1. Introduction
The targets for the reduction of greenhouse gas (GHG) emissions set out in the Paris Agreement of 2015 determine the need for a strong decarbonisation of energy supply systems, in this context great attention is paid to distributed generation with the use of roof-mounted photovoltaic panels [
1,
2]. The International Energy Agency forecasts a large increase in the installation of photovoltaic systems in urban areas, estimating that the residential photovoltaic market could triple its volume by 2030. In the European Union, since the publication in 2009 of the first European Renewable Energy Directive on grid-connected photovoltaic systems, the installed power has increased tenfold, from 11.3 GW at the end of 2008 to over 116 GW at the end of 2018 [
3]. The recent (2018) revised European Renewable Energy Directive has set a target of 32% as a fraction of energy consumption from renewable sources by 2030 [
4]. This is certainly an ambitious target, which will require a significant increase in power generation from renewable sources (at least 65%). Part of it will have to come from photovoltaic generation [
5]. Almost half of the photovoltaic production capacity in Europe comes from rooftop, residential (28%) or commercial (18%) installations. Intensifying the diffusion of photovoltaic modules would facilitate the achievement of the above objectives.A similar development naturally leads one to wonder what the actual potential for photovoltaic production for residential and commercial buildings in the European Community could be, and whether this potential could be compatible with the objectives set.
Following Castellanos et al. [
6] it is possible to divide the methods for the evaluation of photovoltaic potential in urban areas into three levels called
low,
medium and
high. The
low level methods, based on a relationship between population density and the type of construction and roof area available for PV installation, showed low reliability;
medium level methods associate statistical information with spatial information derived from Geographical Information System (GIS) and Light Detection And Ranging (LiDAR)-based methods; finally, the
high level methods use high resolution geographical spatial information about the actual solar radiation, and often include, in addition to the actual rooftop surfaces, their inclination and orientation, and the effects of shading.
According to the classification proposed by Byrne et al. [
7], the methodologies for the evaluation of potential are divided on the basis of the employed data: (i)
sampling-based methodologies, in which the potential for a sample area is accurately assessed and the result is extrapolated to the surrounding areas; (ii)
multivariate sampling-based methodologies, in which correlations are identified between statistical data, such as population density and type of buildings, and the availability of suitable roof area; (iii)
complete census methodologies, in which all available high resolution geographical information on the examined area is used, without extrapolation from sampling.
Such assessments in metropolitan areas have already been carried out in the literature, thanks to the progressive growth in computational resources and the availability of increasingly detailed geographical data. Nguyen et al. outlined a LiDAR-based procedure to identify buildings suitable for panel installation and applied it to a very small area of about 50 buildings in the city of Kingston, ON (Canada) [
8]. Two other similar LiDAR-based techniques have been applied over Knox County, TN (United States) by Kodysh et al. [
9] and over the city of Georgetown in Malaysia by Latif et al. [
10]. Through the LiDAR reconstruction it is also possible to take into account the slope of the roof surfaces when estimating the generation potential. The availability of satellite imagery and the development of automatic image segmentation techniques allowed Khan et al. to identify the available surfaces for a district of Karachi in Pakistan, and to estimate the production potential based on monthly averages of irradiance data [
11].
The increased availability of computational power and high resolution geographical data has allowed for analyses to be carried out on larger metropolitan areas, such as the Gangnam district of the city of Seoul in South Korea [
12]. Hong et al. derived building data from a high resolution GIS database and through coupling with hourly irradiance data, assessed the potential for photovoltaic generation, by taking also into account the shading of buildings. The potential of the entire city of Seoul was already been assessed in [
7], but without taking shading into account and with an estimate of the usable rooftop area obtained on a statistical basis and verified by cartographic analysis. The estimate for the city of Mumbai in India is instead conducted by Singh et al. with a
multivariate sampling-based methodology [
13]. Another technique of the same class, based on sample analysis and extrapolation from 128 major cities in the United States, was presented by Margolis et al. [
14]. Bodis et al. presented an assessment of the potential for the entire European territory, based on a
complete census methodology where a 100 m resolution grid has been analyzed [
5].
The studies described above and the vast majority of tools available for the precise evaluation of the energy potentially produced by a photovoltaic system are based on the estimation of the radiation on the Plane Of Array (POA) from a combination of the direct and diffuse components of solar radiation and from the panel orientation. These evaluations allow an assessment of the production over a typical year, using data available from public services such as PVGIS (Photovoltaic Geographical Information System) [
15] or NREL (National Renewable Energy Laboratory) [
16]. However, given the amount of calculations required for the application of an irradiation-based procedure for very large domains, as is the case for the whole territory of a city, approximate methods have been proposed which only aim at a full year assessment of the energy that can be produced.
The Sky View Factor (SVF), used as an irradiation indicator, i.e., a function of the obstruction of the horizon by surrounding buildings on an exposed surface, is introduced by Robinson et al. [
17]. Moreover, with the aim of reducing the time requested by calculations, Rodriguez et al. proposed another methodology based on a
clear sky model and a three-dimensional representation of buildings [
18].
More recently, Calcabrini et al. introduced an approximate methodology for the assessment of the production potential even in the presence of shading due to complex urban
skylines, based on an elaboration of the SVF and validated through actual production data of some PV plants [
19]. Walch et al. proposed a study on the producibility of the entire territory of Switzerland, also addressing the problem of estimating the temporal variability of production [
20]. Finally, Tiwari et al. propose a procedure for the estimation of the rooftop solar energy photovoltaic potential using LiDAR scans and ortho-rectified aerial photography, without the need to use cadastral data, and apply it to a small city in Israel [
21].
In this work we want to propose a high level and complete census methodology for a fast and accurate evaluation of rooftop photovoltaic production potential by mapping the radiation time series according to the apparent position of the Sun, and by quickly calculating the shading in urban areas from building elevation data. We apply the proposed procedure to a real domain and validate it by comparison with a reference procedure.
The article is structured as follows:
Section 2 presents the adopted methodologies and
Section 3 describes the application of the proposed approach to the city of Cagliari in Italy, while the results are presented in
Section 4 and discussed in
Section 5. Finally,
Section 6 presents the conclusions of this work and its possible developments.
4. Results
As a point example of application of the proposed approach, we consider the case of a building in the city center that, even if located in front of a square, is also surrounded by taller buildings that determine shading effects, as shown in
Figure 3.
As described in
Section 3.5, the irradiation time series for the area in question are first obtained through the PVGIS API, then the values of the irradiance components are grouped by discretized azimuth and zenith values, and are finally added up; the result is shown in
Figure 4. For each apparent position of the Sun for which the irradiation components are not null, it is, therefore, possible to evaluate whether or not the area under examination is in the shade;
Figure 5 shows the effect of shading on GHI as a function of the apparent position of the Sun for a typical year. It appears evident the reduction of radiation due to the effect of the shadows brought by the buildings.
The estimation of the average monthly irradiance on a horizontal surface is shown in
Figure 6 for the proposed approach and for the benchmark methodology described in
Section 3.6. Such quantities are calculated by grouping and averaging the values of the time series output by PVGIS for both the CMSAF and SARAH datasets. The graphs show the monthly average values and the lower and upper extremes during the 10 and 12 years, respectively, for which the time series where available. The latter demonstrate the variability of the irradiation for each month across the years. The estimated GHI for the two methodologies substantially match with minimal deviations. For the considered location the two techniques provide an estimate of the annual irradiance on a horizontal surface equal to 1647 kWh/m
2/year for the proposed method and to 1660 kWh/m
2/year for the reference method, with a relative deviation equal to 0.8% (CMSAF), and respectively of 1602 kWh/m
2/year and 1617 kWh/m
2/year, with a difference of 0.9%, for the SARAH dataset.
The entire city domain, which includes almost 9000 buildings for a theoretically available roof area of
m
2, was then processed with the proposed methodology. After filtering with the procedure described in
Section 2.2, the useful surface has been reduced by 21% to
m
2, i.e., the 79% of the total rooftops.
With reference to the CMSAF dataset on the filtered surface, an average annual irradiance on horizontal surface of 1736 kWh/m2/year is estimated considering the shading effect, while, not taking shading into account, a nominal value of 1842 kWh/m2/year is calculated, thus indicating a loss of 5.75% due to shading. According to our estimate, therefore, the total irradiation on a horizontal surface for all the rooftop surfaces suitable for the installation of photovoltaic panels and within the city of Cagliari is equal to about 6.42 TWh/year. The corresponding values for the SARAH dataset are 1659 kWh/m2/year for the average irradiance on horizontal surface and 1747 kWh/m2/year when the effect of shading is not accounted for, a reduction due to shading equal to 5.03% and 6.13 TWh/year of total average annual irradiation for surfaces suitable for installation of photovoltaic panels.
The average error between the two approaches is quantified, in the case of the CMSAF dataset, with a mean absolute error (MAE) of 28.25 kWh/m2/year and a root mean squared error (RMSE) of 60.72 kWh/m2/year for an average GHI of 1736 kWh/m2/year. The MAE is 1.6% and the RMSE is the 3.5% of the average value. When using the SARAH model data, we have a better match between the estimates: MAE = 22.95 kWh/m2/year (1.4%) and RMSE = 51.47 kWh/m2/year (3.1%), with GHI averaging 1659 kWh/m2/year.
Figure 7 shows the monthly horizontal surface irradiation averaged over the entire domain for both the CMSAF and the SARAH datasets; in this case too there is a substantial match of the GHI estimates for the proposed methodology and the reference technique. The deviation between the two methods is minimal and mainly limited to the winter months when the CMSAF data are used and to the summer months when the SARAH dataset is employed.
Figure 8 shows a detail of the GHI distribution over the building rooftops: shading might result in significant reductions in irradiation. The lower values of GHI are obtained for buildings surrounded by taller buildings, but in general the GHI radiation values appear distributed uniformly and close to the maximum value obtainable for a free surface.
The accurate estimation of the irradiation on the surface of the panels is necessary for evaluating the potential generation from a photovoltaic system. In this approach we assume that panels are always built-in with the rooftop surfaces, i.e., respecting the slope of the surface on which they lie on, and therefore will be characterized by the tilt and the orientation of the rooftop on which they are mounted.
Unfavorably oriented surfaces such as north-facing rooftops, have a meager profitability due to the low average annual irradiance received (
Figure 9).
The total irradiance on the plane of the panels for the municipality of Cagliari, considering the inclination of the panels themselves in addition to the effect of shadows between buildings has an annual technical potential equal to approximately 6.89 TWh/year according to the irradiation provided by the CMSAF dataset and of about 6.54 TWh/year when processing the measurements provided by the SARAH dataset. The electric power potential can be estimated assuming a conversion efficiency for the PV panels.
The left plot in
Figure 10 shows the distribution of GHI values calculated according to the proposed method for the whole set of filtered rooftop surfaces with a resolution of 1 m
2. The distribution is characterized by the presence of different peaks: this is in fact the result of the composition of the GHI distributions for each macro-area (approximately 80% of the surface area of the buildings is actually distributed over just 5 macro-areas, out of a total of 32, in the case of the PVGIS-CMSAF dataset: the buildings within the city of Cagliari are in fact grouped in a quite small area compared to the whole municipal territory.). The most evident peaks correspond to the macro-areas with the highest number of buildings, i.e., the most surfaces potentially usable for the installation of photovoltaic panels. These differences are due to the solar irradiation data of the PVGIS datasets: the values can be significantly different as the result of distinct average weather conditions due, for example, to the proximity to the sea or another body of water, or the presence of a high grounds within the macro-area. The irradiance on the POA surfaces is characterized by the distribution shown in
Figure 10, where a larger ratio of building rooftops suffer from a reduction in actual irradiance, compared to the result for GHI. The effect is mainly related to the panels facing north which collect less radiation. However, the effect is mitigated by the more favorably oriented tilted roofs, which enjoy not only a superior POA irradiation, but also a larger exposed area due to the rooftops inclination.
The assessment of the roof surface area available for installation is based, as mentioned, on two threshold values. The first is related to the value of the slope computed from the DSM, beyond which the portion of the surface in question is considered unsuitable for the installation of panels, because at that point there are technical obstacles due, for example, to on-roof installations. A threshold of 45
was chosen. The second is represented by the value of the minimum usable surface area for a given building, below which the installation of a system is not technically justified. A threshold of 30 m
2 was set. We verified the sensitivity of the estimates obtained for GHI and POA irradiation to the variation of these threshold values, the results are shown in
Figure 11. From the values shown it can be seen that as the threshold value for the minimum available area increases, the constraint becomes more stringent and the smaller, unattractive areas are excluded from the total count. On average there is a reduction in the total irradiation both for GHI and POA, while the average value on the surfaces is almost unchanged. As the threshold on the slope increases, the total area considered acceptable increases, as the constraint becomes less stringent, and consequently the total irradiation collected increases. Also in this case the average values of the GHI and the POA irradiations on the resulting suitable surfaces do not show significant changes.
In addition to assessing the technical feasibility of installing a photovoltaic panel on a given surface, it would also be necessary to consider the cost-effectiveness of such installations. Since, with the same surface area of the panel, its profitability is directly proportional to the POA irradiation it can collect, it is possible to think of a threshold on the POA irradiation above which it is economically convenient to install the panel.
Figure 12 shows the fraction of roof surfaces that exceed this economic threshold as a function of the average irradiance collected. Both the fraction of the total surfaces and the portion exceeding the technical constraints (equal to 79% of the total) are shown. By imposing an additional constraint on the minimum allowable irradiance, the total potential can clearly be reduced. An in-depth economic-commercial analysis is beyond the scope of this work.
Summarizing, the proposed procedure is fast when compared to the benchmark technique, with calculation times for both methods reported in
Table 1. The calculation of horizons with GRASS GIS is computationally expensive and, in order to be tackled in an acceptable timeframe, requires the parallelization of the process on a distributed cluster; on the contrary, the proposed methodology can be completed in about half an hour on a low-performance laptop. For the benchmark methodology it is also necessary to use a database to store the large amount of downloaded and processed data; in the proposed methodology the whole procedure can instead be completed without the need to store on disk the irradiation time series and the arrays containing the heights of the horizons.
5. Discussion
The proposed methodology produces reliable results when compared with a state-of-the-art and widely used technique, and furthermore allows for a rapid and effective assessment of the complete census photovoltaic potential in urban areas.
The procedure has been applied to the case of the city of Cagliari, which is of particular interest for the projects in which this work is included. A direct comparison of the results with those obtained in other cities is clearly difficult, due to the differences in the type of buildings, the geographical position and the meteorological peculiarities of the site. The most directly comparable data is that relating to the rooftop surface actually exploitable.
The production potential for the city of Seoul was analyzed in [
7], where the estimation of the rooftop surface available for the installation of photovoltaic modules was based on cadastral data and usage coefficients; moreover, the effect of the panel inclination on a flat surface was evaluated. The authors estimated a surface actually available for the installation of photovoltaic panels equal to about 38% of the available surface in case of zero slope and equal to 21% for an inclination of
corresponding to the optimal inclination for an isolated installation.
In [
12], the generation potential for the Gangnam District of Seoul is evaluated including the effects of shading and estimating an overall availability of rooftop surfaces suitable for the installation of photovoltaic panels equal to the 66% of the total rooftop areas.
In [
14] the production potential for several cities in the United States is evaluated by including the effects of shading and roof orientation: the 32% of the total building rooftop area is estimated as suitable for the installation of photovoltaic modules.
In the case of the recent Europe-wide study in [
5], the radiation processing was carried out on a 100 m resolution raster grid in order to evaluate the technical and economic potential of photovoltaic production for the whole European territory. In this work, the portion of rooftop surface actually available for the installation of panels is estimated to be 60% of the total.
According to our calculations for the city of Cagliari, we would obtain a surface area suitable for the installation of photovoltaic panels equal to about 60% of the total surface area of the building rooftops, in line with the results available in the literature, if it would be economically advantageous to settle for a production equal to 95% of the average value of production for that area. In reality, the suitable surface area should be equal to almost 80% of the total rooftop area if it were sufficiently profitable to have a production equal to about 80% of the average production, as shown in
Figure 12.