Image Analysis Method for Quantifying

Snow Losses on PV Systems

Jennifer L. Braid Daniel Riley Joshua M. Pearce Laurie Burnham
PV and Materials Tech. PV and Materials Tech. Materials Science and Eng. PV and Materials Tech.
Sandia National Laboratories Sandia National Laboratories Michigan Technical University Sandia National Laboratories
Albuquerque, NM USA Albuquerque, NM USA Houghton, MI, USA Albuquerque, NM USA
jlbraid@sandia.gov driley@sandia.gov pearce@mtu.edu lburnha@sandia.gov

Abstract—Modeling and predicting snow-related power loss is years, however, both centralized and distributed PV systems
important to economic calculations, load management and system are seeing exponential growth in northern latitudes. As a result,
optimization for all scales of photovoltaic (PV) power plants. This investors and asset owners are increasingly concerned about
paper describes a new method for measuring snow shedding from
fielded modules and also describes the application of this method the impact of snow and ice on lifetime rates-of-return and are
to a commercial scale PV power plant in Vermont with two sub- asking the research community for better performance data
systems, one with modules in portrait orientation and the other and predictive models [3].
in landscape. The method relies on time-series images taken
at 5 minute intervals to capture the dynamics of module-level
snow accumulation and shedding. Module-level images extracted
from the full-field view are binarized into snow and clear areas,
allowing for the quantification of percentage snow coverage,
estimation of resulting module power output, and temporal
changes in snow coverage. Preliminary data from the Vermont
case study suggests that framed modules in portrait orientation
outperform their framed counterparts in landscape orientation
by as much as 24% energy yield during a single shedding event.
While these data reflect a single event, and do not capture snow
shedding behavior across diverse temperature and other climatic
conditions, the study nonetheless demonstrates that 1) module
orientation and position in the array influence shedding patterns;
2) the start of power production and bypass diode activation
differ for portrait and landscape module orientations at similar
percentages and orientations of snow coverage; and 3) system
Fig. 1. Leading edge of the modules in landscape orientation suggests that
design is an important factor in snow mitigation and increased
delayed snow shedding introduces reliability concerns when the snow partially
system efficiency in snowy climates. melts and then freezes, creating physical stress on the frame and module.
Index Terms—snow losses, snow shedding, module orientation,
design optimization
Such concerns appear justified: energy losses attributed
to snow adhesion on the surface of a PV module can be
I. I NTRODUCTION
significant, not only because the snow blocks light but because
Snow and ice accumulation on the front surface of photo- the reflectivity of the snow reduces its incident radiation [3],
voltaic (PV) modules is a recognized but poorly understood [4]. In regions of the world that regularly see snow, published
contributor to lifetime PV performance and reliability, not only estimates of snow losses range from as high as 100 percent
in areas such as northern Europe, Canada, and the US that over the course of a single month to 10 percent averaged
see persistent snow throughout the winter, but also at lower annually [5]. Previous studies have shown annual energy losses
latitudes where infrequent but record-breaking snow and hail from snow as high as 30% [6], but they are typically less than
storms can severely damage and impair solar installations [1], 10% annually [5], [7], [8]. These snow-related losses can drive
[2]. For decades, snow loading was largely overlooked in large the economic viability of a solar PV project. For example, the
part because so little solar was deployed in northern regions. 4.25% energy loss due to snow expected on an 8MW solar
With the precipitous drop in the cost of installed solar in recent farm in Ontario could result in annual loss of $140,000 in
solar electric revenue [9].
This work is funded in part or whole by the U.S. Department of Energy
Solar Energy Technologies Office, under Award Number 34363. JLB is These estimates, however, are largely based on small sys-
supported by the U.S. Department of Energy (DOE) Office of Energy tems and have not been fully modeled, nor has the impact
Efficiency and Renewable Energy administered by the Oak Ridge Institute of system design and architectural parameters, including tilt
for Science and Education (ORISE) for the DOE. ORISE is managed by
Oak Ridge Associated Universities (ORAU) under DOE contract number DE- angle, on snow losses been comprehensively investigated and
SC0014664. quantified. One reason for the latter is that contributing factors
are difficult to determine [10]. In addition to an immediate
decrease in power output, heavy snow and ice accumulation
on modules can cause physical damage to modules and/or ac-
celerate degradation of the inner components through moisture
ingress, freeze/thaw cycles, and mechanical stress (Figure 1).
As this paper demonstrates, greater efficiency is possible when
the specific factors that affect snow shedding and snow-related
energy losses, such as module architecture and orientation, are
considered in system design.
This work is part of a larger, multi-institutional, multi-
faceted effort to address the physical and meteorological con-
tributors to snow shedding as well as the overall performance
and reliability of northern PV systems [11]. Although our
specific focus in this paper is to quantify the impact of module
orientation on snow shedding and associated power loss, our
research more broadly aims to 1) identify mitigation strategies,
such as architectural choices, that could result in a quantifiable
reduction in snow losses, and 2) develop system-level design
recommendations for PV installations that regularly see snow.
A new method for detecting, quantifying, and analyzing
snow shedding rates, and for modeling the energy yield of
fielded PV systems, based on that snow-shedding data, is
described here. This method represents an important tool for
qualitatively and quantitatively characterizing the performance
of PV systems in snowy regions, as well as for evaluating
new snow shedding technologies and snow-related energy loss
mitigation strategies for commercial and utility-scale solar
installations.
In a limited demonstration study, we leveraged our method
to conduct a case study of module orientation, using time
series of images collected at 5 minute intervals to compare
various metrics for snow shedding and energy yield from
modules in landscape and portrait orientations.
II. M ETHODOLOGY
A. Data Collection
The snow shedding quantification method presented here is
based on time-series digital images of the modules, similar
to previous work completed on the Open Solar Outdoors Test
Field [12] to quantify the impacts of coatings on snow-related
PV system energy losses [13]. Although automated image
processing of PV modules is not new, the technique has mainly
been applied to electroluminescence (EL) images [14], [15].
Collecting field images requires installation of a camera
capable of capturing images at a set frequency (in this case,
we set the shutter frequency for 5-minute intervals). Other data
that add value to the analysis of performance include back-of-
module temperature, ambient temperature, plane-of-array irra-
diance data from a heated pyranometer and/or inverter power
data. Data can be collected and stored onsite or transmitted via
cellular modem to an offsite server. The latter approach allows
for regular monitoring of the data to ensure its availability and
is thus the preferred approach. The hardware used in our case
study is given in Table I.
TABLE I
DATA ACQUISITION SYSTEM FOR EXPERIMENTAL PV SYSTEM IN
V ERMONT
Data Make/Model Data description
Digital Go-Pro Hero8 Images collected and stored at an
images interval of 5 minutes during day-
light hours
Back-of- Omega, Type T One sensors affixed to each of two
module designated mid-row modules for
temperature each module type
Datalogger Campbell CR- Data collected at a frequency of 15
1000 secs; averaged per minute
B. Digital Image Analysis
Digital images were processed using a customized script in
the Python programming language v.3.6. Our method requires
extraction and planar indexing of individual modules from
point-in-time site images, similar to previous PV image pro-
cessing approaches [14], [15]. Extraction and characterization
of individual modules, as opposed to array-level analysis,
is important for snow loss evaluation because the power
production of a snowy array is not directly related to the
percent coverage. The series connections of cells and modules,
as well as the bypass diodes within a module, mean that
power production is dependent on the specific snow coverage
geometry on a module, as opposed to the overall percent
coverage on an array. Analyzing the individual modules in
the array, and cell strings in the module, allow for a more
accurate estimation of the array current, power, and energy
yield. Module-level images also allow for detailed analysis of
module snow shedding and power production as a function of
module position within the array.
Our snow loss quantification method is divided into four
main steps:
Step 1: Create Time Series of Images for Each Module
• Stabilization of full-field images. Image registration is to
compensate for camera movement.
• Selection of snow-free image. A full-field image is
selected without snow cover or sky reflection on the
modules.
• Color segmentation of snow-free image. The image is
converted to hue-saturation-value (HSV) color space, and
the image is binarized into cell and non-cell colors
• Determination of module boundaries. A contour algo-
rithm is used to approximate the edges of the binarized
image.
• Approximation of contours as quadrilaterals.
• Filtering of quadrilaterals based on expected module area.
• Planar-index transformation. The four corners of each
selected quadrilateral are used to map the module areas
from each image in the time series into an image propor-
tional to the actual module dimensions. This produces a
time series of images for each module.
This process is visualized in Figure 2.
 a
b
c current, where the open circuit voltage of one string is one
d n
Fig. 2. Steps of module extraction from the time series. (a) The image with
uncovered modules used for segmentation. (b) Module areas detected by HSV
color. (c) Contours created on module areas filtered by size and approximated
as quadrilaterals. (d) Quadrilaterals used to extract module areas from other
images in the time series.
Step 2: Map Snow Coverage on Each Module Image
In order to segment each module-level image into snow-
covered and clear regions, each image was again converted to
HSV color, and the color space was divided into snow and non-
snow regions, which were then used to binarize the images.
An example of this is given in Figure 3.
Step 3: Model Power Output
From the time series of module-level binary snow coverage
maps, we calculate simple percentage snow coverage and
percentage snow loss per minute. Using domain knowledge
of the modules electrical layout, we also model the power
output of each module as a function of the geometry of the
snow coverage. For example, a 60 cell module is divided into
3 strings of 20 cells (2 rows of 10 cells), each with its own
Fig. 3. Original RGB (left) and binarized (right) images of a portrait oriented
module during snow shedding. Snow from the module above can be seen
shedding onto the module, as well as snow shedding off the bottom of the
module.
bypass diode. Due to the series connection of Si-PV cells, the
current of a cell string is equal to the lowest current produced
by any cell. The percent of short circuit current of any given
string is equal to the complement of the highest percent
coverage for any cell in that string (100 − %coveragemax ).
In the module-level I-V curve, the string-level I-V curves are
added together in voltage, in order of decreasing short circuit
third of the module voltage. Assuming an equal fill factor
for the module-level and each string-level I-V curves, the
percent power of each string-level I-V curve is proportional
to its modeled short circuit current. Therefore by ranking the
cell string percent short circuit currents, the local maximum
percent power points for each step of the module-level I-V
curve can be approximated as:
%Pn = × %ISCn (1)
3
n
= × (100 − %coveragemax−n ) (2)
3
where n=1 corresponds to the highest cell string short circuit
current, and n=3 the lowest. The largest of these local maxima
is recorded as the percent power output for the module at the
time step of the image.
Step 4: Quantify Shedding Rates
From the time series of percent coverage and percent power,
we can generate several metrics. While it is possible to
generate an instantaneous shedding rate (% coverage shed per
minute), this is not particularly useful due to the commonality
of snow shedding from higher modules onto modules below,
and subsequent re-shedding of the lower modules. Therefore
we evaluate shedding rates in terms of time scales for clearing
and power generation. The time series begins at a fixed time
before shedding begins, and ends after all modules have fin-
ished shedding and return to full power output. By integrating
the modeled power over time, we obtain the energy yield for
each module over this period. We also capture several time
metrics related to performance: shedding start time, power
production start time, 90% power production time, shedding
finish time. The relative differences in these times are useful
for comparing the shedding behavior of different modules and
arrays.
III. M ODULE O RIENTATION : A C ASE S TUDY
The data presented in this paper are from a commercial-
scale, 132kW PV site in Vermont that provides a near ideal
test-bed for measuring the impact of module orientation on
snow losses. The site has two adjacent PV systems that feed
into the same grid tie: a 58kW array (190W poly c-Si mod-
ules), arranged in landscape orientation, and a 74kW system a
(255W poly c-Si modules) mounted in portrait orientation (see
Figure 2a). The areas of the modules are 1.49 and 1.63 m2 ,
respectively, and are from different manufacturers, but both
types are 60-cell poly c-Si. Both are ground-mount systems,
installed at a fixed-tilt angle of 30 degrees.
With permission from the site owner, we have installed a
monitoring system (see Table I) to collect irradiance and back-
of-module temperature data; we have also installed a camera
to collect visual images of each array at 5-minute intervals,
enabling us to obtain quantitative data on snow shedding
needed to validate our new snow-shedding method.
The camera had a field of view that included four partial
arrays (Figure 2a). Although installed in mid-February, Ver- b
mont had an unusually dry winter and did not see significant Fig. 4. Time series of percent snow coverage, percent snow shed per minute,
snow until a 9 inch snowfall on March 24th, 2020, as reported and modeled percent power for (a) a portrait oriented module (b) a landscape
oriented module.
by NOAA weather site in nearby Corinth, VT. We were able
to process images from the subsequent snow shedding on
March 25, 2020, however, using our image analysis method From the module time-series images, we quantified time
to segment the landscape and portrait-oriented module images scales related to shedding for each of the two module ori-
into snow- and non-snow areas. Time-series Graphic Interface entations. The time series were subset to a 445 minute win-
Format images (GIFs) of the original and binarized module dow around the shedding events for all modules. Subsetting
images, as well as time-series graphs of snow coverage per- to common times allows for intercomparability of the time
centage and modeled power are available online [11]. We show scales related to snow shedding and power production. The
time-series graphs for one module of each orientation in Figure median results for the populations of all extracted portrait and
4. While we have confidence in our data for this one event, it landscape oriented modules are given in Table II.
must be noted that the conclusions drawn below represent a
specific set of climate conditions and are not expected to apply TABLE II
to all snow events, which vary in temperature, snow depth, M EDIAN PERCENT MAXIMUM ENERGY YIELD AND TEMPORAL SHEDDING
METRICS FOR ALL PORTRAIT AND LANDSCAPE ORIENTED MODULES . 445
humidity, etc. (and may therefore be reversed or significantly MINUTE DATA WINDOW BEGINS BEFORE ANY MODULES SHED SNOW, AND
changed when we examine a full winters worth of data). ENDS AFTER ALL MODULES HAVE FINISHED SHEDDING .
Future work is clearly necessary to use our method in many
snow locations to develop a generalized model, which can be Orientation Energy Time Time to Time Time to
yield to start produce to 90% finish
done with a relatively small investment in camera imaging and (%) shed power power shed
logging. (min) (min) (min) (min)
Also, the nominal module operating temperature (NMOT) Landscape 59.3 47.5 65 310 310
can vary among modules, even of the same brand and model. Portrait 73.5 30 55 185 212.5
In this study we monitored the backsheet temperature of
3 modules (2 landscape and 1 portrait), and the ambient
temperature, throughout the shedding event. There was no Our snow-shedding results show a significant difference
appreciable difference in temperature between the two module in power output between the two module orientations. The
types, so we omitted temperature data from further analysis in landscape orientation modules produced less than 60% of their
this study. However a baseline temperature could be important maximum power during the measurement window, while the
for future studies. portrait orientation modules produced 73.5%. This implies that
a system would produce nearly 1.5 kWh/kW more for this
event if the modules were oriented in portrait orientation ver-
sus landscape. Moreover, landscape oriented modules began
shedding 7.5 minutes later, began producing power 10 minutes
later, and reached 90% and full power 125 and 97.5 minutes
later than portrait oriented modules.
Those calculations, however, can be further refined based
on a module’s position in an array. Modules at the edge of an
array may be more susceptible to mini-climatic effects such as
wind and ambient temperature differences, and also have fewer
physical impediments to shedding, as they are not surrounded
on all sides by other modules. We therefore considered the
modules’ position in gathering the metrics in Table III. We
include the vertical position of the module as well (top, middle,
or bottom row),because snow shedding from higher modules
onto lower ones, and adjacent modules acting as impediments,
also affect shedding.
As shown in Table III, the portrait oriented modules out-
performed their landscape counterparts in energy yield for
all corresponding positions in the array. The reason is that
all portrait oriented modules, regardless of location, start and
finish snow shedding before the landscape oriented modules.
What is most interesting, however, is that module position –
not orientation – was the biggest factor in the start of power
generation. While the portrait modules on the bottom of the
array were the first to produce power, the portrait modules
at the top of the array were the last. Similarly, landscape
modules on the top row were the last of their orientation to
produce power. For both orientations, the top row of modules
began shedding snow and producing power later than their
counterparts. Additionally, the top row of portrait modules
were the last to produce power, but the first to reach full power.
Our time-series graphs explain why.
Portrait and landscape oriented modules on the bottom row
tended to shed their snow in a few large sheets, clearing
themselves earlier than other modules. However, top-row
modules would shed snow more slowly as a sheet, with pieces
occasionally breaking off and traversing the lower modules.
These pieces would cause shading and sometimes stick to
the lower modules until melting. As a result, the bottom
row of modules had an overall higher energy yield, despite
completing shedding after the upper modules were clear. This
behavior is demonstrated in Figure 4a.
The impact of shedding behavior on power production
reflects the internal circuitry of the modules, including the
location of the bypass diodes. The modeled power for a portrait
oriented module, for example, would remain zero until less
than one full cell width along the bottom of the module was
covered with snow. In contrast, landscape oriented modules on
the top row experiencing the same shedding mechanism and
rate would produce energy much earlier, as the snow would
allow for cell-level strings to begin producing power as the
snow moved vertically down the module. This behavior is
demonstrated by the step-like power gains in Figure 4b.
IV. D ISCUSSION
A. Error in image analysis
In analyzing images as described above, numerous errors in
either snow coverage or modeled power can arise. The angle
of the camera relative to the module area affects not only the
perceived coverage of the module, but also the color of the
vertical faces of deep snow as pieces of the sheet break off and
shed, as shown in Figure 2d. These factors affect the accuracy
of the snow detection as they can cause clear areas to be
classified as snowy, and vice versa. Module or array movement
is a similar source of error. While we can stabilize images for
camera movement, we cannot correct for movement of the
arrays. As some of these modules are mounted on variable
angle racking, shedding of large amounts of snow results in a
slight change in the array angle. This causes the module frame
to be included in the extracted image. Because the frame is
lightly colored, our segmentation step may classify the frame
area as snow, which decreases the modeled power. These errors
can be decreased, however, by choosing a camera angle near
normal to the face of the modules.
B. Expansion of this work
Our intent next year is to improve the snow-loss measure-
ment methodology described here, and apply it more broadly
to multiple PV sites and technologies over the full winter
season. In addition to modifying our camera setup to improve
image quality and reduce the analysis errors described above,
we will incorporate temperature, irradiance, and inverter or
micro-inverter data for improved shedding quantification and
energy yield validation. When considering systems with dif-
ferent NMOTs or different locations, we will use module and
ambient temperature data to normalize shedding statistics. We
will also use module temperature and plane of array irradiance
data to improve on our power prediction model, which we will
validate with module or array-level inverter data. Eventually,
this work could adapt a modified version [16] of the Sandia
array performance model [17] to make a complete model for
all PV systems in snowy environments.
The results above suggest that 1) module orientation is
a significant factor in snow losses and 2) the snow losses
attributable to orientation are quantifiable and can be modeled
based on module design, system design, and meteorological
data. Our case study in Vermont demonstrates not only that
one can quantify snow losses based on image analysis, but
the interplay of snow shedding and coverage mapping is an
important factor in the design of high-efficiency PV systems
in snowy regions. A limitation of this study is that we do not
know the front-side glass coatings or texturing (anti-reflection
or anti-soiling) used by the different module manufacturers.
These factors may affect the rate of module snow shedding and
will be considered in future studies of this and other systems.
We also intend to build on our earlier works that consider
the rear irradiance on bifacial modules [18], and the frame as a
contributor to snow losses [19]. Our visual data from Vermont
suggests the frame creates a significant physical impediment
 TABLE III
M EDIAN PERCENT MAXIMUM ENERGY YIELD AND TEMPORAL SHEDDING METRICS FOR PORTRAIT AND LANDSCAPE ORIENTED MODULES SORTED BY
MODULE POSITION THROUGH A 445 MINUTE DATA WINDOW.

Orientation Position Edge Energy Time Time to Time Time to

yield to start produce to 90% finish
(%) shed power power shed
(min) (min) (min) (min)
Portrait Top Edge 80.2 25 72.5 97.5 125
Portrait Bottom Edge 74.7 25 35 230 240
Portrait Top Non-Edge 64.7 30 115 185 225
Portrait Bottom Non-Edge 72.8 30 37.5 182.5 230
Landscape Top Edge 54.8 45 70 275 280
Landscape Middle Edge 60.5 45 45 285 285
Landscape Bottom Edge 66.0 45 55 285 290
Landscape Top Non-Edge 55.0 57.5 72.5 310 310
Landscape Middle Non-Edge 54.1 37.5 65 357.5 357.5
Landscape Bottom Non-Edge 68.1 55 65 330 335

to snow shedding, but the ratio of frame length to height
(in landscape orientation) is a far bigger contributor than the
presence or absence of a frame. We intend to quantify this
factor and include it in a revised snow-shedding model [10].
It has already been established that the spectral albedo [20]
can play a substantial role in PV performance particualrly
in snowy environments [21]. With the rapid technical rise
of bifacial PV [22] it is important to incorporate both the
positive (albedo [20], [21] and rapid snow shedding) and
negative (snow losses) [18], [23], [24] to enable more straight-
forward financing [25] and eventually securitization [26]–[28]
for mass-scale deployment.
Our research does not take into account the physics of
sliding snow and the contributions of sliding distance in
generating momentum [29], but that analysis is less germane
to our premise that module orientation is a primary factor in
determining energy losses attributable to snow adhesion on the
module surface and along the lower edge of the frame. Nor
do we investigate tilt angle as a factor as this has already been
studied [30]. This study also does not take into account the
impact of retained snow on module reliability. As suggested by
Figure 1, reduced rates of snow shedding can introduce risk
from freeze-thaw cycles, which form ice dams at the lower
edge of an array. These dams, which can persist for days (or
for suboptimal systems in heavy snow environments – months
[6]) and be impacted by additional snow load, may stress the
integrity of the edge seal and also induce cell cracking. We
will carefully track the latter in future investigations.
V. C ONCLUSION
We have presented a method for quantifying snow coverage
of PV modules and modeling the corresponding loss of module
power, and temporal metrics for comparing snow shedding
rates between modules and systems. While our limited data
to date does not allow us to draw statistical conclusions
about the role of module orientation on snow shedding and
module power, we have explored both the limitations, and
potential applications and expansions of this work for studying
the behavior of PV systems in snowy climates and testing
snow-related power loss mitigation strategies. Being able to
quantify the impact of module orientation on energy losses in
regions that consistently see snow accumulation in winter has
significant value to a broad community of manufacturers, asset
owners, investors, etc. Our results provide a technical rationale
for establishing a set of design recommendations/standards
that recognize climate-specific performance challenges. The
anticipated outcome of such mitigation strategies will be
greater system efficiencies and decreased degradation rates at
northern latitudes, helping reduce the levelized cost of energy
and furthering the expansion of solar energy.
ACKNOWLEDGMENT
The authors would like to thank Skip Metayer of FarmWay,
Inc, and Kevin McCollister of Catamount Solar for supporting
this study and providing technical assistance.
Sandia National Laboratories is a multimission laboratory
managed and operated by National Technology and Engineer-
ing Solutions of Sandia LLC, a wholly owned subsidiary of
Honeywell International Inc. for the U.S. Department of Ener-
gys National Nuclear Security Administration under contract
DE-NA0003525.
R EFERENCES
[1] R. E. Pawluk, Y. Chen, and Y. She, “Photovoltaic electricity generation
loss due to snow–A literature review on influence factors, estimation,
and mitigation,” Renewable and Sustainable Energy Reviews, vol. 107,
pp. 171–182, 2019.
[2] R. W. Andrews, A. Pollard, and J. M. Pearce, “The effects of snowfall
on solar photovoltaic performance,” Solar Energy, vol. 92, pp. 84–97,
2013.
[3] E. Andenæs, B. P. Jelle, K. Ramlo, T. Kolås, J. Selj, and S. E. Foss,
“The influence of snow and ice coverage on the energy generation from
photovoltaic solar cells,” Solar Energy, vol. 159, pp. 318–328, 2018.
 [4] C. Good, H. Persson, Ø. Kleven, W. Sulkowski, K. Ellingsen, and [24] L. Kreinin, A. Karsenty, D. Grobgeld, and N. Eisenberg, “Pv systems
T. Boström, “Solar cells above the Arctic Circle-Measuring character- based on bifacial modules: Performance simulation vs. design factors,”
istics of solar panels under real operating conditions,” in Proc 24th in 2016 IEEE 43rd Photovoltaic Specialists Conference (PVSC). IEEE,
European Photovoltaic Solar Energy Conference, 2009, pp. 21–5. 2016, pp. 2688–2691.
[5] T. Townsend and L. Powers, “Photovoltaics and snow: An update [25] J. Meydbray, “Barriers to financing bifacial pv projects,” Present. 5th
from two winters of measurements in the Sierra,” in 2011 37th IEEE Bifi PV Work. Denver, Color, pp. 1–15, 2018.
Photovoltaic Specialists Conference. IEEE, 2011, pp. 003 231–003 236. [26] T. Alafita and J. M. Pearce, “Securitization of residential solar photo-
[6] N. Heidari, J. Gwamuri, T. Townsend, and J. M. Pearce, “Impact of snow voltaic assets: Costs, risks and uncertainty,” Energy Policy, vol. 67, pp.
and ground interference on photovoltaic electric system performance,” 488–498, 2014.
IEEE Journal of Photovoltaics, vol. 5, no. 6, pp. 1680–1685, 2015. [27] D. Hyde and P. Komor, “Distributed pv and securitization: Made for
[7] T. Sugiura, T. Yamada, H. Nakamura, M. Umeya, K. Sakuta, and each other?” The Electricity Journal, vol. 27, no. 5, pp. 63–70, 2014.
K. Kurokawa, “Measurements, analyses and evaluation of residential [28] T. Lowder and M. Mendelsohn, “Potential of securitization in solar pv
pv systems by japanese monitoring program,” Solar energy materials finance,” National Renewable Energy Lab.(NREL), Golden, CO (United
and solar cells, vol. 75, no. 3-4, pp. 767–779, 2003. States), Tech. Rep., 2013.
[8] L. Powers, J. Newmiller, and T. Townsend, “Measuring and modeling [29] J. Bogenrieder, C. Camus, M. Hüttner, P. Offermann, J. Hauch, and
the effect of snow on photovoltaic system performance,” in 2010 35th C. J. Brabec, “Technology-dependent analysis of the snow melting and
IEEE Photovoltaic Specialists Conference. IEEE, 2010, pp. 000 973– sliding behavior on photovoltaic modules,” Journal of Renewable and
000 978. Sustainable Energy, vol. 10, no. 2, p. 021005, 2018.
[9] R. W. Andrews and J. M. Pearce, “Prediction of energy effects on [30] A. Granlund, J. Narvesjö, and A. M. Petersson, “The Influence of Tilt
photovoltaic systems due to snowfall events,” in 2012 38th IEEE on Snow Shadowing of Frameless Bifacial Modules,” in 36th Euro-
Photovoltaic Specialists Conference. IEEE, 2012, pp. 003 386–003 391. pean Photovoltaic Solar Energy Conference and Exhibition, Marseille,
[10] B. Marion, R. Schaefer, H. Caine, and G. Sanchez, “Measured and September 9-13, 2019, 2019, pp. 1646–1650.
modeled photovoltaic system energy losses from snow for colorado and
wisconsin locations,” Solar Energy, vol. 97, pp. 112–121, 2013.
[11] “The Impact of Snow on PV Performance.” [Online].
Available: https://energy.sandia.gov/programs/renewable-energy/
solar-energy/photovoltaics/pv-systems-and-reliability/
snow-as-a-factor-in-photovoltaic-performance-and-reliability/
[12] J. Pearce, A. Babasola, and R. Andrews, “Open solar photovoltaic
systems optimization,” in Proceedings of the 16th Annual National
Collegiate Inventors and Innovators Alliance Conference, 2012, pp. 1–7.
[13] R. W. Andrews, A. Pollard, and J. M. Pearce, “A new method to
determine the effects of hydrodynamic surface coatings on the snow
shedding effectiveness of solar photovoltaic modules,” Solar energy
materials and solar cells, vol. 113, pp. 71–78, 2013.
[14] A. M. Karimi, J. S. Fada, M. A. Hossain, S. Yang, T. J. Peshek,
J. L. Braid, and R. H. French, “Automated pipeline for photovoltaic
module electroluminescence image processing and degradation feature
classification,” IEEE Journal of Photovoltaics, vol. 9, no. 5, pp. 1324–
1335, 2019.
[15] A. M. Karimi, J. S. Fada, N. A. Parrilla, B. G. Pierce, M. Koyutürk,
R. H. French, and J. L. Braid, “Generalized and Mechanistic PV Module
Performance Prediction from Computer Vision and Machine Learning on
Electroluminescence Images,” IEEE Journal of Photovoltaics, vol. 10,
no. 3, pp. 878–887, 2020.
[16] R. W. Andrews, A. Pollard, and J. M. Pearce, “Improved parametric
empirical determination of module short circuit current for modelling
and optimization of solar photovoltaic systems,” Solar Energy, vol. 86,
no. 9, pp. 2240–2254, 2012.
[17] D. L. King, J. A. Kratochvil, and W. E. Boyson, Photovoltaic array
performance model. United States. Department of Energy, 2004.
[18] L. Burnham, D. Riley, B. Walker, and J. M. Pearce, “Performance of
bifacial photovoltaic modules on a dual-axis tracker in a high-latitude,
high-albedo environment,” in 2019 IEEE 46th Photovoltaic Specialists
Conference (PVSC). IEEE, 2019, pp. 1320–1327.
[19] D. Riley, L. Burnham, B. Walker, and J. M. Pearce, “Differences in Snow
Shedding in Photovoltaic Systems with Framed and Frameless Modules,”
in 2019 IEEE 46th Photovoltaic Specialists Conference (PVSC). IEEE,
2019, pp. 0558–0561.
[20] M. Brennan, A. Abramase, R. W. Andrews, and J. M. Pearce, “Effects of
spectral albedo on solar photovoltaic devices,” Solar Energy Materials
and Solar Cells, vol. 124, pp. 111–116, 2014.
[21] R. W. Andrews and J. M. Pearce, “The effect of spectral albedo
on amorphous silicon and crystalline silicon solar photovoltaic device
performance,” Solar Energy, vol. 91, pp. 233–241, 2013.
[22] T. S. Liang, M. Pravettoni, C. Deline, J. S. Stein, R. Kopecek, J. P.
Singh, W. Luo, Y. Wang, A. G. Aberle, and Y. S. Khoo, “A review
of crystalline silicon bifacial photovoltaic performance characterisation
and simulation,” Energy & Environmental Science, vol. 12, no. 1, pp.
116–148, 2019.
[23] E. Molin, B. Stridh, A. Molin, and E. Wäckelgård, “Experimental yield
study of bifacial pv modules in nordic conditions,” IEEE Journal of
Photovoltaics, vol. 8, no. 6, pp. 1457–1463, 2018.