1. Introduction
The world population is estimated to reach nearly 8.1 billion in 2040, with an average global economic growth of 3.4% per year. As a result, energy scenarios forecast a remarkable increase in global energy demand of about 30% by 2040 [
1]. In line with the ratified agreements for worldwide sustainable development [
2], a significant step up of renewables over conventional fossil fuels is expected in the global energy mix. The average net capacity addition per year of renewable energy is foreseen to grow steadily in the near future, led mainly by electricity production via solar photovoltaic (PV) conversion (see
Figure 1a).
In recent years, the PV market in Europe has experienced a tremendous expansion, owing to the EU directive 2009/28/EC on the promotion of energy from renewable sources. Such initiative aims at fulfilling
of the overall energy needs of Europe with renewables by 2020 [
4], and
by 2030 [
5]. Considering that building-integrated installations have the advantage to preserve natural landscape with respect to large free-standing installations (
PV farms), they have been promoted by local governments via specific policies, which led to a continuously increasing number of installations [
6]. On the other hand, despite the outstanding potential in terms of solar energy and thus of sustainable electricity production [
7], investments are still at the beginning in Africa [
8]. In this sense, proper economic arguments on this potential would help to deploy investments, with beneficial effects particularly in rural areas, where no connection to the electric power grid is available. Photovoltaic systems could also favor the democratization of the energy availability [
9], especially in Sub-Saharan Africa [
10]. To promote and drive investments in Europe and Africa in this rapidly expanding market, one key aspect is the availability of fast and reliable screening tools for estimating: (i) the intensity of the available solar resource, which depends on the geographical location and climate; (ii) the potential energy output, which depends on the PV module technology, installation and local ambient conditions.
Methods for estimating the annual energy yield of PV systems can be classified into direct and indirect approaches [
11]. The former evaluates the electrical energy output directly from the solar irradiation (i.e., insolation), whereas the latter obtains the energy output from the solar irradiation, ambient temperature, and some additional ancillary parameters [
3,
12]. The common starting point of these methods is a reliable measure or estimate of the available solar potential for specific geographical locations. In general, solar irradiation is best obtained via experimental measures by transducers (e.g., pyranometers and pyrheliometers in dedicated stations). However, this implies installation and monitoring difficulties in remote areas, especially those characterized by poorly developed technological access. Furthermore, such measurement stations are not suitable for collecting data in large-scale territories, because of their high capital and operating costs.
For this reason, a growing number of (online) tools and databases with solar irradiation data have been developed lately, for instance PVWatts [
13], PVGIS [
14], Global Atlas [
15] or Solargis [
16]. In particular, PVWatts is a web application implemented by the National Renewable Energy Laboratory (NREL) that can be used to predict the electricity generation from PV systems given the geographical position of the installation site and some technical details of the plant (e.g., size, module type, array type, losses, tilt, azimuth). This tool has been developed to be easily accessible and usable by both non-experts and advanced users. The prediction errors of the PVWatts model are claimed to be in the range
for annual energy totals [
13]. PVGIS, instead, is an online calculator of potential electricity production by PV systems developed by the Joint Research Center (JRC) [
17]. In addition to the annual electricity production, PVGIS is also capable to provide estimations of the monthly and hourly ones. The required inputs are again the geographical location of the PV system and its technological characteristics. The database used by PVGIS for solar radiation in Europe and Africa (see e.g., the total in-plane solar irradiation in
Figure 1b) demonstrated a mean bias error of about
against satellite data [
17].
However, such online tools sometimes provide low-resolution data (i.e., in the range of only a few kilometers), which may be scattered and require multiple queries to the online databases in case of large-scale analysis. Thus, various models have been developed for forecasting the local solar resource using only the most widespread climatic data [
18]. Both empirical and non-empirical models have been implemented for estimating monthly, daily, and hourly global solar irradiation [
19,
20]. While short-term estimations methods deserve greater attention [
21,
22] for dealing with the conversion efficiency [
3,
23], long-term estimations are essential to analyze possible investment scenarios with respect to the PV solar energy potential [
24,
25]. These can be obtained using either data pattern analysis [
26,
27] or physical modeling approaches [
28,
29].
In this work, we propose a simple yet effective model to estimate the yearly solar irradiation per unit area, only taking into account the latitude, altitude, and average temperature of a certain location as input parameters. The yearly PV potential output is then obtained considering the different climatic and technological factors affecting electricity production, namely the temperature, reflection, module, and installation efficiencies, as well as the non-optimal azimuthal angle of the module. This procedure is first formulated and then validated against well-established databases for both free-standing and building-integrated systems. This approach may be particularly beneficial when only scattered data of solar irradiation are available over large territories but high-resolution analysis would be required. Furthermore, the presented methodology—which is now implemented for Europe and Africa—does not require querying online databases and thus has the capability to operate offline.
The layout of this work is as follows. The proposed models for the yearly sum of global irradiation and PV energy potential are introduced in
Section 2. The parametrization of these models for Europe and Africa is reported in
Section 3, together with a comparison of the estimated solar irradiation and electricity generation by PV systems with those of a well-established online database, using standard error metrics. In
Section 4, the final conclusions are drawn and an outlook on the perspective developments and applications of the present work are given.
2. Methodology
The aim of the proposed methodology is to provide: (i) a fast means for evaluating the global energy irradiation in locations where no or scattered data are available; (ii) a fast and easy-to-use screening tool for the preliminary assessment of the perspective PV energy yield over large territories. As sketched in
Figure 2, the approach relies on three basic and easily accessible data types, namely the latitude (
), altitude above sea level (
h), and average daily ambient temperature (
) for a given site. We remark that, for the purpose of the analysis, the average daytime temperature would be a more suitable parameter as it is most closely related to sunlight; however, this information is generally less accessible. Therefore, we use
values because they are more easily retrievable for a large number of sites. Clearly, this methodology takes into account local weather conditions only implicitly, since wind, clouds, humidity, fog, pollution, or other environmental conditions indirectly affect the global energy irradiation estimate through the average daytime temperature. Additionally, system-dependent factors such as installation type (free-standing or building-integrated), azimuth of the module respect to the optimal orientation (
), solar exposed surface (
), PV module (
), and installation (
) efficiencies are also required for the final estimation of the yearly PV energy output. The multiple-regression method allowed the correlation of these inputs with the yearly solar irradiation (
) and PV energy output (
) for the considered installation.
First, based on the three selected geodata types, the following general equation is found to provide an accurate estimation of the average yearly sum of global solar irradiation:
In the above equation, the latitude
is considered in absolute value, meaning that the dependence of
on this parameter is symmetrical with respect to the equator. The model also reports that the yearly sum of global solar irradiation depends linearly on the altitude of the selected location and quadratically on the average temperature. This evidence is physically meaningful, since insolation has been repeatedly found to rely on latitude [
31], altitude [
32], and climate conditions (e.g., average temperature) [
33]; whereas, no effect of longitude (
) is typically noticed. Finally,
(
) are the weights (or coefficients) of the polynomial model, which will be fitted in
Section 3 over an extensive set of geographic locations throughout Europe and Africa (see
Figure 3).
Second, to predict the specific electrical output by a PV system at a given location, one must consider the various climatic and technological losses and the corresponding efficiencies . In particular, we consider: (i) the efficiency resulting from losses due to temperature and low-irradiance effects ; (ii) the efficiency resulting from losses due to angular reflectance effects ; (iii) the efficiency of the module technology itself , and (iv) the additional efficiency resulting from losses due to system installation (e.g., inverter and cables). The final correction factor applied to the available solar insolation is the product of the aforementioned efficiencies, namely
Among the various correlations available in the literature [
34,
35], here the temperature and low-irradiance efficiency is considered to have the following form [
24]:
where the model coefficients
depend on the installation type (free-standing or building-integrated) and PV technology. Similarly, one of the possible correlations [
36] to model the reflectance efficiency is a fourth-order polynomial using the azimuthal angle between the module orientation and either the South (northern hemisphere) or North (southern hemisphere) direction as independent variable [
24]:
Note that the model coefficients
in Equation (
3) and
in Equation (
4) will be obtained in
Section 3 by fitting the data of the geographic locations reported in
Figure 3. The remaining coefficients, namely
and
, depend on user choice for the PV technology and installation, respectively.
One further correction of the PV energy output must be considered when non-optimal azimuthal angles of the PV installation are imposed. In fact,
refers to the yearly sum of global solar irradiation under optimal orientation of the PV module, namely
. Usually, this mounting angle is not always possible (especially in building-integrated installations), thus a correction factor
must be included (note that the tilt angle of the modules may also be considered; however, in this work we assume optimal tilt). This scaling factor for the non-optimal azimuth can be obtained as a function of the azimuthal angle only as [
24]
being
the model coefficients. In the next Section, we will show that proper parametrizations of this model allow to obtain well-representative functions even for large geographical extents.
Finally, the electrical energy output of the PV system is computed as
being
the net exposed surface of the installed modules.
In summary, as schematically depicted in
Figure 2, the proposed methodology for fast estimation of solar irradiation (Equation (
1)) depends directly on geographical data (latitude, altitude) and indirectly on climatic data (average daytime temperature, which is implicitly influenced by wind, clouds, humidity, fog, pollution or other local environmental conditions). Instead, the estimation of photovoltaic energy potential (Equations (
2) and (
6)) relies on both climatic data (average daytime temperature, Equation (
3)) and installation characteristics of the considered PV system (photovoltaic technology, Equation (
2); azimuth, Equations (
4) and (
5); installation type—that is building-integrated or free-standing).
3. Results
In this Section, the proposed methodology is tuned, validated, and then adopted to provide possible scenarios of the yearly PV production over large territories in Europe and Africa. First, the yearly sum of global solar irradiation is obtained according to Equation (
1), which requires as inputs the latitude
, the altitude
h and the mean temperature
for a given geographical location. Then, the temperature and reflectance efficiencies are computed according to Equations (
3) and (
4) for optimal tilt and azimuthal angles of the modules. These efficiencies are then combined with the user-defined efficiencies
and
, to compute the total efficiency
according to Equation (
2). The yearly electrical energy output is eventually estimated via Equation (
6), based on the net surface of the installed PV modules. The analysis is finally carried out also for module installations that present non-optimal azimuthal angles, by considering the correction factor
.
3.1. Solar Irradiation Model
The model coefficients for Equation (
1) are fitted to a random set of 80 different geographical locations in Europe and Africa, see
Figure 3 and
Table A1 and
Table A2, with latitudes covering a range from about 60
to −30
. Note that the exact geographical coordinates have been chosen based on the availability of the required data, so they do not generally coincide with the center of the indicated cities. These cities are indeed indicated as the most significant nearby location to the considered geographical point. The altitude, average temperature in the 24 h and the yearly sum of global solar irradiation for each location are extracted from the online database PVGIS [
37] of the Joint Research Centre [
17,
38,
39]. The solar irradiation database of PVGIS is based on data records of the Satellite Application Facility on Climate Monitoring (CM-SAF) [
40]. The average local temperature (
) is provided by the interactive map, which covers Europe and part of the Northern Africa. All the temperature data for the locations in Africa with latitude lower than 32.5
are obtained from Berkeley Earth™ [
41]. The
values are extracted from the PVGIS database considering both optimal azimuthal angle and optimal tilt angle.
The full set of data for these 80 random locations is split in two different subsets, namely a training and a validation set. The model coefficients of Equation (
1) are initially fitted on the training data set; then, the best-fitted model is used to predict the responses for the observations in the second data set (the validation one), which provides an unbiased evaluation of the model fitting. Here, the training data set is taken as 70% (i.e., 56 locations) of the whole data set; whereas, the validation one includes 24 locations. The training/validation process is iteratively repeated 10,000 times, with a random distribution of locations in the two subsets, and standard error metrics are computed with respect to the
values provided by the PVGIS tool [
42]. The distribution of the mean absolute percent error (MAPE) for the 10,000 repetitions is reported in
Figure 4a, for both the training and the validation steps. The MAPE distribution shows a Gaussian shape, with average values equal to 4.3% (training) and 4.9% (validation).
The fitted coefficients of Equation (
1) per each training step are averaged over the 10,000 repetitions, and the resulting mean values and standard deviations are reported in
Table 1, where units for
and
are indeed given per degree of latitude (the unit for latitude is the decimal degree, as recommended by ISO 31, instead of using degrees, minutes and seconds [
43]). These optimal coefficients are then used to predict
for the 80 considered locations in Europe and Africa. The predicted global sum of solar irradiation is compared to that provided by the PVGIS tool, and the relative percent error is reported in the last column of
Table A1 and
Table A2.
Figure 4b shows a graphical comparison between the
values obtained by the proposed model (Equation (
1) with the optimal coefficients in
Table 1) and those given by the PVGIS tool for all the 80 considered locations. The difference in the global irradiation for Europe (from about 60
to 37
latitude) and Africa (from about 37
to −30
latitude) can be clearly appreciated; nevertheless, no systematic errors with the latitude are noticeable. Results show that the maximum relative error of the model prediction is 13.6% in absolute value, which is within the standard range of estimation errors for the yearly global irradiation. For instance, Solargis [
44] reports a data accuracy in the range of ±4% to ±8% for the global horizontal irradiation, and ±8% to ±15% for the direct normal irradiation [
45]. The inset of
Figure 4b, presents the cumulative distribution function of the percent error between the predicted and the PVGIS values of
. The MAPE is 4.4% and the normalized root mean square error (NRMSE) in percentage is 5.5%. These errors are similar to the ones typically obtained in other solar irradiation models from the literature. For example, methodologies based on artificial neural networks or inverse distance weighting algorithm have shown mean absolute percent errors equal to
,
and
at a nationwide level (Malaysia [
46], Indonesia [
47], and Taiwan [
48], respectively). Notably, such discrepancies are all consistent with investors requests, namely average errors within 5% in prediction accuracy [
49].
This evidence shows that the number of chosen locations is sufficient to train adequate values of () coefficients and, therefore, that an accurately representative model for the yearly sum of global solar irradiation in Europe and Africa has been obtained.
3.2. PV Energy Model (Optimal Azimuth)
The proposed methodology is now applied to estimate the potential electrical energy output by PV system installations for the whole set of considered locations. Initially, and thus are computed considering installations with optimal tilt and azimuth, namely and thus , and crystalline silicon solar cells. The methodology is implemented in MATLAB®, resulting in a simple and fast algorithm that can be straightforwardly applied for the analysis of large data sets.
First, the model coefficients for the temperature and low-irradiance efficiency in Equation (
3) are obtained by fitting the values extracted from PVGIS tool over the 80 locations in Europe and Africa. Best-fitted values for free-standing and building-integrated installations are listed in
Table 2. The coefficients of determination of these fittings, that is R
2 = 0.92 (free-standing) and R
2 = 0.90 (building-integrated), demonstrate that Equation (
3) is an accurate correlation between
and
, as also evident in
Figure 5a. Second, the reflectance efficiency of Equation (
4) is fitted on the 80 European and African locations as well. Since, in this case, we consider optimal azimuthal angle, Equation (
4) simplifies to
, being
the best-fitted value. As reported in
Figure 5b, the
values from the PVGIS tool range from 0.967 to 0.978, namely within a
from
. In fact,
demonstrates only a slight dependence on latitude.
The
values are finally computed by Equation (
6), considering an indicative efficiency for the crystalline silicon solar cells equal to
[
50], a system installation efficiency equal to
, and a unitary surface exposed to the sun. The predicted
values for the 80 locations in Europe and Africa are compared to those provided by PVGIS tool, and the relative percent error is reported in
Figure 6a,b (free-standing installations), and in
Figure 6c,d (BI installations), respectively. Results show that the maximum relative error for FS installations is 14.8% in absolute value, MAPE is 4.3% and NRMSE is 5.5%. Similar errors are obtained for BI solutions, namely a maximum relative error equal to 15.4% in absolute value, MAPE equal to 4.3% and NRMSE to 5.5%. These values are essentially coherent with the prediction errors of
discussed in the previous Section, which therefore represent the main source of uncertainty in the
estimation. These uncertainties are consistent with the general accuracy provided by other tools for estimating the PV potential throughout large territories. For example, PVWatts tool shows errors as high as
for annual PV electricity production, with actual performance in a specific year up to
respect to long-term average [
13]. These uncertainties are also similar to those indicated by JRC (i.e., from
to
) when the effect of irradiation and temperature on PV module performance are both considered [
3].
3.3. Effect of Non-Optimal Azimuth
In this Section we analyze the effect of non-optimal azimuthal angles on the reflectance efficiency
of Equation (
4) and on the scaling factor
of Equation (
5) for various geographical locations in Europe and Africa. The locations are specifically selected to be representative of the whole extent of each continent, namely maximum, minimum, and intermediate latitudes. Using this criterion, the following spots are chosen from the data set: St. Petersburg (Russia), Edinburgh (UK), Rome (Italy) and Sevilla (Spain) for Europe; Rabat (Morocco), Aswan (Egypt), Karthoum (Sudan) and Kampala (Uganda) for Africa. Note that the selected spots also differ significantly in longitude; however, no appreciable effect of this latter parameter has been noticed in the analysis. The raw data for the reflectance efficiency and insolation is obtained for each location using standard queries to the online PVGIS tool, for azimuthal angles varying in the range
.
The results are shown in
Figure 7a,b, where we also compare the solution given by Equations (
4) and (
5), respectively. The parametrization of these two latter equations was previously obtained for Italy, using the raw data of seven cities throughout the peninsula for the reflectance efficiency and the only data of Rome for the scaling factor [
24]. We note that the parametrization for Italy gives well-representative curves also for the other locations in Europe: the maximum relative error is 2.1% on the reflectance efficiency and 13.8% on the scaling factor at the maximum azimuthal angles considered (
). We can then assert that, given the geographical location of Italy at intermediate continental latitudes, this parametrization can be assumed to be representative also for Europe—and the coefficients are reported in
Table 3. On the other hand, the results for Africa require a specific treatment, as the decrease in latitude towards the equator significantly modifies both the reflectance efficiency and the scaling factor with respect to those of the European spots. We then proceed with a new parametrization of the models by regression, using the data of the four considered locations in Africa. We obtain the coefficients reported in
Table 3. In this case, the maximum relative error is 0.3% on the reflectance efficiency and 6.4% on the scaling factor at the maximum azimuthal angles. Considering that
is the most unfavorable and limiting situation for an installation, the proposed models provide acceptable approximations for the reflectance efficiency and scaling factor. In particular, according to the results obtained, the parametrization for Europe can be adopted in the range of latitudes
, while that for Africa from
to the equator (
). Symmetry applies for locations in Africa below the equator (negative latitudes).
An example application of the proposed models and related analysis is reported in
Figure 8a,b for a BI installation in the city of Rome (Italy), which has azimuthal angle
,
for crystalline silicon solar cells [
50],
and unitary exposed surface. The main contribution to the total losses is given by the module efficiency, then by temperature and installation losses. Note that the losses due to reflectance effects and azimuthal angle contribute only for a small share on the total losses. For Europe, a similarly small contribution of the reflectance losses would be obtained for any azimuthal angle, as they consist in any case of a few percent (see
Figure 7a). The contribution of losses due to the azimuthal angle can be expected to be more consistent for large azimuthal angles, e.g., for
, where losses could reach nearly 10% (see
Figure 7b). For Africa, both contributions to the total losses tend to be smaller for any azimuth.
4. Concluding Remarks
In this work, we have proposed a simple methodology for a fast estimation of the yearly sum of solar irradiation in Europe and Africa and the resulting potential PV energy output. The model thus generated is intended to provide: (i) high spatial (continuous) resolution and (ii) a fast means for pre-screening of the PV potential over large-scale territories for any investor. In fact, the presented method relies on just a small set of input parameters. For instance, estimation of the yearly sum of global solar irradiation requires only the knowledge of a few basic geodata, such as latitude, altitude, and average daily temperature. One such tool operates offline and provides a much faster screening than systematic calls to online databases, and can be easily implemented with just a few lines of code.
Limitations of this work include the lack of information at local level, for instance, about the shading effect of buildings or other elements nearby, as well as horizon height. Thus, the reported tool should be considered as a fast (macroscopic) screening of the solar and PV potentials over large-scale territories with tolerable errors in the range
, which should then be followed by a more accurate (microscopic) analysis considering also local elements for a limited set of locations. Furthermore, both the solar irradiation and the potential photovoltaic electricity generation are here estimated on an annual basis, namely they consider the cumulative value over the average year. The reason is that the yearly electricity output is a standard figure of merit needed to assess the techno-economic feasibility of a possible photovoltaic system in a given location [
31]. Clearly, this information would not be enough to size the whole photovoltaic system, since detailed estimations of daily and hourly variability are required to design, for example, the energy storage system [
51]. In perspective, our multiple-regression model could be further improved to consider daily variability as well, for instance by including well-established models from the literature [
52].
This approach may find application in the design of micro-grids [
53], polygeneration systems [
54], BI PV solutions [
55] or flexible storage photovoltaic systems [
56]. Furthermore, this methodology could also support the large-scale potential assessment of sustainable technologies other than photovoltaics, for example, solar thermal systems [
57], solar greenhouses or dryers for food production or conservation [
58], and desalination plants driven by solar source [
59].