Accepted Manuscript

Impact of tropical desert maritime climate on the performance of a PV grid-

connected power plant

Daha Hassan Daher, Léon Gaillard, Mohamed Amara, Christophe Ménézo

PII: S0960-1481(18)30308-2

DOI: 10.1016/j.renene.2018.03.013

Reference: RENE 9882

To appear in: Renewable Energy

Received Date: 06 September 2017

Revised Date: 05 March 2018

Accepted Date: 06 March 2018

Please cite this article as: Daha Hassan Daher, Léon Gaillard, Mohamed Amara, Christophe
Ménézo, Impact of tropical desert maritime climate on the performance of a PV grid-connected
power plant, Renewable Energy (2018), doi: 10.1016/j.renene.2018.03.013

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

1 Impact of tropical desert maritime climate on the performance of a PV grid-connected

2 power plant
3

4 Daha Hassan Daher a, b,*, Léon Gaillard c, Mohamed Amara b, Christophe Ménézo c
5 a Centre d’Etudes et de Recherche de Djibouti, Laboratoire des Energies Nouvelles et
6 Renouvelables, Djibouti
7 b Université de Lyon, CNRS, INSA-Lyon, Université Claude Bernard Lyon 1, CETHIL
8 UMR5008, F-69621, Villeurbanne, France
9 c Université Savoie Mont Blanc, CNRS, Polytech'Annecy-Chambéry, LOCIE UMR 5271, F-
10 73376, Le Bourget-du-Lac, France
11 *daha.enea@gmail.com
12

13 ABSTRACT
14 This paper provides experimental results on the performance of a grid-connected PV power
15 plant operating under dusty, desert maritime climate conditions using data from the first
16 installation of its kind in Djibouti. The first 4 years of operation were evaluated in terms of
17 IEC 61724 measures, and the impact of climate factors was estimated using a novel
18 combination of analysis techniques. The monthly average daily array yield and final yield
19 were 5.1 kWh/kWp and 4.7 kWh/kWp, respectively. The average performance ratio for
20 respective PV arrays and the global grid-connected system were 90% and 84%, corresponding
21 to monthly average daily PV module and system efficiencies of 12.68% and 11.75%. The
22 seasonal variation in PV module efficiency was found to follow a funnel-shape with a sharp
23 minimum centred on July. The impact of ambient temperature and soiling-induced losses
24 were evaluated, revealing a reduction in the performance ratio by 0.7% for each 1°C rise in
25 daily ambient temperature. Losses due to soiling varied from 0.03% following rainfall events,
26 to 14.23% during dry dusty periods. Finally, to maintain the modules performances losses less
27 than 5%, a cleaning schedule is recommended every two weeks.
28 Keywords: grid-connected PV system, performance ratio, desert maritime climate, soiling
29 loss.
30

31

32

33

34

1
 ACCEPTED MANUSCRIPT

35 Nomenclature
36 Aa PV module area (m2)
37 CF capacity factor (%)
38 EAC AC energy output (kWh)
39 EDC DC energy output (kWh)
40 G0 solar irradiance under standard test conditions (kW/m2)
41 IPOA plane of array irradiance (W/m2)
42 LC array capture losses (kWh/kWp)
43 LS system losses (kWh/kWp)
44 P power output (kW)
45 P0 PV rated power (kWp)
46 PR performance ratio (%)
47 PRTamb performance ratio corrected for ambient temperature (%)
48 PRTmod performance ratio corrected for module temperature (%)
49 PV photovoltaic
50 STC standard test conditions
51 Tamb ambient temperature (°C)
52 Tmod module temperature (°C)
53 Ya array yield (kWh/kWp)
54 Yf final yield (kWh/kWp)
55 Yr reference yield (kWh/kWp)
56 Ws wind speed (m/s)
57 ηPV array efficiency (%)
58 ηSYS system efficiency (%)
59 ηinv inverter efficiency (%)
60

61 Subscripts
62 AC alternative current
63 DC direct current
64 p peak
65

66

67

68

2
 ACCEPTED MANUSCRIPT

69 1. Introduction
70 The performance of PV power plants arises from on-site weather conditions including
71 irradiation, ambient temperature, humidity and wind speed [1]. Several studies have been
72 reported on the performance of grid-connected solar PV power plants operating in a range of
73 climate zones, for various mounting configurations and PV technologies [1-21]. The
74 performance analysis of PV power plants under Temperate climatic conditions were reported
75 in [2-6]. A performance comparison with PV systems from other European countries
76 (Northern Ireland [2], Germany [3], Poland [4]) has shown that the annual final yield and
77 performance ratio are higher for the system in Dublin, Ireland [5]; as a result of lower average
78 ambient temperature and high wind speed for that site. A preliminary assessment of a rooftop
79 grid-connected under Norwegian climate was reported by Adaramola et al. [6], which showed
80 a closer agreement in performance than those observed in the same climate zone [5]. Other
81 studies investigate the performance under Mediterranean climate [7-8], and overall PV
82 systems are found to perform better under Mediterranean than Temperate climate. This
83 observation indicates that solar insolation is a more important factor than operating
84 temperature in these cases.
85 Several studies on performance evaluation have been undertaken under tropical and
86 hot climatic conditions [9-21]. An analytical performance monitoring of a 142.5 kWp grid-
87 connected rooftop building integrated photovoltaics (BIPV) system in Singapore was carried
88 out by Wittkopf et al. [9]. The performance ratio was 81%, array yield and final yield was
89 respectively 3.86 kWh/kWp and 3.12 kWh/kWp. The system and array efficiency were
90 respectively estimated at 11.2% and 11.8%. The performance comparison of different PV
91 technologies shows that HIT (Heterojunction with Intrinsic Thin layer) and a-Si perform
92 better than p-Si under composite climate conditions (hot and humid in summer and cooler in
93 winter) [10] and p-Si perform better under hot climate than a-Si [11]. The analysis of climatic
94 parameters on module efficiencies, net energy production per unit area and performance ratio
95 with PV modules made from crystalline silicon-based (p-Si) and thin film technologies-based
96 have been undertaken by Ahmed et al. [12]. The results show that the cadmium telluride
97 (CdTe) technology performs better under Kuwait climate conditions. Padmavathi et al. [13]
98 evaluated the performance of a 3 MWp grid-connected solar PV power plant in Karnataka,
99 India. The performance ratio was found to vary between 61% and 78% from June to
100 February 2011. The annual reference and final yield were respectively 5.36 kWh/kWp and
101 3.73 kWh/kWp. The capacity factor for this PV power plant was 12.38% for 2010 and
102 15.69% for the year 2011. In [14-15] the performances of PV systems have been evaluated

3
 ACCEPTED MANUSCRIPT

103 under the hot and desert climate of Sohar, Oman and Abu Dhabi, United Arab Emirates.
104 These works reported final yields of respectively 5.14 kWh/kWp for Sohar, and a range from
105 4.5 kWh/kWp to 5.571 kWh/kWp for Abu Dhabi. Also, in the same region of Arabic
106 peninsula, Al-Otaibi et al [16] presented a 12-month performance evaluation of two (85.05
107 kWp and 21.6 kWp) rooftop grid-connected PV systems in Kuwait, wherein a range in
108 performance ratio 74% - 85% was reported and an associated annually averaged daily final
109 yield of 4.5 kWh/kWp. To maintain this level of performance, the PV module was equipped
110 with an automated cleaning system to overcome soiling due to the dusty conditions of the site.
111 Numerous performance analyses have been undertaken in different parts of the globe in
112 Nouakchott, Mauritania [17], Maseru, Lesotho [18], Sivagangai [19] and Rajkot [20], India
113 and Fortaleza, Brazil [21] and are categorized by climatic zone in the table 1. The
114 performance of each of the above-mentioned studies are summarised in table 1. Through these
115 various studies it is possible to infer the expected performance per climate zone, although it is
116 also evident that widely varying levels can be observed for different systems in similar areas.
117 Overall, plants operating in hot and dusty climate zones attain the highest performances due to
118 favourable solar radiation conditions.
119 Furthermore, dust accumulation on PV panels is highly variable and correlates to a
120 multitude of factors: characteristics of the installation site, panel inclination, site meteorology
121 (ambient temperature, humidity and wind speed) and the properties of the dust [22]. Soiling
122 decreases the solar photovoltaic system performance. In order to quantify the effect of soiling
123 over time, it is necessary to monitor the performance of the PV modules during this time
124 period. Previous studies have been conducted to evaluate the performance loss due to PV
125 module soiling [23-26]. In a study conducted by Kleissl et al. [23] on soiling losses for 186
126 solar PV systems in California, it was revealed that the average losses were 0.051% per day in
127 relative solar conversion efficiency. Kimber et al. [24] studied the effect of the soiling on
128 large grid-connected PV systems in California and the Southwest region of the United States.
129 It was found that PV system efficiency declines by an average of 0.2% per day without
130 rainfall in dry climates. This daily loss finding equates to an annual energy loss between 1.5
131 and 6.2% depending on system location. In a study conducted in the Atacama Desert (Chile)
132 on different technologies, it was reported by Fuentealba et al. [25] that the performance ratio
133 (PR) decreased by 4.8% per month for thin film technology and at a rate of 6.2% per month
134 for multi-crystalline technology, due to the accumulation of dust and extreme temperatures. In
135 a study conducted by Pavan et al. [26] in the southern Italian countryside where two 1 MWp
136 solar plants were monitored for 8 weeks. Performance losses of 6.9% and 1.1% were observed

4
 ACCEPTED MANUSCRIPT

137 respectively for the plant built on sandy soil and the plant built on a more compact soil. The
138 soiling loss estimation can help to further improve cleaning schedule.
139 Few data are available regarding the actual performance of PV systems in the coastal
140 regions of the Horn of Africa or similar climate zones. This lack of data is a hindrance to the
141 development of solar energy in a region which otherwise has a substantial solar potential of
142 about (for example, 2200 kWh/m2 per year total horizontal radiation in Djibouti), and a
143 rapidly growing electricity consumption [27-29]. To overcome this shortcoming, in 2012 a
144 pilot 302.4 kWp grid-connected PV plant was installed at the "Centre d’Etudes et de
145 Recherche de Djibouti" (CERD), in Djibouti. The installation is unique in the region as it
146 represents a full scale system dedicated to research and development, under the supervision of
147 the Laboratory of New and Renewable Energies of the CERD. Moreover, the plant is the first
148 grid-connected PV system in Djibouti. In this paper, a performance evaluation of the facility
149 is presented with the aim of providing some insight into the operation of a grid-connected PV
150 power plant in harsh environment characterised by hot, arid and saline dusty conditions with
151 very low rainfall. The installation has been monitored continuously from 2012 to 2015, with
152 minute-wise and hourly measurements of electrical performance and environmental
153 conditions.

154 Table 1. PV system performance evaluation for different climate zone.

Location PV type- Final PV System Performanc Reference
(mounting) yield module efficiency e Ratio -
(kWh/k efficiency (%) (Capacity
Wp) (%) factor) (%)
Temperate climate
Ballymena, mc-Si 1.69 7.54 6.39 61 [2]
Northern (rooftop)
Ireland
Northern --- 1.86 ----- ---- 66.5 [3]
Germany (rooftop)
Warsaw, a-Si 2.27 4.5-5.5 4.0-5.0 60-80 [4]
Poland (rooftop)
Dublin, p-Si 2.4 14.9 12.6 81.5 [5]
Ireland (rooftop) (10.1)
Ås, Norway p-Si 2.55 12.7 11.6 83.03 [6]
(rooftop) (10.58)
Mediterranean climate

5
 ACCEPTED MANUSCRIPT

Crete Island, p-Si 1.96- ----- ----- 67.36 [7]

Greece (ground) 5.07 (15.26)

Nicosia, ----- 4.33 ----- ----- ----- [8]

Cyprus
Tropical and hot climate
Singapore p-Si 3.12 11.8 11.2 81 [9]
(rooftop,
BIPV)
Gandhinagar p-Si 2.79- 11.07 10.52 75.3 [11]
, India (ground) 5.14
Gandhinagar a-Si 2.62- 6.56 6.06 70.8 [11]
, India (ground) 4.84
Karnataka, mc-Si 3.73 10.1-13.25 9.8-12.3 70 (12.38- [13]
India (ground) 15.69)
Sohar, p-Si 5.14 ----- ----- 84.6 [14]
Oman (rooftop) (21)
Abu-Dubai p-Si 4.507- 10.3-11.1 ----- ----- [15]
5.571
Kuwait CIGS 4.5 ----- ----- 70-85 [16]
(rooftop)
Nouakchott a-Si\μa-Si 4.27 ----- ----- 67.96 [17]
,Mauritania (ground)
Maseru, p-Si 3.43 10.93 9.58 67 [18]
Lesotho (ground) (17.2)
Fortaleza, p-Si 4.6 13.3 12.6 82.9 [19]
Brazil (ground) (19.2)
Sivagangai, a-Si 4.81 6.08 5.08 89.15 [20]
India (ground)
Rajkot, p-Si 4.49 ----- ----- 74 [21]
India (rooftop)

155

156 2. The CERD 300kW demonstration PV facility in Djibouti

157 Djibouti is located in the Horn of Africa at a latitude 11°30' N and longitude 43°00' E.
158 The region is classed as a hot desert maritime climate [30], with a cooler season from October
159 to April, with small episodes of rain and a temperature range of 21°C to 33°C, a hot season
160 from May to September, with a temperature range of 35°C to 45°C, and a sustained dust-
161 laden dry wind phenomenon called Khamsin, which comprises an increase in dust at ground
162 level and suspended dust in the atmosphere. Fig.1 shows monthly averages of daily global
163 solar radiation and daily ambient temperature (Tamb), at the CERD demonstration test facility,
164 both taken over daylight hours during the monitoring period of February 2012 to December
165 2015. Note that radiation measurements were made with sensors inclined to the plane of the
166 PV array (IPOA). A clear seasonal trend is visible, with an increase of Tamb and decrease of

6
 ACCEPTED MANUSCRIPT

167 IPOA during summer months and a cooler period with generaly higher IPOA. The low solar
168 radiation intensities observed from June to August are caused by a heightened atmospheric
169 opacity during the Khamsin seasonal phenomenon.

170

171 Fig. 1. Monthly average daily plane of array irradiance and daylight ambient temperature.
172

173 The solar photovoltaic power plant Fig.2 was installed by the Japan International
174 Cooperation Agency (JICA) at the CERD as part of a bilateral cooperation between the
175 Republic of Djibouti and the Japanese government to promote economic development through
176 solar energy technologies. The installation is a 302.4 kWp grid-connected facility, comprising
177 a fixed ground-based array of 1,440 polycrystalline silicon PV modules (Model: KD210GH-
178 2P), oriented due south and inclined to an optimal 12° for the site’s latitude, and covering an
179 area of 2138.4 m². The photovoltaic panels, each of rated power 210 Wp, are arranged into
180 thirty 10 kWp arrays of forty-eight modules, interconnected in parallel at junction boxes. The
181 system is distributed across three 100 kWp power conditioners (inverters) that convert direct
182 current (DC) output of the solar panels into alternating current (AC) synchronized with the
183 grid.

184 The installation was monitored continuously for the first four years of operation
185 (1st February 2012 to 31st December 2015), with data recorded on a minute-wise basis.
186 Electrical measurements (including EDC and EAC) were provided by the internal data
187 acquisition system of each power conditioner, to which were also added one in-plane

7
 ACCEPTED MANUSCRIPT

188 pyranometer (IPOA) and an ambient temperature sensor (Tamb). Prior to their analysis, these
189 data were pre-processed to remove outliers caused by anomalous behaviour, including
190 instrumentation and data acquisition failures, grid outages, and voluntary stoppages of the PV
191 plant for maintenance or unscheduled interruptions.

192

193 Fig. 2. PV system installation.

194

195 3. Methodology
196 3.1. Performance evaluation
197 The IEC 61724 standard (International Electrotechnical Commission) specifies the set
198 of measures proposed by the International Energy Agency (IEA) for the assessment of PV
199 installations [31]. The daily performance measures considered in this study are summarized in
200 Table 2 and in particular these include the reference yield (Yr), array yield (Ya), final yield
201 (Yf), the performance ratio (PR), system losses (LS) and array capture losses (LC). These
202 quantities describe the system performance as a whole in relation to energy production, solar
203 resources and the overall effect of the PV system losses. As mentioned by Lee et al. [32] a PR
204 greater than 80% corresponds to a system with performance approaching the ideal
205 performance under STC and a system with a PR lower than 70% should be suspected of
206 failure or malfunction of components from the system (panels, inverters, etc.) or
207 environmental factors (shade nearby, excessive dust panels, etc.).

8
 ACCEPTED MANUSCRIPT

208 Table 2. Performance measures.

Performance measure Definition Eq

Reference yield 𝑌𝑟 = 𝐼𝑃𝑂𝐴/ 𝐺0 (kWh/kWp) (1)

Array yield 𝑌𝑎 = 𝐸𝐷𝐶/ 𝑃0 (kWh/kWp) (2)

Final yield 𝑌𝑓 = 𝐸𝐴𝐶/ 𝑃0 (kWh/kWp) (3)

Performance ratio 𝑃𝑅 = 𝑌𝑓/𝑌𝑟 (%) (4)

System losses 𝐿𝑆 = 𝑌𝑎 ‒ 𝑌𝑓 (kWh/kWp) (5)

Array capture losses 𝐿𝐶 = 𝑌𝑟 ‒ 𝑌𝑎 (%) (6)

Array efficiency 𝜂𝑃𝑉 = 100 × 𝐸𝐷𝐶/𝐼𝑃𝑂𝐴 × 𝐴𝑎 (%) (7)

System efficiency 𝜂𝑆𝑌𝑆 = 100 × 𝐸𝐴𝐶/𝐼𝑃𝑂𝐴 × 𝐴𝑎 (%) (8)

Inverter efficiency 𝜂𝑖𝑛𝑣 = 100 × 𝐸𝐴𝐶/𝐸𝐷𝐶 (%) (9)

Capacity factor 𝐶𝐹 = 𝐸𝐴𝐶/(𝑃0 × 24 × 365) = 𝑌𝑓/8760 (10)

209

210 3.2. Climate-corrected performance ratings

211 The performance measures described in the previous section are all sensitive to the
212 environmental conditions of the site and hence vary as a function of weather and the seasons.
213 Linear models can be used to estimate the installed power of PV installations by transposing
214 observed power generation to reference conditions, thereby providing a measure of real
215 performance corrected for variations in the local environement. These models typically
216 include correlation coefficients for each environmental factor that must be estimated by
217 regression. The Photovoltaic for Utility Scale Application (PVUSA) rating model is routinely
218 used to assess installations, and was initially developed for the climate zones of North
219 America [33]. The rating comprises a simple model of power as function of plane-of-array
220 irradiance, ambient temperature and wind speed Eq. (11).
P = IPOA × (A + B × IPOA + C × Tamb + D × WS) (11)

221 As wind data were not available for the whole monitoring period, the final term in
222 Eq.11 was dropped for the present analysis. According to Myers, the wind term typically

9
 ACCEPTED MANUSCRIPT

223 contributes 0.4% of the PVUSA rating [34]. Therefore an uncertitude of this magnitude was
224 assumed for the analysis of CERD data.
P = IPOA × (A + B × IPOA + C × Tamb) (12)

225 Eq. (11) is derived from Eq. (13), which describes the relationship between power
226 output, plane-of-array irradiance and module temperature. If measurements of the latter are
227 available, Eq. 13 can also be used to provide an estimate of installed rated power.
P = IPOA × (K1 + K2 × Tmod) (13)

228 Using a training dataset corresponding to a period with clean PV modules, multiple linear
229 regression is performed in order to determine constants A, B, C, K1 and K2 (eq. 12 and 13)
230 under nominal conditions. Once tuned, the models can then be used to estimate the expected
231 performance at other times of the year, in the absence of dust or other change to the thermal
232 and electrical properties of the system. The two models presented above were hence used to
233 predict the performance ratio corrected for ambient temperature PRTamb and corrected for
234 module temperature PRTmod and thus infer the effect of soiling.
235

236 4. Results and discussions

237 4.1. Performance analysis
238 The monthly and cumulative energy generated from the PV system is shown in Fig.3
239 for the 47 months spanning the monitoring period. Monthly energy generation was found to
240 range from 31.3 MWh in July 2013 to 46.7 MWh in October 2015, with a mean output of
241 41 MWh. The accumulated energy generation for the first 12 months was 494 MWh, some
242 7.4% above the 460 MWh predicted during the prefeasibility study.

10
 ACCEPTED MANUSCRIPT

243

244 Fig. 3. Monthly and cumulated energy generation of the CERD power plant, 2012 – 2015.
245

246 Fig. 4 illustrates monthly average daily final yield (Yf), system losses (LS) and PV
247 array capture losses (LC). The monthly average daily final yield varied from a minimum of
248 3.68 kWh/kWp in July 2014 to a maximum of 5.43 kWh/kWp in February 2013 with an
249 average of 4.69 kWh/kWp, slightly out of phase with the reference yield (not shown), which
250 varied from 4.59 kWh/kWp in July 2013 to 6.36 kWh/kWp in April 2015. A seasonal drop in
251 Yf from June to August is visible each year, and is caused by the decrease in received
252 radiation during the period.
253 The system losses LS were relatively stable throughout the period with an average of
254 0.37 kWh/kWp, and a range from a 0.31 kWh/kWp in July 2013 to 0.41 kWh/kWp in
255 February 2013.
256 In contrast, array capture losses LC were more varied and more pronounced, ranging
257 from 0.12 kWh/kWp in December 2012 to 1.22 kWh/kWp in July 2015 with an average of
258 0.54 kWh/kWp. Indeed for 20 of the 47 months of monitoring, these losses exceeded
259 0.6 kWh/kWp.
260

11
 ACCEPTED MANUSCRIPT

261

262 Fig. 4. Monthly average daily: final yield (Yf), system losses (LS) and PV array capture losses
263 (LC)
264

265 Fig.5 presents the monthly performance ratio of the PV array (PRDC, red circle) and
266 system (PRAC, blue cross) on axes of monthly average daily array and final yield versus
267 reference yield. The data points are contained within the interval 70% – 100% as indicated by
268 the overlaid solid lines, and are centred on a mean PR of 90% and 84% for the PV arrays and
269 the system respectively. The PR is high compared to the results of the work cited in [9-19].
270 The maximum value of PR is reached in December 2012 with respective values of 98% for
271 PRDC and 91% for PRAC. The poorest PR was observed in July 2015, with 77% for PRDC and
272 71% for PRAC, a period that was characterised by elevated temperatures coinciding with high
273 dust levels during Khamsin.
274

12
 ACCEPTED MANUSCRIPT

275

276 Fig. 5. PV array and system performance ratio.

277

278 The monthly average of daily PV module efficiency is reported in Fig. 6. Apart from
279 the first year when efficiency remained elevated during the hot season, the annual variation in
280 efficency followed a funnel-shaped trend varying from 10.87% in July 2015 to 13.82% in
281 December 2012, with a mean value of 12.68% (by comparison, the rated PV module
282 efficiency was 14.14% at STC). A spurious variation in efficiency of ~2% was observed for
283 periods with similar incident radiation and ambient temperature, such as July of each year.
284 The additional variation is due to other environmental conditions, namely dust effects for the
285 case of Djibouti. Overall, efficiencies decreased towards the latter part of the monitoring
286 period: the mean annual efficiencies varied from 12.96% to 12.35%, from 12.03% to 11.43%
287 and from 92.81% to 92.6% between 2012 and 2015, respectively for the efficiencies of PV
288 modules, PV system and the inverters.

13
 ACCEPTED MANUSCRIPT

289

290 Fig. 6. Monthly average daily PV efficiency.

291

292 Fig.7 presents the monthly average daily capacity factor for the four years of operation
293 of the 302.4 kWp solar PV plant. CF varies between 12.76% in July 2014 and 22.11% for the
294 month of October 2014 with a four-year CF average of 16.35%. A decline in CF occurs
295 during summer months from June to August; a similar trend is observed each year,
296 characterised by a seasonal variation of ~3%. This result is generaly consistent with the local
297 climate (Fig.1), since CF is lowest during the period of maximum ambient temperature and
298 weakest in-plane solar radiation, although other factors such as dust also appear to contribute
299 significantly to the losses.

14
 ACCEPTED MANUSCRIPT

300

301 Fig. 7. Capacity factor (2012 to 2015).

302

303 In summary, the results summarised in Fig.4 to Fig.7 indicate that the solar PV power
304 plant performed well during most months of the monitoring period, with the exception of
305 June to August. For this harsh summer period the decline in performance is attributed to a
306 lower intensity of solar radiation (due to atmospheric dust brought by the Khamsin), higher
307 ambient temperatures, and dust accumulation on PV modules. Refering to Table 1, the
308 performance of the CERD facility is relatively high compared to most of PV plants operating
309 in hot and tropical climates, and was similar to certain sites in Oman and Abu-Dubai [14-15].
310 These facilities also exhibit sharp decreases in output during the summer period. In the next
311 section, the impact of the climate factors on the performance of the PV power plant will be
312 quantified.
313

314 4.2. Effect of weather conditions on performance and temperature correction

315 Fig.8 represents the hourly AC power output over the monitoring period as a function
316 of the plane-of-array solar irradiation. In addition to a dominant positive and narrow
317 correlation between generated power and incident solar radiation, the data are distributed in
318 two groups, as highlighted by the dashed ellipse. The smaller of the two populations
319 corresponds to a reduced efficiency coinciding with a period of high dust accumulation on the
320 PV array from June to August when dust deposition was observed to be higher than at other
321 times of the year. During this period, the dust layer on each module filtered a significant

15
 ACCEPTED MANUSCRIPT

322 fraction of incident radiation, thereby resulting in a fall in energy generation. Furthermore,
323 considering the apparent shape of the two populations in Fig.8, the gradient of the correlation
324 between power and radiation is shallower during dusty periods. Therefore, the presence of
325 dust may also affect the thermal properties of the PV modules.

ts
e ven
ling
Soi

326

327 Fig. 8. Hourly AC energy output against plane of array irradiance.

328

329 The variation in PR provides a further indication of the relationship between system
330 performance and environmental conditions including, wind, ambient temperature and dust
331 accumulation [1]. Fig.9 presents the dependence of daily AC performance ratio as a function
332 of mean daytime ambient temperature, with a colour scale to indicate average daytime
333 radiation. Peak PR can be seen to occur during the coolest part of the year, and the system
334 regularly exceeds PR=90% when the ambient temperature is below 33ºC annual mean
335 ambient temperature of the site. In contrast, mean PR falls to 70.4% during the hottest days. A
336 decrease of 0.71% of daily PR is observed for each 1°C in the daytime ambient temperature.
337 As was illustrated by Fig.1, periods of high temperature tend to occur when incident radiation
338 is at a minimum.

16
 ACCEPTED MANUSCRIPT

339

340 Fig. 9. Daily AC performance ratio against daylight ambient temperature.

341

342 For days with temperatures in excess of 36°C the spread in PR is notably more dispersed. In
343 order to explore this feature further, a modified PR can be calculated using on-site
344 meteorological data. Several approaches exist and are proposed by NREL [35], IEA-PVPS
345 [36]. Here, the modified PVUSA model described in section 3.2 was applied to estimate the
346 performance ratio corrected for the ambient temperature PRTamb and performance ratio
347 corrected for module temperature PRTmod.
348 Data were filtered to extract a sample where the PV array had minimal soiling,
349 following a significant episode of rain. Three days were selected with incident radiation
350 higher than 5 kWh/m2/d, greater than 3 kWh/m2/day threshold recommended by Dierauf et al.
351 [35], with temperatures close to the average observed during the cool season. The data sample
352 was used to estimate the coefficients of the models described by equations (12) and (13) by
353 multiple linear regression. The results are summarised in Table 3. The negative coefficients C
354 and K2 indicate that PR decreases with increasing Tamb and Tmod.

355 Table 3. Regression coefficients.

A B C

Ambient temperature-corrected PR (PRTamb) 95.94 0.55 -0.20

K1 K2 ---
Module temperature-corrected PR (PRTmod) 117.36 -0.66 ---

17
 ACCEPTED MANUSCRIPT

356 The estimated coefficients were used to calculate PR modified for module temperature
357 and ambient temperature during 2014. Figure 10.(a) presents the measured PR using
358 equation (4) for 2014 (black curve), and modified PR, corrected for ambient temperature (red
359 curve) and module temperature (green curve). Daily PR varies by 18.6% throughout the year,
360 and the various features in the PR curve arise due to several phenomena. A stepwise increase
361 in PR during consecutive days corresponds to cleaning by the team of technicians. In contrast,
362 a sudden increase in the PR results from a rain event. A sudden drop in PR would ordinarily
363 be attributed to inverter failure, however as such outages were filtered from the dataset, these
364 are not visible in the figure.
365 The variations in modified PR for ambient and module temperature are less
366 pronounced than the uncorrected PR. Indeed, the corrections were found to remove much of
367 the seasonal variation over the year. The average daily absolute differences between measured
368 and corrected performance ratios were 5.5% and 4.5% for ambient temperature and module
369 temperature corrections respectively. Moreover, the standard deviation of the absolute daily
370 differences is 0.6% higher with ambient temperature than with module temperature (Table 4).
371 PRTmod appears therefore to provide a more complete correction of ambient conditions of
372 module performance than PRTamb. This is to be expected, because the use of module
373 temperature implicitly includes all factors affecting the thermal coupling of the PV modules to
374 the environment, such as wind speed.

375 Table 4. Average daily absolute differences and standard deviation.

Average daily absolute Standard deviation of

difference (%) daily difference (%)

Measured vs ambient temperature 5.5 3.6

corrected PR

Measured vs module temperature 4.5 3.0

corrected PR

18
 ACCEPTED MANUSCRIPT

376

377 Fig. 10. (a) daily PR (black), ambient temperature corrected PR (red), module temperature
378 corrected PR (green) (b) soiling losses (purple), daily rainfall (blue bar) and manual cleaning
379 (vertical dashed lines) two vertical lines represent one cleaning.
380

381 At CERD, dust accumulation on PV panels was observed to be at its fastest during the
382 hot and dry period of the year. It was found that PR of PV modules decline by an average of
383 0.31% to 0.36% per day in dry periods without rainfall. As discussed by Adinoyi et al. [37],
384 the frequency of rainfall is a signiciant factor in the overall performance of a PV plant
385 operating in dusty conditions. Guo et al. [38] suggested that 5 mm of rainfall is sufficient to
386 clean PV modules, which is consistent with observations at CERD. In addition, successive
387 periods of light rain of less than 5 mm per day were also found to thoroughly clean the PV
388 panels. An initial estimate of the impact of soiling upon PR was achieved by comparing the
389 performance ratio of the plant before and after significant rainfall events. Comparing the PR
390 for 5 days after rainfall with the 5 preceding days, an increase of 6.8% was observed due to
391 the cleaning affect of the rain during the hot and dry season. Smaller rises in PR ranging from
392 0.87% to 4.33% were observed following rainfall events during the cooler season, due to the
393 less dusty conditions during this period.
394 Fig 10.(b) shows the daily absolute difference between corrected and uncorrected PR and
395 the daily rainfall. Assuming other physical properties of the PV array remain constant, the
396 curve in Fig 10.(b) provides a quantitative estimate of losses due to soiling, because the

19
 ACCEPTED MANUSCRIPT

397 corrected PR was tuned for a clean PV array. Measured in terms of reduced PR, soiling loss
398 varies between 0.03% shortly after a rain episode, to 14.23% for dry periods with no rainfall.
399 Losses tend to increase incrementally at a rate that appears to vary between seasons,
400 presumably due to varying rates of dust accumulation. After two weeks without PV cleaning,
401 the PR losses due to soiling may increase by 4.3% to 5.0%. Abrupt reductions in losses to
402 near zero are the result of rainfall. More gradual decreases in soiling losses by 0.37% to
403 0.84% per day result from cleaning by the maintenance team, because the installation is too
404 large to be cleaned in a single day by the current team. In order to maintain the losses below
405 5%, it would be necessary to clean the array every two weeks, with each intervention
406 spanning sufficient days to clean the whole array. Given the rate of cleaning currently
407 possible for the CERD facility, such a programme would require either the recrutement of
408 more technical staff, or the introduction of enhanced cleaning equipment including some
409 degree of automation.
410

411 5. Conclusion
412 The performance of a 302.4 kWp grid-connected PV system in Djibouti was evaluated
413 and the impact of climatic factors has been assessed. The analysis of the first 4 years of
414 operation provides new insights into the viability of PV technologies for coastal, desert
415 maritime climate zones. The monthly average daily reference yield, array yield and final yield
416 were 5.6 kWh/kWp, 5.1 kWh/kWp, and 4.7 kWh/kWp, respectively. The average
417 performance ratio for respective PV arrays and system are 90% and 84%. Comparing the
418 CERD plant to the reported results of other installations operating in various harsh operating
419 conditions, a desert maritime climate may prove less detrimental.
420 A new approach has been undertaken to correct the performance according to the ambient
421 temperature and module temperature. The method works as long as the data for clean modules
422 is available and preferably for a short period. Furthermore, the correction were found to
423 remove the performance ratio seasonal variation over the year. The effect of the Djibouti
424 climate on performance were characterised as follows:
425  A reduction of the performance ratio by 0.7% for each 1°C increase in ambient
426 temperature.
427  In terms of reduced PR, soiling losses vary between 0.03% after a rainfall that has
428 cleaned the PV modules surfaces to 14.23% for dry periods without rainfall.

20
 ACCEPTED MANUSCRIPT

429  In the absence of rain, leaving the installation two weeks without cleaning can result in
430 a fall in PR by 4.3% to 5.0%. A recommendation was made in this direction to clean
431 the panels every two weeks.
432 The outcome of this research shows that this 302.4 kWp PV power plant works efficiently
433 under harsh environmental conditions of Djibouti and maintains its performance relatively
434 well over time. This could be an incentive for investors for the future development of grid-
435 connected solar PV plant. The data provided in this work will be a baseline for comparison to
436 further PV power plants installed in similar climate conditions, and in particular the coastal
437 region of the Horn of Africa.
438 The CERD PV facility is currently being analysed to provide a first estimate of
439 degradation rate for the climate of Djibouti. In parallel, the results of this experimental study
440 are being used to improve the reliability of PV sizing studies and to enhance numerical
441 models with which to predict the thermal and electrical behaviour of the PV systems in Sun
442 Belt regions.
443

444 References
445 [1] Marion, B., Adelstein, J., Boyle, K., Hayden, H., Hammond, B., Fletcher, T., ... & Rich,
446 G. (2005, January). Performance parameters for grid-connected PV systems. In Photovoltaic
447 Specialists Conference, 2005. Conference Record of the Thirty-first IEEE (pp. 1601-1606).
448 IEEE.
449 [2] Mondol, J. D., Yohanis, Y., Smyth, M., & Norton, B. (2006). Long term performance
450 analysis of a grid connected photovoltaic system in Northern Ireland. Energy conversion and
451 management, 47(18), 2925-2947.
452 [3] Decker, B., & Jahn, U. (1997). Performance of 170 grid connected PV plants in northern
453 Germany—analysis of yields and optimization potentials. Solar Energy, 59(4-6), 127-133.
454 [4] Pietruszko, S. M., & Gradzki, M. (2003). Performance of a grid connected small PV
455 system in Poland. Applied energy, 74(1), 177-184.
456 [5] Ayompe, L. M., Duffy, A., McCormack, S. J., & Conlon, M. (2011). Measured
457 performance of a 1.72 kW rooftop grid connected photovoltaic system in Ireland. Energy
458 conversion and management, 52(2), 816-825.
459 [6] Adaramola, M. S., & Vågnes, E. E. (2015). Preliminary assessment of a small-scale
460 rooftop PV-grid tied in Norwegian climatic conditions. Energy Conversion and
461 Management, 90, 458-465.

21
 ACCEPTED MANUSCRIPT

462 [7] Kymakis, E., Kalykakis, S., & Papazoglou, T. M. (2009). Performance analysis of a grid
463 connected photovoltaic park on the island of Crete. Energy Conversion and
464 Management, 50(3), 433-438.
465 [8] Makrides, G., Zinsser, B., Norton, M., Georghiou, G. E., Schubert, M., & Werner, J. H.
466 (2010). Potential of photovoltaic systems in countries with high solar irradiation. Renewable
467 and Sustainable energy reviews, 14(2), 754-762.
468 [9] Wittkopf, S., Valliappan, S., Liu, L., Ang, K. S., & Cheng, S. C. J. (2012). Analytical
469 performance monitoring of a 142.5 kWp grid-connected rooftop BIPV system in
470 Singapore. Renewable Energy, 47, 9-20.
471 [10] Sharma, V., Kumar, A., Sastry, O. S., & Chandel, S. S. (2013). Performance assessment
472 of different solar photovoltaic technologies under similar outdoor conditions. Energy, 58,
473 511-518.
474 [11] Tripathi, B., Yadav, P., Rathod, S., & Kumar, M. (2014). Performance analysis and
475 comparison of two silicon material based photovoltaic technologies under actual climatic
476 conditions in Western India. Energy Conversion and Management, 80, 97-102.
477 [12] Ali, A. H. H., Zeid, H. A. S., & AlFadhli, H. M. (2017). Energy performance,
478 environmental impact, and cost assessments of a photovoltaic plant under Kuwait climate
479 condition. Sustainable Energy Technologies and Assessments, 22, 25-33.
480 [13] Padmavathi, K., & Daniel, S. A. (2013). Performance analysis of a 3 MWp grid
481 connected solar photovoltaic power plant in India. Energy for Sustainable
482 Development, 17(6), 615-625.
483 [14] Kazem, H. A., Khatib, T., Sopian, K., & Elmenreich, W. (2014). Performance and
484 feasibility assessment of a 1.4 kW roof top grid-connected photovoltaic power system under
485 desertic weather conditions. Energy and Buildings, 82, 123-129.
486 [15] Al-Sabounchi, A. M., Yalyali, S. A., & Al-Thani, H. A. (2013). Design and performance
487 evaluation of a photovoltaic grid-connected system in hot weather conditions. Renewable
488 energy, 53, 71-78.
489 [16] Al-Otaibi, A., Al-Qattan, A., Fairouz, F., & Al-Mulla, A. (2015). Performance evaluation
490 of photovoltaic systems on Kuwaiti schools’ rooftop. Energy Conversion and
491 Management, 95, 110-119.
492 [17] Sidi, C. E. B. E., Ndiaye, M. L., El Bah, M., Mbodji, A., Ndiaye, A., & Ndiaye, P. A.
493 (2016). Performance analysis of the first large-scale (15MWp) grid-connected photovoltaic
494 plant in Mauritania. Energy Conversion and Management, 119, 411-421.

22
 ACCEPTED MANUSCRIPT

495 [18] Mpholo, M., Nchaba, T., & Monese, M. (2015). Yield and performance analysis of the
496 first grid-connected solar farm at Moshoeshoe I International Airport, Lesotho. Renewable
497 Energy, 81, 845-852.
498 [19] Sundaram, S., & Babu, J. S. C. (2015). Performance evaluation and validation of 5MW p
499 grid connected solar photovoltaic plant in South India. Energy Conversion and
500 Management, 100, 429-439.
501 [20] Dobaria, B., Pandya, M., & Aware, M. (2016). Analytical assessment of 5.05 kWp grid
502 tied photovoltaic plant performance on the system level in a composite climate of western
503 India. Energy, 111, 47-51.
504 [21] de Lima, L. C., de Araújo Ferreira, L., & de Lima Morais, F. H. B. (2017). Performance
505 analysis of a grid connected photovoltaic system in northeastern Brazil. Energy for
506 Sustainable Development, 37, 79-85.
507 [22] Catelani, M., Ciani, L., Cristaldi, L., Faifer, M., Lazzaroni, M., & Rossi, M. (2012,
508 September). Characterization of photovoltaic panels: The effects of dust. In Energy
509 Conference and Exhibition (ENERGYCON), 2012 IEEE International (pp. 45-50). IEEE.
510 [23] Mejia, F. A., & Kleissl, J. (2013). Soiling losses for solar photovoltaic systems in
511 California. Solar Energy, 95, 357-363.
512 [24] Kimber, A., Mitchell, L., Nogradi, S., & Wenger, H. (2006, May). The effect of soiling
513 on large grid-connected photovoltaic systems in California and the southwest region of the
514 United States. In Photovoltaic Energy Conversion, Conference Record of the 2006 IEEE 4th
515 World Conference on (Vol. 2, pp. 2391-2395). IEEE.
516 [25] Fuentealba, E., Ferrada, P., Araya, F., Marzo, A., Parrado, C., & Portillo, C. (2015).
517 Photovoltaic performance and LCoE comparison at the coastal zone of the Atacama Desert,
518 Chile. Energy Conversion and Management, 95, 181-186.
519 [26] Pavan, A. M., Mellit, A., & De Pieri, D. (2011). The effect of soiling on energy
520 production for large-scale photovoltaic plants. Solar energy, 85(5), 1128-1136.
521 [27] Renewables Readiness Assessment: Djibouti, IRENA, May 2015
522 [28] Least Cost Electricity Master Plan, Djibouti, Parsons Brinckerhoff, November 2015
523 [29] National Energy Plan, Institut Supérieur d’Etudes et de Recherche Scientifiques et
524 Techniques de Djibouti, United Nations Development Programme (DJI/86/012 ; DJI/86/U71),
525 P138, July 1987.
526 [30] Dunham, D. Building for the maritime desert: Climate, construction, and energy in
527 Djibouti. Arlington, Virginie: VITA, 1983.

23
 ACCEPTED MANUSCRIPT

528 [31] Standard, B. (1998). Photovoltaic system performance monitoring—Guidelines for

529 measurement, data exchange and analysis. BS EN, 61724.
530 [32] Lee, J. B., Park, J. W., Yoon, J. H., Baek, N. C., & Shin, U. C. (2014). An empirical
531 study of performance characteristics of BIPV (Building Integrated Photovoltaic) system for
532 the realization of zero energy building. Energy, 66, 25-34.
533 [33] Dows, R. N., & Gough, E. J. (1995). PVUSA procurement, acceptance, and rating
534 practices for photovoltaic power plants(No. DOE/AL/82993--21). Pacific Gas and Electric
535 Co., San Ramon, CA (United States). Dept. of Research and Development; Bechtel Corp.,
536 San Francisco, CA (United States).
537 [34] Myers, D. R. (2009). Evaluation of the Performance of the PVUSA Rating Methodology
538 Applied to Dual Junction PV Technology: Preprint (Revised) (No. NREL/CP-550-45376).
539 National Renewable Energy Laboratory (NREL), Golden, CO..
540 [35] Dierauf, T., Growitz, A., Kurtz, S., Cruz, J. L. B., Riley, E., & Hansen, C.
541 (2013). Weather-corrected performance ratio (No. NREL/TP-5200-57991). National
542 Renewable Energy Laboratory (NREL), Golden, CO..
543 [36] Nordmann, T., Clavadetscher, L., van Sark, W., & Green, M. (2015). Analysis of long-
544 term performance of PV systems.
545 [37] Adinoyi, M. J., & Said, S. A. (2013). Effect of dust accumulation on the power outputs of
546 solar photovoltaic modules. Renewable energy, 60, 633-636.
547 [38] Guo, B., Javed, W., Figgis, B. W., & Mirza, T. (2015, March). Effect of dust and weather
548 conditions on photovoltaic performance in Doha, Qatar. In Smart Grid and Renewable Energy
549 (SGRE), 2015 First Workshop on (pp. 1-6). IEEE.

24
 ACCEPTED MANUSCRIPT

Highlights

 Performance analysis of full-scale grid-connected PV solar power plant

 Effect of weather conditions on performance and temperature correction

 Soiling losses estimation using performance ratio

 Efficient PV system operating under desert maritime climate of Djibouti