Content uploaded by Diane Palmer
Author content
All content in this area was uploaded by Diane Palmer on Jan 05, 2016
Content may be subject to copyright.
DETECTION OF ROOF SHADING FOR PV BASED ON LIDAR DATA USING A MULTI-MODAL APPROACH
Diane Palmer *, Ian Cole, Brian Goss, Thomas Betts and Ralph Gottschalg
d.palmer@lboro.ac.uk
Centre of Renewable Energy Systems and Technology (CREST), Loughborough University, LE11 3TU, UK,
* Corresponding Author. Tel. +44 (0) 1509 635 604
Topic 5, Subtopic: 5.1 Operation of PV Systems and Plants.
ABSTRACT: There is a current drive to increase rooftop deployment of PV. Suitable roofs need to be located, especially as
regards shading. A shadow cast on one small section of a solar panel can disproportionately undermine output of the entire
system. Nevertheless, few shading figures are available to researchers and developers. This paper reviews and categorizes a
number of methods of determining shade losses on photovoltaic systems. Two existing methods are tested on case study
areas: shadow simulation from buildings and ambient occlusion. The first is conceptually simple and was found to be useful
where data is limited. The second is slightly more demanding in terms of data input and mathematical models. It produces
attractive shadow maps but is intended for speed and represents an approximation to ray-tracing. Accordingly, a new model
was developed which is fast, flexible and accurately models solar radiation.
Keywords: Modelling, Photovoltaic, Ray Tracing, Solar Radiation, Shading
1 INTRODUCTION AND BACKGROUND
Many countries are currently focusing on roof-top
PV installations. There is a global trend towards
distributed energy production where electricity is
generated at point of use, on the roofs of houses, public
and commercial buildings. Some roofs are more suitable
than others in terms of size, azimuth, tilt and shade
received. Decentralised energy requires load
management from electricity grid operators so it is
essential to identify those roofs which are most suitable
for PV, including susceptibility to shading, for potential
future deployment.
Photovoltaic system performance is very sensitive to
shading. The output of a single cell is reduced in relation
to the amount of shade falling upon it. However, due to
the fact that cells in a module are connected in series and
that the number of bypass diodes per module is generally
low, shading just one cell causes a disproportionately
large decrease in the current in the bypassed block
(Kirchoff’s laws). With particular types of shading
pattern, a small amount of shade covering only a few
cells may reduce the power output of the module to near
zero.
Despite the impact of shading on photovoltaic output,
there is little detailed data available regarding shading
profiles. Commercial data suppliers sometimes retail
roof maps as simple binary shaded/not shaded outline
polygons, based on manual inspection of aerial
photographs. Other possibilities include expensive on-
site measurement or customised surveys of just one or
two buildings. The latter often produces results by
simple projection of building and tree shadows onto the
ground. Online tools such as Solar Census Surveyor [1]
and Sun Area [2] take LiDAR data as input, but include
shade in a PV suitability value determined via patented
algorithms.
For accurate PV installation decisions the minimum
data requirement is percentage shading of roof. Shade
polygons on roofs with area and location are even more
useful because they allow the most suitable roof panes to
be identified.
The total shade falling on a roof may be categorised
as arising from one or more of three causes: (1) Distant
topographical features such as hills, cliffs etc; (2) Near
features e.g. neighbouring taller buildings, trees, pylons
etc; and (3) Self-shading from chimneys, dormers, cross-
gables, aerials, vents. This research aims to identify a
method which delivers the whole shade from all three
causes.
2 REVIEW OF METHODS OF MODELLING
SHADE
Several methods have been applied to study shading
losses in PV systems. These may be grouped according
to the input data they require as follows: shadow
simulation from buildings, LiDAR methods and image
methods.
Buildings methods need data in the form of building
polygons linked to a height attribute. Height information
is often obtained by sophisticated analysis of LiDAR
data but may also originate from building plans or
manual estimations. Sketchup (bought by Trimble in
2012) and ArcGIS Sun Shadow Volume are two
examples of this simple method. Both base shadow
calculations on latitude and longitude of the site of
interest; and time and date. The sun is treated as a point
source. In other words only beam effects are reproduced.
Building shadow simulation models produce the location
of shading only (shadow polygons). The effect of shade
on incident irradiance must be computed separately.
Dereli et al (2013) [3] applied Sketchup to a case study
area, investigating shading from trees on PV in the US
MidWest.
LiDAR methods may be subdivided into: hillshade-
type, horizon angle computation and sky-dome models.
Elementary hillshade models are found in various
software, for instance, Hillshade in ArcGIS or Shaded
Relief in GRASS. They are more generally utilised to
enhance the visualisation of terrain. In the context of
photovoltaics, it is the hillshade of the individual roof
which is generated, that is, self-shading from aspect and
cross-gables. Again, as with the building simulation
models, only shadows are produced and not their
influence on irradiation. Kassner et al [4] applied the
hillshade algorithm to 13 buildings on a university
campus. Interestingly, there is no record in the literature
of the algorithm being applied twice, once to calculate
the shade on the roof from surrounding hills and once to
estimate self-shading from the building itself.
Horizon angle computation is employed by a number
of solar analysis tools but its implementation as r.sun /
r.horizon / r.sunmask in GRASS software is one of the
best known. The minimum input requirements for this
function are location and date/time to compute the sun
position and a gridded elevation map, usually LiDAR.
The model calculates beam irradiation for clear-sky
conditions from extraterrestrial irradiance. Accuracy
may be improved by including a local atmospheric
turbidity factor (Linke). The incoming insolation may
also be divided into its direct and diffuse fractions by
incorporating a coefficient derived from ground-station
measurements. Shadowing effects may be pre-calculated
as follows. For a given solar incidence angle, the
calculation starts on e.g. a roof pane and moves along the
path of the incoming sunray. Each height point is
checked to ascertain whether it intersects or is higher
than the sunray, in which case it will cast shade. In
contrast to the previously described building shadow
simulation and hillshade techniques, r.sun directly
calculates global horizontal irradiation, taking into
account the consequences of shading. These GRASS
algorithms are computationally intensive but have been
successfully applied in both urban [5] and mountainous
landscapes [6].
Sky-dome models are found in ray-tracing routines in
games and graphics software and in architectural
applications whose purpose is to include daylight in
building design, as well as photovoltaic applications. A
typical sky-dome model comprises a hemispherical
dome, divided into segments, positioned over the PV
array. Diffuse irradiation originating from various parts
of the sky (segments of the dome) may then be modelled.
Image methods apply a completely different
approach to the functionality just described. They
capture actual shadows as recorded by aerial
photography rather than modelling the theoretical
location of shadows from sun position. For instance,
Bergamasco and Asinari (2011) drew up conditions to
define clusters of dark pixels as shadow in images of the
whole city of Turin [7]. Image methods will not be
considered by this paper. They require powerful
computing facilities and automatic detection of shadows
visible to the human eye is challenging. Furthermore,
“what-if scenarios” of proposed buildings cannot be
added.
3 COMPARISON OF RESEARCH METHODS AND
FINDINGS
This paper will compare and contrast two shadow
modelling methods: building height and sky-dome. A
new method, “solarscene.xyz”, (developed at CREST)
which improves on previous sky-dome based techniques
will then be presented.
Case studies of two specific (1km x 1km) areas will
be used: Loughborough University West Campus and
Prestwich Village, Manchester. The University Campus
was selected primarily for ease of validation. The area is
landscaped to a slight rise and it encompasses a mix of
building types from large lecture theatres and sports halls
to small residential homes. There are also wooded areas.
Prestwich was selected as the second area because of the
range of LiDAR data available there. The gently
undulating topography is characterized by the central
Roman road and there are many large domestic
residences surrounded by generous gardens with mature
trees.
3.1 ArcGIS Sun Shadow Volume for Buildings
This facility models shadows as polyhedrons
thrown by each building. Sun position is calculated
from time of day and year. The building heights data
used was Landmap Features Earth Observation
Collection [8] which derives heights from LiDAR or
aerial imagery. Initial results appear promising (Fig. 1).
Figure 1: Shadow volume cast each hour 10am – 3pm
22 December 2015 by cube-shaped Henry Ford College,
Loughborough
However, each building is treated as a simple cuboid
i.e. shading from sloping roofs is not accounted for.
Shade from trees and hills is not considered, although
these could be modelled separately and the results
combined. More difficult to resolve is the fact that this
model creates shadows with a set length which is not
terminated by any other feature. That is, the shadows
may pass through and under other buildings, through
hills and into the ground. Hence shade is over-estimated
or falsely identified. The tool is also computationally
demanding. Therefore, although the Sun Shadow
Volume is useful in terms of its simplicity and easily
obtained input data, more sophisticated tools were
trialed.
3.2 SAGA-GIS Analytical Hillshading sky-dome
application
This tool modifies the work of Tarini et al [9]. It
utilizes the ambient occlusion global lighting technique.
Height data (of terrain, buildings, trees etc) is obtained
from LiDAR. For each LiDAR point (see later) on a
given roof, rays are projected in every compass direction
for a positive z value i.e. upwards. Rays which travel
unrestricted to the “sky” provide irradiance. Those
which strike another object are given an irradiance value
of zero (shadow). The light-blocking effect of objects is
exponentially weighted according to the distance of the
object. Light is treated as uniform. It is allocated the
same value from all directions. The effect is that of
cover by a cloudy sky. Ambient occlusion is an
approximation to ray-tracing. It was originally
developed to visualize graphics at high frame rates and
uses as few computational resources as possible.
The height of the roof points is obtained from
LIDAR (Light Detection and Ranging) data. This is
produced from a sensor mounted on an aircraft. The
sensor pulses a laser downwards and times its return.
The number of returns per square metre defines the
resolution of the data. In the UK, LiDAR data is now
available from the Environment Agency as Open Data
(free of charge and restriction) [10]. 2m, 1m, 50cm and
25cm resolution datasets are supplied. However, the
datasets were flown in different years and 25cm data
covers less than 4% of England and Wales.
The results produced by the SAGA-GIS Analytical
Hillshading module using 1m LiDAR appear generally
realistic when checked against local observations,
although spurious features can be discerned due to the
approximation of the method (Fig. 2 and Fig. 3).
Detailed percentage shade-on-roof maps may be
generated (Fig. 4). Additionally, SAGA software is fast
because it is programmed to maximise use of all
available computer cores. Shade maps were produced
for the 1 sqkm case study areas in under 1 second.
Figure 2: Shade (grey) on roofs of Pilkington Library
(square building) and Village Restaurant, at foot of hill,
West Campus, Loughborough University, noon 22
December 2015
Figure 3: Shade (grey) on roofs of northwest-facing
Burleigh Court Hotel, Loughborough, noon 22 December
2015
Figure 4: Percentage shading on roofs in Prestwich
Village, Manchester, noon 22 December 2015.
The SAGA-GIS Analytical Hillshading module
yields lifelike results but there are openings for
improvement. It calculates the portion of the sky-dome
over each LiDAR point which is not blocked by another
feature (building, tree, hill etc). Nonetheless, light is
considered to be uniform, recreating conditions only.
3.3 Improved Sky-dome method for LiDAR:
solarscene.xyz
This new model develops on the skydome
modelling technique described by Goss et al [11] and the
ray tracing methodology of Cole & Gottschalg, as used
in [12]. The model can accept any form of gridded xyz
data and any set of meteorological data with separable
beam and diffuse irradiance components at a user defined
temporal resolution. The default data inputs, and those
used here, are LiDAR data and hourly global horizontal
irradiation, produced by interpolation of UK Met. Office
data [13]. The model calculates total global in-plane
irradiation (beam and diffuse) based on actual sky
conditions from measured values.
Previous sky-dome techniques have adopted one of
two alternative approaches:
1. Divide the hemisphere into almost equal area
segments with varying spacing.
2. Divide the hemisphere into patches equally
spaced in elevation and azimuth with different
sizes.
The solarscene.xyz model offers a variable input sky-
dome. User defined sky-domes and resolutions can be
used, the model corrects for sky-patch weightings by
means of solid angle calculation and correction. The sky
dome used here is the Tregenza dome, as explained in
[14].
The model first iterates over each timestamp in the
meteorological dataset, computing the relative sun
position and distributing the solar irradiation over the
sky-dome accordingly. The net irradiation sky dome for
the user defined measurement campaign is then held in
RAM and the sky-patch locations and intensities used in
the ray tracing algorithm, resulting in an extremely fast
and efficient model.
In the ray tracing algorithm, solarscene.xyz iterates
over each sky-patch in the hemisphere and calculates the
longest subjective path of intersections through the 3D
environment. This path is then objectified and used as a
reference path for the other traces through the 3D
environment. This method optimizes the trace procedure
as there is only one explicit ray path calculation per sky-
patch. The calculation is performed such that, for an
individual sky-patch to pixel interaction, the cosine
corrected irradiation from the pixel 3D surface normal to
the sky-patch vector is either: shaded, half-shaded or
unshaded. The same routine is performed for each sky-
patch in the sky-dome and the irradiation falling on each
pixel in the 3D environment is summed.
solarscene.xyz handles both beam and diffuse
irradiance whilst overcoming problems arising from
division of the sky-dome. Preliminary results appear
very satisfactory (Fig. 5). These were verified locally by
comparing with pyranometer readings from CREST
meteorological station. Pyranometers are mounted
horizontally and south-facing at both 35° and 45°
inclination angles. Initial model results checked against
these pyranometer readings were found to be within 5%
of net annual energy values. Wide area validation is
currently underway.
Figure 5: Total Global Horizontal Irradiation for 2014
(kWh) on roofs Prestwich Village, Manchester
4 DETERMINATION OF APPROPRIATE LIDAR
RESOLUTION
An experiment was carried out to investigate whether
high resolution LiDAR is necessary for shade modelling,
or if acceptable results may be produced with medium
resolution data. For reasons of brevity, the SAGA-GIS
Analytical Hillshading tool was employed, but further
investigations will be carried out using solarscene.xyz in
the future.
Binary gridded shade maps were created for 10am 22
December 2015 for Prestwich Village using 25cm, 50cm
and 1m LiDAR and the results compared. This time was
selected because this gives some of the longest shadows
in the year. Mathematical comparison of the resultant
maps delivered the following outcomes:
• 50cm has 5.7% more grid cells (pixels) shaded
but also 2.7% less than 1m (total difference
8.4%).
Unfortunately, the available LiDAR data were
flown on different dates. The 25cm and 50cm LiDAR
were flown on 27 January 2010 and the 1m LiDAR on 1
February 2013. Consequently, it would be invalid to
directly compare this data with respect to resolution due
to the significant environmental changes likely to occur
over 3 year period. For this reason, the 25 cm and 50cm
LiDAR datasets were averaged to 1m resolution to
manufacture data for the same year. Differences between
the original 1m data and 1m data aggregated from 50 cm
are:
• Original 1m has 3.4% more pixels shaded but
also 2.2% less than generated 1m (total
difference 5.6%)
The differences between the original 1m data and 1m
data aggregated from 25 cm are virtually identical to the
1m/aggregated 50cm results, differing by just 3 pixels, as
would be expected since these datasets were flown in the
same year. Measured 50cm and original 1m data differ
by 8.4%, whilst original 1m data and 1m data
manufactured by aggregation differ by 5.6%. Therefore
about two-thirds of the variation in shadows generated
between 50cm and 1m data is not due to resolution but to
the three years’ time interval which would give
opportunity for tree growth and building construction. In
fact, only 2.8% of the difference may be explained by the
finer resolution.
Further comparisons reveal:
• 25cm has 5% more and 1.4% less grid cells
shaded than 50cm (total difference 6.4%).
• 25cm has 8.3% more and 1.4% less pixels
shaded than 1m (total difference 9.7%).
So it may be seen that moving from 50cm to 1m
resolution produced a 2.8% difference in pixels shaded
(0.001% per sqcm). Contrasting 25cm and 50cm data
results in 6.4% variation (0.01% per sqcm). This is a
much greater relative change.
25cm LiDAR produces more shading than 50cm
using the SAGA tool but it is thought that much of this is
due to rogue shadowing effects generated by the ambient
occlusion technique. The extra shadows may be false.
They often appear as tiny shade polygons which could be
caused by trees casting dabbled shade or could be a
source of error, related to inherent approximations in the
method. However, for this method, it seems that 25cm
data is no more likely to capture small potential shade
producing features such as chimneys than 50cm data.
5 ANALYSIS OF FINDINGS
Tables I, II and III detail similarities and
differences between the shading methods reviewed and
appraised in this paper. Numbers 1, 4 and 6 underwent
extensive trials. The others were tested or reviewed.
Table I: Shading methods: Comparison of Inputs
Model
Input Data
Extra
Inputs
3D
build-
ings
LiDAR
1
Building
shadow
simulation
Yes
No
No
2
Hillshade
No
Yes
No
3
r.sun
No
Yes
Yes
4
Ambient
Occlusion
No
Yes
No
5
Image
methods
No
No
Yes
6
solarscene.
xyz
Yes
Yes
Yes
Table II: Shading methods: Details of Included Models
and Options
Model
Include
self-
shading,
hills and
trees
Models
diffuse
irradiation
Choice
of input
data &
sky-
dome
1
Building
shadow
simulation
No
No
No
2
Hillshade
No
No
No
3
r.sun
Yes
Yes
No
4
Ambient
Occlusion
Yes
No
No
5
Image
methods
Yes
No
No
6
solarscene
.xyz
Yes
Yes
Yes
Table III: Shading methods: Comparison of Outputs and
Computer Requirements
Model
Output
Comput-
er Speed
Shade
Polygons
Global
Horizontal
Irradiation
1
Building
shadow
simulation
Yes
No
No
2
Hillshade
Yes
No
Yes
3
r.sun
Yes
Yes
Yes
4
Ambient
Occlusion
Yes
No
Yes
5
Image
methods
Yes
No
No
6
New
solarscene.
xyz
Yes
Yes
Yes
Each of the models has advantages and disadvantages.
Some of the models create shadow polygons only. Of
these, shadow simulation from buildings (e.g. ArcGIS
Sun Shadow Volume) is a simple technique which
neglects several sources of shade and is very slow to
compute. On the other hand, it is easy to understand and
the necessary input data is straightforward to obtain and /
or prepare. Hillshade techniques also overlook some
shade-producing features. Likewise, these techniques are
conceptually simple but have the advantage of greater
speed. LiDAR data which, in general, must be purchased
or commissioned is a pre-requisite. Image methods too
need an aircraft to capture data. Image recognition
includes all sources of shade but despite the functionality
being relatively clear, are computationally intensive and
difficult to implement successfully. Ambient occlusion
is a little more demanding upon the user in terms of the
mathematical models which form its basis. Here again,
LiDAR is a pre-requisite. Ambient occlusion is fast and
produces visually attractive results but is inclined to
generate spurious shadows. All of the techniques
mentioned so far may be implemented via a graphical
user interface.
Turning to methods which generate global horizontal
irradiation figures (r.sun and solarscene.xyz), they need
LiDAR data or other gridded height data, as well as
further inputs (e.g. Met. Office data, atmospheric
coefficients) which may not be easy to obtain. R.sun
runs from the command-line. It may be run with limited
data by accepting included default values but accuracy is
increased by over-riding with measured values. GUI
development for solarscene.xyz is planned for the future.
These models which output irradiation data provide the
most natural results but understanding how they work
can be challenging for the layman.
6 CONCLUSION
With the present focus on increasing the penetration
of rooftop PV, there is a need for installers and energy
companies to identify the most suitable buildings. The
amount of shade falling on a roof greatly influences its
appropriateness for PV installation. This paper has
examined a number of shade-estimation techniques and
presented the new solarscene.xyz model.
The findings demonstrate that choice of shading
model depends on:
• The user - programming skill, technical
knowledge and amount of time available to
dedicate to the task.
• Type of data available.
• Computer resources.
• Outputs required.
The new sky-dome model from CREST,
solarscene.xyz, is flexible as regards data inputs, choice
of sky-dome model, time-step and outputs. It may take
either building heights data or LiDAR, This flexibility as
regards elevation data allows what-if analysis. For
instance, the impact on a national electricity grid may be
estimated if a new housing estate with roof-top PV is
constructed. solarscene.xyz relies on ground-based
meteorological data. Accurate ray tracing is employed
with minimal approximations – those governed by the
resolution and accuracy of the input data. It is fast
because it is optimized to differentiate between simple
and complex terrain. Currently, UK Met. Office data is
only obtainable with an hourly timestamp. However, the
model accepts user defined datasets at any time
resolution. Should there be an opportunity for automated
extraction of meteorological data at high temporal
resolution, the model will be updated accordingly.
solarscene.xyz offers the user a choice of sky-dome
model. It provides a sophisticated treatment of diffuse
radiation shadow losses as well as manufacturing beam-
derived shadows. Lastly, there is the possibility of
alternative outputs, either global in-plane irradiation
maps or shadow polygons.
Future work will continue validation of the new
model. Further investigation will be pursued to
determine the optimal LiDAR resolution for use with
solarscene.xyz.
References
[1] ‘Sun Area’, http://www.sun-
area.net/index.php?id=103, accessed 14 August 2015
[2] ‘Solar Census’, http://www.solarcensus.com/,
accessed 14 August 2015
[3] Z. Dereli, C. Yücedağ and J. M. Pearce, Simple and
Low-Cost Method of Planning for Tree Growth and
Lifetime Effects on Solar Photovoltaic Systems
Performance, Solar Energy, 95, pp.300-307 (2013).
Source: http://www.academia.edu/4074627/Simple_and_
low-
cost_method_of_planning_for_tree_growth_and_lifetime
_effects_on_solar_photovoltaic_systems_performance, a
ccessed: 2/9/2015
[4] R. Kassner, W. Koppe, T. Schüttenberg, G. Bareth,
Analysis of the Solar Potential of Roofs by using official
LiDAR data, The International Archives of the
Photogrammetry, Remote Sensing and Spatial
Information Sciences, 37, B4, Beijing 2008
[5] H. T. Nguyen and J. M. Pearce, Incorporating
Shading Losses in Solar Photovoltaic Potential
Assessment at the Municipal Scale, Solar Energy 86(5),
pp. 1245–1260 (2012). Source:
http://www.academia.edu/1499891/Incorporating_Shadin
g_Losses_in_Solar_Photovoltaic_Potential_Assessment_
at_the_Municipal_Scale, accessed: 2/9/2015
[6] F. Nexa, F. Remondinoa, G. Agugiaroa, R. De Filippi,
M. Poletti, C. Furlanellob, S. Menegon, G. Dallagoc, S.
Fontanari, 3D SolarWeb: A Solar Cadaster in the Italian
Alpine Landscape, International Archives of the
Photogrammetry, Remote Sensing and Spatial
Information Sciences, Volume XL-7/W2, 2013.
Source: http://www.int-arch-photogramm-remote-sens-
spatial-inf-sci.net/XL-7-W2/173/2013/isprsarchives-XL-
7-W2-173-2013.pdf, accessed: 2/9/2015
[7] L. Bergamasco and P.Asinari, Scalable methodology
for the photovoltaic solar energy potential assessment
based on available roof surface area: Further
improvements by ortho-image analysis and application to
Turin (Italy), Solar Energy, 85, pp. 2741–2756 (2011).
[8] Landmap (2014): Landmap Features Earth
Observation Collection. NERC Earth Observation Data
Centre, accessed:
2/9/2015. http://catalogue.ceda.ac.uk/uuid/42bcf75ae7f0b
2a12d84dfa2216c31e5
[9] M. Tarini, P. Cignoni, and C. Montani, Ambient
Occlusion and Edge Cueing to Enhance Real Time
Molecular Visualization, IEEE Transactions on
Visualization and Computer Graphics, 12, no. 5, , pp.
1237-1244, September/October 2006
[10] Geomatics, Environment
Agency, https://www.geomatics-
group.co.uk/GeoCMS/Order.aspx, http://environment.dat
a.gov.uk/ds/survey#/, accessed: 3/9/2015.
[11] B. Goss, I. Cole, T. Betts and R. Gottschalg,
Irradiance modelling for individual cells of shaded solar
photovoltaic arrays, Solar Energy, 110, pp. 410–419
(2014).
[12] I.R. Cole and R. Gottschalg, Optical modelling for
concentrating photovoltaic systems: insolation transfer
variations with solar source descriptions, IET Renewable
Power Generation, 9(5), pp.412-419, (2015), ISSN:
1752-1416. DOI: 10.1049/iet-rpg.2014.0369
[13] UK Meteorological Office. MIDAS Land Surface
Stations data (1853-current), [Internet]. NCAS British
Atmospheric Data Centre, 2006 - 2015. Available
from http://badc.nerc.ac.uk/view/badc.nerc.ac.uk__ATO
M__dataent_ukmo-midas
[14] P.R. Tregenza, Subdivision of the sky hemisphere
for luminance measurements, Lighting Research and
Technology, 19 (1), pp. 13-14, (1987). DOI:
10.1177/096032718701900103