2. 2
available in Qatar until now. There is also a need to study the
relationship between impact of dust and environmental
conditions such as airborne particulate matter concentration,
wind, temperature, and humidity [17].
The objective of the study was to obtain data of PV system
performance degradation due to dust deposition, to obtain data
of ambient dust and weather conditions, and to determine the
correlation between the former and the latter. In the following
sections the methods and results of this study are described,
following which the conclusions are presented.
II. METHODS
Data collection of this study was carried out in the Solar
Test Facility located at the Qatar Science & Technology Park
(QSTP), Doha, Qatar. Data collection for this study occurred
in June 01 through December 31, 2014.
Fig. 1. A DustTrak® airborne dust concentration monitor
installed at the Solar Test Facility
A. Measurement and Calculation of PV Module Performance
Three PV arrays were used in this study, each comprising
eight 220 Wp polysilicon PV modules, tilted at 22° and facing
due South, in a single string connected to identical grid-tied
inverters. The arrays’ DC electrical parameters and module
back surface temperatures were measured at maximum power
point condition once per minute. DC power, voltage and
current were measured via transducers with +/- 0.5%
accuracy. Module temperatures were measured via
permanently attached thermocouple sensors, with unspecified
accuracy.
One array was cleaned every week (“high wash”), one
every second month (“medium wash”), and one every sixth
month (“low wash”). During the test period, the “low wash”
event occurred on 25th
June 2014, and the “medium wash”
events on 25th
June, 2nd
September, and 4th
November 2014.
There were two significant rain events, which occurred on 24th
November and 1st
December 2014.
A “cleanness index” was used in this study as a metric for
the effect of soiling on PV performance ratio. It is defined as
the ratio of a PV module’s temperature-corrected performance
ratio to that of a clean PV module. Its physical meaning is
similar to the “soiling ratio” that has used by other researchers
[3]. The temperature-corrected performance ratio of a PV
module is determined as:
PRT _corr =
PDC _i
1+δ Tcell _i −TSTC( )i
∑
PSTC
GPOA_i
GSTC
#
$
%
&
'
(
i
∑
(1)
Where:
The summation is over ever 24-hour day, from the first
minute after midnight to the last minute before midnight.
PDC_i is the array’s power at maximum power point in the
ith
minute of a day [kW].
PSTC is the array’s power rating at maximum power point,
at standard test conditions (STC), from flash-test data
[kW].
GPOA_i is the measured plane of array (POA) irradiance in
the ith
minute of the day [kW/m2
].
GSTC is the irradiance at the standard test conditions (1 kW
m-2
).
Tcell_i is the average array temperature in the ith
minute of
the day [˚C].
TSTC is the temperature at the standard test conditions (25
˚C).
δ is the temperature coefficient for power of the arrays
(-0.485 % ˚C-1
)
The temperature-corrected performance ratio is similar in
concept to the “weather-corrected performance ratio” defined
in a NREL report [18]. The temperature-corrected
performance ratio in this study uses the PV module’s DC
power output, and is corrected to the temperature at STC (25
˚C). In contrast, the “weather-corrected performance ratio”
uses the PV module’s AC power output, and is corrected to a
locality-dependent temperature based on the project weather
file [18].
The cleanness index of a PV module, in a 24-hour day, is
then calculated as follows:
CI =
PRT _corr
PRT _corr _clean
(2)
Where:
PRT_corr is the temperature-corrected performance ratio of
the PV module whose cleanness index is being evaluated.
PRT_corr_clean is the temperature-corrected performance ratio
of a “clean” PV module. Based on the average PRT_corr of a
weekly-cleaned PV module during the test period, a
constant value of 0.88 was assigned for PRT_corr_clean.
The metric CI is a measure of a PV module’s cleanness. Its
value decreases as the PV module’s soiling level increases. It
takes into account the effect of soiling on module temperature.
It was found through experiment that more heavily soiled
modules tended to be several degrees cooler than clean
modules, presumably because the deposited dust served as a
thermal barrier from the sun’s irradiation. A clean PV module
should have a CI value of unity. Because a constant value of
0.88 was used for the clean module’s temperature-corrected
performance ratio, some daily CI values were slightly greater
3. than unity due to measurement uncertainty. However, this
should have no effect on the objectives and conclusions of this
study.
The daily change of CI for each day was calculated using
the following equation:
∆ CIn = CIn −CIn−1 (3)
Where:
∆CIn is the change in cleanness index of a PV module
attributed to the nth
day.
CIn is the cleanness index of the PV module on the nth
day.
CIn-1 is the cleanness index of the PV module on the (n-1)th
day.
B. Measurement of Ambient Dust and Weather Conditions
Ambient dust concentration (mg m-3
) in term of PM10 was
continuously measured by using a TSI 8533EP DustTrak®
DRX Aerosol Monitor (TSI Inc., Shoreview, MN, USA) with
a temporal resolution of 2 min, stationed at the Solar Test
Facility over the entire study period. This instrument is a
continuous 90° light-scattering laser photometer that produces
size-segregated mass fraction concentrations corresponding to
PM1, PM2.5, PM4, PM10 size fractions. It has a minimum
detectable particle size of 0.1 µm and a sensitivity of 0.001 mg
m-3
. It uses a constant sample flow rate of 3 l m-1
,
automatically controlled by an external pump. The instrument
was set to auto-zero once every 15 minutes using an external
zeroing module, in order to minimize the effect of zero drift.
The instrument was placed inside an environmental enclosure,
which was mounted on a tripod at a height of 1.5 m above
ground.
Ambient air temperature, relative humidity, wind speed and
wind direction were recorded at one minute intervals
continuously every day during the test period. Daily average
of dust concentration, temperature, relative humidity and wind
speed was computed using the usual arithmetic mean for all
data points within a 24-hour day. Daily wind direction was
computed by treating all angular measurements as point on the
unit circle and computing the resultant vector of the unit
vectors determined by data points [19].
C. Data Processing by Multi-Variable Regression
In this study, a multi-variable linear regression model was
used to examine the correlation between daily change of the
cleanness index cleanness index and the daily ambient
environmental conditions. Daily ∆CI was used as the
dependent variable. Three daily average ambient
environmental parameters, namely dust concentration, wind
speed, and relative humidity were used as the independent
variables. The regression model predicts the dependent
variable as a linear function of the independent variables:
∆ CIPre = β0 + β1PM10 + β2WS + β3RH (4)
Where:
∆CIPre is the predicted value of ∆CI on a 24-hour day, which
is described in Eqn. (3).
PM10 is the 24-hour average concentration of particles
smaller than 10 µm in aerodynamic diameter in ambient air,
measured experimentally as described in previous sections.
WS is the 24-hour average of wind speed based on
experimental measurement, as described in previous
sections.
RH is the 24-hour average of relative humidity measured
experimentally as described in previous sections.
β0, β1, β2, β3 are coefficients to be determined using the
experimental data, by minimizing the sum of squares of the
error.
III. RESULTS
The cleanness index of the “low wash” and “medium wash”
PV arrays decreased substantially over the course of this
study. Ambient dust concentration had a significant effect on
the daily change of cleanness index. Weather conditions also
affected the daily change of the cleanness index.
A. Cleanness Index and Daily Change
Fig. 2 shows the cleanness index of the three test PV arrays
in the months of June through December 2014. On average,
the cleanness index decreased 0.0042 (standard deviation
0.0080) per day and 0.0045 (standard deviation 0.0091) per
day over the study period (seven months) for the “low wash”
and “medium wash” PV arrays, respectively.
Two individual dust episodes on the days of 25th
August
and 12th
December caused the cleanness index to decrease
0.027 and 0.024, respectively (for both “low wash” and
“medium wash” arrays).
The PV arrays were also naturally cleaned by rain, on 24th
November and 1st
December 2014, which restored the test
arrays’ cleanness index to around unity.
Fig. 2. The cleanness index of the PV arrays with different
cleaning frequencies
B. Correlation with Dust and Meteorological Data
Examination of the data revealed that, dust concentration
(PM10), wind speed (WS), and relative humidity (RH) had the
most significant correlation with the daily ∆CI. The
correlation coefficient between any single environmental
variable with ∆CI was lower than 0.5, suggesting the
complexity of the soiling process. The daily ∆CI after a
cleaning or rain event was not included in the correlation
analysis. The mean and standard deviation values for PM10,
WS, and RH are given in Table I, which will be referred to in
the following sections.
As shown in Fig. 3, daily ∆CI (excluding cleaning and rainy
days) and daily PM10 both varied significantly from day to day
4. in the months studied. As can be seen from Fig. 3, ∆CI is in
fact positive on many days, suggesting that the level of soiling
actually reduced on those days. By examining the weekly
moving averages of daily ∆CI and daily PM10, one may see
that the trends of these two variables are generally opposite to
each other. In other words, in periods when ∆CI was
increasing, PM10 would generally decrease. This correlation
between ∆CI and PM10 may also be observed in Fig. 4.
However, it may also be observed that the relation between
daily ∆CI and daily PM10 was complex, suggesting there was
interaction with other variables as well.
Fig. 3. Daily ∆CI of the “medium wash” PV array and daily
dust concentration PM10 (lines are weekly moving average of
the daily values)
The correlation between wind speed and daily ∆CI may be
seen in Fig. 5. In general, the daily loss of PV performance
loss due to soiling is greater at lower wind speeds (i.e., daily
∆CI is more negative). On days with high wind speed, it is
more likely to see positive daily ∆CI (partial performance
recovery of soiled PV modules), except during a dust storm.
One explanation for this observation is that higher wind
speeds cause higher re-suspension of deposited dust on PV
panels [20]. Therefore it is possible for the dust deposition of a
PV module to decrease on a high-wind day, and hence the PV
performance would actually recover. The effect of wind speed
on dust deposition on solar surfaces has been noted by other
researchers, albeit under laboratory conditions that are very
different from this study, using Belgian Brabantian loess dust
at high concentrations (0.56 – 2.25 g m-3
) and a narrower
range of wind speeds (0.63 – 2.59 m s-1
) [21, 22].
It should be pointed out that higher wind speeds may cause
PV performance ratio to increase due to the stronger cooling
of PV under high wind conditions [18, 23]. However, since the
daily ∆CI metric uses temperature-corrected performance
ratio, the effect of wind speed on module temperature is taken
into account. Therefore, the wind speed effect on daily ∆ CI
observed in this study should be attributed to the role of wind
speed in dust deposition and re-suspension of deposited dust.
Fig. 4. Daily ∆CI and daily average dust concentration
Fig. 5. Daily ∆CI and daily average wind speed
Fig. 6. Daily ∆CI and daily average relative humidity
The relation between daily ∆CI and daily average relative
humidity is shown in Fig. 6, which suggests that relative
humidity has some impact on the PV soiling. Overall, daily
∆CI was more negative on days with higher relative humidity
levels. This is intuitively consistent with the perception that
higher relative humidity causes dust particles to more likely
“stick” to the PV module, and less likely to be re-suspended
by wind. In other words, with increasing higher relative
TABLE I
STATISTICS OF AMBIENT CONDITION VARIABLES
Variable Mean Standard Deviation
PM10 (mg m-3
) 0.094 0.032
WS (m s-3
) 2.0 0.78
RH 49% 14%
5. humidity, PV soiling is likely to be more severe, provided
other parameters are kept the same.
It should be noted that the opposing diurnal patterns wind
speed and relative humidity might have enhanced the effect of
wind speed on PV surface soiling. The daily peak of wind
speed was found to occur at the same time as the daily
minimum of the relative humidity (data no shown). This
suggests that the dust resuspension effect of high winds is
enhanced when the deposited dust particles contain the lowest
moisture, which make them less sticky and more likely to be
carried away by the wind.
Fig. 7. Daily ∆CI and daily average wind direction (Note: the
inside solid circular line represent a ∆CI value of zero.)
Fig. 8. Wind rose showing distribution of wind speed and
direction
The relation between wind direction and daily ∆CI is
complex, as can be seen in Fig. 7. The daily ∆CI is mostly
negative when the wind comes from the south; when the wind
comes from the north, both negative and positive values are
possible for ∆CI. The prevailing wind is from the northwest,
as shown in Fig. 8. The prevailing wind covers the entire
spectrum of wind speed, but winds from other directions
appear to be only available at relatively low wind speeds. In
other words, wind direction and wind speed are not
independent of each other. On the other hand, the PM10 and
wind direction plot (Fig. 9) shows that, the prevailing wind is
associated with the entire range of dust concentration, but
wind from other directions is generally associated with
medium-to-high dust concentrations. In other words, wind
direction was not included in the multi-variable regression in
this study. Due to the fact that wind direction is not
independent of wind speed and dust concentration, it is not
included in the multivariable regression.
Fig. 9. Plot of daily average PM10 and daily average wind
direction
C. Multivariable Regression Results
The coefficients for Eqn. (4) derived from the multivariable
regression analysis are given in Table II.
Using Eqn. (4) and the set of coefficients in Table II, one
can calculate the predicted ∆CI under various ambient
conditions. Using the mean values from Table I, we can
calculate the predicted ∆CI under “Mean Ambient
Conditions”. We can see that the predicted ∆CI under “Mean
Ambient Conditions” is significantly larger in magnitude
(more negative) than the mean measured ∆CI. This
discrepancy may be partly attributed to the fact that ∆CI is
apparently not a linear function of the ambient environmental
variables, and hence the predicted ∆CI under “Mean Ambient
Conditions” should not be expected to equal the mean
experimental ∆CI. Nevertheless, as a semi-quantitative
approximation, the linear regression model may be used to
assess each environmental variable’s contribution to the
variation of the daily ∆CI. Using the standard deviation values
from Table I, one can calculate the predicted ∆CI with each
variable increased or decreased by one standard deviation,
while keeping the other variables at their mean values. Such
results are shown in the “Varied by One Standard Deviation”
rows in Table III. We can see that a one standard deviation
change in WS or RH causes significantly greater variation in
the predicted ∆CI, than a one standard deviation change in
PM10 would. This suggests that the variation of ∆CI observed
in this study may have been caused by wind speed or relative
humidity variation more than by dust concentration variation.
The multivariable regression model is only a preliminary
approximation of the relationship between ∆CI and the
environmental variables. Dust deposition velocity, which
relates dust deposition flux and ambient dust concentration,
6. has a non-linear dependence on wind speed and particle size
[24], and likely has a non-linear relationship with relative
humidity. Therefore, additional work is needed to have a
better understanding of such relationships, so that one may be
able to more accurately predict the effect of dust and weather
conditions of PV performance loss over long periods of time.
IV. CONCLUSIONS
The results of this study show that surface soiling due to
dust deposition causes significant loss in PV power output in
Doha, Qatar. A “cleanness index”, defined as the ratio of
temperature-corrected performance ratio of a soiled PV
module to that of a clean module placed in identical position
and environment, was introduced to quantify the soiling effect
on PV power output. On average, the cleanness index of a PV
module cleaned every second month may decrease by 0.45
percentage points per day, or 10-20 percentage points per
month, due to dust deposition alone. Dust concentration, wind
speed, and relative humidity are the most important factors
affecting surface soiling. The mathematical relationship
between daily change of the cleanness index and the ambient
environmental variables is yet to be determined through
further research.
REFERENCES
[1] E. Kymakis, S. Kalykakis, and T. M. Papazoglou, "Performance
analysis of a grid connected photovoltaic park on the island of
Crete," Energy Conversion and Management, vol. 50, pp. 433-438,
Mar 2009.
[2] H. Qasem, T. Betts, and H. Miillejans, "Dust effect on PV
modules," presented at the Proceedings of 7th photovoltaic science
application and technology conference (PVSAT-7), Edinburgh,
UK, 2011.
[3] M. Gostein, B. Littmann, J. R. Caron, and L. Dunn, "Comparing
PV power plant soiling measurements extracted from PV module
irradiance and power measurements," in Photovoltaic Specialists
Conference (PVSC), 2013 IEEE 39th, Tampa, FL, 2013, pp. 3004 -
3009.
[4] J. Zorrilla-Casanova, M. Piliougine, J. Carretero, P.
Bernaola-Galvan, P. Carpena, L. Mora-Lopez, et al., "Losses
produced by soiling in the incoming radiation to photovoltaic
modules," Progress in Photovoltaics, vol. 21, pp. 790-796, Jun
2013.
[5] J. R. Caron and B. Littmann, "Direct Monitoring of Energy Lost
Due to Soiling on First Solar Modules in California," Ieee Journal
of Photovoltaics, vol. 3, pp. 336-340, Jan 2013.
[6] T. Townsend and P. Hutchinson, "Soiling Analyses at PVUSA,"
presented at the American Solar Energy Society Annual
conference, Madison, WI, 2000.
[7] H. A. AlBusairi and H. J. Möller, "PERFORMANCE
EVALUATION OF CdTe PV MODULES UNDER NATURAL
OUTDOOR CONDITIONS IN KUWAIT," presented at the 25th
European Photovoltaic Solar Energy Conference and Exhibition /
5th World Conference on Photovoltaic Energy Conversion,
Valencia, Spain, 2010.
[8] S. A. M. Said, "EFFECTS OF DUST ACCUMULATION ON
PERFORMANCES OF THERMAL AND PHOTOVOLTAIC
FLAT-PLATE COLLECTORS," Applied Energy, vol. 37, pp.
73-84, 1990.
[9] M. Ibrahim, B. Zinßer, H. El-Sherif, E. Hamouda, G. Makrides, G.
E. Georghiou, et al., "Advanced Photovoltaic Test Park in Egypt
for Investigating the Performance of Different Module and Cell
Technologies," 2009.
[10] A. Hasan and A. A. Sayigh, THE EFFECT OF SAND DUST
ACCUMULATION ON THE LIGHT TRANSMITTANCE,
REFLECTANCE, AND ABSORBENCY OF THE PV GLAZING.
Oxford: Pergamon Press Ltd, 1992.
[11] S. Canada. (2013) Impacts of Soiling on Utility-Scale PV System
Performance. SolarPro. Available:
http://solarprofessional.com/articles/operations-maintenance/impac
ts-of-soiling-on-utility-scale-pv-system-performance
[12] S. Pathak, "200MW Solar Power Plant to Be Ready in 8 Years:
Kahramaa," in Qatar Tribune, ed. Doha: Qatar Tribune, QIM
Group, 2013.
[13] Qatar Electricity & Water Company (Kahramaa), "Annual Report
2013," Doha, Qatar2014.
[14] R. Davies. (2012, October 9, 2013). Can Qatar's food security plan
ripen? Available:
http://www.aljazeera.com/indepth/features/2012/12/201212212373
38571.html
[15] GreenGulf. (2015). Project Development. Available:
http://www.green-gulf.com/experience/project-development
[16] S. S. Khatri. (2011, October 9). QSA: Harmful air particles
continue to surpass acceptable limits in Qatar. Available:
http://dohanews.co/post/44366912970/qsa-harmful-air-particles-co
ntinue-to-surpass
[17] T. Sarver, A. Al-Qaraghuli, and L. L. Kazmerski, "A
comprehensive review of the impact of dust on the use of solar
energy: History, investigations, results, literature, and mitigation
approaches," Renewable & Sustainable Energy Reviews, vol. 22,
pp. 698-733, Jun 2013.
[18] T. Dierauf, A. Growitz, S. Kurtz, J. L. B. Cruz, E. Riley, and C.
Hansen, "Weather-Corrected Performance Ratio,"
NREL/TP-5200-57991. 2013.
[19] S. Jammalamadaka and U. Lund, "The effect of wind direction on
ozone levels: a case study," Environmental and Ecological
Statistics, vol. 13, pp. 287-298, 2006/09/01 2006.
[20] W. C. Hinds, Aerosol Technology: Properties, Behavior, and
Measurement of Airborne Particles. New York: John Wiley &
Sons, Inc., 1999.
[21] D. Goossens, Z. Y. Offer, and A. Zangvil, "WIND-TUNNEL
EXPERIMENTS AND FIELD INVESTIGATIONS OF EOLIAN
DUST DEPOSITION ON PHOTOVOLTAIC SOLAR
COLLECTORS," Solar Energy, vol. 50, pp. 75-84, Jan 1993.
[22] D. Goossens and E. Van Kerschaever, "Aeolian dust deposition on
photovoltaic solar cells: The effects of wind velocity and airborne
dust concentration on cell performance," Solar Energy, vol. 66, pp.
277-289, Jul 1999.
[23] S. Mekhilef, R. Saidur, and M. Kamalisarvestani, "Effect of dust,
humidity and air velocity on efficiency of photovoltaic cells,"
Renewable & Sustainable Energy Reviews, vol. 16, pp. 2920-2925,
Jun 2012.
[24] Y. Tang and B. Guo, "Computational fluid dynamics simulation of
aerosol transport and deposition," Frontiers of Environmental
Science & Engineering in China, vol. 5, pp. 362-377, 2011/09/01
2011.
TABLE II
MULTIVARIABLE REGRESSION RESULTS
Coefficient Value and Unit
2.3×10-3
-5.7×10-2
m3
mg-1
3.5×10-3
s m-1
-2.0×10-1
TABLE III
PREDICTED ∆CI UNDER VARIOUS AMBIENT CONDITIONS
Independent Variables Predicted ∆CI
Mean Ambient Conditions -0.0058
Single Variable Varied by
One Standard Deviation
PM10 -0.0076 to -0.0040
WS -0.0086 to -0.0031
RH -0.0086 to -0.0030