].

All planning actions aimed at promoting initiatives such as Renewable Energy Communities (RECs)—particularly those with a solidarity-oriented approach, such as Renewable and Supportive Energy Communities (RSECs)—represent a recent and promising trend, which can be regarded either as an autonomous approach or as one integrated with urban regeneration interventions. Facilitating the development of RECs, among other benefits, can enable households experiencing EP to gain direct or indirect access to clean energy [26,27,28,29,30,31].

2. Literature Review and Aim

Since the topic of EP was introduced in the UK some thirty years ago, it has gradually attracted scholars and policymakers in Europe and beyond [32].

Recently, the EU’s Energy Poverty Advisory Hub (EPAH) developed a handbook on understanding and how to deal with EP. This manual is structured in three parts, corresponding to three main phases of work, considered effective for local governments to tackle EP:

(1)

Assessment of energy poverty at an identified local level—‘Diagnosis’ phase;

(2)

Development of an informed plan—‘Planning’ phase;

(3)

Execution of an impactful energy poverty project—‘Implementation’ phase [19].

In this framework, for each of the three phases, EPAH’s ‘Atlas’ platform compiles information on ongoing initiatives, including national and international projects, policies, and measures addressing EP at the international, national, regional and local levels [33].

As regards the diagnosis phase, the manual confirms the ‘causes’ of energy poverty previously identified by the EU and mentioned above:

-

Low income levels;

-

Low household energy efficiency and energy performance of buildings;

-

High energy prices.

In addition, the manual identifies a non-exhaustive list of ‘vulnerability factors’, i.e., those factors that can make residents more vulnerable to EP: socio-demographic factors such as rented residents, people with low levels of education, and ethnic minorities; factors related to the composition of the family unit, such as single-parent families, and people with disabilities; health factors, such as people with illnesses related to indoor thermal conditions; factors related to energy education, such as lack of awareness of the problem and solutions to address it; cultural factors, related to residents’ behavioral habits [19].

In general, scholars agree that EP is a multidimensional phenomenon, therefore not quantifiable by a single variable but rather through a set of indicators/indices. There is still a debate in the scientific community about what the EP components are, the appropriate methods to measure it and read its geographies at the territorial level [34].

However, the most recent studies on the subject can be classified mainly according to the spatial scale investigated, namely national (I), municipal (II), and infra-municipal or local (III). Most studies concern the national scale [35,36,37,38,39,40,41], while few studies address the issue at the municipal scale [42,43,44], and only a limited number focus on the local level [45,46,47,48,49].

2.1. Measuring and Mapping Energy Poverty

Leaving aside the national scale (I), which is beyond the scope of the research, at the municipal level (II) the study by Gouveia et al. proposes a composite index of vulnerability to EP, organized into two sub-indices, referring to the energy performance of residential buildings and the ability of the population to implement measures to combat EP. The first sub-index is derived from indicators relating to both energy performance and final energy consumption. The second sub-index is constructed by aggregating indicators related to unemployment, housing tenure, education, monthly income, population age, and the state of conservation of buildings [42].

With particular reference to the local scale (III), which is the focus of this paper, Robinson et al. use a multidimensional index of vulnerability to EP, starting from socioeconomic, demographic, housing and energy indicators. Since the context of the study is England, the data useful for measuring these indicators are mainly derived from the 2011 Census. The spatial unit of mapping is the Lower Super Output Area, which is broadly comparable to the neighborhood scale. The composite index is obtained using Principal Component Analysis (PCA), to identify the determining vulnerability factors [45]. This work is taken as a reference by Zardo et al., who apply the same method to Italian metropolitan cities, albeit on a municipal scale [44].

Castaño-Rosa et al. propose an Index of Vulnerable Homes, developed using four sub-indices: monetary poverty, an indicator that assesses the monetary vulnerability of the household based on household net income; energy vulnerability, estimated according to the median energy consumption required for the type of dwelling, established on the basis of the Typology Approach for Building Stock Energy Assessment (TABULA) project; thermal comfort, using the percentage of hours in which living spaces are outside the set comfort range; associated health costs, defined by attributing a range of monetary values to the Quality-Adjusted Life Year. The model was tested in a city centre area of the Borough of Salford, UK. Although the reference to urban planning is not explicit, the work intends to support local policy makers to direct energy renovation interventions. The proposed standard renovation intervention consists of the installation of a combination of energy efficiency measures: thermal insulation of the walls, insulation of the attic and double-glazed windows. The authors highlight that low-income households may still struggle to pay their energy bills, despite improving the energy efficiency of their homes [46].

Recent research, conducted by Clavijo-Núñez et al., proposes a methodology designed to support targeted and participatory regeneration interventions in vulnerable neighborhoods, applied to neighborhoods A and B of Polígono San Pablo, in the city of Seville, Spain. To measure energy vulnerability, an analysis extended to the entire city, the study uses indicators derived from both statistical data (building age, presence of an elevator, building condition, electricity consumption, heating, occupancy, income) and questionnaires administered to residents of some of the most critical neighborhoods (residents’ state of health, obsolete domestic elements). Also in this study it is suggested to pay attention to gender-related issues. However, the authors highlight that the indicators of vulnerability to EP proposed at the European level are difficult to measure at the local scale, due to the lack or outdated nature of publicly available data in Spain [47].

Italian research conducted in Bologna, in the Emilia-Romagna Region, proposes the mapping of EP using an index of vulnerability to EP, obtained from two sub-indices, relating to energy and socio-economic conditions of the population. The sub-indices are aggregated using weights, assigned to each variable according to its importance with respect to the phenomenon, based on the Budget Allocation Process. The authors note that, since a weight must be assigned to each variable, the number of variables should not exceed 10–12. The explicit reference to urban planning is interesting, as the authors suggest incorporating the resulting vulnerability maps into both the General Urban Plan (GUP) and the Sustainable Energy and Climate Action Plan (SECAP) of Bologna Municipality, in order to spatially target and refine strategies to combat energy poverty [48].

More recently, Calama-González et al. assessed energy vulnerability at the local scale in southern Spain, using an open access tool based on GIS. The methodology integrates public and open source data to estimate both the energy performance of the residential building stock and the socio-economic vulnerability of its occupants. Also in this study, the two dimensions are combined in a composite vulnerability index. The method has been tested in Seville on a dwelling scale, but an extension of the model on an urban scale is proposed by the authors, with application to several municipalities in Andalusia [49].

2.2. Current Challenges and Objectives of the Work

The analysis of the state of the art in the field of EP shows that this phenomenon is mostly analyzed on a national scale, while studies exploring the diagnosis of EP at the infra-urban scale are still limited.

In these studies, EP is interpreted in terms of the socio-spatial vulnerability of households, potentially exposed to the risk of experiencing the phenomenon, rather than in absolute terms, based on the number of energy-poor households.

Unlike studies on a national or sub-national scale, which use the municipal base as the minimum spatial unit for the measurement of EP indicators and the consequent mapping, at the local level the spatialization of the chosen variables takes place by census sections or neighborhoods. In fact, census sections generally represent the most detailed spatial level at which the relevant open data—data required to measure socio-economic indicators or indicators related to the characteristics of buildings and homes—are available from official national statistical institutes.

The choice of variables is not straightforward, but depends first of all on the definition of EP adopted, which may be different on a case-by-case basis, and secondly on the availability of data necessary to measure the indicators at the spatial level investigated and for the territorial context under consideration. The transferability of the indicators proposed in previous works to the Italian geographical context is limited by the lack of open data, normally accessible in municipal planning processes, to measure them. Although the data provided by the statistical offices are similar, they are not fully comparable. Therefore, a first research question (RQ1) is: which indicators can provide a multidimensional diagnosis of EP vulnerability at the intra-urban scale using open and institutional data available for urban planning?

The approaches adopted in the cited studies to construct composite indices include: simple arithmetic mean; weighted arithmetic mean; PCA; spatial autocorrelation techniques.

When it comes to measuring EP, it should be considered that indicators can be correlated with each other, with the risk of overestimating some dimensions of the phenomenon relative to others. For example, according to Franco et al., the energy consumption of buildings in Europe is significantly correlated with the following five key variables: living floor area, height, construction year, climatic conditions, building functionality [50]. Thus, if energy consumption is included together with these five additional variables, often mentioned in local-scale studies of EP, vulnerability is likely to be overestimated.

Similarly, in Italian metropolitan areas, household income is significantly correlated with low levels of education and high levels of unemployment [51,52]. In this framework, the PCA method may represent a suitable technique to limit redundancy of information when constructing composite EP indices. Conversely, if the statistical characteristics of the data do not support it, PCA may not be an appropriate technique [53]. Thus, a second research question (RQ2) is: which aggregation approach can provide a robust synthesis of the selected indicators while limiting information redundancy?

The planning implications of the constructed indices, which are consistently represented through thematic maps produced using Geographic Information System (GIS), are neglected. The maps obtained are discussed with reference to generic local policies to combat energy poverty, overlooking the importance of incorporating these policies into urban planning, both general and energy planning, mentioned in a single study [48]. So, a third research question (RQ3) is: how can the spatial diagnosis of EP vulnerability support the targeting and integration of energy, social, and urban regeneration actions within local planning processes?

Ultimately, what emerges from the state of the art is that mapping and tackling EP at a sub-municipal scale remain challenging. According to the identified research questions (RQs), the specific objectives of this work are, in order, the following:

-

propose an easily practicable methodology for diagnosing and mapping EP vulnerability within Italian urban planning processes, using data normally available to municipalities, and demonstrate its application to a case study;

-

discuss the implications of the results obtained for tackling EP vulnerability at local scale, with particular reference to urban planning.

To this end, Section 3 below describes the proposed methodology, which has been applied to the case study of Milan, Lombardy Region, Italy, as specified in Section 4, while Section 5 contains the results obtained from this application, and a sensitivity analysis aimed at verifying methodology robustness. Section 6 discusses the findings in relation to the literature reviewed in the current Section 2 and the RQs emerged. Section 7 concludes the paper by outlining its limitations and future research developments.

3. Materials and Methods

This paper proposes a methodology for the diagnosis and mapping of Energy Poverty Vulnerability (V_ep) at the local scale in the Italian geographical context, to support the choice of actions to tackle V_ep, to be implemented in urban planning.

Vulnerability, in accordance with the broader literature on territorial risk, is understood as systemic, i.e., as the set of characteristics of a system responsible for its overall response to one or more hazards [54]. In line with previous studies on EP, referred to in Section 2, the dwelling-occupant system is considered vulnerable with respect to two main issues: energy vulnerability, related to the dwelling energy performance and the demand for basic energy services; the ability of occupants to implement EP alleviation measures. The V_ep is evaluated with reference to the LIV_ep composite index, Local Index of Vulnerability to Energy Poverty (Figure 1).

Figure 1. Methodological scheme to construct LIV_ep, Local Index of Vulnerability to Energy Poverty.

This index is based on various indicators, selected by systematizing those proposed by the studies that address the phenomenon of EP vulnerability at the local level. Among the variables identified by the relevant literature, those measurable with Census data and SIAPE data, normally accessible in urban planning processes, were selected for a total of 12 variables. This choice allows for a reliable reading of the phenomenon at a local scale (Section 3.1).

In order to aggregate the indicators into the LIV_ep composite index, they must be normalized, to harmonize their units of measurement, not the polarity, which is the same for all (Section 3.2).

As mentioned, the indicators can be correlated with each other, therefore, while ensuring the possibility of reading the representative maps of the individual indicators, it is necessary to assess the PCA applicability in the calculation of the overall vulnerability index. PCA is applied not so much to reduce the initial set of indicators, but to avoid overestimating the EP vulnerability phenomenon by using redundant information (Section 3.3).

A sensitivity analysis can be performed to assess the robustness of LIV_ep to the aggregation method by comparing the PCA-based index with an alternative one (Section 3.4).

3.1. Indicators Selection and Data Collection

With reference to energy vulnerability, energy performance can be assessed through the Global Non-Renewable Performance Index (EP_gl,nren), according to the GIS-based methodology proposed by Grimaldi & Marra [30]. Starting from census data, this method allows a value of EP_gl,nren to be assigned to each census section throughout the Italian territory using data retrieved from SIAPE. By fixing climatic zone, residential use and building age class, SIAPE returns the corresponding EP_gl,nren value, obtained from the average values declared in the APEs of dwellings having the specified characteristics. Since climatic zone is constant within the municipality, the differentiation among census sections is obtained from the number of residential buildings and the composition of the building stock by age class. Specifically, the EP_gl,nren value for each census section is calculated as a weighted average of the SIAPE values associated with the different building age classes, using the number of buildings belonging to each age class as weights, according to the procedure detailed in [30].

The EP_gl,nren index thus obtained, expressed in kWh/m² per year, is multiplied by the useful floor area of the dwellings within each census section to estimate the overall demand for basic energy services, expressed in kWh per year [30].

The ability of occupants to implement EP alleviation measures depends on multiple vulnerability factors. The ratio of rented dwellings to total dwellings is a particularly relevant EP indicator, as it is associated with a higher incidence of absolute poverty and lower incentives for energy efficiency, as also suggested by the Italian National Integrated Energy and Climate Plan (PNIEC) [25]. Empty or sporadically occupied dwellings lead to possible deterioration due to lack of maintenance, with negative effects in terms of energy efficiency. Similarly, this happens if the state of conservation of the buildings is mediocre or very bad, a condition indicative of the degree of degradation. The elderly and children, who spend more time at home on average and need stable thermo-hygrometric conditions, are more vulnerable according to the studies examined in the literature reviewed. The widespread presence of foreigners from low-income countries, or families living in overcrowded conditions, is equivalent to a greater probability of being reduced to absolute poverty, consequently also to EP.

The selection of the socio-economic, demographic, and housing-related indicators was additionally supported by the framework proposed by ISTAT within the Italian Parliamentary Commission of Inquiry on urban safety and degradation [52], which provides an institutional basis for diagnosing multidimensional urban vulnerability using census data at the intra-urban scale (Table 1).

Table 1. Selected infra-urban scale Vulnerability to Energy Poverty (V_ep) indicators. Definition, units of measurement, formulas for their measurement and data available for census section useful for calculation.

Indicator	Definition and Units of Measurement	Formula	Census Data
I1—Global Non-Renewable Energy Performance Index (EPgl,nren)	Annual non-renewable primary energy requirement of a building, in relation to its useful surface area, to meet energy services (heating, cooling, ventilation, domestic hot water)	Calculated according to Formula (1) in [30]	E8—Buildings constructed before 1919 1
			E9—Buildings constructed from 1919 to 1945 1
			E10—Buildings built from 1946 to 1960 1
			E11—Buildings built from 1961 to 1970 1
			E12—Buildings built from 1971 to 1980 1
			E13—Buildings built from 1981 to 1990 1
			E14—Buildings built from 1991 to 2000 1
			E15—Buildings built from 2001 to 2005 1
			E16—Buildings constructed after 2005 1
I2—Demand for basic energy services	Product of the Global Non-Renewable Energy Performance Index and the useful surface area of dwellings (kWh per year)	${EP}_{gl,nren}\times A44$ [30]	A44—Useful surface area of dwellings occupied by at least one resident 1
I3—Active unemployment rate	Ratio of unemployed population seeking new employment to population in the total labour force (-)	${\displaystyle \frac{P62}{P60}}$ [52]	P60—Total population aged 15 and over in the total labour force
			P62—Total population aged 15 years and over unemployed looking for new employment
I4—First-time job seekers rate	Ratio of persons seeking their first employment to the total labour force (-)	${\displaystyle \frac{P60-P61-P62}{P60}}$ [52]	P60
			P61—Total population aged 15 and over employed
			P62
I5—Low school attendance rate	Ratio of the population with an educational qualification equal to a lower secondary school diploma (middle school) or less to the total resident population aged 6 and over (-)	${\displaystyle \frac{P49+P50+P51+P52}{P46}}$ [52]	P46—total of 6 years and older 1
			P49—Population with middle school diploma
			P50—Population with primary school leaving certificate
			P51—Literate people without a degree
			P52—Illiterate people
I6—Incidence of children	Ratio of population of age less than 5 years and the total resident population (-)	${\displaystyle \frac{P14}{P1}}$	P1—Total population
			P14—Population under 5 years of age
I7—Old-age index	Ratio of population aged 65 years and over to the population aged 0–14 years (-)	${\displaystyle \frac{P27+P28+P29}{P14+P15+P16}}$ [52]	P14
			P15—Population aged 5–9 years
			P16—Population aged 10–14 years
			P27—Population aged 65–69 years
			P28—Population aged 70–74 years
			P29—Population over 74 years of age
I8—Incidence of large families	Ratio between the number of households with 6 or more members and the total number of households (-)	${\displaystyle \frac{PF8}{PF1}}$ [52]	PF1—Total resident households
			PF8—Resident households with 6 or more members
I9—Vulnerable migration rate	Ratio of foreigners from low-income countries to the total of the resident population (-)	${\displaystyle \frac{ST10+ST12}{P1}}$ [52]	ST10—Foreigners residing in Italy—Africa 2
			ST12—Foreigners residing in Italy—Asia 2
			P1—Total population
I10—Building Degradation Index	Ratio between residential buildings in poor and mediocre state of conservation and the total number of residential buildings (-)	${\displaystyle \frac{E30+E31}{E3}}$ [52]	E3—Buildings for total residential use
			E30—Residential buildings with a poor state of conservation 1
			E31—Residential buildings with a very poor state of conservation 1
I11—Incidence of empty or sporadically occupied dwellings	Ratio of empty dwellings and dwellings occupied only by non-residents to total dwellings (-)	${\displaystyle \frac{A6+A7}{A2+A6+A7}}$	A2—Dwellings occupied by at least one resident
			A6—Empty dwellings
			A7—Dwellings occupied only by non-residents
I12—Housing rate for rent	Ratio of rented dwellings to total dwellings occupied by at least one resident (-)	${\displaystyle \frac{A46}{A2}}$ [52]	A2
			A46—Rental housing 1

¹ Data available only at the 2011 Population and Housing Census of Italian National Institute of Statistics (ISTAT), not in 2021. ² The technical report taken as a reference for the formula of I₉ [52] specifies that this is a proxy variable, not an individual vulnerability, since the available census data do not allow a more detailed disaggregation of the foreign population by country of origin.

3.2. Normalization

The normalization of the m basic indicators ($I_{\mathrm{m}}$, with m = 1, …, 12) is carried out through the “Min-Max” method. Among the possible normalization techniques, this procedure allows to relate the unique values of a single input indicator ($x_{m,i}$) to the maximum ($x_{{m,i}_{max}}$) and to the minimum ($x_{{m,i}_{min}}$) value that the indicator itself assumes in the territory under study, consisting of n census sections (with $i$ = 1, …, n). In the chosen set, all indicators have positive polarity, i.e., an increase in their value corresponds to an increase in V_ep. Therefore, the formula that obtains the respective normalized values ($y_{m,i}$) can be written as follows:

$$y_{m,i}=\left\{\begin{array}{cc}0& x_{m,i}\le x_{{m,i}_{min}}\\ {\displaystyle \frac{x_{m,i}-x_{{m,i}_{min}}}{x_{{m,i}_{max}}-x_{{m,i}_{min}}}}& \hspace{1em}\hspace{1em}\hspace{1em}{x}_{{m,i}_{min}}<x_{m,i}<x_{{m,i}_{max}}\hspace{1em}\hspace{1em}\hspace{1em}\\ 1& x_{m,i}\ge x_{{m,i}_{max}}\end{array}\right.$$

The domain of each normalized indicator is made up of the interval to which it belongs [0, 1], where 1 corresponds to the maximum degree of criticality, while 0 represents the minimum.

3.3. PCA Suitability, Weights Choice and Aggregation

Since some of the selected indicators may convey partially overlapping information, the applicability of Principal Component Analysis (PCA) is assessed before their aggregation.

As a preliminary diagnostic, linear relationships among the normalized indicators are explored through the Pearson correlation matrix. For example, for the normalized indicators $y_{1,i}$ and $y_{2,i}$ the index is calculated as follows:

$$r_{12}={\displaystyle \frac{{\sum }_{i=1}^n\left(y_{1,i}-{\overline{y}}_1\right)\left(y_{2,i}-{\overline{y}}_2\right)}{\sqrt{{{\sum }_{i=1}^n(y_{1,i}-{\overline{y}}_1)}^2}\sqrt{{{\sum }_{i=1}^n(y_{2,i}-{\overline{y}}_2)}^2}}}$$

where ${\bar{y}}_1$ and ${\bar{y}}_2$ represent the average values of the two indicators calculated on the set of n census sections of the study area.

The correlation matrix is examined to identify the presence and extent of shared information among the indicators; however, no minimum pairwise correlation is imposed as a necessary condition for PCA, since the overall suitability of the data structure can be assessed through the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity. A KMO value above 0.5 and a statistically significant Bartlett’s test with p-value < 0.05 are considered evidence supporting the applicability of PCA [53].

When these conditions are satisfied, PCA is applied to the normalized indicators. Component retention is primarily based on the Kaiser criterion (eigenvalue ≥ 1), complemented by inspection of the scree plot and by consideration of the cumulative explained variance and substantive interpretability of the extracted components. The percentage of explained variance is therefore used as a diagnostic measure of the information retained by the solution rather than as a fixed exclusion threshold. To facilitate component interpretation, the retained solution is subjected to Varimax rotation with Kaiser normalization. Component scores are estimated using the regression method.

The extracted components are first interpreted as complementary dimensions of V_ep on the basis of their rotated loadings. Their inclusion in the composite LIV_ep additionally requires that the component can be interpreted according to a consistent vulnerability direction, so that increasing component scores can be unambiguously associated with increasing vulnerability.

The composite index ${LIV}_{ep}$, for each i-th census section, will be given by the weighted sum of the Principal Components extracted that satisfy these conditions, as follows:

$${LIV}_{ep,i}={\displaystyle \sum _{k=1}^K}{w}_k·{PC}_{k,i}$$

where $K$ is the number of components selected for aggregation and $w_k$ is the weight assigned to each component.

Since Varimax rotation redistributes the variance explained among the retained components, the weights are calculated from the Rotation Sums of Squared Loadings of the components included in LIV_ep, as follows:

$$w_k={\displaystyle \frac{{\lambda }_k}{{\sum }_{k=1}^K{\lambda }_k}}$$

where ${\lambda }_k$ is the variance explained by the single component after rotation, and the denominator of the formula represents the cumulative variance.

Since the weighted sum of standardized orthogonal component scores does not necessarily have unit variance, the thresholds used to classify the individual standardized component scores are not directly applicable to ${LIV}_{ep}$.

Thus, ${LIV}_{ep}$ is subsequently classified with the Natural Breaks method [55], which identifies class boundaries by minimizing within-class variance and maximizing differences between classes. This allows a practical use of the final map of ${LIV}_{ep}$ for urban planning purposes.

3.4. Sensitivity Analysis

As emerged from the literature review, several techniques can be used to aggregate the selected indicators as alternatives to PCA. Among these, the arithmetic mean with equal weights represents one of the simplest and most transparent approaches. Therefore, an alternative equal-weighted index (LIV_ep-EW) is calculated as the arithmetic mean of the twelve normalized indicators, without applying PCA.

To assess the sensitivity of LIV_ep to the aggregation method, the two indices are compared using Spearman’s rank correlation coefficient. Furthermore, both indices are classified into four vulnerability classes using the Jenks Natural Breaks method, and the agreement between census sections classified as High Vulnerability is assessed.

4. Case of Study

The municipality of Milan is the capital of the Lombardy Region, the most populous in Italy, with over 10 million inhabitants. The city of Milan has 1,365,698 inhabitants in 2025, according to ISTAT data. The census sections identified by ISTAT for the city in 2011 are 6079, of which 68 have no data. In addition, the census sections are aggregated to constitute the Local Identity Nuclei (NIL), equal to 88. The NILs were introduced by the first GUP, which is called the Territorial Government Plan (PGT) in the Lombardy Region, and correspond to the recognizable districts of the city (Figure 2).

Figure 2. Territorial framework of the municipality of Milan, in Lombardy Region, Italy.

To comply with the commitments made under the ‘C40 Cities’ initiative, the Municipality is currently drafting the ‘Plan for the fight against energy poverty and precariousness’, thus representing an optimal case study for testing the proposed methodology. Through this plan, the municipality intends to address both vulnerability to EP and the risk factors contributing to energy poverty and precariousness, structuring its strategy around four lines of action [56].

The first line of action (Action 1) concerns the establishment of a Municipal Observatory on Energy Welfare and specifically focuses on mapping households experiencing EP, which represent the target of measures aimed at tackling the phenomenon, while the subsequent lines of action define these measures. The mapping methodology consists of an initial measurement of EP based on a traditionally used indicator, namely the ratio of energy expenditure to household income. According to this approach, households are considered to be in EP if the energy expenditure exceeds 10% of household income. In addition, the plan envisages a more detailed assessment in the future, aimed at mapping vulnerable areas that better reflect the multidimensional nature of the phenomenon, in order to identify priority areas for intervention within the municipal territory. This assessment will be performed through the integration of additional economic, energy, climate and socio-demographic indicators (Low Income-High Costs, arrears in bills, energy prices, thermal comfort, age, education and unemployment). The data necessary to measure these indicators will be obtained from both institutional sources and ad hoc surveys. Data can be collected at various spatial scales, involving municipal technical staff, external professionals, and stakeholders. However, the spatial unit for mapping chosen is the census section, which is considered more suitable for the official communication of results, as it allows the data to be anonymized while providing an appropriate spatial resolution for understanding the territorial distribution of the phenomenon.

The measures for tackling EP concern economic support measures, aimed at improving access to energy efficiency incentives and promoting RSECs, in line with the Air and Climate Plan (Action 2); awareness-raising and user accompaniment activities (Action 3); governance tools (Action 4), aimed at strengthening coordination between energy, social and urban policies, within the framework of the Air and Climate Plan and other municipal strategies.

Although there is no explicit reference to the PGT, the municipal office responsible for urban regeneration is mentioned among the actors involved in Action 4.

The current PGT, approved in 2020, identifies areas where integrated actions aimed at promoting widespread urban regeneration are to be prioritized. These priority areas are the following: Buildings in outlying Historic Settlements; Urban Renewal Areas (ARU); Environmental Regeneration Areas; Squares; Interchange nodes; Major Urban Functions Areas. These areas partially overlap with some existing Public housing areas (ERP), which are therefore not all identified as priorities for intervention (Figure 3).

Figure 3. Priority areas for urban regeneration according to the Territorial Government Plan (PGT) of Milan (own elaboration on data from the map “R.02—Urban Planning Indications” of the PGT).

Energy efficiency interventions and measures aimed at increasing carbon neutrality, on the other hand, apply to the entire municipal territory rather than only to priority areas [57].

5. Results

The application of the proposed methodology to the Milan case study used the following software, used with respect to the workflow in Figure 1:

-

QGIS (v. 3.40.4; QGIS Association, Laax, Switzerland) to prepare the input census data, calculate the basic indicators per census section and normalize them [58];

-

IBM SPSS Statistics (v. 32.0; IBM Corp., Armonk, NY, USA), for the verification of the PCA applicability, and the subsequent extraction of the Principal Components [59];

-

Again QGIS (v. 3.40.4) to classify the output data and visualize the results via choropleth maps.

5.1. Verification of the PCA Applicability

The analysis was conducted on a large number of observations: 6011 census sections, for 12 variables.

The PCA applicability was assessed by examining the Pearson correlation matrix of the twelve normalized indicators (Table A1—Appendix A) and performing the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity.

The value of KMO is equal to 0.706; while Bartlett’s test was statistically significant (χ² = 8287.058; df = 66; p < 0.001).

These results support the suitability of the dataset for PCA, despite the generally modest pairwise correlations observed among several indicators.

5.2. Extraction and Mapping of the Principal Components

The number of Principal Components to be retained was determined primarily according to the Kaiser criterion (eigenvalue ≥ 1), complemented by inspection of the Scree plot (Figure 4), and consideration of the cumulative explained variance, as well as components interpretability.

Figure 4. Scree plot of the initial eigenvalues.

The number of components that meet the eigenvalues criterion ≥ 1 is four (Figure 4), and the overall percentage of explained variance is 52.3% (Table 2). Although this percentage indicates that a substantial proportion of the original variability remains unexplained, the four-component solution was retained because it allows distinct and interpretable dimensions of V_ep to be identified. The explained variance is therefore considered a measure of the information represented by the extracted components rather than a fixed threshold for excluding PCA.

Table 2. Rotation sums of squared loadings and percentage of total variance explained by the four retained Principal Components after Varimax rotation.

Principal Component	Rotation Sums of Squared Loadings	% Variance	% Cumulative Variance
PC1	2.380	19.831	19.831
PC2	1.500	12.500	32.331
PC3	1.317	10.978	43.309
PC4	1.082	9.014	52.323

As previously specified, the retained components were subjected to Varimax rotation with Kaiser normalization to facilitate their interpretation. Table 2 reports the Rotation Sums of Squared Loadings and the corresponding explained variance after rotation.

By observing the matrix of rotated components (Table 3) it is possible to proceed with the interpretation of the extracted Principal Components and recommend possible associated planning strategies, while the extraction communalities of the twelve indicators are reported in Appendix A (Table A2).

Table 3. Matrix of rotated components for the four extracted Principal Components.

	PC1	PC2	PC3	PC4
I1	0.408	0.715	0.004	0.085
I2	−0.067	0.780	−0.115	−0.107
I3	0.656	0.034	−0.109	0.006
I4	0.442	−0.089	0.257	−0.366
I5	0.783	0.173	0.009	−0.098
I6	−0.057	0.542	0.554	0.125
I7	0.236	0.152	−0.614	−0.146
I8	0.223	−0.053	0.556	−0.387
I9	0.404	0.106	0.464	0.212
I10	0.370	−0.098	0.094	0.497
I11	−0.058	0.015	0.083	0.658
I12	0.746	0.011	0.050	0.082

With regard to the First Principal Component (PC1), observing the matrix of rotated components, the value of I₃, I₅ and I₁₂ is greater than 0.65. The values of I₁, I₄, I₉ and I₁₀ are close to 0.4, while the others are negligible. The signs are positive, so high positive values of PC1 identify areas with greater criticality, low negative values areas with less criticality. PC1 is therefore characterized by high loads associated with the Active unemployment rate (I₃), the Low schooling rate (I₅) and the Housing rate for rent (I₁₂), mainly configuring itself as a Socio-Economic Vulnerability (Figure 5). Additional contributions, albeit to a lesser extent, are provided by the Global Non-Renewable Energy Performance Index (I₁), the First-time job seekers rate (I₄), the Vulnerable migration rate (I₉), and the Building Degradation Index (I₁₀). This makes this component interpreted as an expression of a multidimensional fragility, even if the socio-economic aspect is predominant. Including both social and energy aspects, PC1 describes a systemic vulnerability. The strategy associated with the most critical areas is Integrated Urban Regeneration, which can provide for a multiplicity of actions, capable of addressing the different dimensions of fragility that this component represents.

Figure 5. Map of Socio-Economic Vulnerability (PC1).

The Second Principal Component (PC2) has Global Non-Renewable Energy Performance Index (I₁) and Demand for basic energy services (I₂) above 0.7. The value of the Incidence of children (I₆) is greater than 0.5, while all the others are negligible. The signs are positive, so high positive values of PC2 identify areas with greater criticality, low negative values areas with less criticality. Predominantly depending on indicators I₁ and I₂, PC2 stands as an Energy Vulnerability (Figure 6). Therefore, the strategy associated with the most critical areas is the promotion of Energy Retrofit interventions.

Figure 6. Map of Energy Vulnerability (PC2).

With regard to the Third Principal Component (PC3), the value of the Old-Age Index (I₇) is the only one higher than 0.6. Its loading is negative. The values of I₆ and I₈ are greater than 0.5, while the value of I₉ is close to 0.5. The signs are positive. The other variables are negligible.

Low, negative values of PC3 identify the areas with elderly populations, while high, positive values concern areas with more children and large families.

PC3 highlights a contrast between contexts characterized by a higher incidence of young population and large families and areas with a high Old age index, configuring itself as a dimension of Demographic Vulnerability (Figure 7).

Figure 7. Map of Demographic Vulnerability (PC3).

With reference to the Fourth Principal Component (PC4), the Incidence of empty or sporadically occupied dwellings (I₁₁) is the only one higher than 0.6 and only the Building Degradation Index (I₁₀) is close to 0.5. The signs are both positive. The values of the First-time job seekers rate (I₄) and the Incidence of large families (I₈) are close to 0.4 and are negative. The other variables are negligible. The signs are contrasting, therefore low, negative values of PC4 indicate the areas with the greatest criticality in terms of large families and unemployed people; high, positive values of PC4 identify the areas with the greatest criticality in terms of empty dwellings and building degradation.

PC4 is characterized by high loads associated with the incidence of empty or sporadically occupied dwellings and building degradation, configuring itself as a dimension of physical-settlement vulnerability and underuse of the housing stock, in short a Housing Vulnerability (Figure 8).

Figure 8. Map of Housing Vulnerability (PC4).

The maps obtained were classified into four ranges (<−1; −1–0; 0–1; >1), using the mean (0) and ±1 standard deviation as thresholds. This classification is possible because the scores of the Principal Components are standardized, with a mean of 0 and a standard deviation of 1. So, the thresholds adopted make it possible to distinguish the units with more marked deviations from the mean (more than ±1 standard deviation) from those characterized by intermediate values.

5.3. Measurement and Mapping of the LIV_ep Index

The first two main components (PC1 and PC2) have a clear interpretative consistency with respect to the polarity of the variables considered, i.e., the explanatory variables have a positive sign of loading. The subsequent components (PC3 and PC4), while contributing to the total variance, show contrasting loading signs that do not allow an unambiguous interpretation of increasing vulnerability. Therefore, they were mapped separately but excluded from the construction of LIV_ep.

From the correlation analysis of the PC1 and PC2 components it is clear that there is no correlation but they are orthogonal, or uncorrelated. This result was expected, having used Varimax’s rotation, which maintains orthogonality. However, it is interesting that even geographically PC1 and PC2 represent two distinct dimensions of the V_ep phenomenon, as is evident from the almost total absence of spatial overlap. Specifically, critical areas for PC1 are more pronounced in peripheral districts, while critical areas for PC2 also concern privately developed housing in central districts (Figure 9).

Figure 9. Spatial overlap between the critical areas of PC1 and PC2.

These two separate sub-indices of energy poverty, spatialized in the GIS environment, make it possible to obtain the relative maps, useful for territorializing contrast strategies: integrated urban regeneration, in the first case, or energy retrofit, in the second case.

To create the LIV_ep composite index, useful for knowing the areas with the greatest V_ep, priority for intervention, the weighted sum of the PC1 and PC2 components is performed, where the weights take into account the variance explained by the single component with respect to the cumulative variance, given considering both components:

$${LIV}_{ep,i}=w_1·{PC}_{1,i}+w_2·{PC}_{2,i}$$

where

w₁ = 19.831/32.331 = 0.613

w₂ = 12.500/32.331 = 0.387

Higher values of the LIV_ep composite index indicate increasing levels of criticality, resulting from the combination of socio-economic and energy vulnerability (Figure 10).

Figure 10. Map of LIV_ep composite index.

5.4. Sensitivity Analysis of the LIV_ep Index

A sensitivity analysis was performed to assess the robustness of LIV_ep to the aggregation method by comparing the PCA-based index with an alternative equal-weighted index (LIV_ep-EW), obtained as the arithmetic mean of the twelve normalized indicators. The values of LIV_ep-EW were spatially represented in a GIS environment, obtaining the corresponding map (Figure 11).

Figure 11. Map of LIV_ep-EW composite index.

The comparison between the two indices showed a strong positive correlation, with Spearman’s rank correlation coefficient of 0.876 (p < 0.001), indicating a high degree of consistency in the ranking of census sections according to their vulnerability levels.

However, some differences emerged in the identification of the most vulnerable areas: 588 census sections were classified as High Vulnerability according to LIV_ep, compared with 557 according to LIV_ep-EW. Of these, 350 census sections were classified as High Vulnerability by both indices, corresponding to 59.5% of those identified by LIV_ep.

These results indicate that the overall ranking of census sections is relatively stable with respect to the aggregation method, although some differences emerge in identifying the most vulnerable areas.

5.5. Implications of the Results Obtained for Local Planning

Among the areas of the consolidated urban fabric that the PGT considers priorities for regeneration, represented in Figure 3, the Urban Renewal Areas (ARU) are of particular interest to discuss the implications of the results obtained for local planning, as well as Public housing areas (ERP). Thus, spatial overlap was calculated as the percentage of the total surface area of ERP and ARU falling within census sections characterized by critical vulnerability conditions according to PC1, PC2, and LIV_ep, respectively.

A total of 54.54% of the ERP surface area—as defined in the Milan PGT—falls within census tracts characterized by critical conditions according to PC1, compared with 4.80% according to PC2 (Figure 12). Furthermore, 49.78% of the total ERP surface area falls within census tracts classified as High Vulnerability according to LIV_ep (Figure 13).

Figure 12. Spatial overlap between Public housing areas (ERP) and critical areas identified by PC1 and PC2.

Figure 13. Spatial overlap between ERP areas and critical areas identified by LIV_ep.

Regarding ARUs, 16.68% and 6.26% of their total surface area fall within census tracts characterized by critical conditions according to PC1 and PC2, respectively (Figure 14), while 8.36% falls within census tracts classified as High Vulnerability according to LIV_ep (Figure 15). This highlights the potential contribution of V_ep mapping to the integration of energy vulnerability considerations into urban regeneration strategies.

Figure 14. Spatial overlap between Urban Renewal Areas (ARU) and critical areas identified by PC1 and PC2.

Figure 15. Spatial overlap between ARU and critical areas identified by LIV_ep.

The drafting of the Plan for the fight against energy poverty and precariousness can benefit from the knowledge thus produced, in order to enhance the potential benefits of urban regeneration, already envisaged by the PGT, also in terms of tackling V_ep (Action 4), and to spatially target measures aimed at facilitating access to energy efficiency incentives (Action 2).

Furthermore, the results suggest that the critical conditions affecting ERP areas are predominantly associated with socio-economic vulnerability (PC1). This finding is consistent with the fact that the presence of public housing may itself influence the socioeconomic composition of census tracts. However, vulnerability represented by PC1 is particularly pronounced in peripheral districts, where most ERP areas are located.

Energy vulnerability (PC2), on the other hand, also affects privately developed housing in central districts.

For the implementation of Action 2, which concerns the promotion of the RSECs, in line with the current Air and Climate Plan, and Action 3, relating to awareness-raising and user accompaniment activities, this knowledge is useful for identifying priorities for intervention.

With this awareness, the mapping of V_ep envisaged by the Plan under Action 1 should be carried out.

More generally, these results confirm the importance of complementing the analysis of the energy efficiency of the building stock with the related social and economic aspects, in order to effectively counteract all dimensions of V_ep.

6. Discussion

Previous studies on the EP subject show that in the scientific community there is a general agreement on the causes of EP, while there is still a lack of a shared vision on vulnerability indicators at the local scale.

With reference to the first research question (RQ1), previous studies addressing EP vulnerability at the intra-urban scale are still limited, probably due to the difficulty of finding the data necessary to measure EP indicators at such a detailed level [47]. In Italy, the datasets available in open format do not allow the measurement of all the variables identified by the relevant studies at an infra-urban scale, limiting their transferability to other geographical contexts. This work proposes a method of diagnosis of V_ep entirely based on open data provided by public institutions, therefore easily practicable within urban planning processes.

This choice represents a strength of the work in terms of its transferability, but it implies that some indicators proposed by the relevant studies have not been selected due to the lack of data available in open format at an infra-urban scale for the entire Italian territory. This is the case, for example, of the ‘income’ variable and the indicators related to gender vulnerability.

The inclusion of income, recognized as one of the main drivers of EP, is supported by numerous studies addressing EP at the local level [42,46,47,48]. In the Italian geographical context, the territorial variation of per capita income is generally not known at an infra-urban scale. In fact, the Ministry of Economy and Finance provides these data with reference to a lower spatial detail, i.e., for each Italian municipality, with the exception of some large cities. To overcome this problem, it is possible to use other variables, measurable at the local level, that are indicative of potential socio-economic hardship among resident households, according to the technical-scientific literature on urban poverty. These variables, so-called ‘proxies’, have been proposed in order to identify measures other than income, since incomes tend to fluctuate over time, and related data may also be affected by under-reporting, especially with reference to individual income [60]. Nevertheless, similar problems may also arise when using income proxy variables. The market value of a dwelling is widely considered a proxy variable for estimating the income of its inhabitants. In Italy, these data can be derived from the property values provided by the Real Estate Market Observatory (OMI), through which the Revenue Agency has mapped urban and rural areas according to their market values across the entire Italian territory [61]. However, the level of spatial detail is not sufficient to differentiate within urban centres and smaller settlements, i.e., on an infra-urban scale.

The gender issue has been raised by relevant and very recent studies, which recommend considering the greater vulnerability of the female population to EP when the phenomenon is investigated on an infra-urban scale [47,62]. However, the proposed indicators, such as those relating to single women with children or elderly women living alone, cannot be measured using open data available at the infra-urban scale across the entire Italian territory. Manaresi and Santangelo, for example, report that they obtained such data from the municipality of Bologna, although at a less detailed spatial scale than the census section [48].

Further indicators were not considered, not only due to the lack of data, but also because they were already partly taken into account in the energy vulnerability assessment. One of these variables is the average outdoor temperature, proposed by Robinson et al. [45], whereas Zardo et al. use degree days [44]. From an operational perspective, it is problematic to find temperature datasets at the infra-urban level. Values are usually available for large areas, therefore at an infra-regional scale. The knowledge of temperature at the local scale requires advanced specialized processing, for example through the analysis of satellite data conducted for a specific study area [63]. It should also be noted that this variable is already taken into account through the climate zone, which is used to estimate energy vulnerability.

Data limitations can be addressed by integrating the available datasets with information directly provided by local authorities or citizens. The knowledge offered by the data already available can also serve as a basis for involving citizens in providing their data, thus allowing the results of the proposed approach to be compared with those obtained through a participatory one.

The second research question (RQ2) concerns the methods used to construct the composite indices based on the indicators identified to measure the selected vulnerability factors. As for the approaches adopted in previous studies to construct composite indices, the choice depends both on the objectives of the work and on the applicability of these methods. For example, PCA requires an adequate number of observations and a suitable correlation structure among the selected variables, as assessed through appropriate statistical tests. Therefore, its applicability must be verified in relation to the characteristics of the territorial context under study. The weighted average, on the other hand, is affected by the choice of weights, especially if this does not rely on statistical criteria, but depends on a subjective judgment. In the latter case, the method may be valid for the context examined but it is not automatically suitable for territories that do not have similar characteristics. These issues have already been highlighted by scholars regarding the application of these methods to other urban phenomena [64].

In the Milan case study, the large number of observations and the adequate statistical properties of the data supported the application of PCA. Among the extracted components, Socio-Economic Vulnerability (PC1) and Energy Vulnerability (PC2) were aggregated into LIV_ep, providing a synthetic representation of these dimensions intended to support spatial planning decisions. Conversely, PC3 and PC4, related to demographic and housing vulnerability, were retained as complementary diagnostic dimensions but excluded from LIV_ep, since their contrasting loading structures do not allow an unambiguous direction of vulnerability to be identified without introducing additional assumptions. Consequently, their direct translation into planning strategies is less straightforward. Notably, PC1 and PC2 correspond to the two core dimensions of energy poverty vulnerability recurrently identified in the reviewed literature. Nevertheless, since PC1 and PC2 together explain 32.331% of the original variance, LIV_ep should be interpreted as a planning-oriented synthesis of selected vulnerability dimensions rather than an exhaustive measure of vulnerability to energy poverty. Furthermore, since I₂ measures an absolute quantity of energy, its contribution to PC2 may also reflect differences in residential floor area across census sections.

The sensitivity analysis compared LIV_ep with an alternative index based on the arithmetic mean of the twelve normalized indicators, assigning equal weights to all indicators to avoid context-specific weighting choices and facilitate the transferability of the method beyond the Milan case study. The results showed a good degree of robustness of LIV_ep in terms of the overall ranking of census sections, although the areas identified as having the highest vulnerability did not fully overlap. It should be noted that this comparison also reflects differences in the information retained in the final indices, since LIV_ep is based on Principal Components PC1 and PC2, whereas LIV_ep-EW includes all twelve original indicators. Therefore, the resulting maps of priority areas should be interpreted with caution. Overall, these findings highlight the importance of assessing both the applicability of PCA and the influence of the aggregation method when transferring the proposed methodological workflow to other territorial contexts.

The third research question (RQ3) concerns how the spatial diagnosis of V_ep can support the targeting and integration of energy, social, and urban regeneration actions within local planning processes.

To combat EP through sustainable urban planning, it is possible to act on energy vulnerability on the one hand, and on the ability of occupants to implement alleviation measures on the other.

This approach translates into spatial strategies for tackling EP on a local scale, which can be framed in two main lines of action:

-

redeveloping the building stock as a priority where there is a greater problem (e.g., high energy consumption, poor energy performance of buildings), which can be addressed by facilitating private initiative through the rules of the general urban plan (e.g., by favoring volumetric increases, making urban planning parameters more flexible) or through local taxation (e.g., reductions in property and waste taxes);

-

identifying situations where residents cannot cope with such interventions, due to low income or other factors of social deprivation or vulnerability, i.e., where more incisive public action is needed, through integrated urban regeneration interventions.

A concrete example of the implementation of such strategies is the Metropolitan Urban Development Plan of the Metropolitan Area of Barcelona [65]. Although the concept of energy poverty is not expressly mentioned, the plan makes explicit reference to the phenomena of energy vulnerability and social vulnerability. For residential buildings in conditions of “energy deficit” or “energy vulnerability”, the plan provides for the “rehabilitate” strategy, while residential buildings that fall within urban areas where there is high social vulnerability are included in areas of urban regeneration, called “ARU-Re”, for which more integrated strategies are needed.

Urban regeneration is integrated, in the sense that it can provide for a plurality of actions, both aimed at the built environment and of a social and economic nature, intercepting the objectives of any existing sectoral planning. In this sense, the case of Milan is emblematic, as the drafting of the Plan for the fight against energy poverty and precariousness provides for a specific action (Action 4) aimed at systematizing energy, social and urban policies as part of the Air and Climate Plan and other municipal strategies, incorporating this perspective of integration.

However, the results obtained highlight the importance of integrating a spatially explicit diagnosis of V_ep into urban planning processes.

The comparison with the PGT forecasts provides an initial assessment of the spatial relationships between the areas already identified as priorities for urban renewal and those characterized by critical vulnerability conditions. Moreover, the overlap with public housing areas highlights the relevance of socio-economic vulnerability in defining local intervention priorities. In this context, LIV_ep can support the identification of priority areas, while the maps of the individual components provide complementary information for differentiating interventions according to the prevailing dimensions of vulnerability. This knowledge can contribute to the implementation of Actions 2 and 3 of the municipal Plan for tackling energy poverty, within the broader policy coordination envisaged by Action 4.

7. Conclusions

This work proposes the construction of a knowledge base of local vulnerability to energy poverty, in order to support urban planning processes, informing measures aimed at tackling the phenomenon. The maps obtained, both for the LIV_ep composite index and V_ep components, allow to spatially identify the most critical areas and, through PCA, where applicable, to distinguish the main dimensions of vulnerability. The knowledge produced aims to orient and coordinate energy, social and urban regeneration policies, to be translated into both general and sectoral urban planning tools, as demonstrated by the application to the case study of the municipality of Milan (Italy). The approach underlying the definition of the knowledge base is quantitative and based on datasets available throughout Italy, supporting the replicability of the method in other Italian municipalities.

The findings provide answers to the three research questions formulated after a literature review. First, a multidimensional diagnosis of V_ep at the intra-urban scale can be developed using a set of indicators derived from open and institutional data available for municipalities (RQ1). Second, the proposed workflow allows the aggregation method to be selected according to the statistical characteristics of the data, using PCA where its applicability is supported, in order to limit information redundancy in the construction of the composite index. The sensitivity analysis showed a strong consistency in the overall ranking of census sections, although the identification of the most vulnerable areas remains sensitive to the aggregation method (RQ2).

Third, the spatial knowledge obtained can support the targeting of actions and their integration across energy, social, and urban regeneration policies within urban planning processes (RQ3).

Accordingly, the main contribution of the study is a spatial decision-support framework, based on open and institutional data and GIS, that links the multidimensional diagnosis of V_ep at the intra-urban scale with the spatial targeting of planning actions.

The main limitation of the work presented concerns the outdated nature of the census data used to measure the proposed indicators, derived from the 2011 ISTAT Census. This was necessary because the input census variables are not yet available in the data released by the most recent censuses, although they are likely to become available in the future.

In addition, the energy data derived from SIAPE refer to different time periods, while, ideally, they should refer to the same reference year as the census data. Furthermore, SIAPE provides data derived from APEs, which are issued on the occasion of leases, sales, or renovations; dwellings without an APE are therefore underrepresented in the database. However, currently SIAPE represents the most suitable Italian source for retrieving the type of energy data required by the proposed methodology, being an institutional, open-access source with data available for the entire national territory.

Therefore, these aspects limit the ability to use the results as representative of current conditions. However, the aforementioned limitations could be overcome through future updates of the census datasets, or the availability of more recent local data sources with an equivalent level of spatial detail, which would allow the diagnosis to be updated. While keeping the proposed methodological structure substantially unchanged, some adjustments related to the indicators could be made, having adequate data that allow it, i.e., the indicator I₂ could be constructed in such a way as to be independent of I₁, while the indicator I₉ could better reflect the foreign population actually living on low incomes.

With reference to the specific application to the Milan case study, future developments should include further testing of alternative normalization methods and component-retention criteria within the PCA framework.

Further progress of the work could also include the validation of the results against independent measures, where available, such as aggregated data on energy bonus beneficiaries or payment arrears, as well as the vulnerability maps to be produced as part of Action 1.

A further limitation concerns the spatial comparison between critical V_ep areas and existing planning geographies, which is based on the observed spatial overlaps. Future developments could complement this analysis with permutation or Monte Carlo tests to assess whether the observed overlaps differ significantly from those expected under random spatial allocation.

Moreover, the proposed methodology does not explicitly account for the spatial dependence among the territorial units of analysis. Future research could therefore investigate the spatial autocorrelation of V_ep and its components, in order to assess whether similar vulnerability conditions tend to cluster spatially and to identify statistically significant spatial patterns.

The methodological framework is potentially transferable to geographical contexts other than Italy, where comparable census and energy-related data are available. In this direction, the indicators and data sources may require adaptation to local data availability and planning contexts.

Future developments of the work also concern the extension of the proposed diagnosis to vulnerability to Summer Energy Poverty (SEPOV), a phenomenon expected to become increasingly relevant for sustainable urban planning due to climate change [62,63]. Finally, while the knowledge produced is useful in supporting decision-making, planning strategies will ultimately result from technical, political and social choices, also informed by citizen engagement. Innovative methodologies for public participation include Digital Twins, AI-based approaches and immersive visualization tools [66,67,68].