Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Applied Energy 339 (2023) 120981

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Direct and diffuse shading factors modelling for the most representative
agrivoltaic system layouts
Sebastian Zainali a, *, Silvia Ma Lu a, Bengt Stridh a, Anders Avelin a, Stefano Amaducci b,
Michele Colauzzi b, Pietro Elia Campana a, *
a
Mälardalen University, Dept. of Sustainable Energy Systems, Box 883, 721 23 Västerås, Sweden
b
Università Cattolica del Sacro Cuore, Dept. of Sustainable Crop Production, Emilia Parmense 84, 291 22 Piacenza, Italy

H I G H L I G H T S

• Shadings at crop level under different agrivoltaic system configurations are modelled.
• Photosynthetically active radiation (PAR) distribution at the crop level is modelled.
• The agrivoltaic configurations showed a variation of light homogeneity from 86 % to 95 %.
• Yearly cumulative PAR reduction varies from 11 % to 21 %.
• The two-axis agrivoltaic system showed the highest light homogeneity and lowest yearly PAR reduction at one location.

A R T I C L E I N F O A B S T R A C T

Keywords: Agrivoltaic systems are becoming increasingly popular as a crucial technology for attaining multiple sustainable
Agrivoltaics development goals, such as affordable and clean energy, zero hunger, clean water and sanitation, and climate
Beam Shading Factor action. However, a comprehensive understanding of the shading effects on crops is essential for choosing an
Diffuse Shading Factor
optimal agrivoltaic system, as an incorrect choice can result in significant crop yield reductions. In this study,
Photosynthetically Active Radiation
fixed vertical, one-axis tracking, and two-axis tracking photovoltaic arrays were developed for agrivoltaic ap­
Photovoltaics
Tracking plications to analyse the shading conditions on the ground used for crop production. The models demonstrated
remarkable accuracy in comparison to commercial software such as PVsyst® and SketchUp®. These models will
help to reduce crop yield uncertainty under agrivoltaic systems by providing accurate photosynthetically active
radiation distribution at the crop level. The photosynthetically active radiation distribution was further analysed
using a light homogeneity index, and the results showed that homogeneity and photosynthetically active radi­
ation reduction varied significantly depending on the agrivoltaic system design, ranging from 86% to 95%, and
11% to 22%, respectively. Studying the effect of shading with distribution analysis is crucial for identifying the
most suitable agrivoltaic system layout for specific crops and geographical locations.

of shading on the field’s design. The shadow is the most critical


parameter in PV designs and its shape is dependent on dimensional
1. Introduction parameters such as collector height, spacing between collectors, row
length, and area. It is common to have more complex situations than a
Shadings on photovoltaic (PV) modules cause significant losses in field of solar collectors. Hence, Cascone et al. [3] developed a calcula­
electricity production, especially in cases where there are no diodes as tion procedure for calculating the shading factors under complex
the performance is decreased approximately proportionally to the irra­ boundary and sky condition. The procedure can obtain the instanta­
diance reduction [1]. Hence, research has focused on developing algo­ neous, daily average, or monthly average shading factor values.
rithms to accurately estimate energy losses produced by shadings. In Nowadays, several commercial software solutions are used to calculate
1987, Bany and Appelbaum [2] developed algorithms to calculate the the shading factors, such as Aurora®, System Advisor Model (SAM)®,
shadow on a field of solar collectors during the day and assess the effect

* Corresponding authors.
E-mail addresses: sebastian.zainali@mdu.se (S. Zainali), pietro.campana@mdu.se (P.E. Campana).

https://doi.org/10.1016/j.apenergy.2023.120981
Received 9 August 2022; Received in revised form 17 February 2023; Accepted 13 March 2023
Available online 31 March 2023
0306-2619/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
S. Zainali et al. Applied Energy 339 (2023) 120981

Nomenclature(List of symbols) PAR Photosynthetically active radiation


PARb,S Shaded beam PAR (W/m2)
AST Apparent solar time (h) PARd,S Shaded diffuse PAR (W/m2)
Ashade Shaded area obtained by the projection given on the plane Rαγ Radiance (W/m2sr)
(m2) s Solar vector (-)
Ashade,i Shaded area from each respective PV module (m2) SF Shaded fraction (-)
Atot Total reference area for the crop between two PV module SL Shadow length (m)
rows (m2) W Width of the panel (m)
BHIS Shaded beam horizontal irradiance (W/m2) xi Total PAR including shadings from the agrivoltaic system
DHIS Shaded diffuse horizontal irradiance (W/m2) (W/m2)
Eext Extraterrestrial irradiance (W/m2) x Average yearly PAR (W/m2)
fb Beam shading factor (-) αZ Angle to adjust system deviations for different azimuth
fd Diffuse shading factor (-) angles (◦ )
Gcs Clear sky global horizontal irradiance (W/m2) β Solar altitude angle (◦ )
Is Total diffuse irradiance on the ground plane (W/m2) βIT Ideal secondary tracking angle (◦ )
IH Diffuse irradiance that traverses the shaded areas (W/m2) βITC Corrected secondary tracking angle (◦ )
(s)
kd Satellite-derived diffuse fraction (-) βZ Axis tilt from the horizontal plane (◦ )
kt Clearness index (-) γ Azimuth angle (◦ )
kd Diffuse fraction (-) θ Angle of incidence (◦ )
L Length of the panel (m) Ω Solid angle (sr)
LHI Light homogeneity index (-) ωIT Ideal primary tracking angle (◦ )
LEW Distance between the axes from east to west (m) ωITC Corrected primary tracking angle (◦ )
LNS Distance between the axes from north to south (m) ωC Back-tracking correction angle (◦ )
n Normal vector (-) δ Orientation angle (◦ )
P PV module coordinates (m) σ Tilt angle (◦ )
P’ Projection onto the plane along a straight line (m)

SketchUp®, and PVsyst®, to name a few [4,5]. However, these products at a cost in computational time. Asgharzadeh et al. [15], benchmarked
often have limitations, such as a lack of control over the timestep or four bifacial irradiance models with ray-tracing and view factors. In
angular displacement needed for shading factor calculations. To provide their study, the ray-tracing simulation dependent on the level of accu­
more flexibility and control over the shading factor calculation process, racy of the modelled 3D scene had approximately 104 to 105 greater
Silva et al. [6] developed a model in Matlab® that calculates the beam computational time than the other models.
and diffuse shading factors at 5-minute timesteps rather than hourly. In Shading factors are used frequently for PV systems’ performance
addition, their model can calculate the shading factor with any timestep assessment. Nevertheless, more research still needs to be conducted on
and various degrees of shading discretisation resolutions. The model new emerging PV systems’ layouts, especially for agrivoltaic (AV) ap­
could potentially improve the shading calculations capabilities of cur­ plications. AV systems can increase renewable energy capacity and
rent open-source libraries like PVLIB [7]. Their model was validated produce food simultaneously [17]. Adeh et al. [18] estimated that
with PVsyst® and SAM® using obstacles shading the fixed tilted PV converting less than 1 % cropland into AV systems can offset the global
array. Their model has shown similar results as SAM®. However, the energy demand. AV systems also show significant synergies with live­
PVsyst® comparison result differed significantly from their model, with stock raising [19]. Additionally, there are several mutual benefits, such
an R2 of 46 % when analysing one module shaded by one infinite row. as reduced PV module heat stress, reduction of plant drought stress, and
The shading factor simulation over a long period requires high compu­ increased food production under certain climatic conditions [20].
tational power. To reduce the time complexity, Melo et al. [8] developed Despite all the benefits, a tremendous lack of knowledge exists on how
a method to estimate the shading factor and irradiation received on a the properties of the crop (i.e., yield, energy content, and morphology)
surface using bi-linear interpolation, which reduced the simulation time underneath the PV modules are affected due to shadings [20–22].
by 95.9 % without sacrificing accuracy. The view factor approach is Therefore, accurate shading calculations on the ground are necessary to
another frequently used technique for modelling incident solar irradi­ understand better the effects of combining PV systems and agriculture.
ance [9]. The view factor for beam radiation is defined as cos(θ)/cos(Z), Campana et al. [23] analysed one vertically mounted AV system located
where θ is the incidence angle (◦ ) and Z is the zenith angle (◦ ) [10]. in Sweden. In their study, diffuse and beam shading factors are calcu­
Instead of calculating the PV geometry’s projected area on the ground as lated to get the photosynthetically active radiation (PAR) distribution on
done for shading factors, the view factors from each ground gridded cell the ground. The PAR is needed for the crop model Environmental Policy
to the sky must be calculated for each object blocking the sky portions Integrated Climate (EPIC) developed by Williams et al. [24] to calculate
[11]. There are several view factor modelling approaches to accurately the actual crop yield. The model is robust in identifying crop produc­
estimate the diffuse radiation reaching the PV modules, such as Liu- tivity, especially in open-field conditions [23]. Amaducci et al. [25]
Jordan’s [12], Badescu’s [13], and Tian’s [14]. An emerging technique developed a shading and radiation model to compute the crop yield for
to model incident solar irradiance is ray-tracing, due to its capability of an AV system using two-axis trackers in North Italy. Their model in­
modelling complicated 3D scenes [15]. The ray-tracing approach has dicates that AV systems support clean energy conversion, crop yield, and
gained increased interest as the need for accurately estimating the rear water saving. Nevertheless, there is still a lack of studies analysing the
irradiance for the emerging bifacial PV modules. Pelaez et al. [16] shading that occurs under the panels of an AV system. This knowledge
analysed the impact of a torque tube behind a one-axis module row and gap needs to be filled.
found that the primary peak in shading loss directly behind the tube In this study, a model is developed to accurately calculate the PAR
resulted in a rear irradiance reduction of 15 % to 20 %. On the other distribution reaching the crops for fixed, one-axis, and two-axis AV
hand, accurate irradiance estimation for complex 3D geometries comes systems. Most current studies on shadings for PV system assessment

2
S. Zainali et al. Applied Energy 339 (2023) 120981

have focused on shading calculations on PV modules. However, this DHIS = DHI(1 − fd ), (2)
study develops shading calculations at ground level, which is necessary
for accurate crop yield estimations in AV systems. This study bench­ where, fb is the beam shading factor, fd is the diffuse shading factor, BHIS
marks three standard AV designs to show the variation of PAR depen­ is the shaded beam horizontal irradiance, and DHIS is the shaded diffuse
dent on design choice, one of the main contributions of this study. horizontal irradiance. The input parameters used for the vertical, one-
Another contribution is the analysis of PAR distribution at crop level, axis, and two-axis AV system designs are presented in Table 1. Due to
including a related decomposition model integration, since other studies construction limitations, the one-axis and two-axis maximum tilt angles
have typically focused only on global horizontal irradiance (GHI) dis­ are assumed to be − 60◦ and + 60◦ .
tribution. The developed model can be used as a starting point for The projection calculation procedure for the panel coordinates with
accurately estimating crop yields at any location under the most com­ different tilt angles for all three AV systems is presented in Appendix A.
mon AV system designs. It considers the agricultural part of AV systems In Appendix B, an extended analysis with additional design scenarios
and can be integrated with existing PV power models to analyse crop has been considered.
yield and power production. In AV systems equipped with one- or two-
axis trackers, controlling the irradiance on the crops grown underneath 2.1. Beam shading factor
the PV system is common [26]. Therefore, the developed model can be
used to assess PAR distribution at the crop level and as a starting point The projected coordinates on the ground plane are used to calculate
for managing PAR. The paper is structured as follows. Section 2 de­ the beam shading factor as follows:
scribes the methodology used to calculate the beam and diffuse shading
factors for the fixed, one-axis, and two-axis AV systems. Section 3 pre­ fb =
Ashade
, (3)
sents the model comparisons and PAR distribution at different locations. Atot
Section 4 summarises the results of this study.
where Ashade is the shaded area obtained by the projection given on the
2. Methodology plane (m2) and Atot (m2) is the total reference area for the crop between
two PV module rows. In this study, the area between the PV rows (i.e.,
The schematic of the modelling chain of this study is presented in the crop area as defined in Table 1) is the reference area. In a case with
Fig. 1. The shading factors are first calculated with our developed multiple PV modules, there is a chance of overlapping shadings. To
Matlab® model. Secondly, the shading factors are validated with two
commercial products, PVsyst® and SketchUp®. The developed model is Table 1
then tested by using three ICOS weather stations in Europe to gather Design parameters for the vertical, one-axis, and two-axis AV systems.
PAR data. The PAR data is decomposed to beam and diffuse PAR using a Input parameters Vertical One-axis Two-axis
decomposition model. At last, the model is presented in distribution Panel width [m] 1 1 1
plots at these locations. Panel length [m] 2 2 2
The shadow effect of the solar modules includes direct and diffuse Number of panels 40 40 40
shading factors. The direct component depends on the module geome­ Total panel area [m2] 80 80 80
Number of rows 2 2 2
try, obstacles, and the sun’s position. The diffuse component only de­ Row spacing [m] 10 10 10
pends on module geometry and obstacles. There is also a reflected Row length [m] 20 20 20
component reduced by obstacles, reducing the albedo [6]. The albedo Crop area [m2] 200 200 200
effect is not included in this work. The shading factors are used to Pitch [m] – – 2
Height [m] 0 3 3
calculate the shaded beam horizontal irradiance (BHIS), and the shaded
Fixed tilt angle [◦ ] 90 – –
diffuse horizontal irradiance (DHIS) as follows: Azimuth angle [◦ ] − 90 − 90 − 90
Maximum tilt angle [◦ ] 60 60
(1)

BHIS = BHI(1 − fb ),
Minimum tilt angle [◦ ] – − 60 − 60

Fig. 1. Modelling schematic.

3
S. Zainali et al. Applied Energy 339 (2023) 120981

avoid calculating overlapping shading on the ground, a union summa­ 2.3. Fixed vertical PV
tion is used as follows [1]:
The fixed vertical AV system PV panel geometry is defined by four

n
Ashade = Ashade,i , (4) points Pi in the plane corresponding to the vertexes of the PV module or
i=1 array. The projections of those vertices on the horizontal plane are used
to calculate the shaded area on the ground at a given time step. It is
where, n is the number of PV modules, and Ashade,i is the shaded area assumed that grass growing close to the supporting structure does not
from each respective PV module. allow sunlight to pass through, similar to Campana et al. [23]. An
illustration of the fixed vertical AV system is depicted in Fig. 2.
2.2. Diffuse shading factor

2.4. One-axis and two-axis tracking


The diffuse shading factor fd is constant for fixed geometries. How­
ever, in the case of one- and two-axis tracking systems, the diffuse
In this study, one-axis and two-axis tracking and back-tracking an­
shading factor has to be determined for each timestep [6]. The diffuse
gles have been calculated using similar algorithms as presented by
irradiance can be calculated from the total diffuse irradiance on the
Lorenzo et al.[28]. The coordinates for the PV panel can be defined as
ground plane Is and the amount of the diffuse irradiance that traverses
described previously for the fixed vertical PV. An illustration of the one-
the shaded areas IH of the hemisphere and is given by [3,23,27]:
axis and two-axis tracking systems used in this study is presented in
fd =
IS
, (5) Fig. 3 and Fig. 4, respectively.
IH
2.5. Model comparison
where IS can be obtained by integrating the surface over the shaded
areas on the hemisphere. The diffuse shading factor integration is then
The commercial software PVSyst® and SketchUp® are used to vali­
rewritten as follows:
date the shading factors of the model developed in this study. The
∫ π2 ∫ 2π
f R cosθdΩ commercial software PVsyst® does not have any standard way of
β=0 γ=0 b βγ
fd = ∫ π ∫ 2π , (6)
2
R cosθdΩ
β=0 γ=0 βγ

where β and γ are the sun altitude and azimuth angles, Rαγ is the radiance
(W/m2sr), θ is the angle of incidence (◦ ), which is the angle between the
sky element and the normal to the surface, and Ω is the solid angle (sr).
The solid angle can also be expressed as ∂Ω = cos(α)⋅∂α∂γ. The angle of
incidence is used to only account for the radiance component normal to
the ground. To simplify the diffuse shading factor calculation and
remove the radiance Rαγ , an isotropic sky must be considered which
means that the radiance Rαγ is constant for all sky directions. A shading
table for the direct component is used to discretize the sky dome with
altitude angles from 0◦ to 90◦ and azimuth angles from − 180◦ to + 180◦
of 1◦ . The discretized sky dome can be numerically computed as follows
[6]:
∑91 ∑361 ( )
j=1 fbij cos θij cos(αi )
(7)
i=1
fd = ∑91 ∑361 ( ) ,
i=1 j=1 cos θij cos(αi )

Fig. 3. One-axis tracking AV system.

Fig. 2. Vertical fixed AV system. Fig. 4. Two-axis tracking AV system.

4
S. Zainali et al. Applied Energy 339 (2023) 120981

Fig. 5. Schematics using PVsyst® shading scene of the three AV systems: vertical (left), one-axis tracking (middle), and two-axis tracking (right).

calculating beam and diffuse shading factors on the ground. It mainly downtown building in Montreal, the authors suggested to use PVsyst®
focuses on conventional PV systems assessment and thus calculates due to the accurate electricity production estimates. The software has a
shading factors only on PV arrays. Similarly, there is no standard way to 3D shading scene to analyse the shadings from a specific PV system
extrapolate the shading factors from the simulations of the shadings design. This study used the 3D shading scene to calculate both the
from SketchUp®. Therefore, we have used the 3D shading scene tools of diffuse shading factor and beam shading factor for the three AV con­
the investigated software, and we have manually adjusted the shading figurations. To calculate the shading factors on the ground in PVsyst®
scene for each hour to calculate the shading factors. The shading factors instead of on the PV array as it is conventionally performed, the scene
in PVsyst® are calculated using a flat PV array. Given this limitation for had to be designed in a specific way as the software does not have any
the compared commercial software, we have considered only four AV/ground shading scene solution. The crop area was designed as a
representative days, one for each season (i.e., the equinox of spring, the horizontal PV array, and the vertical and tracker modules were con­
solstice of summer, the equinox of autumn, and the solstice of winter), to verted to shading objects. The shading factors can be calculated hourly
carry out the validation. Otherwise, the validation process would have by re-constructing the shading scene for the trackers to adjust their
been extremely time-consuming. The shading factors comparison refers tracking position. Fig. 5 shows a schematic of the different shading
to Kärrbo Prästgård (Latitude 59.6099◦ N, Longitude 16.5448◦ E, Alti­ scenes in PVsyst®.
tude 20 m), Västerås, Sweden. In this location, the first AV system in
Sweden was installed in 2021. The time resolution for validation is 1- 2.5.2. SketchUp®
hour. The methodology employed to obtain the shading factors from SketchUp® is a 3D geometry modelling tool used by architects, en­
both commercial software is briefly presented in this section. gineers, and designers. The software contains a real-time shadow engine
that can be used, for instance, for sun exposure analysis. In this study,
2.5.1. PVsyst® SketchUp® Make 2017 is used with the TIG Shadow Projector v7.0 [35]
PVsyst® is one of the most powerful software to model and simulate plugin to obtain the amount of shading on a surface for a specific time
PV systems and is designed for architects, researchers, and engineers and location. Melo et al. [8] developed a shading factor model called
[29]. PVsyst® has been validated using several PV systems measured Solar3DBR within SketchUp 8 environment to reduce time spent on long
data in Geneva for over 2–3 years. The predicted power output had an simulation periods and allow easy numerical integration over the sky
annual accuracy of ± 5 % [30]. The estimation accuracy varies dome to obtain the diffuse shading factors. The developed polygon
depending on the location. In high-latitude locations such as Kiruna in projection model was validated with a reduced dimension reference cell.
Sweden, the PVsyst® estimation accuracy was 15 % compared to mea­ A difference of 0.3 % between the measured and simulated daily values
surements [31]. However, in three other locations in Sweden at lower was obtained during a clear sky day. Consequently, the beam shading
latitudes, the accuracy varied between − 2.7 % to 2 %. The power esti­ factors for the three AV configurations are derived. The real-time
mation accuracy does not only vary on geographical location. It depends shadow engine from SketchUp® simulates the sun’s position at a spe­
on partial shading or other influential weather conditions, such as cific time of the year. Therefore, the engine visualizes the shadings on
clouds. In Porto Alegre, located in southern Brazil, PVsyst® has been the ground created by the solar modules of the different AV configura­
used to analyse the difference between the measurements and estima­ tions. Except for the vertical configuration, which is fixed, the geome­
tions on clear-sky and cloudy days. During the entire measurement tries (rotation angles) of the one-axis and two-axis AV systems are
period, a difference of 9 % was noticed. However, during cloudy days manually adjusted at every time step according to the optimized values
the estimations made by PVsyst® were close to measurements with an from the tracking systems. Fig. 6 shows the schematics of the different
error of less than 1 % [32]. In partial shading situations, the PVsyst® had systems in SketchUp® and an example of the results displayed by the
a deviation to the measurements between 4 % and 8 % [33]. Overall, plugin. Concerning the diffuse shading factor, unfortunately, the vali­
PVsyst® shows high accuracy and is, therefore, one of the reference dation could not be performed using SketchUp®.
software for the PV sector today. Siraki et al. [34] compared Ecotect®,
PVsyst® and HOMER® for urban PV applications. By investigating a

Fig. 6. Schematics using SketchUp® Make 2017 and the plugin TIG Shadow Projector of the three investigated AV systems: vertical (left), one-axis tracking (middle),
and two-axis tracking (right). The area of analysis is the grass-coloured rectangle.

5
S. Zainali et al. Applied Energy 339 (2023) 120981

Table 2 creating a non-homogenous PAR/diffuse PAR distribution during the


ICOS stations. day [37]. More details regarding the decomposition model and the
Station Latitude (◦ ) Longitude (◦ ) Elevation (m) calculation procedure can be found in Appendix C. PAR is decomposed
in diffuse PAR using the Yang2 model [38], as the best performing model
Lanna, Sweden 58.33 13.1 75
Estrees-Mons, France 49.87 3.02 85 evaluated by Ma Lu et al. [37]. The Yang2 model, besides measured data
Klingenberg, Germany 50.89 13.52 478 needs GHI and DHI satellite derived data for performing the decompo­
sition. The satellite data is obtained through the Copernicus Atmosphere
Monitoring Service radiation service (CAMS), which has a spatial
2.6. Model testing coverage of − 66◦ to + 66◦ in both latitudes and longitudes and a
temporal resolution of 15-min [39].
The model developed in Matlab® is tested at three different sites in The PAR/diffuse PAR reaching the crops are calculated by rewriting
Europe with all three developed configurations. Table 2 summarises the Eq. (1) and Eq. (2) as follows:
coordinates of the selected sites. These sites were chosen because are
located on cropland. The data on PAR and GHI are collected at these PARb,S = PARb (1 − fb ), (8)
sites from the Integrated Carbon Observation System (ICOS) stations and
refers to year 2018 [36]. PARd,S = PARd (1 − fd ), (9)
The PAR distribution can be calculated using the shading factor
where PARb,S is the shaded beam PARb and PARd,S is the shaded diffuse
calculation procedure. However, as the ground area now is discretised
PARd .
by 0.25 m × 0.25 m, the shading factors for each gridded cell will vary.
In this study, the calculated PAR reaching the crops does not account
Therefore, the shading calculation procedure is conducted on each
for PAR reflections from the PV modules and structures, which can affect
gridded cell for the whole year with hourly temporal resolution. The
the crops’ growth as the incoming PAR increases. PAR reflections are
discretisation makes it possible to analyse the PAR variation between the
studied in a parallel study by Campana et al. [36], where a Monte Carlo
rows with high spatial resolution. The PAR is decomposed into its diffuse
approach is used to assess the reflections. Furthermore, the model has
and direct components. This decomposition of PAR is critical in AV
undergone additional cross-validation with the findings from Tahir and
systems due to the variable shadings caused by the panels on the crops,

Fig. 7. Beam shading factor time series for vertical, one-axis, and two-axis AV systems for four representative days. Comparison between the Matlab® model
developed in this study and the models developed in PVsyst® and SketchUp®.

6
S. Zainali et al. Applied Energy 339 (2023) 120981

Table 3 resolution. To our best knowledge, current commercial software like


Beam shading factor error metrics between Matlab®, PVsyst®, and SketchUp®. PVsyst® does not have the capabilities to calculate shading factors at the
Beam shading factor ground level and does not consider PAR. With interest in AV systems, the
need for shading calculations at ground level will increase tremen­
System PVsyst® SketchUp®
dously. The National Renewable Energy Laboratory (NREL) only
MBE RMSE MBE RMSE recently added spatial albedo and ground irradiance calculation in the
Vertical 0.0008 0.0040 0.0011 0.0039 SAM®. This update strengthens the importance of having accurate
One-axis 0 0.0006 0 0.0012 ground calculations and improved shading analysis for AV systems. The
Two-axis 0.0003 0.0006 0.0026 0.0055
light homogeneity index (LHI) can be used to compare different AV
systems in terms of PAR distribution within the AV field. LHI is defined
Butt’s [40] analysis of spatio-temporal shading in AVs under vertical and as the ratio of sample standard deviation to mean and is given by the
one-axis configurations. The vertical and one-axis model was designed following equation [43]:
with a row spacing of 4 m instead of 10 m, the PV module array are set to ⎛ ⎞
1 m2, and the heights were adjusted to both be 1 m. In their study, ⎜ ∑
1 n
(xi − x)2 ⎟
typical meteorological conditions in Lahore (31.5204◦ N, 74.3587◦ E) ⎜
LHI = 100 • ⎜1 − n− 1 n− 1

⎟, (10)
⎝ x ⎠
were considered, however, no information was provided on how the
PAR data was obtained from this location. To make a valid comparison
with our model, global horizontal and diffuse horizontal irradiance data
where, n is the number of discretized areas on the ground, xi is the total
were acquired in Lahore (31.5204◦ N, 74.3587◦ E) using Heliosat
PAR reaching the ground including shadings from the AV system, x is the
(SARAH-2), a satellite-based climate data record derived from satellite
average yearly PAR. A LHI of 100 % corresponds to a PAR light distri­
observations [41]. Our aim is to determine whether the reduction of PAR
bution that is equal throughout the ground area and can in other terms
is of similar magnitude between the designs. To achieve this, a PAR/GHI
be defined as a complete homogenous light distribution.
ratio was used to estimate PAR values throughout the year. Ma et al.
[42] reported that this ratio ranges from 0.35 to 0.58 in various locations
3. Results and discussion
worldwide. Therefore, a PAR/GHI ratio of 0.5 was assumed in our cross-
validation.
In this section, the shading factors validation is first presented and is
This work aims to develop and benchmark approaches to accurately
then followed by presenting the PAR distribution for the vertical, one-
calculate the incoming PAR, including shadings from the PV modules
axis, and two-axis AV systems in Lanna, Estrees-Mons, and Klingenberg.
from different AV systems configurations with high temporal and spatial

Fig. 8. Diffuse shading factor time series for one axis and two axis AV systems. Comparison between the Matlab® model developed in this study and the model
developed in PVsyst®.

7
S. Zainali et al. Applied Energy 339 (2023) 120981

Table 4 in this study is the lack of a graphical user interface as available com­
Diffuse shading factor error metrics between the model developed in this study mercial tools have. On the other hand, compared to existing commercial
in Matlab® and PVsyst®. and open-source packages, an advantage of the current modelling suite
Diffuse shading factors is that it can be easily combined into AV systems integrated modelling
System PVsyst®
tools (i.e., to perform simulation and optimization of AV systems
considering both energy and agricultural performances) such as those
MBE RMSE
presented by Campana et al. [23] and Amaducci et al. [25].
Vertical 0.0018 0.0018 For the assumptions made concerning the isotropy of the sky, the
One-axis − 0.0002 0.0004 diffuse shading factor must be determined only once for the fixed tilt AV
Two-axis − 0.0002 0.0004
system layout. Nevertheless, for the one-axis and two-axis AV systems
layouts, the diffuse shading factor must be determined hourly due to the
3.1. Shading factors validation involved rotations. Fig. 8 presents the diffuse shading factor time series
for the one- and two-axis AV systems for all the representative days. The
The beam shading factor validation results for the vertical, one-axis, diffuse shading factors from the Matlab® model are accurately calcu­
and two-axis AV systems are depicted in Fig. 7. The MBE and RMSE for lated compared to PVsyst®. The diffuse shading factor was obtained by
the beam shading factors are summarized in Table 3. The Matlab® manually adjusting the near shading scene in PVsyst®.
model shows an extremely good accuracy with both PVsyst® and The MBE and RMSE for the diffuse shading factors are summarized in
SketchUp® in all the investigated AV layouts. Since the models use the Table 4. The model developed in Matlab® is estimating the diffuse
same temporal distribution and geometries similar results for the beam shading factors accurately for all four representative days, for both the
shading factors were expected. The shading calculations are performed one- and two-axis tracker as seen in Fig. 8.
by using the topocentric solar position defined by geodetic latitude, The validation of the developed model in Matlab® has shown a good
longitude, altitude, and universal time to define the shading on the accuracy compared to existing commercial software and can therefore
ground from the modules by using vector projections [3,44,45]. be used for accurate shading calculations under AV systems. In this
As presented in this study, there are existing tools for calculating the study, we have focused mostly on model development and employed
shading factors accurately. However, the calculation of the shading
factors on the ground, which are of paramount importance for AV sys­
tems, are limited in commercial software as they are intended for con­ Table 5
ventional PV systems assessments and thus calculate shadings only on Light homogeneity index for the vertical, one-axis, and two-axis configurations
for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany.
the PV arrays. Moreover, commercial PV performance prediction tools
do not consider PAR and PAR diffuse in their models, which is necessary LHI
for crop yield estimations. Lanna Estrees-Mons Klingenberg
Since it is intended for AV systems, the model developed in this study Vertical 86.37 % 86.45 % 86.52 %
calculates the shading factors on the ground at any given temporal and One-axis 94.07 % 93.28 % 93.34 %
spatial resolution. A limitation of the current modelling suite presented Two-axis 95.15 % 94.68 % 94.64 %

Fig. 9. Yearly PAR distribution for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany.

8
S. Zainali et al. Applied Energy 339 (2023) 120981

Fig. 10. Yearly LHI distribution for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany.

Sweden, Estrees-Mons in France, and Klingenberg in Germany is


Table 6 depicted in Fig. 9.
Total yearly PAR reduction for the vertical, one-axis, and two-axis configura­ The PAR distribution can be used to analyse several AV systems
tions for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany. layouts to identify the most suitable layout for a specific crop or crop
Total yearly PAR reduction rotation at any location. As seen in Fig. 9, the PAR varies depending on
Lanna Estrees-Mons Klingenberg the choice of system configuration and geographical location. The two-
axis tracking system showed the highest yearly PAR. For instance, in
Vertical 22.49 % 20.62 % 20.42 %
One-axis 15.79 % 15.93 % 15.67 %
Estrees-Mons the mean PAR was 492 (kWh/m2/year). The yearly mean
Two-axis 11.01 % 11.85 % 12.00 % PAR varies significantly dependent on system configuration. In Estrees-
Mons, the difference in yearly mean PAR between the two-axis and the
vertical system was 49 (kWh/m2/year). It is also essential to have a low
commercial software for validation. Currently, our research group is variation of PAR throughout the reference crop area, as a high variation
performing model validation on the vertical AV system in the experi­ of PAR could heavily affect the crop water stress and thus the yield.
mental facility at Kärrbo Prästgård. The PAR distribution model vali­ To avoid a system configuration with high PAR heterogeneity, the
dation has been presented by Campana et al.[4]. At the same time, our LHI is used for all three system configurations to assess PAR homoge­
research group is enlarging the current experimental AV system into a neity. In Table 5, the LHI for the vertical, one-axis, and two-axis con­
facility with different AV technologies to provide more in-depth vali­ figurations are presented. 100 % corresponds to a homogenous
dation of the models presented in this study. distribution and 0 % corresponds to a non-homogenous distribution.
In the PAR distributions, it could be noticed that the vertical AV
3.2. PAR distribution and light homogeneity system shows a larger variation between the rows. This could also be
noted by calculating the LHI. The vertical AV system had a significantly
In AV systems, PAR and light homogeneity distribution are two key larger variation of PAR reaching the crops than the two-axis and one-
parameters to improve the understanding of how various configurations axis AV systems. In Lanna, the LHI was 8.78 % higher for the two-axis
might affect crop yields at specific geographical locations. Improving the tracker AV system than the vertical AV system. The LHI can provide a
understanding of light reduction with any kind of AV system will better understanding of the variation of PAR over the field by investi­
significantly reduce uncertainty and support decision-making processes. gating the gridded cells, as presented in Fig. 10. A high variation in light
In existing commercial products for PV systems assessments, PAR dis­ homogeneity could reduce crop yields in specific areas within an AV
tribution on the ground cannot be provided. In the developed model, the system. The LHI between the rows is not symmetrical, and there are
PAR is first decomposed to account for both diffuse and direct compo­ specific locations with a significant reduction of light during the year.
nents. In addition, the model discretises the ground with a spatial res­ Therefore, it is required to configurate a system with a light reduction
olution of 0.25 m × 0.25 m and an hourly temporal resolution to see the that does not reach the minimum light requirement needed for optimal
variation of PAR at each gridded cell between the AV rows. The spatio- crop yields. In addition, reducing light in specific areas could also delay
temporal resolution can easily be adjusted to carry out the same calcu­ crop growth [22]. A delay of crop growth at specific areas within the AV
lation procedure at other resolutions. The yearly PAR distribution for the system can drastically affect the harvests as there will be unbalanced
three different AV configurations developed in this study for Lanna in growth between the AV rows and will require new harvest approaches

9
S. Zainali et al. Applied Energy 339 (2023) 120981

Fig. 11. Yearly LHI distribution for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany.

for this issue. two-axis) will result in the highest PAR reduction. Therefore, using an
However, it is still essential to know the total reduction of PAR for AV tracker design with similar elevation as the vertical system would
the specific system as a complement to the LHI (Table 6). The total correspond to a lower amount of PAR reaching crops. This highlights the
yearly PAR reduction can be calculated by fractioning the PAR without complexity of choosing an appropriate design.
shading on the ground and the PAR with the respective AV system. However, a low yearly PAR reduction does not necessarily corre­
The yearly PAR reduction varies significantly depending on the spond to an optimal AV system configuration. The crops’ photosynthetic
choice of the AV configuration. The choice of an AV system is funda­ rate is a function of light intensity, and it varies between species.
mental as the shading significantly impacts the amount of yearly PAR Different crops can tolerate crop stresses induced by PAR and GHI dis­
reaching the crops. In this case, the variation of yearly PAR reduction tributions differently. Moreover, several crops will be grown underneath
could vary from 11 % to 22 %, dependent on the choice of AV system. It the AV system during its lifetime. Thus, it is essential to accurately
can be noticed that the two-axis AV system had the lowest yearly PAR design the AV system by optimizing the synergies between the crops and
reduction at all studied sites. The vertical AV system has a large PV system productions during its lifetime. The Matlab® model devel­
reduction of PAR close to the PV modules, where it could reach up to a oped in this study can be used as a universal model for AV systems at any
reduction of 50 %, as seen in Fig. 11. The significant PAR reduction close global location. The model gives the freedom to analyze shading factors
to the PV modules in the vertical system can significantly affect crop and how these affect the PAR distribution at the crop level. Additionally,
yield and growth at those locations. Therefore, AV systems equipped it can be integrated into modelling platforms such as those developed by
with trackers could be used instead to mitigate this issue. In the case of Campana et al. [23] and Amaducci et al. [25] to maximize crop yield and
less shade-tolerant crops, one-axis and two-axis trackers AV systems are electricity production and PV modules orientation, especially for those
more suitable due to less PAR reduction and high LHI. However, Tahir AV systems layouts equipped with tracking systems.
and Butt [40] found that a one-axis tracker system had a higher reduc­
tion in PAR reaching the crops than a vertical system. This is due to the 4. Conclusions
dimensional and geometric differences in the designs used to analyse
PAR reaching the crops. Tahir and Butt [40] predicted a PAR reduction In this study, mathematical models are developed in Matlab® for
of approximately 15 % using a vertical AV system with a row distance of accurately calculating the beam and diffuse shading factors produced by
4 m and height of 1 m. In comparison, the one-axis AV system had an three different AV system configurations. The models can be employed
additional PAR reduction of approximately 4 % compared to the vertical to calculate the shaded beam and diffuse shaded horizontal irradiance at
AV system. In our cross-validation, the vertical system with similar ge­ ground level. This modelling effort has been carried out to accurately
ometries had a PAR reduction of 15 %, similar to Tahir and Butt’s [40] depict the PAR distribution and reduction at ground level, which is
study. Moreover, the prediction for the reduction of PAR in the one-axis essential in assessing AV system effects on crop yield. The mathematical
agrivoltaic system was estimated to be 20 %, which is a 5 % increase models simulate fixed vertical, one-axis, and two-axis tracking PV sys­
compared to the 4 % reduction observed by Tahir and Butt [40]. Our tems for AV applications. The models were validated with the com­
results suggest that the amount of PAR reaching crops is strongly mercial software PVsyst® and SketchUp®. The main conclusions that
influenced by the dimensions used to design the AV system. Thus, it is can be drawn from this study are as follows:
not possible to definitively state which AV design (fixed, one-axis, or

10
S. Zainali et al. Applied Energy 339 (2023) 120981

• All mathematical models developed in this study showed high ac­ Writing – review & editing, Funding acquisition.
curacies compared to PVsyst® and SketchUp® and can be considered
robust for shading calculations under AV systems.
• Besides comparing the hourly model results with commercially Declaration of Competing Interest
available products, the models development has also focused on the
high temporal and spatial resolution of PAR distribution at the crop The authors declare the following financial interests/personal re­
level. This assessment is of fundamental importance for studying the lationships which may be considered as potential competing interests:
effects of shading on crops and crop rotation to identify the most The company European Energy is financing half of Sebastian Zainali’s
suitable AV system layout depending on crops and geographical Ph.D. salary.
locations.
• The light homogeneity and PAR reduction could vary significantly Data availability
depending on AV system configurations, from 86 % to 95 % and 11 %
to 22 %, respectively. The two-axis AV system showed the highest Data will be made available on request.
light homogeneity and lowest yearly PAR reduction in Lanna. The
light homogeneity index was 95.15 %, and the annual PAR reduction Acknowledgments
was 11.01 %.
The authors acknowledge the financial support received from the
The models developed in Matlab® are universal models that can be Swedish Energy Agency through the project “SOLVE solar energy
easily adapted to other types of AV system layouts. The models can also research center”, grant number 52693-1. The main author also ac­
be easily integrated with existing AV system modelling platforms to dual knowledges the financial support received from European Energy
assess electricity and crop production. Sverige AB. The author Pietro Elia Campana acknowledges Formas - the
Swedish Research Council for Sustainable Development, for the funding
CRediT authorship contribution statement received through the early career project “Avoiding conflicts between
the sustainable development goals through agro-photovoltaic systems”,
Sebastian Zainali: Conceptualization, Methodology, Data curation, grant number FR-2021/0005. The authors also acknowledge the finan­
Validation, Formal analysis, Writing – original draft, Writing – review & cial support received from the Swedish Energy Agency through the
editing. Silvia Ma Lu: Writing – original draft, Writing – review & project “Evaluation of the first agrivoltaic system in Sweden”, grant
editing. Bengt Stridh: Funding acquisition, Writing – review & editing. number 51000-1. The authors also acknowledge Dr. Frank de Ruijter’s
Anders Avelin: Writing – review & editing. Stefano Amaducci: Writing feedback, which improved the manuscript extensively.
– review & editing. Michele Colauzzi: Writing – review & editing. Likewise, the authors acknowledge ICOS for provisioning the
Pietro Elia Campana: Conceptualization, Writing – original draft, weather station data.

Appendix A. – Projection and PV coordinates

The coordinates of the PV module are calculated with simple trigonometric relationships as follows:
⎡ ⎤
x0
P1 = ⎣ y0 ⎦, (A1)
z0
⎡ ⎤
x0

P2 = y0 + W ⎦, (A2)
z0
⎡ ⎤
x0 + L
P3 = ⎣ y0 ⎦, (A3)
z0
⎡ ⎤
x0 + L
P4 = ⎣ y0 + W ⎦, (A4)
z0

where, x0, y0, and z0 are the initial PV module coordinates located in the plane, L is the length of the PV panel (m), and W is the width of the PV panel
(m). The tilt angles are calculated with the equations defined by Lorenzo et al. [28]. First, an ideal tracking angle ωIT including mutual shadowing must
be calculated and is given by:
x
tan(ωIT ) = , (A5)
z
where x, y, and z are the Cartesian coordinates of the Sun, which is referred to as a reference system with the z-axis pointing to the zenith, the y-axis
pointing south, and the x-axis pointing west. The components of the solar vector are given as follows [28]:

⎨ x = cos(β)sin(γ)
s = y = cos(β)cos(γ), (A6)

z = sin(β)

where, γ is the solar azimuth angle (◦ ), and β is the solar elevation angle (◦ ). The solar vector components are calculated according to a panel-oriented

11
S. Zainali et al. Applied Energy 339 (2023) 120981

Fig. A1. One-axis horizontal N–S oriented tracking. (a) ideal tracking with mutual shadowing, (b) corrected ideal tracking (back-tracking). Depicted from (Lorenzo
et al. [28]).

coordinate system, which is defined with a north–south axis direction. To adjust the system deviations for different azimuth angles αZ relative to the
north to south direction and axis tilt from the horizontal plane βZ , the system must be referred to as follows:
⎧ ′
⎨ x = xcos(αZ ) − ysin(αZ )
(A7)

y = xcos(βZ )sin(αZ ) + ycos(βZ )cos(αZ ) − zsin(βZ ),
⎩ ′
z = xsin(βZ )sin(αZ ) + ysin(βZ )cos(αZ ) + zcos(βZ )
The shaded fraction SF and shadow length SL are given by:
( )]
LEW
SF = max[0, 1 − , (A8)
SL

1
SL = , (A9)
cos(ωIT )

where, LEW is the distance between axes from east to west. The geometry of a one-axis horizontal tracking is depicted in Fig. A.1.
To avoid self-shading between the PV panels, a back-tracking calculation can be employed where the corrected ideal tracking angle ωITC is given
by:
ωC = ωIT − ωITC , (A10)

cos( ωC ) = LEW cos(ωIT ), (A11)

where, ωC is the back-tracking correction angle. For a fixed PV system, the tilt angle is fixed. The rotation around the y-axis is done by using the
rotation matrix defined as follows [46]:
⎡ ⎤
cos(ωc ) 0 sin(ωc )
Ry = ⎣ 0 1 0 ⎦, (A12)
− sin(ωc ) 0 cos(ωc )

The two-axis tracking PV system is rotated in the y-direction similarly as done for the one-axis systems. In Fig. A.2, the geometry of the two-axis
horizontal tracker is presented.
However, the second rotation follows the solar azimuth angle and the ideal tracking angle βIT for the x-axis is given by:
y
tan(βIT ) = , (A13)
(x2 + y2 )1/2
The corrected ideal tracking angle βITC is given by:
LNS
cos(βITC ) = cos(βIT ), (A14)
W

12
S. Zainali et al. Applied Energy 339 (2023) 120981

Fig. A2. Two-axis horizontal N–S oriented axis tracking, (a) ideal tracking with mutual shading, (b) corrected ideal tracking (back-tracking). Depicted from Lorenzo
et al. [25].

where, LNS is the distance between the axes from north to south, and W is the panel width. The second rotation is done by using the rotation matrix
around the x-axis as follows [46]:
⎡ ⎤
1 0 0
Rx = ⎣ 0 cos(ω1c ) − sin(ω1c ) ⎦, (A15)
0 sin(ω1c ) cos(ω1c )

The final position of the coordinates of the PV modules is obtained after the rotation around both the y-axis and x-axis and one translational
displacement. The rotations are performed by multiplying each panel point by a rotation matrix. The translational displacement is performed by first
subtracting the current position of the panel’s coordinate with rotation center points. The mathematical expression of the rotations and translational
displacement is given by [47]:
⎡ ⎤ ⎡ ⎤
Pix − xc xc
Pi = Rx Ry Piy − yc ⎦ + ⎣ yc ⎦, i ∈ {1, 2, 3, 4},
⎣ (A16)
Piz − zc zc

where, Pi are the points defining the PV module, Pix , Piy , and Piz are the coordinates in the x-axis, y-axis, and z-axis respectively for the PV module
points; xc , yc , and zc are the coordinates describing the centre of the PV module. The equation of the normal vector n can be retrieved from the
orientation and the tilt of the PV modules plane [3]:
⎧ a = cos(δ)sin(σ)

n = b = sin(δ)sin(σ ), (A17)

c = cos(σ )

where δ is the orientation angle and σ is the tilt angle. Once the normal vector n and solar vector s are defined, the projection of the PV module points
that define the shading on the desired plane can be solved by calculating the straight-line projections and can be determined as follows [3]:

⎨ Px + x⋅t
(A18)

P = Py + y⋅t ,

Pz + z⋅t

The coordinates of point P’, projection onto the plane along the straight line are given by the intersection of the straight line and the plane and can
be solved by calculating parameter t [3]:
aPx + bPy + cPz
t= − , (A19)
ax + by + cz

Appendix B. – Sensitivity analysis of agrivoltaic configurations

In this section, a sensitivity analysis has been conducted concerning the configuration parameters for all three AV designs. The designs considered
in Table 1 do not consider row-to-row shadings. In addition, in large-scale AV systems the amount of PAR reaching the crops will be different in the
center of a row as compared to the edge of the row due to edge shading effects. In the AV design defined in Table 1, there are edge shading effects as the

13
S. Zainali et al. Applied Energy 339 (2023) 120981

Fig. B1. S2 yearly PAR distribution for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany.

Fig. B2. S3 yearly PAR distribution for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany.

14
S. Zainali et al. Applied Energy 339 (2023) 120981

Table B1
Total yearly PAR reduction for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany for the
investigated scenarios.
Total yearly PAR reduction

Lanna Estrees-Mons Klingenberg

Vertical 22.49 % 20.62 % 20.42 %


S1 One-axis 15.79 % 15.93 % 15.67 %
Two-axis 11.01 % 11.85 % 12.00 %
Vertical 22.49 % 20.62 % 20.42 %
S2 One-axis 17.87 % 17.22 % 16.85 %
Two-axis 11.78 % 12.61 % 12.80 %
Vertical 24.00 % 21.92 % 21.72 %
S3 One-axis 18.13 % 18.29 % 18.01 %
Two-axis 12.95 % 13.84 % 14.08 %
Vertical 24.00 % 21.92 % 21.72 %
S4 One-axis 20.96 % 20.05 % 19.65 %
Two-axis 14.13 % 15.01 % 15.31 %
Vertical 22.49 % 20.62 % 20.42 %
S5 One-axis 18.19 % 17.43 % 17.03 %
Two-axis 11.92 % 12.74 % 12.94 %

PV row has the same length as the ground area studied. Therefore, to analyse the impact of these effects (i.e., row-to-row shadings and edge shading
effects), four new configuration scenarios have been considered (Scenario 1 (S1) is the base case as defined in Table 1): scenario 2 (S2) adds two
additional rows with a row distance of 10 m placed on each side to keep the area of interest centered; scenario 3 (S3) increases the length of the rows by
20 m and the area of interest is centered between the row to fully exclude edge effects; scenario 4 (S4) includes both the two additional rows and
increases the length of the rows with 20 m; and scenario 5 (S5) adds four additional rows in a similar way as in S2 to keep the area of interest centered.
In general, the vertical AV systems produce a large amount of shading with only two rows as it is fixed and with a low elevation as compared to the
AV designs equipped with trackers. On the other hand, AV designs with trackers usually produce shades on other rows due to their elevated heights.
Therefore, by increasing the number of rows to include row-to-row shadings could reduce the yearly PAR. However, including two additional rows
does not significantly reduce the yearly PAR distribution in the same reference area for the tracker configurations, as observed in Fig. B.1. The fraction
of shading produced by the vertical system remains higher. In addition, the diffuse PAR reaching the ground level is lower for the vertical system due
to the high diffuse shading factors on each gridded cell.
By comparing S1 and S2, it can be noted that edge shading increased the yearly PAR reaching the crops. It is, therefore, important to analyse the
differences between these occurrences to better understand the variation of PAR reaching the crops throughout the rows. Increasing the row lengths
affects the one-axis and vertical AV configuration more compared to the two-axis AV design (see Fig. B.2). In Lanna, the vertical AV configuration had
an additional PAR reduction of 1.51 % by excluding edge shading effects (Table B.1). Thus, the reduction of PAR due to exclusion of edge shading
effects is still advantageous for the crops as the LHI is generally higher at all locations for all analysed AV configurations (Table B.2).
The one-axis AV configuration suffers the most from adding additional rows as well as excluding edge shading effects. Therefore, in S4, it could be
noted that the one-axis configuration has a significant reduction of yearly PAR reaching the ground as compared to both the vertical and two-axis AV
designs (Fig. B.3). The one-axis configuration had the most significant reduction of PAR in Lanna, with an additional reduction of 5.17 % compared to
S1 (Table B.1).
The reduction of PAR by including additional rows was significant as noted in S2, corresponded in an additional reduction of 2.08 % observed in
Lanna. However, in S5 (Fig. B.4), the additional reduction of PAR as compared to S1 was 2.40 %. As the number of rows increases, the decrease in PAR
from adding more rows become less significant. For example, when comparing S2 to S5, only a 0.32 % decrease in PAR is seen.

Table B2
Light homogeneity index for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany for the
investigated scenarios.
LHI

Lanna Estrees-Mons Klingenberg

Vertical 86.37 % 86.45 % 86.52 %


S1 One-axis 94.07 % 93.28 % 93.34 %
Two-axis 95.15 % 94.68 % 94.64 %
Vertical 86.37 % 86.45 % 86.52 %
S2 One-axis 93.49 % 92.91 % 93.00 %
Two-axis 94.93 % 94.46 % 94.40 %
Vertical 86.32 % 86.30 % 86.36 %
S3 One-axis 94.35 % 93.39 % 93.42 %
Two-axis 95.23 % 94.73 % 94.72 %
Vertical 86.32 % 86.30 % 86.36 %
S4 One-axis 93.83 % 93.02 % 93.07 %
Two-axis 94.96 % 94.46 % 94.43 %
Vertical 86.37 % 86.45 % 86.52 %
S5 One-axis 93.34 % 92.82 % 92.93 %
Two-axis 94.90 % 94.43 % 94.38 %

15
S. Zainali et al. Applied Energy 339 (2023) 120981

Fig. B3. S4 yearly PAR distribution for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany.

Fig. B4. S5 yearly PAR distribution for the vertical, one-axis, and two-axis configurations for Lanna in Sweden, Estrees-Mons in France, and Klingenberg in Germany.

16
S. Zainali et al. Applied Energy 339 (2023) 120981

Appendix C. – PAR decomposition model

The decomposition model used in this study is YANG2 and is given by [38]:
1− C
kd = C + (s)
+ β5 kde , (C1)
β0 +β1 kt +β2 AST+β3 Z+β4 Δktc +β6 kd
1+e

GHI
kt = , (C2)
Eext

Gcs
Δktc = − kt , (C3)
Eext
( )
Gcs
kde = max 0, 1 − , (C4)
GHI

where the model coefficients used in this study have been fitted to three ICOS stations from Sweden [37] being C = 0.0888, β0 = − 26258, β1 =
7.2506, β2 = − 0.0458, β3 = 0.0099, β4 = − 0.0839, β5 = 0.5002, β6 = − 2.1731. kd is the satellite-derived diffuse fraction, Gcs is the clear sky GHI
(s)

(W/m2), AST is the apparent solar time (h), Eext is the extraterrestrial radiation (W/m2), and kt is the clearness index. The kd value from the YANG2
model can be used in the relationship defined by Spitters et al. [48] to get the correct PAR estimates:
[ ( ( )2 ) ] YANG2
PARdiffuse 1 + 0.3 1 − kdYANG2 kd
PAR
kd = = ( ( )2
) , (C5)
PARtotal 1 + 1 − kYANG2 cos2 (90 − β)cos3 (β)
d

where, β is the solar elevation angle (◦ ).

References [17] Proctor K, Murthy G, Higgins C. Agrivoltaics Align with Green New Deal Goals
While Supporting Investment in the US’ Rural Economy. Sustainability 2020;13:
137. https://doi.org/10.3390/su13010137.
[1] V. Quaschning, R. Dr. Hanitsch, SHADE CALCULATIONS IN PHOTOVOLTAIC
[18] Adeh EH, Good SP, Calaf M, Higgins CW. Solar PV Power Potential is Greatest Over
SYSTEMS; 1995.
Croplands. Sci Rep 2019;9:11442. https://doi.org/10.1038/s41598-019-47803-3.
[2] Bany J, Appelbaum J. The effect of shading on the design of a field of solar
[19] Pascaris AS, Handler R, Schelly C, Pearce JM. Life cycle assessment of pasture-
collectors. Sol Cells 1987;20:201–28. https://doi.org/10.1016/0379-6787(87)
based agrivoltaic systems: Emissions and energy use of integrated rabbit
90029-9.
production. Clean Responsible Consum 2021;3:100030. https://doi.org/10.1016/j.
[3] Cascone Y, Corrado V, Serra V. Calculation procedure of the shading factor under
clrc.2021.100030.
complex boundary conditions. Sol Energy 2011;85:2524–39. https://doi.org/
[20] Barron-Gafford GA, Pavao-Zuckerman MA, Minor RL, Sutter LF, Barnett-Moreno I,
10.1016/j.solener.2011.07.011.
Blackett DT, et al. Agrivoltaics provide mutual benefits across the
[4] Campana P, Staaf Scragg J, Ma Lu S, Zainali S, Stridh B, Amaducci S, et al. Solar
food–energy–water nexus in drylands. Nat Sustain 2019;2:848–55. https://doi.
irradiance distribution under vertically mounted agrivoltaic systems – Model
org/10.1038/s41893-019-0364-5.
development, validation, and applications for microclimate assessment.
[21] Agrivoltaic Systems, A Promising Experience, Energy Ind. Rev; 2021. https://
Agriculture 2022. https://doi.org/10.31223/X5G07D.
energyindustryreview.com/analysis/agrivoltaic-systems-a-promising-experience/
[5] (11) Shadow Analysis. Tools and software | LinkedIn, (n.d.). https://www.linkedin.
[accessed June 2, 2022].
com/pulse/shadow-analysis-tools-software-eduardo-rodriguez-e-i-t-/ [accessed
[22] Tazawa S. Effects of various radiant sources on plant growth (Part 1). Jpn Agric Res
December 14, 2022].
Q 1999;33:163–76.
[6] Silva M, Roberts JJ, Prado PO. Calculation of the Shading Factors for Solar Modules
[23] Campana PE, Stridh B, Amaducci S, Colauzzi M. Optimisation of vertically
with MATLAB. Energies 2021;14:4713. https://doi.org/10.3390/en14154713.
mounted agrivoltaic systems. J Clean Prod 2021;325:129091. https://doi.org/
[7] PV Performance Modeling Collaborative | PV_LIB Toolbox, (n.d.). https://pvpmc.
10.1016/j.jclepro.2021.129091.
sandia.gov/applications/pv_lib-toolbox/ [accessed December 9, 2022].
[24] Williams JR, Jones CA, Kiniry JR, Spanel DA. The EPIC crop growth model. Trans
[8] Melo EG, Almeida MP, Zilles R, Grimoni JAB. Using a shading matrix to estimate
ASAE 1989;32:497–0511.
the shading factor and the irradiation in a three-dimensional model of a receiving
[25] Amaducci S, Yin X, Colauzzi M. Agrivoltaic systems to optimise land use for electric
surface in an urban environment. Sol Energy 2013;92:15–25. https://doi.org/
energy production. Appl Energy 2018;220:545–61. https://doi.org/10.1016/j.
10.1016/j.solener.2013.02.015.
apenergy.2018.03.081.
[9] Arias-Rosales A, LeDuc PR. Comparing View Factor modeling frameworks for the
[26] Edera.digital, REM Tec - La soluzione per il fotovoltaico legata all’agricoltura;
estimation of incident solar energy. Appl Energy 2020;277:115510. https://doi.
2021. https://remtec.energy/en/agrovoltaico [accessed December 31, 2021].
org/10.1016/j.apenergy.2020.115510.
[27] Li DHW, Lam JC. Predicting solar irradiance on inclined surfaces using sky
[10] Reindl DT, Beckman WA, Duffie JA. Evaluation of hourly tilted surface radiation
radiance data. Energy Convers Manag 2004;45:1771–83. https://doi.org/10.1016/
models. Sol Energy 1990;45:9–17. https://doi.org/10.1016/0038-092X(90)90061-
j.enconman.2003.09.020.
G.
[28] Lorenzo E, Narvarte L, Muñoz J. Tracking and back-tracking: Tracking and Back-
[11] Yusufoglu UA, Lee TH, Pletzer TM, Halm A, Koduvelikulathu LJ, Comparotto C,
Tracking. Prog Photovolt Res Appl 2011;19:747–53. https://doi.org/10.1002/
et al. Simulation of Energy Production by Bifacial Modules with Revision of Ground
pip.1085.
Reflection. Energy Procedia 2014;55:389–95. https://doi.org/10.1016/j.
[29] PVsyst, PVsyst; 2022. https://www.pvsyst.com/ [accessed January 4, 2022].
egypro.2014.08.111.
[30] A. Mermoud, Use and validation of PVSYST, a user-friendly software for PV-system
[12] B.L. Liu, R. Jordan, Daily insolation on surfaces tilted towards equator, in: 1961.
design, H.S. Stephens, 1995. https://archive-ouverte.unige.ch/unige:119365
[13] Badescu V. 3D isotropic approximation for solar diffuse irradiance on tilted
[accessed December 4, 2022].
surfaces. Renew Energy 2002;26:221–33. https://doi.org/10.1016/S0960-1481
[31] Stridh B, Campana PE, Sandra A, Thomas C, Tomas L, Nordlander E, Vadiee A.
(01)00123-9.
Förbättrad beräkning av solelproduktion i Sverige; 2020. http://urn.kb.se/resolve?
[14] Tian YQ, Davies-Colley RJ, Gong P, Thorrold BW. Estimating solar radiation on
urn=urn:nbn:se:mdh:diva-53184 [accessed December 10, 2022].
slopes of arbitrary aspect. Agric For Meteorol 2001;109:67–74. https://doi.org/
[32] Chepp ED, Gasparin FP, Krenzinger A. Accuracy investigation in the modeling of
10.1016/S0168-1923(01)00245-3.
partially shaded photovoltaic systems. Sol Energy 2021;223:182–92. https://doi.
[15] Asgharzadeh A, Toor F, Bourne B, Anoma MA, Hoffman A, Chaudhari C, Bapat S,
org/10.1016/j.solener.2021.05.061.
Perkins R, Cohen D, Kimball GM, Riley D, A Benchmark and Validation of Bifacial
[33] Carigiet F, Baumgartner FP, Sutterlueti J, Allet N, Pezzotti M, Haller J. Verification
PV Irradiance Models, in: 2019 IEEE 46th Photovolt. Spec. Conf. PVSC, IEEE,
of Measured PV Energy Yield Versus Forecast and Loss Analysis, 28th Eur.
Chicago, IL, USA; 2019: pp. 3281–3287. 10.1109/PVSC40753.2019.8981272.
Photovolt. Sol. Energy Conf. Exhib. 3922-3927. (2013) 6 pages, 5478 kb. 10.4229/
[16] Pelaez SA, Deline C, Greenberg P, Stein JS, Kostuk RK. Model and Validation of
28THEUPVSEC2013-5BV.4.25.
Single-Axis Tracking With Bifacial PV. IEEE J Photovolt 2019;9:715–21. https://
[34] Siraki AG. Comparison of PV System Design Software Packages for Urban
doi.org/10.1109/JPHOTOV.2019.2892872.
applications; (n.d.).

17
S. Zainali et al. Applied Energy 339 (2023) 120981

[35] Shadow Projector | SketchUcation. (n.d.). https://sketchucation.com/plugin/ and Measurements. J. Nat. Resour. 2007;22:673. https://doi.org/10.11849/
1292-tig_shadowprojector [accessed August 9, 2022]. zrzyxb.2007.05.001.
[36] Integrated Carbon Observation System, ICOS; (n.d.). https://www.icos-cp.eu/ [43] Trommsdorff M. Performance Indices for Parallel Agriculture and PV, (n.d.).
[accessed June 2, 2022]. [44] Koblick D. Vectorized Solar Azimuth and Elevation Estimation; 2022. https://se.
[37] Ma Lu S, Zainali S, Stridh B, Avelin A, Amaducci S, Colauzzi M, et al. mathworks.com/matlabcentral/fileexchange/23051-vectorized-solar-azimuth-a
Photosynthetically active radiation decomposition models for agrivoltaic systems nd-elevation-estimation [accessed December 13, 2022].
applications. Sol Energy 2022. https://doi.org/10.1016/j.solener.2022.05.046. [45] Computing planetary positions - a tutorial with worked examples; (n.d.).
[38] Yang D, Boland J. Satellite-augmented diffuse solar radiation separation models. http://stjarnhimlen.se/comp/tutorial.html#5 [accessed December 13, 2022].
J Renew Sustain Energy 2019;11:023705. https://doi.org/10.1063/1.5087463. [46] Evans PR. Rotations and rotation matrices. Acta Crystallogr D Biol Crystallogr
[39] Data | Copernicus; (n.d.). https://atmosphere.copernicus.eu/data [accessed June 2001;57:1355–9. https://doi.org/10.1107/S0907444901012410.
2, 2022]. [47] linear algebra - Rotation around non-zero point, Math. Stack Exch; 2016. https://
[40] Tahir Z, Butt NZ. Implications of spatial-temporal shading in agrivoltaics under math.stackexchange.com/questions/1964905/rotation-around-non-zero-point
fixed tilt & tracking bifacial photovoltaic panels. Renew Energy 2022;190:167–76. [accessed December 31, 2021].
https://doi.org/10.1016/j.renene.2022.03.078. [48] Spitters CJT, Toussaint HAJM, Goudriaan J. Separating the diffuse and direct
[41] Pfeifroth U, Kothe S, Müller R, Trentmann J, Hollmann R, Fuchs P, Werscheck M. component of global radiation and its implications for modeling canopy
Surface Radiation Data Set - Heliosat (SARAH) - Edition 2; 2017. 7.1 TiB. photosynthesis Part I. Components of incoming radiation. Agric For Meteorol 1986;
10.5676/EUM_SAF_CM/SARAH/V002. 38:217–29. https://doi.org/10.1016/0168-1923(86)90060-2.
[42] Ma Jin-yu WB, Liu Jing-miao, Li Shi-kui, Liang Hong, Jiang Chao-yang. Study on
the Features of the Photosynthetic Active Radiation (PAR) with Experimentations

18

You might also like