Spatial Dependence of Solar Photovoltaic Systems: Data Gathering
Process, Related Issues and Preliminary Results
Sergio Copiello
Department of Design and Planning, IUAV University of Venice, Dorsoduro 2206, Venice, Italy
Keywords: Solar Photovoltaic, Spatial Energy, Spatial Data, Peer Effects, Neighborhood Effects.
Abstract: In a previous study (Copiello and Grillenzoni, En. Proc., 2017), we have proven the solar photovoltaic
capacity in Italy to be characterized by spatial dependence. In that research, the units of analysis were the
Italian provinces, which correspond to level 3 of the European NUTS (Nomenclature of territorial units for
statistics) classification. Here we focus on new data encoded according to the Italian townships, namely, the
municipalities corresponding to level 2 of the European LAU (Local administrative units) classification.
The change of scale is a huge challenge, due to both the difficulty to find reliable information and the time-
consuming definition of the proximity structure of the units: while the provinces are about 100, the Italian
municipalities are several thousands, and each one shares the borders with many others. In particular, three
neighboring regions - Veneto, Trentino-Alto Adige, and Friuli-Venezia Giulia, in North-eastern Italy - and
their 1,121 towns are considered in this study, which primarily aims to delve into the issues related to the
data gathering process. As far as the preliminary findings are concerned, we find more clues about the role
played by the so-called neighborhood and peer effects.
1 INTRODUCTION
During the last four decades, in the Western
economies, the energy production and consumption
model has faced several changes, which imply that
producers and consumers have experienced shifts in
the energy mix. For instance, it deserves mentioning
the progressive substitution of oil products with
natural gas, which nowadays is the primary source to
produce electricity, as well as to heat buildings, in
several countries (Copiello, 2017). Moreover, it is
worth recalling the ongoing transition toward the
renewables. Under this framework, the last ten years
have seen a sizeable increase in the amount of solar
photovoltaic (PV) generation, which is about to
supply a 10% share of the primary energy used in
the residential sector (Copiello, 2017). The upward
trend in PV energy production is expected to go on
during the next years. According to theShort-Term
Energy Outlook published by the Energy
Information Administration (July 2017), in the U.S.,
the large-scale PV electricity generation should
increase by 38% in 2017 and 19% in 2018, while the
small-scale PV electricity generation will experience
a growth of 32% and 29%, respectively. As far as
long-term trends are concerned, the 2014 edition of
Technology Roadmap: Solar Photovoltaic Energy
published by the International Energy Agency
envisions that 16% of total electricity generation will
be met by PV systems in 2050, in comparison to 2%
in 2020 and 7% in 2030.
The ever-greater role played by PV systems has
drawn the attention of the scholarly research, which
has been engaged in analyzing the determinants of
their adoption and deployment. Following a
promising research strand focusing on neighborhood
and peer effects, in a previous study we proven the
spatial dependence that characterizes the installation
of PV capacity in Italy (Copiello and Grillenzoni,
2017b). In that research, the units of analysis were
the Italian provinces, which correspond to level 3 of
the European NUTS (Nomenclature of territorial
units for statistics) classification. Here we focus on
new data encoded according to the Italian townships,
namely, the municipalities corresponding to level 2
of the European LAU (Local administrative units)
classification. In particular, three neighboring
regions - Veneto, Trentino-Alto Adige, and Friuli-
Venezia Giulia, in North-eastern Italy - and their
1,121 towns are considered (Figure 1). The dataset
consists of all the PV systems that have been
installed - both by households and companies, on the
Copiello, S.
Spatial Dependence of Solar Photovoltaic Systems: Data Gathering Process, Related Issues and Preliminary Results.
DOI: 10.5220/0006665400150022
In Proceedings of the 7th International Conference on Smart Cities and Green ICT Systems (SMARTGREENS 2018), pages 15-22
ISBN: 978-989-758-292-9
Copyright
c
2019 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
15
Figure 1: Area of analysis: the municipalities in North-eastern Italy.
buildings’ rooftop or on the ground - during the
period 2005-2016, thanks to the subsidies provided
by the Italian laws named “Conto Energia” (Palmer
et al., 2015).
The main purpose of this study is to delve into
the issues related to the data gathering process,
particularly the stage meant to define the proximity
structure characterizing the units of analysis.
Moreover, we aim to discuss the preliminary
empirical evidence, as we find more clues about the
role played by the so-called neighborhood and peer
effects.
The remainder of this paper is organized as
follows. Section 2 provides a brief literature review
about the drivers of the adoption of PV systems,
with a specific focus on the few studies dealing with
the topic of spatial patterns. Section 3 is devoted to
discuss the data gathering process and the related
issues, particularly as regards the proximity structure
of the observations. Section 4 describes the
preliminary results we achieve, stressing
the additional clues of spatial dependence. Finally,
Section 5 outlines the conclusions of the
analysis.
2 LITERATURE REVIEW
The literature argues that the choice to adopt PV
systems depend on a set of influential parameters.
Balcombe et al. (2013) provide a summary of 18
earlier and contemporary studies that relate to the
motivations and barriers for the adoption of
microgeneration energy technologies, including both
solar thermal and solar PV. Half of these studies
concerns the UK, and most of the remaining
involves continental Europe’s countries. The
reviewed literature agrees in identifying the role
played by environmental concerns and financial
aspects. As far as the latter are concerned, the will to
save money due to lower energy bills is a significant
incentive, although counteracted by the expectation
of high upfront and operating costs, not to mention
long payback times and unclear impact on property
SMARTGREENS 2018 - 7th International Conference on Smart Cities and Green ICT Systems
16
value. It looks like other motivations and barriers do
matter, although there is a lack of consensus about
their importance. Additional determinants emerging
from the literature review performed by Balcombe et
al. (2013) are as follows: age; household size;  home
ownership or tenancy; social class; income;
education.
The survey performed by Sardianou and Genoudi
(2013) focusing on the residential sector confirms
some of the above findings: the consumers’
willingness to adopt renewable energy sources is
affected by age, education, income, electricity cost,
and perceived installation and maintenance costs.
The same authors claim that a tax deduction is more
likely to support the acceptance of the renewables
than an energy subsidy. However, it should be
considered that the above results stem from a small
sample, and are characterized by a low goodness of
fit.
Within the domain of the renewable energies, the
research strand that focuses on the adoption of PV
points out the significance of the following factors to
distinguish between early innovators, potential
adopters, majority adopters, and rejecters: the per-
capita income more than sunlight intensity (Schaffer
and Brun, 2015); the costs to be incurred and their
ratio to the expected benefits (Vasseur and Kemp,
2015); the built environment as well as the property
ownership structure (Schaffer and Brun, 2015;
Graziano and Gillingham, 2015; Balta-Ozkan et al.,
2015; Sommerfeld et al., 2017).
Alongside the above empirical evidence, another
phenomenon came to light following specific
studies. The literature suggests that the adoption of
the renewables, and especially the deployment of
solar PV across a country, may be encouraged by a
kind of emulation within communities and between
neighbors. Let us quote the Schelly’s (2014) words:
“Adoption of technological innovations is arguably
promoted through [a] form of informal information
sharing. [...] it is not simply information, but
particular communities of information. [...] For
some, individuals within their neighbourhood or
community provided inspiration” (p. 188). Actually,
during the last few years, a promising research
strand has focused on the occurrence of peer effects
and neighborhood effects in order to explain the
adoption of renewable energy sources, and
especially PV systems. That research branch sinks
its roots in the idea that spatial dependence is a key
driver for the diffusion of technological innovations
across territories and regions (Anselin, 1988; Keller,
2002; Schaffer and Brun, 2015).
Bollinger and Gillingham (2012) found that
social interactions - namely, peer effects - play a
major role in explaining the diffusion of PV panels
in California. Their analysis points to the
significance of two phenomena that occur within the
same zip code area and give rise to social spillovers:
the visibility of the PV panels is the former, the
influence of word of mouth is the latter. Other
studies show evidence that PV adoption is affected
by the number of similar systems that have been
previously installed in the same area or, more to the
point, in the recent past and in the immediate
surroundings (Müller and Rode, 2013; Schaffer and
Brun, 2015; Graziano and Gillingham, 2015; Balta-
Ozkan et al., 2015; Palm, 2016; Rode and Weber,
2016; Dharshing, 2017; Zhao at al., 2017, Copiello
and Grillenzoni, 2017b). Let us consider the words
of Müller and Rode (2013) that get to the heart of
the matter: “imitation of spatially close precursors is
indeed an explaining factor in PV adoption; [...]
results confirm a localized peer effect in the
adoption of PV” (p. 527). Similarly, Graziano and
Gillingham (2015) “find clear evidence of spatial
neighbor effects (often know as ‘peer effects’) from
recent nearby adoptions that diminish over time and
space” (p. 816). Balta-Ozkan et al. (2015),
Dharshing (2017), and Copiello and Grillenzoni
(2017b) confirm the occurrence of regional spillover
effects. Rode and Weber (2016) show the
occurrence of localized emulative behavior. Zhao et
al. (2017) claim that the deployment of PV systems
may be described by clusters that tend to spread in
the surrounding areas.
3 DATA GATHERING PROCESS
AND RELATED ISSUES
3.1 Proximity Structure
In order to investigate the occurrence of
neighborhood and peer effects, the identification of
the proximity structure that characterizes the unit of
analysis is the most time-consuming process we had
to deal with. It relies on the following stages:
use of search engines to find the list of
adjoining municipalities for each analyzed
township;
replacement of the adjoining municipalities’
names with the codes provided by the
National Institute of Statistics;
Spatial Dependence of Solar Photovoltaic Systems: Data Gathering Process, Related Issues and Preliminary Results
17
Figure 2: Excerpt from the proximity structure dataset.
use of the codes to extrapolate the data
concerning the solar photovoltaic capacity in
the adjoining municipalities, which are then
summed to calculate the total (Figure 2).
As far as the first stage is concerned, in North-
eastern Italy, the number of municipalities
surrounding each township highly varies, from a
minimum of 1 to a maximum of 21. On average, the
number of municipalities sharing their boundaries is
equal to 7. Figure 3 shows a selection of complex
neighborhoods. For instance, Verona - one of the
chief town in the Veneto region - is surrounded by
16 medium-sized townships. That situation is
common to other chief towns, but it also occurs in
rural and mountainous areas. Another unusual
feature is that several municipalities are composed
by at least two not contiguous territories. The issue
is further complicated because, contrary to what is
commonly thought, the municipal boundaries are not
stable at all. During the last decade, several changes
have taken place, mainly due to the need to reduce
the number of local administrative units, so as to
achieve saving in public expenditure. In the
Trentino-Alto Adige region alone, 50 municipalities
have disappeared: after having merged themselves,
they have brought 18 new larger townships into
being. Most of the mergers occurred in the last few
years and became effective in January 2016. Instead,
the figures on the installation of solar photovoltaic
systems mainly refer to the ex-ante situation.
Therefore, we had to keep track of both the
following aspects: the photovoltaic capacity installed
in the municipalities according to their former
boundaries (before the mergers), and their currently
neighboring towns (after the mergers). Another
distinguishing feature concerns the Alto Adige area -
namely, the province of Bolzano - where the
municipalities are identified by two names. Since it
is legally designated as a bilingual region, the former
name is in Italian, while the latter is in German.
Unfortunately, several sources use only one of the
names to label the data they provide, hence we had
to face matching problems when assembling the
dataset.
Leaving the above specific problems aside, the
identification of the proximity structure entails, at
least, two other issues, which have wide significance
and strong ability to affect the results. The former is
how we define the concept of proximity, namely,
what is the assumptions - and the measures - which
we rely on to distinguish the near spatial units from
the distant ones. The latter consists in the sort of
truncation the proximity structure is sometimes
subjected to.
As far as the first topic is concerned, to quote the
words of Tobler (1970), “everything is related to
everything else” (p. 234) and, more to the point,
“everything is related to everything else, but near
things are more related than distant things” (p. 236).
In this study, we assume that the energy-related
behavior in a municipality may be affected by what
happens in the adjoining municipalities. Therefore,
here we establish a relationship between proximity
and administrative borders, suggesting to translate
SMARTGREENS 2018 - 7th International Conference on Smart Cities and Green ICT Systems
18
Figure 3: Selection of complex neighborhoods.
the concept of proximity into practice according to
the shared boundaries between the analyzed
municipalities. But why not to assume that the local
behavior may be somehow affected by what happens
in all the surrounding municipalities within a radius
of, let us say, 50 kilometers? And why not to
consider all the municipalities within the same
province or region? Each of the above option is
arbitrary and, although we prefer to adopt the first
solution, the third one could be preferred for
simplicity’s sake. However, it looks hard to sustain
that a specific way to define the proximity structure
should certainly prevail among several available
alternatives. Moreover, one should be aware that the
above remarks are not free of consequences for the
results. In other words, the empirical findings on the
occurrence of spatial dependence phenomena, in
turn, also depend on how the spatial relationships
between the units of analysis are defined.
As regards the second topic, in our case study,
the claim that “everything is related to everything
else is somehow violated by the presence of the
national borders, where the spatial relationships find
an unexpected interruption. For instance, in the
province of Bolzano, the town of San Candido
borders on the Austrian town of Sillian. The two
towns are not situated on the opposite slopes of high
mountains, instead they are both located along the
Drava River in the Puster Valley. Moreover, they are
well linked by a primary road, and border controls
are no more carried out thanks to the Schengen
Agreement, not to mention that more than 80% of
the inhabitants in the Italian town of San Candido
are German native speakers. The same situation can
be found in several other municipalities, especially
in the northern Alto Adige, at the Austrian border,
and in the north-eastern Friuli, at the Austrian and
Slovenian borders. Therefore, there is no reason to
neglect the occurrence of cross-border relationships
and dependencies, except that we have no data on
the installed photovoltaic systems outside of Italy.
Obviously that data can be searched for, but we must
consider that they have a different nature and origin.
Indeed, we are analyzing the photovoltaic systems
that were subsidized according to a sequence of
Italian laws (Palmer et al., 2015). That laws were
stimulated by the European Directive 2001/77/CE.
The same happened in Austria, but according to
different detailed rules, as well as to different timing
and subsidies.
3.2 Other Parameter
The data concerning the installed PV capacity, both
in each municipality and in the adjoining ones, are
juxtaposed with variables belonging to the following
clusters (Table 1): geoclimatic aspects (surface area,
latitude, altitude, solar radiation); demography
(inhabitants and population density); economy
(income); social and behavioral aspects (waste
recycling rate). The underlying hypotheses are as
follows. The PV capacity is expected to be fostered
by a lower latitude and the corresponding higher
solar radiation, while it is expected to be limited by
unfavorable geographic conditions, such as smaller
surface area and higher altitude. The number of
inhabitants and the population density are
anticipated to be positively related to the installed
PV capacity, since individuals and families are
important targets of the policies providing incentives
and subsidies for the renewables. Also, the
disposable income is expected to be positively
related to the installed capacity, because the
adoption of PV systems involves the ability to incur
investment costs, even in presence of public grants.
The waste recycling rate is assumed as a proxy of
the adoption of innovative and responsible behavior,
hence we expect that the more the individuals and
households are prone to recycle, the higher should
be the installed PV capacity.
The above variables have some limitations,
especially with regard to the reference period, which
is not homogeneous. In particular, the data on solar
radiation refer to several years ago. They stem from
a research performed by ENEA, the former Italian
institute for research on nuclear energy, now
Spatial Dependence of Solar Photovoltaic Systems: Data Gathering Process, Related Issues and Preliminary Results
19
National agency for new technologies, energy and
sustainable economic development. The average
radiation on monthly and yearly basis is extrapolated
from EUMETSAT maps acquired during the period
1995-1999. The results are published only for the
towns with more than 10 thousand inhabitants.
However, the yearly solar radiation for different
locations, according to their latitude and longitude,
may be estimated using a web-based calculation
tool.
Table 1: Summary of the parameters.
Code Parameter Unit of
measure
Oipc Overall installed power
capacity
kW
Oipc
s-1
Oipc in the surrounding
towns
kW
Oipc_05-10 Oipc 2005-2010 kW
Oipc_05-10
s-1
Oipc 2005-2010 in the
surrounding towns
kW
Oipc_11-16 Oipc 2011-2016 kW
Oipc_11-16
s-1
Oipc 2011-2016 in the
surrounding towns
kW
Area Municipality surface area km2
Lat Latitude degrees
Alt Altitude m
Rad Global solar radiation MJ/m2
Inhab Number of inhabitants
Dens Population density inhab/km2
Inc Per capita disposable
income
Euros per
capita
Recycl Share of urban wastes
recycled
%
4 PRELIMINARY RESULTS
We base our preliminary findings on the following
regression model, from which we expect useful
suggestions in order to develop further studies:
Ln Oipc =
+
Ln Oipc
s-1
+
Ln X +
(1)
where
is the constant,
and
are the regression
coefficients, X is the vector of the independent
variables, and
is the error term. We use a double
logarithmic model since in the previous study it
proved to fit better the data (Copiello and
Grillenzoni, 2017b). Moreover, it allows dealing
with the possible non-linear relationships between
the parameters. Since the Ordinary Least Squares
(OLS) estimates are affected by heteroscedasticity,
as shown by the cone-shaped scatterplot of the
residuals (Figure 4), we opt for using
heteroskedasticity-robust Weighted Least Squares
(WLS) (Copiello and Grillenzoni, 2017a). The
results are summarized in Table 2.
Figure 4: Cone-shaped scatterplot of the residuals.
Due to their implications, two empirical findings
are worth attention. The first is that, contrary to the
expectations, the deployment of solar PV
installations has little or nothing to do with latitude
and solar radiation. The second is that several clues
of neighborhood and peer effects arise from the
analysis.
Table 2: Summary of the results.
Dependent Oipc
Parameter
T-stat P-value
const. 6.1309 2.375 0.0177
Oipc
s-1
0.4579 13.58 0.0000
Area 0.8493 21.42 0.0000
Dens 0.8313 22.40 0.0000
Inc -1.0487 3.843 0.0001
Adj. R
2
0.6471
The relationship between PV systems and
geoclimatic variables is quite weak with regard to
the installed capacity, on the one hand, and both
latitude and solar radiation, on the other hand. In
Figure 5 the values of these variables are subdivided
into quartiles. The correlation values are -0.38 and
0.45, respectively. It looks like the reason is the
strong development of the PV capacity in Alto-
Adige. Despite being an entirely mountainous region
characterized by a solar radiation of 4,661 MJ/m2 on
average, the more northern area of analysis has an
SMARTGREENS 2018 - 7th International Conference on Smart Cities and Green ICT Systems
20
Figure 5: Relationship between PV systems and geoclimatic variables.
installed photovoltaic capacity of 1,975 kW (54
kW/Km2), which are nearly the same one can find in
the Friuli region (2,151 kW, 77kW/km2), one degree
of latitude to the south. To go to the root cause of
that empirical finding, at least two hypotheses can be
put forward. Firstly, the geoclimatic data may not
tell the whole story, since during the winter a not
negligible share of the solar radiation is lost in the
Po Valley due to the recurrent presence of dense and
persistent fog. Secondly, the propensity to adopt PV
systems in Alto Adige may be ascribed to the
influence of the neighboring Austria, where the
government subsidies have started earlier.
The second hypothesis paves the way to the main
aim of this study, which is to check whether the
deployment of PV capacity is driven by
neighborhood effects, that is to say, whether the
phenomenon is bolstered by emulation. The
relationship between the installed capacity in each
municipality and the corresponding installed
capacity in the adjoining townships is positive and
high. Therefore, if we aim to understand the
deployment of the PV capacity and generation in a
territory, then we should consider not only
geoclimatic and socio-economic factors of that same
territory, but also what happens with regard to the
adoption of PV systems in the surroundings.
5 CONCLUSIONS
In this follow-up study, we analyze data encoded at
Spatial Dependence of Solar Photovoltaic Systems: Data Gathering Process, Related Issues and Preliminary Results
21
the municipal level, hence disaggregated at level 2
of the European LAU (Local administrative units)
classification. Here we find new empirical evidence
of the spatial dependence characterizing the
deployment of PV capacity and generation,
confirming our previous findings and the claims of
the few studies that have so far looked at this
promising research strand. We may conclude that
some energy-related behavior, signally those
concerning the adoption of renewable energy
sources, spread themselves across the space due to
phenomena of emulation between neighbors and
peers that can be caught and expressed according to
proximity measures.
However, further developments are required: by
enlarging the dataset in order to include additional
variables, by testing other proximity measures, and
by defining not only spatial but also spatio-temporal
regression models.
ACKNOWLEDGEMENTS
Statistical analysis is performed using the packages
R v. 3.3.2 and gretl v. 2017b. Spatial data
representation is made using the software QGis v.
2.14.9.
REFERENCES
Anselin, L, 1988. Spatial econometrics: methods and
models. Dordrecht: Springer Kluwer.
Balcombe, P., Rigby, D., Azapagic, A., 2013. Motivations
and barriers associated with adopting microgeneration
energy technologies in the UK. Renewable and
Sustainable Energy Reviews, 22, pp.655–666.
Balta-Ozkan, N., Yildirim, J., Connor, P. M., 2015.
Regional distribution of photovoltaic deployment in
the UK and its determinants: A spatial econometric
approach. Energy Economics, 51, pp.417–429.
Bollinger, B., Gillingham, K., 2012. Peer Effects in the
Diffusion of Solar Photovoltaic Panels. Marketing
Science, 31(6), pp.900–912.
Copiello, S., 2017. Building energy efficiency: A research
branch made of paradoxes. Renewable and
Sustainable Energy Reviews, 69, pp.1064–1076.
Copiello, S. Grillenzoni, C., 2017a. Is the cold the only
reason why we heat our homes? Empirical evidence
from spatial series data. Applied Energy, 193, pp.491–
506.
Copiello, S., Grillenzoni, C., 2017b. Solar photovoltaic
energy and its spatial dependence Energy Procedia,
141, pp. 86–90.
Dharshing, S., 2017. Household dynamics of technology
adoption: A spatial econometric analysis of residential
solar photovoltaic (PV) systems in Germany. Energy
Research & Social Science, 23, pp.113–124.
Graziano, M., Gillingham, K., 2015. Spatial patterns of
solar photovoltaic system adoption: The influence of
neighbors and the built environment. Journal of
Economic Geography, 15(4), pp.815–839.
Keller, W., 2002. Geographic localization of international
technology diffusion. American Economic Review,
92(1), pp.120–142.
Müller, S., Rode, J., 2013. The adoption of photovoltaic
systems in Wiesbaden, Germany. Economics of
Innovation and New Technology, 22(5), pp.519–535.
Palm, A., 2016. Local factors driving the diffusion of solar
photovoltaics in Sweden: A case study of five
municipalities in an early market. Energy Research
and Social Science, 14, pp.1–12.
Palmer, J., Sorda, G., Madlener, R., 2015. Modeling the
diffusion of residential photovoltaic systems in Italy:
An agent-based simulation. Technological Forecasting
and Social Change, 99, 106–131.
Rode, J., Weber, A., 2016. Does localized imitation drive
technology adoption? A case study on rooftop
photovoltaic systems in Germany. Journal of
Environmental Economics and Management, 78,
pp.38–48.
Sardianou, E., Genoudi, P., 2013. Which factors affect the
willingness of consumers to adopt renewable energies?
Renewable Energy, 57, pp.1–4.
Schaffer, A. J., Brun, S., 2015. Beyond the sun—
Socioeconomic drivers of the adoption of small-scale
photovoltaic installations in Germany. Energy
Research & Social Science, 10, pp.220–227.
Schelly, C., 2014. Residential solar electricity adoption:
What motivates, and what matters? A case study of
early adopters. Energy Research and Social Science, 2,
pp.183–191.
Sommerfeld, J., Buys, L., Mengersen, K., Vine, D., 2017.
Influence of demographic variables on uptake of
domestic solar photovoltaic technology. Renewable
and Sustainable Energy Reviews, 67, pp.315–323.
Tobler, A.W.R., 1970. A Computer Movie Simulating
Urban Growth in the Detroit Region. Economic
Geography, 46, pp.234–240.
Vasseur, V., Kemp, R., 2015. The adoption of PV in the
Netherlands: A statistical analysis of adoption factors.
Renewable and Sustainable Energy Reviews, 41,
pp.483–494.
Zhao, T., Zhou, Z., Zhang, Y., Ling, P., Tian, Y., 2017.
Spatio - temporal analysis and forecasting of
distributed PV systems diffusion : A case study of
Shanghai using a data - driven approach. IEEE Access,
5, pp.5135–5148.
SMARTGREENS 2018 - 7th International Conference on Smart Cities and Green ICT Systems
22