- Email: [email protected]
A method for mapping monthly average hourly diffuse illuminance from satellite data in Thailand
Available online at www.sciencedirect.com
ScienceDirect Solar Energy 102 (2014) 162–172 www.elsevier.com/locate/solener
A method for mapping monthly average hourly diffuse illuminance from satellite data in Thailand S. Janjai ⇑, S. Pattarapanitchai, J. Prathumsit, S. Buntoung, R. Wattan, I. Masiri Solar Energy Research Laboratory, Department of Physics, Faculty of Science, Silpakorn University, Nakhon Pathom 73000, Thailand Received 7 May 2013; received in revised form 14 January 2014; accepted 18 January 2014 Available online 11 February 2014 Communicated by: Associate Editor Christian A. Gueymard
Abstract This paper presents a method for mapping monthly average hourly diffuse illuminance from satellite data. The calculation of monthly average hourly diffuse illuminance starts with the estimation of monthly average hourly global illuminance from MTSAT-1R satellite data using an improved satellite-based illuminance model. Next, a diffuse fraction model is developed from ground and satellite-based data which is then used to extract diffuse illuminance from the satellite-derived global illuminance. To assess the performance of the method, modeled diffuse illuminance obtained from this method is compared with that obtained from measurements at four stations in Thailand. There is good agreement between calculated and the measured values of monthly average hourly diffuse illuminance, with the root mean square difference and mean bias difference of 9.7% and 1.4% respectively. The model is used to map monthly average hourly diffuse illuminance for the country. The maps reveal the diurnal and seasonal variations in response to a range of factors including cloud cover, zenith angle and monsoonal effects. Ó 2014 Elsevier Ltd. All rights reserved.
Keywords: Diffuse illuminance; Satellite data; Mapping
1. Introduction A utilization of daylight for illuminating building interior can provide significant saving in building electricity consumption (Chirarattananon, 2005). For this reason, daylight-integrated buildings and daylight equipment have been developed in many countries (Li and Lam, 2001; Ihm et al., 2009; Zain-Ahmed et al., 2002; Singh and Garg, 2010). Daylight consists of direct sunlight and diffuse sky light. As diffuse sky light does not create glare, it is preferable to use diffuse sky light for illuminating building interior. As a result, the amount of diffuse illuminance available at a location is usually required for assessing ⇑ Corresponding author. Tel.: +66 3427 0761; fax: +66 3427 1189.
E-mail address: [email protected] (S. Janjai). http://dx.doi.org/10.1016/j.solener.2014.01.020 0038-092X/Ó 2014 Elsevier Ltd. All rights reserved.
the potential use of daylight at that location. Ideally information on diffuse illuminance should be obtained from a dense network of daylight measuring stations where diffuse illuminance is routinely measured. However, in reality the number of the stations is usually scarce and certainly not sufficient to provide reliable data for daylight applications. As solar illuminance is part of broadband solar spectrum which is derivable from satellite data (Tarpley, 1979; Gautier et al., 1980; Perez et al., 2002; Exell, 1984; Schillings et al., 2004; Vignola et al., 2007; Polo et al., 2011; Ineichen and Perez, 1999; Janjai et al., 2009). Consequently, it is possible to derive illuminance from satellite data. The first attempt to estimate illuminance from satellite data was carried out in the SATEL-LIGHT project (Fontoynont et al., 1997). The estimation technique used in the project consists
S. Janjai et al. / Solar Energy 102 (2014) 162–172
163
Nomenclature AOD d aer Ed Eg I 0k
aerosol optical depth (–) aerosol depletion coefficient (–) diffuse illuminance (lux) global illuminance (lux) extraterrestrial spectral irradiance (Wm2 lm1) kd diffuse fraction of global illuminance (–) kt clearness index relative to global illuminance (–) ma air mass (–) MBD mean bias difference (%) n cloud index (–) R correlation coefficient (–) RMSD root mean square difference (%) Tq total transmission due to scattering (–) Ta total transmission due to absorption (–) VIS visibility (km)
aaer ag ao aT aw qaer qA qc qEA qG qT s
absorption coefficient due to aerosols (–) absorption coefficient due to gases (–) absorption coefficient due to ozone (–) total absorption define by Eq. (1b) (–) absorption coefficient due to water vapor (–) scattering coefficient of aerosols (–) cloud–atmospheric reflectivity (–) maximum cloud reflectivity (–) earth-atmospheric reflectivity (–) surface reflectivity (–) total scattering define by Eq. (1a) illuminance transmission of the atmosphere (–)
Superscript 0 quantity in satellite band 00 quantity in photopic band
of two steps. In the first step, broadband solar radiation is derived from imagery data of METEOSAT satellite. In the second step, it is converted into illuminance by using a luminous efficacy model. For the tropical and subtropical zones, Janjai et al. (2008) and He and Ng (2010) proposed satellite-based methods to estimate global illuminance from geostationary satellite data. Despite the importance of diffuse illuminance, the derivation of diffuse illuminance from satellite data receives less attention, especially in tropical zone where daylight is abundant. Therefore, the objective of this work is to develop a method for mapping diffuse illuminance from satellite data. The study area of the work is focused on a tropical zone. At a time scale of an hour, the cloud field is strongly random. This imposes constraints on the ability of satellites to map illuminance with hourly satellite data. By contrast, cloud regional structure emerges after long-term averaging. Therefore in this work, we choose to examine hourly mean illuminance by averaging each hour of the day over the period of one month. Consequently, this procedure averages short time fluctuation of illuminance caused by rapid spatial and temporal variations of cloud structure. Modeling results will present a climatology of diffuse illuminance for daylight applications. Details of the proposed method are described in the next section. Model performance is assessed in Section 3.1 while Section 3.2 maps diffuse illuminance over Thailand.
data. The method can be schematically shown in Fig. 1 and the details of each step are described as follows.
2. Description of the proposed method
In order to formulate a satellite-based global illuminance model, diffuse fraction model, and validation of the method, it is necessary to have ground-based global and diffuse illuminance data. As Thailand is divided into four geographical regions, our laboratory has stations in each region for monitoring solar radiation in different
The main idea of the proposed method is first to derive global illuminance from satellite data. Then diffuse illuminance is extracted from the satellite-derived global illuminance by using a diffuse fraction model from surface-based
2.1. Processing of satellite data The satellite data employed in this study are obtained from the visible channel (0.55–0.90 lm) of MTSAT-1R satellite, encompassing a six-year period (2006–2011). The satellite belongs to Japan Meteorological Agency (JMA). In general, the satellite gives hourly data. However, only the data from 8:30 am to 4:30 pm are used in this work. The early morning and late afternoon data are not used due to their low quality resulted from the non-Lambertian reflection. When displayed as images, the satellite data cover the entire area of Thailand with a spatial resolution of 3 3 km2. These images are transformed into a cylindrical projection and then navigated by using points on the coastlines as reference. Each navigated image comprises 500 800 pixels. Each pixel has a gray level having a value in the range of 0–255. The gray levels of all pixels are converted into earth-atmospheric reflectivity (q0EA ) using a conversion table provided by the satellite data agency. After correction for incident angle of solar radiation, the satellite data are used as input of a satellite-based illuminance model. All data is grouped and averaged according to hour for each individual month. 2.2. Illuminance measurement
164
S. Janjai et al. / Solar Energy 102 (2014) 162–172
Monthly average hourly earth-atmospheric reflectivity ( ρ'EA ) from MTSAT-1R satellite
Monthly average hourly surface reflectivity ( ρ G )
A satellite-based global illuminance model
Monthly average hourly atmospheric ancillary data
Monthly average hourly global illuminance (E g)
Surface-based monthly average hourly diffuse fraction model
Monthly average hourly diffuse illuminance (E d ) Fig. 1. Schematic diagram of the proposed method for the estimation of diffuse illuminance. All data are presented at the hourly scale, these being averaged over a month.
wavelength bands including global and diffuse illuminance. These stations are located at Chiang Mai (18.78°N, 98.98°E) in the Northern region, Ubon Ratchathani (15.25°N, 104.87°E) in the Northeastern region, Nakhon Pathom (13.82°N, 100.04°E) in the Central region and Songkhla (7.20°N, 100.60°E) in the Southern region (Fig. 2). In the south, Songkhla is dominantly equatorial, with no dry season and experiences rainfall all year round. Nakhon Pathom occupies a position close to Bangkok. It experiences industrial and urban pollution and is marked by both dry and wet seasons. Ubon Ratchathani in the northeast, is in the driest region of the country and experiences a continental climate. Chiang Mai in the north has a mountainous high elevation climate and is characterized by high pollution due to biomass burning in February to April. At each station, global illuminance is measured using a luxmeter (EKO, model ML-020-O). A second luxmeter (EKO, model ML-020-O) equipped with a shade ring (Kipp&Zonen, model CM121) is employed to measure diffuse illuminance. The voltage signals from luxmeters are captured by a datalogger (Yogokawa, model DC100) with a sampling rate of 1 s. These signals are converted into illuminance by using the sensitivity of the corresponding luxmeters. The effect of shade ring on diffuse illuminance was corrected using a correction table provided by Kipp&Zonen. The hourly average global and diffuse illuminance from the four stations for the period of 6 years (2006–2011) are used in this work. These data are partitioned into two data sets: the first set (2006–2009)
for formulating the models and the second set (2010– 2011) for model validation. The final step involves averaging all illuminance data according to hour for each month. 2.3. The satellite-based global illuminance model In general, there are two approaches to derive global illuminance from satellite data, namely direct and indirect approaches. For the direct approach, global irradiance is first estimated from satellite data, then the result is converted to global illuminance using a luminous efficacy model. The direct approach estimates directly global illuminance from satellite data. He and Ng (2010) have shown that the direct approach gave more accurate result than that of the indirect approach. Due to the better accuracy and the availability of illuminance measurements in Thailand, the direct approach was selected to estimate global illuminance in this work. The details of the approach are described as follows. 2.3.1. Formulation of the model The satellite-based global illuminance model proposed by Janjai et al. (2008) is modified by accounting for the multiple reflection and radiation absorption in upwelling path which were ignored in the original model. The schematic diagram of the absorption and scattering of solar radiation in the atmosphere of the improved model is shown in Fig. 3. For clarity we shall summarize the following quantities:
S. Janjai et al. / Solar Energy 102 (2014) 162–172
165
b
b
a
a
A B C
a
a D
b
b
Fig. 2. Pictorial view of instruments for measuring global illuminance (a) and diffuse illuminance (b), and positions of the stations. A, B, C and D indicate the Northern, Northeastern, Central and the Southern regions.
q0T ¼ q0A þ q0aer
ð1aÞ
a0T ¼ a0w þ a0o þ ag þ a0aer
ð1bÞ
T 0q
q0T
ð1cÞ
T 0a ¼ 1 a0T
ð1dÞ
¼1
where symbols q and a denote scattering and absorption respectively in the entire atmospheric path, primes denote spectral values corresponding to the satellite channel, and subscripts w, o, g and aer denote depletion by water vapor, ozone, atmospheric gases and aerosols, respectively. q0A is reflectivity due to cloud and air molecules. Total absorption and scattering are denoted by a0T and q0T respectively and total transmission due to absorption (T 0a ) or scattering (T 0q ) is given as 1 minus the total absorption or scattering respectively, Eqs. (1d) and (1c). According to Fig. 3, a fraction of downwelling solar radiation entering the atmosphere is scattered to space which is given as the sum of scattering due to clouds, atmospheric gases and aerosols (q0T ). The remaining fraction
reaches the surface after absorption due to water vapor, ozone, atmospheric gases and aerosols. Using the nomenclature described in Eq. (1), the fraction of radiation reaching the surface is given by T 0q T 0a . Part of this radiation is reflected by the amount T 0q T 0a q0G where q0G is surface reflectivity in the satellite band. This reflected fraction is further absorbed and scattered by the atmosphere, so that the frac2 2 tion reflected to space is given by T 0q T 0a q0G . There is also multiple scattering between the surface and the atmosphere and we consider that absorption is occurring for each backscattering event in its upwards and downwards path, so that the final percentage of incoming solar radiation escaping to space (q0EA ) is: 2
2
q0EA ¼ q0A þ q0aer þ ðT 0q T 0a Þ q0G ½1 þ ðq0A þ q0aer ÞT 0a q0G 2
4
2
þ ðq0A þ q0aer Þ T 0a q0G þ . . . 2
¼ q0A þ q0aer þ
ðT 0q T 0a Þ q0G 2
1 ðq0A þ q0aer ÞT 0a q0G
ð2Þ
166
S. Janjai et al. / Solar Energy 102 (2014) 162–172
Satellite
Tρ′2 Tα′2 ρ′G
ρ′T
1
Earth-atmospheric reflectivity as seen by satellite ( ρ′EA ).
Tρ′2 Tα′4 ρ′T ρ′G2
Scattering by clouds and air molecule and aerosols ( ρ′T )
Tρ′ Tα′2 ρ′G
Tρ′
Tρ′ Tα′2ρ′T ρ′G
Tρ′ Tα′4 ρ′T2 ρ′G2
Tρ′ Tα′4 ρ′T ρ′G2
α ′T
α ′T
α ′T
α′T
α′T
Absorption by water vapour, ozone, gases and aerosols ( α′T)
Tρ′ Tα′ 3ρ′T ρ′G
Tρ′Tα′
Atmosphere
Tρ′ Tα′ ρ′T2ρ′G2 5
Tρ′ Tα′ ρ′T ρ′G2 3
Tρ′ Tα′ ρ′G Absorption by ground (1 − ρ′G )
First incident solar radiation
Ground
Multiple reflection Fig. 3. Flow diagram showing radiation estimates within, above and below the atmosphere. Fluxes leaving the atmosphere are used in the earthatmosphere reflectivity calculation. Fluxes incident at the surface define the transmissivity term. q0T stands for total scattering, a0T stands for total absorption. Surface reflectivity is denoted as q0G . q0EA is Earth-atmospheric reflectivity. T 0q and T 0a are total transmission due to scattering and absorption, respectively. The prime denotes a radiation quantity in the satellite band, 0.55–0.9 lm.
Except for q0A , the other parameters in Eq. (2) can be obtained from satellite and ground-based estimates to be described in next sub-section. To evaluate the unknown q0A , Eq. (2) is rearranged as follows: q0A ¼
A AC þ ABC þ B2 C C 1 þ BC þ AC
ð3Þ 2
where A ¼ q0aer q0EA , B ¼ 1 q0aer and C ¼ T 0a q0G . Using the same nomenclature we describe the transmission of illuminance s00 from the top of the atmosphere to the surface as measured by our luxmeters. The same processes and symbols are kept as in Eq. (1), but now we denote all absorption and scattering in the photopic band (0.38–0.77 lm) by the double prime. Transmission to the surface is described by a first order transmission T 00q T 00a plus successive backscattering from the atmosphere including absorption: 2
2
2
s00 ¼ T 00q T 00a ½1 þ ðq00A þ q00aer ÞT 00a q00G þ ðq00A þ q00aer Þ T 4a q0G þ . . . ¼
T 00q T 00a
ð4Þ
2
1 ðq00A þ q00aer ÞT 00a q00G
The transmission term s00 is obtained from our surface illuminance measurements and all terms are known with the exception of atmospheric back scattered term in the photopic band (q00A ). Eq. (4) may then be written as: 2
q00A ¼
the country (Eg ¼ Eoh s00 and Eoh is the extraterrestrial illuminance). The final step involves estimating the diffuse fraction of the illuminance by use of an empirical model. The following sections provide details of how the atmospheric coefficients and diffuse fraction were obtained.
T 00a ð1 q00aer Þ s00 ð1 q00aer q00G T 00a Þ T 00a ð1 s00 q00G T 00a Þ
ð5Þ
The essence of our technique consists in relating the largely instrumental estimates of q00A at our four stations with corresponding estimates of q0A from the satellite signal (Eq. (3)). Knowing the relationship between the two variables, q00A ; s00 and illuminance Eg may be estimated throughout
2.3.2. Determination of model coefficients The coefficients of the global illuminance model are calculated as follows: 2.3.2.1. Absorption coefficients for ozone, water vapor and gases. These terms are estimated in the photopic and the satellite bands using a generic expression for absorption, ax : R k2 I 0k sxk dk k ax ¼ 1 R1 k2 ð6Þ I dk k1 0k where x stands for either ozone, water vapor or constituent gases and I 0k represents the extraterrestrial irradiance in the wavelength k. Integration is performed between two wavelength limits which in the case of the photopic band is 0.38– 0.77 lm and for the satellite band is 0.55–0.90 lm. For the case of the coefficients in the photopic band, I 0k in Eq. (6) has to be replaced by I 0k Rk where Rk is the CIE photopic response (Janjai et al., 2008). Transmission coefficients for ozone were obtained from the formula proposed by Iqbal (1983) with the total ozone column obtained from the OMI/Aura satellite. Transmission due to water vapor was estimated using the formula proposed by Leckner (1978) with precipitable water obtained from a technique developed by Janjai et al. (2008). For atmospheric gases sgk is calculated by employing the formula proposed by Leckner (1978).
S. Janjai et al. / Solar Energy 102 (2014) 162–172
2.3.2.2. Absorption and scattering coefficients due to aerosols. Detailed information on aerosol optical depth and related properties may be obtained from sunphotometers and in Thailand there are four stations which record this information as part of AERONET (Janjai et al., 2012). However a wider coverage may be obtained by using visibility observations from 85 meteorological stations throughout the country. Our procedure uses a simple formula for aerosol depletion coefficient described by Janjai et al. (2009): 0:9 0:66 ma
1 0.9
ρ′A′ = 0.7692ρ′A + 0.0974
0.8
0.76 rR==0.76
0.7 0.6
ρ A// 0.5 0.4 0.3
ð7Þ
0.2
where d aer is aerosol depletion coefficient, VIS is visibility (km) and ma is air mass. In turn daer is defined as the percentage diminution of solar radiation with a given aerosol load compared with solar radiation for the same conditions but with no aerosols. The final step involved partitioning daer into absorption and scattering terms. This was accomplished using aerosol single scattering albedo data from 82 AERONET stations located across East and Southeast Asia, including 4 stations in Thailand. These were spatially interpolated for Thailand and seasonal average patterns were obtained.
0.1
d aer ¼ 1 ½1:0358 0:3293ðVISÞ
2.3.2.3. Surface reflectivity. The surface reflectivity (q0G ) for the entire country is estimated using a satellite method proposed by Janjai et al. (2006). Satellite images of q0EA collected at 12:30 h local time is used. Rectified images are examined and those pixels with the lowest gray level value are selected so as to create the cloud-free composite image for that month. All cloud-free composite images are then transformed from earth-atmosphere reflectivity q0EA into surface reflectivity using a conversion table based on the 5S radiative transfer model (Tanre´ et al., 1986).
167
0 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
ρ A/ Fig. 4. Relation between the satellite band atmospheric reflectivity (q0A ) and photopic band atmospheric reflectivity (q00A ). Estimates refer to monthly averages of hourly data. R is correlation coefficient.
2.3.4. Performance of the model To evaluate its performance, this modified model was used to calculate monthly average hourly global illuminance at the four stations for the year 2011 and the results were compared with the corresponding global illuminance measured at these stations. The comparison gives the root mean square difference (RMSD) of 7.0% and mean bias difference (MBD) of 1.5%. The original model (Janjai et al., 2008) presents the RMSD of 8.1% and MBD of 2.6%. Overall, the modified model shows better performance, as compared to the original one. 2.4. Development of a diffuse fraction model
2.3.3. Determination of the atmospheric scattering coefficients in the photopic band (q00A ) Knowing all model input coefficients as described in the previous section, it is possible to convert q0A into q00A using a statistical formula. Illuminance data covering a 4-year period (2006–2009) were obtained from the four solar monitoring stations and converted into q00A employing Eq. (5). Concurrent satellite data for the same stations and the same period are selected and converted into q0A using Eq. (3). Linear regression between the two sets of variables is shown in Fig. 4 and the best fitted equation is written as follows: q00A ¼ 0:7692q0A þ 0:0974 ð8Þ
Diffuse illuminance (Ed) is extracted by applying a diffuse fraction relationship k 00d to the estimated global illuminance (Eg), where k 00d is a ratio of diffuse to global illuminance Ed/Eg. The diffuse fraction for broadband solar radiation may be expressed as a polynomial function of clearness index (e.g. Orgill and Hollands, 1977). However, it may also represented by a sigmoid function of clearness index (Boland et al., 2008). We have chosen to use a sigmoid function in this study as polynomial functions are very sensitive to the number of terms used. In this work we propose to represent the illuminance diffuse fraction (k 00d ) as a sigmoid function of parameters affecting diffuse illuminance as follows:
It is noted that both q0A and q00A are calculated as monthly average hourly basis. To ensure the generality of Eq. (8), the regression for individual station was carried out and it was found that the slope and intercept of each individual regression were within the 95% confidence level. As the four stations are located in the main regions of Thailand, the relationship in Eq. (8) is used for the estimation of q00A throughout the entire country.
k 00d ¼
1 expða0 þ
a1 k 00t
1 þ a2 ma þ a3 AOD þ a4 nÞ
ð9Þ
where AOD is aerosol optical depth at 500 nm, n is satellite-derived cloud index, k 00t is clearness index relative to global illuminance defined as a ratio of surface global illuminance to extraterrestrial illuminance, ma is air mass and a0 , a1 , a2 , a3 and a4 are empirical coefficients of the model.
168
S. Janjai et al. / Solar Energy 102 (2014) 162–172
The cloud index (n), which was first introduced by Cano et al. (1986), was defined as n¼
q0EA q0G q0c q0G
ð10Þ
where q0EA , q0G and q0c are the earth-atmospheric reflectivity, surface reflectivity and maximum cloud reflectivity, respectively. To obtain the values of the empirical coefficients, Eq. (9) was regressed using values of k 00d , k 00t , ma, AOD and n for the 4-year period (2006–2009) at the four stations. k 00d and k 00t are calculated from illuminance data and AOD is obtained from MODIS/Terra satellite. The cloud index (n) is derived from MTSAT-1R satellite data using the method described in Janjai (2013). All parameters used in Eq. (9) are estimated on monthly average hourly basis. All averaging processes are performed over specific hours in the months. The results are as follows: a0 ¼ 1:6288; a1 ¼ 3:0098; a2 ¼ 0:07964; a3 ¼ 0:89073, and a4 ¼ 0:95809. In the final step, the diffuse fraction obtained from preceding section is used to extract global illuminance calculated in Section 2.3 by using the following equation. Ed ¼ k 00d Eg
Fig. 5. Modeled diffuse fraction (k 00d;model ) vs measured diffuse fraction (k 00d;meas ) for illuminance for all four stations during 2010–2011.
70
ð11Þ
Chiang Mai 60
Ubon Ratchathani
3. Results and discussion
Nakhon Pathom
3.1. Performance of the method The method described in the preceding section is now ready for use to calculate diffuse illuminance at all pixels of the satellite data. Although the method is logically developed, it is necessary to test its performance prior to the utilization. The diffuse fraction in the photopic band is first examined for all data (2010–2011) grouped together. Fig. 5 compares hourly modeled fractions with those measured for all data. Results give a regression as shown below: k 00d;model 2
¼
0:794k 00d;meas
þ 0:099
R ¼ 0:84; RMSD ¼ 8:9%; MBD ¼ 2:0%
Ed,mode l (klux)
50
Songkhla 1:1 line
40
30
20
MBD= -1.4% RMSD= 9.7%
10
0 0
10
20
30
40
50
60
70
Ed,meas (klux)
ð12Þ
where k 00d;model and k 00d;meas are diffuse fraction calculated from the model and that obtained from the measurement, respectively. The above statistical analysis was repeated for each individual station. The slopes and intercepts of each individual regression were within the 95% confidence level. Therefore, the diffuse fraction model performs equally for four stations. Measured and modeled monthly averages of hourly diffuse illuminance at the four solar monitoring stations for the period: 2010–2011 are next compared (Fig. 6). It is noticed that the data points are scattered around the 1:1 line. However, the root mean square difference (RMSD) and mean bias difference (MBD) between the calculated and measured diffuse illuminance are 9.7% and 1.4%,
Fig. 6. Comparison between monthly average hourly diffuse illuminance calculated from the model (Ed,model) and that obtained from the measurement (Ed,meas).
respectively. This means that, the two data sets are in good agreement. The regression was also applied to each individual station data and results showed that all slopes and intercepts were also within the 95% confidence interval. 3.2. Mapping diffuse illuminance The result is displayed as maps showing geographical distribution of monthly average hourly diffuse illuminance as depicted in Figs. 7 and 8. The maps show diurnal and seasonal variations of monthly average hourly diffuse illuminance over Thailand.
S. Janjai et al. / Solar Energy 102 (2014) 162–172
169
January
February
March
April
May
June
8:30
9:30
10:30
11:30
12:30
13:30
14:30
15:30
klux 10
15
20
25
30
35
40
45
50
55
60
65
70
Fig. 7. The maps of monthly average hourly diffuse illuminance for January to June.
16:30
170
S. Janjai et al. / Solar Energy 102 (2014) 162–172
July
August
September
October
November
December
8:30
9:30
10:30
11:30
12:30
13:30
14:30
15:30
klux 10
15
20
25
30
35
40
45
50
55
60
65
70
Fig. 8. The maps of monthly average hourly diffuse illuminance for July to December.
16:30
S. Janjai et al. / Solar Energy 102 (2014) 162–172
Apart from the effect of the variation of the position of the sun, the diurnal variation is likely influenced by the variation of cloud amount and type. However, the satellite is not able to give the detail effect of cloud on this variation. The seasonal variation is clearly seen in the maps of 9:30– 14:30. It is noticed that from January to February, the area with high diffuse illuminance expands from the Southern to the Northern regions of this country. This is because the sun path progressively moves northward to the celestial equator during this period, causing the increase of global illuminance. As diffuse illuminance is part of global illuminance, the amount of diffuse illuminance depends strongly on the amount of global illuminance and the diffuse fraction. The diffuse fraction in the Southern region is higher than that of Northern region, being likely due to scattered clouds in this period. In March and April, high diffuse illuminance of over 40 klux features in most parts of the country. This is because the sun is nearly at the overhead position around noon time over the country, causing high global illuminance. In addition the diffuse fraction is also high, due to high aerosol load caused by biomass burning (Janjai et al., 2012). In general, the onset of the southwest monsoon is in mid-May and lasts until mid-October. This monsoon causes cloudy skies and rain for most parts of the country. Although, global illuminance decreases in this period, the diffuse fraction is relatively high due to clouds, keeping the high level of diffuse illuminance over country. Normally, the monsoons change from a southwest to a northeast monsoons in mid-October, and the northeast monsoons brings dry and cool air to the Northern, Northeastern and Central regions. However, while blowing across the Gulf of Thailand, it brings moisture to the Southern region causing cloudy skies and subsequently high diffuse fraction. Although, the Northern, Northeastern and Central regions have a majority of clear days caused by the northeast monsoon, the level of global illuminance in these regions is not high in November and December because the sun path moves southward from the celestial equator in this period. High global illuminance in the southern region is caused by the favorable position of the sun path. Due to high global illuminance and diffuse fraction, diffuse illuminance in this region is constantly high during November–December. 4. Conclusion A method for mapping monthly average hourly diffuse illuminance from satellite data has been developed. In developing the method, a satellite-based global illuminance model is improved. The improved model provides a more complete description of the absorption and scattering of solar radiation in the earth-atmospheric system. The model is used to derive monthly average hourly global illuminance from MTSAT-1R satellite data. Monthly average hourly diffuse illuminance is extracted from the satellite derived-global illuminance by using a diffuse fraction model developed in this work. Monthly average hourly
171
diffuse illuminance obtained from this method agrees well with that obtained from measurements, with RMSD and MBD of 9.7% and -1.4%, respectively. The model is used to generate monthly average hourly diffuse illuminance maps of Thailand. The maps show the diurnal and seasonal variations of monthly average hourly diffuse illuminance with the effect of the northeast and southwest monsoons. Acknowledgements The authors would like to thank the Thai Meteorological Department for supporting the measurements of illuminance and related atmospheric parameters. The authors are also grateful to Dr. Manuel Nunez for his assistance to improve the manuscript. References Boland, J., Ridley, B., Brown, B., 2008. Models of diffuse solar radiation. Renew. Energy 33, 384–575. Cano, D., Monget, J.M., Albuisson, M., Gurnard, H., Regas, N., Wald, L.A., 1986. A method for the determination of the global solar radiation from meteorological satellite data. Sol. Energy 37, 31–39. Chirarattananon, S., 2005. Building for Energy Efficiency. Asian Institute of Technology, Bankgok, Thailand. Exell, R.H.B., 1984. Mapping solar radiation by meteorological satellite. Renew. Energy Rev. J. 6 (1), 27–39. Fontoynont, M., Dumortier, D., Heinemann, D., Hammer, A., Olseth, J., Skatveit, A., Ineichen, P., Reise, C., Page, J., Roche, L., Beyer, H.G., Wald, L., Santos, A., 1997. SATELLIGHT: A European program dedicated to serving daylight data computed for Meteosat images. In: Proceedings of the eighth European Lighting Conference, 11–14 May, Amsterdam. Gautier, C., Diak, G., Masse, S., 1980. A simple physical model to estimate incident solar radiation at the surface from GOES satellite data. J. Appl. Meteorol. 36, 1005–1012. He, J.Z., Ng, E., 2010. Using satellite-based methods to predict daylight illuminance for subtropical Hong Kong. Lighting Res. Technol. 42, 135–147. Ihm, P., Nemri, A., Krarti, M., 2009. Estimation of lighting energy savings from daylighting. Build. Environ. 44, 509–514. Ineichen, P., Perez, R., 1999. Derivation of cloud index from geostationary satellites and application to the production of solar irradiance and daylight illuminance data. Theoret. Appl. Climatol. 64, 119–130. Iqbal, M., 1983. An Introduction to Solar Radiation. Academic Press, New York. Janjai, S., 2013. A satellite-based sky luminance model for the tropics. Int. J. Photoenergy 2013, 1–11. Janjai, S., Wanvong, W., Laksanaboonsong, J., 2006. The determination of surface albedo of Thailand using satellite data. In: Proceedings of the 2nd Joint International Conference on Sustainable Energy and Environment (SEE2006) Bangkok, Thailand, pp. 156–161. Janjai, S., Tohsing, K., Nunez, M., Laksanaboonsong, J., 2008. A technique for mapping global illuminance from satellite data. Sol. Energy 82, 543–555. Janjai, S., Laksanaboonsong, J., Pankaew, P., 2009. A model for calculating hourly global solar radiation from satellite data in the tropics. Appl. Energy 86, 1450–1457. Janjai, S., Nunez, M., Masiri, I., Wattan, R., Buntoung, S., Jantarach, T., Promsen, W., 2012. Aerosol optical properties at four sites in Thailand. Atmos. Climate Sci. 2, 441–453. Leckner, B., 1978. The spectral distribution of solar radiation at the earth’s surface-Elements of a model. Sol. Energy 20, 143–150. Li, D.H.W., Lam, J.C., 2001. Evaluation of lighting performance in office buildings with daylighting controls. Energy Build. 33, 793–803.
172
S. Janjai et al. / Solar Energy 102 (2014) 162–172
Orgill, J.F., Hollands, K.G.T., 1977. Correlation equation for hourly diffuse radiation on horizontal surface. Sol. Energy 19, 357–362. Perez, R., Ineichen, P., Moore, K., Kmiecik, M., Chain, C., George, R., Vignola, F., 2002. A new operational model for satellite-derived irradiances: description and validation. Sol. Energy 73, 307–317. Polo, J., Zarzalejo, L.F., Cony, M., Navarro, A.A., Marchante, R., Martı´n, L., Romero, M., 2011. Solar radiation estimations over India using Meteosat satellite images. Sol. Energy 85, 2395–2406. Schillings, C., Mannstein, H., Meyer, R., 2004. Operational method for deriving high resolution direct normal irradiance from satellite data. Sol. Energy 76, 475–484. Singh, M.C., Garg, S.N., 2010. Illuminance estimation and daylighting energy savings for Indian regions. Renew. Energy 35, 703–711.
Tanre´, D., Deroo, C., Duhaut, P., Herman, N., Morcrette, J.J., Perbos, J., Deschamps, P.Y., 1986. Simulation of the Satellite Signal in the Solar Spectrum. Technical Report, Laboratoire d’Optique Atmospherique, Universite´ des Science et Techniques de Lille, 59655 Villeneuve d’ Ascq Cedet dex, France. Tarpley, J.D., 1979. Estimating incident solar radiation at the surface from geostationary satellite data. J. Appl. Meteorol. 18, 1172–1181. Vignola, F., Harlan, P., Perez, R., Kmiecik, M., 2007. Analysis of satellite derived beam and global solar radiation data. Sol. Energy 81, 768–772. Zain-Ahmed, A., Sopian, K., Othman, M.Y.H., Sayigh, A.A.M., Surendran, P.N., 2002. Daylighting as a passive solar design strategy in tropical buildings: a case study of Malaysia. Energy Convers. Manage. 43, 1725–1736.
ScienceDirect Solar Energy 102 (2014) 162–172 www.elsevier.com/locate/solener
A method for mapping monthly average hourly diffuse illuminance from satellite data in Thailand S. Janjai ⇑, S. Pattarapanitchai, J. Prathumsit, S. Buntoung, R. Wattan, I. Masiri Solar Energy Research Laboratory, Department of Physics, Faculty of Science, Silpakorn University, Nakhon Pathom 73000, Thailand Received 7 May 2013; received in revised form 14 January 2014; accepted 18 January 2014 Available online 11 February 2014 Communicated by: Associate Editor Christian A. Gueymard
Abstract This paper presents a method for mapping monthly average hourly diffuse illuminance from satellite data. The calculation of monthly average hourly diffuse illuminance starts with the estimation of monthly average hourly global illuminance from MTSAT-1R satellite data using an improved satellite-based illuminance model. Next, a diffuse fraction model is developed from ground and satellite-based data which is then used to extract diffuse illuminance from the satellite-derived global illuminance. To assess the performance of the method, modeled diffuse illuminance obtained from this method is compared with that obtained from measurements at four stations in Thailand. There is good agreement between calculated and the measured values of monthly average hourly diffuse illuminance, with the root mean square difference and mean bias difference of 9.7% and 1.4% respectively. The model is used to map monthly average hourly diffuse illuminance for the country. The maps reveal the diurnal and seasonal variations in response to a range of factors including cloud cover, zenith angle and monsoonal effects. Ó 2014 Elsevier Ltd. All rights reserved.
Keywords: Diffuse illuminance; Satellite data; Mapping
1. Introduction A utilization of daylight for illuminating building interior can provide significant saving in building electricity consumption (Chirarattananon, 2005). For this reason, daylight-integrated buildings and daylight equipment have been developed in many countries (Li and Lam, 2001; Ihm et al., 2009; Zain-Ahmed et al., 2002; Singh and Garg, 2010). Daylight consists of direct sunlight and diffuse sky light. As diffuse sky light does not create glare, it is preferable to use diffuse sky light for illuminating building interior. As a result, the amount of diffuse illuminance available at a location is usually required for assessing ⇑ Corresponding author. Tel.: +66 3427 0761; fax: +66 3427 1189.
E-mail address: [email protected] (S. Janjai). http://dx.doi.org/10.1016/j.solener.2014.01.020 0038-092X/Ó 2014 Elsevier Ltd. All rights reserved.
the potential use of daylight at that location. Ideally information on diffuse illuminance should be obtained from a dense network of daylight measuring stations where diffuse illuminance is routinely measured. However, in reality the number of the stations is usually scarce and certainly not sufficient to provide reliable data for daylight applications. As solar illuminance is part of broadband solar spectrum which is derivable from satellite data (Tarpley, 1979; Gautier et al., 1980; Perez et al., 2002; Exell, 1984; Schillings et al., 2004; Vignola et al., 2007; Polo et al., 2011; Ineichen and Perez, 1999; Janjai et al., 2009). Consequently, it is possible to derive illuminance from satellite data. The first attempt to estimate illuminance from satellite data was carried out in the SATEL-LIGHT project (Fontoynont et al., 1997). The estimation technique used in the project consists
S. Janjai et al. / Solar Energy 102 (2014) 162–172
163
Nomenclature AOD d aer Ed Eg I 0k
aerosol optical depth (–) aerosol depletion coefficient (–) diffuse illuminance (lux) global illuminance (lux) extraterrestrial spectral irradiance (Wm2 lm1) kd diffuse fraction of global illuminance (–) kt clearness index relative to global illuminance (–) ma air mass (–) MBD mean bias difference (%) n cloud index (–) R correlation coefficient (–) RMSD root mean square difference (%) Tq total transmission due to scattering (–) Ta total transmission due to absorption (–) VIS visibility (km)
aaer ag ao aT aw qaer qA qc qEA qG qT s
absorption coefficient due to aerosols (–) absorption coefficient due to gases (–) absorption coefficient due to ozone (–) total absorption define by Eq. (1b) (–) absorption coefficient due to water vapor (–) scattering coefficient of aerosols (–) cloud–atmospheric reflectivity (–) maximum cloud reflectivity (–) earth-atmospheric reflectivity (–) surface reflectivity (–) total scattering define by Eq. (1a) illuminance transmission of the atmosphere (–)
Superscript 0 quantity in satellite band 00 quantity in photopic band
of two steps. In the first step, broadband solar radiation is derived from imagery data of METEOSAT satellite. In the second step, it is converted into illuminance by using a luminous efficacy model. For the tropical and subtropical zones, Janjai et al. (2008) and He and Ng (2010) proposed satellite-based methods to estimate global illuminance from geostationary satellite data. Despite the importance of diffuse illuminance, the derivation of diffuse illuminance from satellite data receives less attention, especially in tropical zone where daylight is abundant. Therefore, the objective of this work is to develop a method for mapping diffuse illuminance from satellite data. The study area of the work is focused on a tropical zone. At a time scale of an hour, the cloud field is strongly random. This imposes constraints on the ability of satellites to map illuminance with hourly satellite data. By contrast, cloud regional structure emerges after long-term averaging. Therefore in this work, we choose to examine hourly mean illuminance by averaging each hour of the day over the period of one month. Consequently, this procedure averages short time fluctuation of illuminance caused by rapid spatial and temporal variations of cloud structure. Modeling results will present a climatology of diffuse illuminance for daylight applications. Details of the proposed method are described in the next section. Model performance is assessed in Section 3.1 while Section 3.2 maps diffuse illuminance over Thailand.
data. The method can be schematically shown in Fig. 1 and the details of each step are described as follows.
2. Description of the proposed method
In order to formulate a satellite-based global illuminance model, diffuse fraction model, and validation of the method, it is necessary to have ground-based global and diffuse illuminance data. As Thailand is divided into four geographical regions, our laboratory has stations in each region for monitoring solar radiation in different
The main idea of the proposed method is first to derive global illuminance from satellite data. Then diffuse illuminance is extracted from the satellite-derived global illuminance by using a diffuse fraction model from surface-based
2.1. Processing of satellite data The satellite data employed in this study are obtained from the visible channel (0.55–0.90 lm) of MTSAT-1R satellite, encompassing a six-year period (2006–2011). The satellite belongs to Japan Meteorological Agency (JMA). In general, the satellite gives hourly data. However, only the data from 8:30 am to 4:30 pm are used in this work. The early morning and late afternoon data are not used due to their low quality resulted from the non-Lambertian reflection. When displayed as images, the satellite data cover the entire area of Thailand with a spatial resolution of 3 3 km2. These images are transformed into a cylindrical projection and then navigated by using points on the coastlines as reference. Each navigated image comprises 500 800 pixels. Each pixel has a gray level having a value in the range of 0–255. The gray levels of all pixels are converted into earth-atmospheric reflectivity (q0EA ) using a conversion table provided by the satellite data agency. After correction for incident angle of solar radiation, the satellite data are used as input of a satellite-based illuminance model. All data is grouped and averaged according to hour for each individual month. 2.2. Illuminance measurement
164
S. Janjai et al. / Solar Energy 102 (2014) 162–172
Monthly average hourly earth-atmospheric reflectivity ( ρ'EA ) from MTSAT-1R satellite
Monthly average hourly surface reflectivity ( ρ G )
A satellite-based global illuminance model
Monthly average hourly atmospheric ancillary data
Monthly average hourly global illuminance (E g)
Surface-based monthly average hourly diffuse fraction model
Monthly average hourly diffuse illuminance (E d ) Fig. 1. Schematic diagram of the proposed method for the estimation of diffuse illuminance. All data are presented at the hourly scale, these being averaged over a month.
wavelength bands including global and diffuse illuminance. These stations are located at Chiang Mai (18.78°N, 98.98°E) in the Northern region, Ubon Ratchathani (15.25°N, 104.87°E) in the Northeastern region, Nakhon Pathom (13.82°N, 100.04°E) in the Central region and Songkhla (7.20°N, 100.60°E) in the Southern region (Fig. 2). In the south, Songkhla is dominantly equatorial, with no dry season and experiences rainfall all year round. Nakhon Pathom occupies a position close to Bangkok. It experiences industrial and urban pollution and is marked by both dry and wet seasons. Ubon Ratchathani in the northeast, is in the driest region of the country and experiences a continental climate. Chiang Mai in the north has a mountainous high elevation climate and is characterized by high pollution due to biomass burning in February to April. At each station, global illuminance is measured using a luxmeter (EKO, model ML-020-O). A second luxmeter (EKO, model ML-020-O) equipped with a shade ring (Kipp&Zonen, model CM121) is employed to measure diffuse illuminance. The voltage signals from luxmeters are captured by a datalogger (Yogokawa, model DC100) with a sampling rate of 1 s. These signals are converted into illuminance by using the sensitivity of the corresponding luxmeters. The effect of shade ring on diffuse illuminance was corrected using a correction table provided by Kipp&Zonen. The hourly average global and diffuse illuminance from the four stations for the period of 6 years (2006–2011) are used in this work. These data are partitioned into two data sets: the first set (2006–2009)
for formulating the models and the second set (2010– 2011) for model validation. The final step involves averaging all illuminance data according to hour for each month. 2.3. The satellite-based global illuminance model In general, there are two approaches to derive global illuminance from satellite data, namely direct and indirect approaches. For the direct approach, global irradiance is first estimated from satellite data, then the result is converted to global illuminance using a luminous efficacy model. The direct approach estimates directly global illuminance from satellite data. He and Ng (2010) have shown that the direct approach gave more accurate result than that of the indirect approach. Due to the better accuracy and the availability of illuminance measurements in Thailand, the direct approach was selected to estimate global illuminance in this work. The details of the approach are described as follows. 2.3.1. Formulation of the model The satellite-based global illuminance model proposed by Janjai et al. (2008) is modified by accounting for the multiple reflection and radiation absorption in upwelling path which were ignored in the original model. The schematic diagram of the absorption and scattering of solar radiation in the atmosphere of the improved model is shown in Fig. 3. For clarity we shall summarize the following quantities:
S. Janjai et al. / Solar Energy 102 (2014) 162–172
165
b
b
a
a
A B C
a
a D
b
b
Fig. 2. Pictorial view of instruments for measuring global illuminance (a) and diffuse illuminance (b), and positions of the stations. A, B, C and D indicate the Northern, Northeastern, Central and the Southern regions.
q0T ¼ q0A þ q0aer
ð1aÞ
a0T ¼ a0w þ a0o þ ag þ a0aer
ð1bÞ
T 0q
q0T
ð1cÞ
T 0a ¼ 1 a0T
ð1dÞ
¼1
where symbols q and a denote scattering and absorption respectively in the entire atmospheric path, primes denote spectral values corresponding to the satellite channel, and subscripts w, o, g and aer denote depletion by water vapor, ozone, atmospheric gases and aerosols, respectively. q0A is reflectivity due to cloud and air molecules. Total absorption and scattering are denoted by a0T and q0T respectively and total transmission due to absorption (T 0a ) or scattering (T 0q ) is given as 1 minus the total absorption or scattering respectively, Eqs. (1d) and (1c). According to Fig. 3, a fraction of downwelling solar radiation entering the atmosphere is scattered to space which is given as the sum of scattering due to clouds, atmospheric gases and aerosols (q0T ). The remaining fraction
reaches the surface after absorption due to water vapor, ozone, atmospheric gases and aerosols. Using the nomenclature described in Eq. (1), the fraction of radiation reaching the surface is given by T 0q T 0a . Part of this radiation is reflected by the amount T 0q T 0a q0G where q0G is surface reflectivity in the satellite band. This reflected fraction is further absorbed and scattered by the atmosphere, so that the frac2 2 tion reflected to space is given by T 0q T 0a q0G . There is also multiple scattering between the surface and the atmosphere and we consider that absorption is occurring for each backscattering event in its upwards and downwards path, so that the final percentage of incoming solar radiation escaping to space (q0EA ) is: 2
2
q0EA ¼ q0A þ q0aer þ ðT 0q T 0a Þ q0G ½1 þ ðq0A þ q0aer ÞT 0a q0G 2
4
2
þ ðq0A þ q0aer Þ T 0a q0G þ . . . 2
¼ q0A þ q0aer þ
ðT 0q T 0a Þ q0G 2
1 ðq0A þ q0aer ÞT 0a q0G
ð2Þ
166
S. Janjai et al. / Solar Energy 102 (2014) 162–172
Satellite
Tρ′2 Tα′2 ρ′G
ρ′T
1
Earth-atmospheric reflectivity as seen by satellite ( ρ′EA ).
Tρ′2 Tα′4 ρ′T ρ′G2
Scattering by clouds and air molecule and aerosols ( ρ′T )
Tρ′ Tα′2 ρ′G
Tρ′
Tρ′ Tα′2ρ′T ρ′G
Tρ′ Tα′4 ρ′T2 ρ′G2
Tρ′ Tα′4 ρ′T ρ′G2
α ′T
α ′T
α ′T
α′T
α′T
Absorption by water vapour, ozone, gases and aerosols ( α′T)
Tρ′ Tα′ 3ρ′T ρ′G
Tρ′Tα′
Atmosphere
Tρ′ Tα′ ρ′T2ρ′G2 5
Tρ′ Tα′ ρ′T ρ′G2 3
Tρ′ Tα′ ρ′G Absorption by ground (1 − ρ′G )
First incident solar radiation
Ground
Multiple reflection Fig. 3. Flow diagram showing radiation estimates within, above and below the atmosphere. Fluxes leaving the atmosphere are used in the earthatmosphere reflectivity calculation. Fluxes incident at the surface define the transmissivity term. q0T stands for total scattering, a0T stands for total absorption. Surface reflectivity is denoted as q0G . q0EA is Earth-atmospheric reflectivity. T 0q and T 0a are total transmission due to scattering and absorption, respectively. The prime denotes a radiation quantity in the satellite band, 0.55–0.9 lm.
Except for q0A , the other parameters in Eq. (2) can be obtained from satellite and ground-based estimates to be described in next sub-section. To evaluate the unknown q0A , Eq. (2) is rearranged as follows: q0A ¼
A AC þ ABC þ B2 C C 1 þ BC þ AC
ð3Þ 2
where A ¼ q0aer q0EA , B ¼ 1 q0aer and C ¼ T 0a q0G . Using the same nomenclature we describe the transmission of illuminance s00 from the top of the atmosphere to the surface as measured by our luxmeters. The same processes and symbols are kept as in Eq. (1), but now we denote all absorption and scattering in the photopic band (0.38–0.77 lm) by the double prime. Transmission to the surface is described by a first order transmission T 00q T 00a plus successive backscattering from the atmosphere including absorption: 2
2
2
s00 ¼ T 00q T 00a ½1 þ ðq00A þ q00aer ÞT 00a q00G þ ðq00A þ q00aer Þ T 4a q0G þ . . . ¼
T 00q T 00a
ð4Þ
2
1 ðq00A þ q00aer ÞT 00a q00G
The transmission term s00 is obtained from our surface illuminance measurements and all terms are known with the exception of atmospheric back scattered term in the photopic band (q00A ). Eq. (4) may then be written as: 2
q00A ¼
the country (Eg ¼ Eoh s00 and Eoh is the extraterrestrial illuminance). The final step involves estimating the diffuse fraction of the illuminance by use of an empirical model. The following sections provide details of how the atmospheric coefficients and diffuse fraction were obtained.
T 00a ð1 q00aer Þ s00 ð1 q00aer q00G T 00a Þ T 00a ð1 s00 q00G T 00a Þ
ð5Þ
The essence of our technique consists in relating the largely instrumental estimates of q00A at our four stations with corresponding estimates of q0A from the satellite signal (Eq. (3)). Knowing the relationship between the two variables, q00A ; s00 and illuminance Eg may be estimated throughout
2.3.2. Determination of model coefficients The coefficients of the global illuminance model are calculated as follows: 2.3.2.1. Absorption coefficients for ozone, water vapor and gases. These terms are estimated in the photopic and the satellite bands using a generic expression for absorption, ax : R k2 I 0k sxk dk k ax ¼ 1 R1 k2 ð6Þ I dk k1 0k where x stands for either ozone, water vapor or constituent gases and I 0k represents the extraterrestrial irradiance in the wavelength k. Integration is performed between two wavelength limits which in the case of the photopic band is 0.38– 0.77 lm and for the satellite band is 0.55–0.90 lm. For the case of the coefficients in the photopic band, I 0k in Eq. (6) has to be replaced by I 0k Rk where Rk is the CIE photopic response (Janjai et al., 2008). Transmission coefficients for ozone were obtained from the formula proposed by Iqbal (1983) with the total ozone column obtained from the OMI/Aura satellite. Transmission due to water vapor was estimated using the formula proposed by Leckner (1978) with precipitable water obtained from a technique developed by Janjai et al. (2008). For atmospheric gases sgk is calculated by employing the formula proposed by Leckner (1978).
S. Janjai et al. / Solar Energy 102 (2014) 162–172
2.3.2.2. Absorption and scattering coefficients due to aerosols. Detailed information on aerosol optical depth and related properties may be obtained from sunphotometers and in Thailand there are four stations which record this information as part of AERONET (Janjai et al., 2012). However a wider coverage may be obtained by using visibility observations from 85 meteorological stations throughout the country. Our procedure uses a simple formula for aerosol depletion coefficient described by Janjai et al. (2009): 0:9 0:66 ma
1 0.9
ρ′A′ = 0.7692ρ′A + 0.0974
0.8
0.76 rR==0.76
0.7 0.6
ρ A// 0.5 0.4 0.3
ð7Þ
0.2
where d aer is aerosol depletion coefficient, VIS is visibility (km) and ma is air mass. In turn daer is defined as the percentage diminution of solar radiation with a given aerosol load compared with solar radiation for the same conditions but with no aerosols. The final step involved partitioning daer into absorption and scattering terms. This was accomplished using aerosol single scattering albedo data from 82 AERONET stations located across East and Southeast Asia, including 4 stations in Thailand. These were spatially interpolated for Thailand and seasonal average patterns were obtained.
0.1
d aer ¼ 1 ½1:0358 0:3293ðVISÞ
2.3.2.3. Surface reflectivity. The surface reflectivity (q0G ) for the entire country is estimated using a satellite method proposed by Janjai et al. (2006). Satellite images of q0EA collected at 12:30 h local time is used. Rectified images are examined and those pixels with the lowest gray level value are selected so as to create the cloud-free composite image for that month. All cloud-free composite images are then transformed from earth-atmosphere reflectivity q0EA into surface reflectivity using a conversion table based on the 5S radiative transfer model (Tanre´ et al., 1986).
167
0 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
ρ A/ Fig. 4. Relation between the satellite band atmospheric reflectivity (q0A ) and photopic band atmospheric reflectivity (q00A ). Estimates refer to monthly averages of hourly data. R is correlation coefficient.
2.3.4. Performance of the model To evaluate its performance, this modified model was used to calculate monthly average hourly global illuminance at the four stations for the year 2011 and the results were compared with the corresponding global illuminance measured at these stations. The comparison gives the root mean square difference (RMSD) of 7.0% and mean bias difference (MBD) of 1.5%. The original model (Janjai et al., 2008) presents the RMSD of 8.1% and MBD of 2.6%. Overall, the modified model shows better performance, as compared to the original one. 2.4. Development of a diffuse fraction model
2.3.3. Determination of the atmospheric scattering coefficients in the photopic band (q00A ) Knowing all model input coefficients as described in the previous section, it is possible to convert q0A into q00A using a statistical formula. Illuminance data covering a 4-year period (2006–2009) were obtained from the four solar monitoring stations and converted into q00A employing Eq. (5). Concurrent satellite data for the same stations and the same period are selected and converted into q0A using Eq. (3). Linear regression between the two sets of variables is shown in Fig. 4 and the best fitted equation is written as follows: q00A ¼ 0:7692q0A þ 0:0974 ð8Þ
Diffuse illuminance (Ed) is extracted by applying a diffuse fraction relationship k 00d to the estimated global illuminance (Eg), where k 00d is a ratio of diffuse to global illuminance Ed/Eg. The diffuse fraction for broadband solar radiation may be expressed as a polynomial function of clearness index (e.g. Orgill and Hollands, 1977). However, it may also represented by a sigmoid function of clearness index (Boland et al., 2008). We have chosen to use a sigmoid function in this study as polynomial functions are very sensitive to the number of terms used. In this work we propose to represent the illuminance diffuse fraction (k 00d ) as a sigmoid function of parameters affecting diffuse illuminance as follows:
It is noted that both q0A and q00A are calculated as monthly average hourly basis. To ensure the generality of Eq. (8), the regression for individual station was carried out and it was found that the slope and intercept of each individual regression were within the 95% confidence level. As the four stations are located in the main regions of Thailand, the relationship in Eq. (8) is used for the estimation of q00A throughout the entire country.
k 00d ¼
1 expða0 þ
a1 k 00t
1 þ a2 ma þ a3 AOD þ a4 nÞ
ð9Þ
where AOD is aerosol optical depth at 500 nm, n is satellite-derived cloud index, k 00t is clearness index relative to global illuminance defined as a ratio of surface global illuminance to extraterrestrial illuminance, ma is air mass and a0 , a1 , a2 , a3 and a4 are empirical coefficients of the model.
168
S. Janjai et al. / Solar Energy 102 (2014) 162–172
The cloud index (n), which was first introduced by Cano et al. (1986), was defined as n¼
q0EA q0G q0c q0G
ð10Þ
where q0EA , q0G and q0c are the earth-atmospheric reflectivity, surface reflectivity and maximum cloud reflectivity, respectively. To obtain the values of the empirical coefficients, Eq. (9) was regressed using values of k 00d , k 00t , ma, AOD and n for the 4-year period (2006–2009) at the four stations. k 00d and k 00t are calculated from illuminance data and AOD is obtained from MODIS/Terra satellite. The cloud index (n) is derived from MTSAT-1R satellite data using the method described in Janjai (2013). All parameters used in Eq. (9) are estimated on monthly average hourly basis. All averaging processes are performed over specific hours in the months. The results are as follows: a0 ¼ 1:6288; a1 ¼ 3:0098; a2 ¼ 0:07964; a3 ¼ 0:89073, and a4 ¼ 0:95809. In the final step, the diffuse fraction obtained from preceding section is used to extract global illuminance calculated in Section 2.3 by using the following equation. Ed ¼ k 00d Eg
Fig. 5. Modeled diffuse fraction (k 00d;model ) vs measured diffuse fraction (k 00d;meas ) for illuminance for all four stations during 2010–2011.
70
ð11Þ
Chiang Mai 60
Ubon Ratchathani
3. Results and discussion
Nakhon Pathom
3.1. Performance of the method The method described in the preceding section is now ready for use to calculate diffuse illuminance at all pixels of the satellite data. Although the method is logically developed, it is necessary to test its performance prior to the utilization. The diffuse fraction in the photopic band is first examined for all data (2010–2011) grouped together. Fig. 5 compares hourly modeled fractions with those measured for all data. Results give a regression as shown below: k 00d;model 2
¼
0:794k 00d;meas
þ 0:099
R ¼ 0:84; RMSD ¼ 8:9%; MBD ¼ 2:0%
Ed,mode l (klux)
50
Songkhla 1:1 line
40
30
20
MBD= -1.4% RMSD= 9.7%
10
0 0
10
20
30
40
50
60
70
Ed,meas (klux)
ð12Þ
where k 00d;model and k 00d;meas are diffuse fraction calculated from the model and that obtained from the measurement, respectively. The above statistical analysis was repeated for each individual station. The slopes and intercepts of each individual regression were within the 95% confidence level. Therefore, the diffuse fraction model performs equally for four stations. Measured and modeled monthly averages of hourly diffuse illuminance at the four solar monitoring stations for the period: 2010–2011 are next compared (Fig. 6). It is noticed that the data points are scattered around the 1:1 line. However, the root mean square difference (RMSD) and mean bias difference (MBD) between the calculated and measured diffuse illuminance are 9.7% and 1.4%,
Fig. 6. Comparison between monthly average hourly diffuse illuminance calculated from the model (Ed,model) and that obtained from the measurement (Ed,meas).
respectively. This means that, the two data sets are in good agreement. The regression was also applied to each individual station data and results showed that all slopes and intercepts were also within the 95% confidence interval. 3.2. Mapping diffuse illuminance The result is displayed as maps showing geographical distribution of monthly average hourly diffuse illuminance as depicted in Figs. 7 and 8. The maps show diurnal and seasonal variations of monthly average hourly diffuse illuminance over Thailand.
S. Janjai et al. / Solar Energy 102 (2014) 162–172
169
January
February
March
April
May
June
8:30
9:30
10:30
11:30
12:30
13:30
14:30
15:30
klux 10
15
20
25
30
35
40
45
50
55
60
65
70
Fig. 7. The maps of monthly average hourly diffuse illuminance for January to June.
16:30
170
S. Janjai et al. / Solar Energy 102 (2014) 162–172
July
August
September
October
November
December
8:30
9:30
10:30
11:30
12:30
13:30
14:30
15:30
klux 10
15
20
25
30
35
40
45
50
55
60
65
70
Fig. 8. The maps of monthly average hourly diffuse illuminance for July to December.
16:30
S. Janjai et al. / Solar Energy 102 (2014) 162–172
Apart from the effect of the variation of the position of the sun, the diurnal variation is likely influenced by the variation of cloud amount and type. However, the satellite is not able to give the detail effect of cloud on this variation. The seasonal variation is clearly seen in the maps of 9:30– 14:30. It is noticed that from January to February, the area with high diffuse illuminance expands from the Southern to the Northern regions of this country. This is because the sun path progressively moves northward to the celestial equator during this period, causing the increase of global illuminance. As diffuse illuminance is part of global illuminance, the amount of diffuse illuminance depends strongly on the amount of global illuminance and the diffuse fraction. The diffuse fraction in the Southern region is higher than that of Northern region, being likely due to scattered clouds in this period. In March and April, high diffuse illuminance of over 40 klux features in most parts of the country. This is because the sun is nearly at the overhead position around noon time over the country, causing high global illuminance. In addition the diffuse fraction is also high, due to high aerosol load caused by biomass burning (Janjai et al., 2012). In general, the onset of the southwest monsoon is in mid-May and lasts until mid-October. This monsoon causes cloudy skies and rain for most parts of the country. Although, global illuminance decreases in this period, the diffuse fraction is relatively high due to clouds, keeping the high level of diffuse illuminance over country. Normally, the monsoons change from a southwest to a northeast monsoons in mid-October, and the northeast monsoons brings dry and cool air to the Northern, Northeastern and Central regions. However, while blowing across the Gulf of Thailand, it brings moisture to the Southern region causing cloudy skies and subsequently high diffuse fraction. Although, the Northern, Northeastern and Central regions have a majority of clear days caused by the northeast monsoon, the level of global illuminance in these regions is not high in November and December because the sun path moves southward from the celestial equator in this period. High global illuminance in the southern region is caused by the favorable position of the sun path. Due to high global illuminance and diffuse fraction, diffuse illuminance in this region is constantly high during November–December. 4. Conclusion A method for mapping monthly average hourly diffuse illuminance from satellite data has been developed. In developing the method, a satellite-based global illuminance model is improved. The improved model provides a more complete description of the absorption and scattering of solar radiation in the earth-atmospheric system. The model is used to derive monthly average hourly global illuminance from MTSAT-1R satellite data. Monthly average hourly diffuse illuminance is extracted from the satellite derived-global illuminance by using a diffuse fraction model developed in this work. Monthly average hourly
171
diffuse illuminance obtained from this method agrees well with that obtained from measurements, with RMSD and MBD of 9.7% and -1.4%, respectively. The model is used to generate monthly average hourly diffuse illuminance maps of Thailand. The maps show the diurnal and seasonal variations of monthly average hourly diffuse illuminance with the effect of the northeast and southwest monsoons. Acknowledgements The authors would like to thank the Thai Meteorological Department for supporting the measurements of illuminance and related atmospheric parameters. The authors are also grateful to Dr. Manuel Nunez for his assistance to improve the manuscript. References Boland, J., Ridley, B., Brown, B., 2008. Models of diffuse solar radiation. Renew. Energy 33, 384–575. Cano, D., Monget, J.M., Albuisson, M., Gurnard, H., Regas, N., Wald, L.A., 1986. A method for the determination of the global solar radiation from meteorological satellite data. Sol. Energy 37, 31–39. Chirarattananon, S., 2005. Building for Energy Efficiency. Asian Institute of Technology, Bankgok, Thailand. Exell, R.H.B., 1984. Mapping solar radiation by meteorological satellite. Renew. Energy Rev. J. 6 (1), 27–39. Fontoynont, M., Dumortier, D., Heinemann, D., Hammer, A., Olseth, J., Skatveit, A., Ineichen, P., Reise, C., Page, J., Roche, L., Beyer, H.G., Wald, L., Santos, A., 1997. SATELLIGHT: A European program dedicated to serving daylight data computed for Meteosat images. In: Proceedings of the eighth European Lighting Conference, 11–14 May, Amsterdam. Gautier, C., Diak, G., Masse, S., 1980. A simple physical model to estimate incident solar radiation at the surface from GOES satellite data. J. Appl. Meteorol. 36, 1005–1012. He, J.Z., Ng, E., 2010. Using satellite-based methods to predict daylight illuminance for subtropical Hong Kong. Lighting Res. Technol. 42, 135–147. Ihm, P., Nemri, A., Krarti, M., 2009. Estimation of lighting energy savings from daylighting. Build. Environ. 44, 509–514. Ineichen, P., Perez, R., 1999. Derivation of cloud index from geostationary satellites and application to the production of solar irradiance and daylight illuminance data. Theoret. Appl. Climatol. 64, 119–130. Iqbal, M., 1983. An Introduction to Solar Radiation. Academic Press, New York. Janjai, S., 2013. A satellite-based sky luminance model for the tropics. Int. J. Photoenergy 2013, 1–11. Janjai, S., Wanvong, W., Laksanaboonsong, J., 2006. The determination of surface albedo of Thailand using satellite data. In: Proceedings of the 2nd Joint International Conference on Sustainable Energy and Environment (SEE2006) Bangkok, Thailand, pp. 156–161. Janjai, S., Tohsing, K., Nunez, M., Laksanaboonsong, J., 2008. A technique for mapping global illuminance from satellite data. Sol. Energy 82, 543–555. Janjai, S., Laksanaboonsong, J., Pankaew, P., 2009. A model for calculating hourly global solar radiation from satellite data in the tropics. Appl. Energy 86, 1450–1457. Janjai, S., Nunez, M., Masiri, I., Wattan, R., Buntoung, S., Jantarach, T., Promsen, W., 2012. Aerosol optical properties at four sites in Thailand. Atmos. Climate Sci. 2, 441–453. Leckner, B., 1978. The spectral distribution of solar radiation at the earth’s surface-Elements of a model. Sol. Energy 20, 143–150. Li, D.H.W., Lam, J.C., 2001. Evaluation of lighting performance in office buildings with daylighting controls. Energy Build. 33, 793–803.
172
S. Janjai et al. / Solar Energy 102 (2014) 162–172
Orgill, J.F., Hollands, K.G.T., 1977. Correlation equation for hourly diffuse radiation on horizontal surface. Sol. Energy 19, 357–362. Perez, R., Ineichen, P., Moore, K., Kmiecik, M., Chain, C., George, R., Vignola, F., 2002. A new operational model for satellite-derived irradiances: description and validation. Sol. Energy 73, 307–317. Polo, J., Zarzalejo, L.F., Cony, M., Navarro, A.A., Marchante, R., Martı´n, L., Romero, M., 2011. Solar radiation estimations over India using Meteosat satellite images. Sol. Energy 85, 2395–2406. Schillings, C., Mannstein, H., Meyer, R., 2004. Operational method for deriving high resolution direct normal irradiance from satellite data. Sol. Energy 76, 475–484. Singh, M.C., Garg, S.N., 2010. Illuminance estimation and daylighting energy savings for Indian regions. Renew. Energy 35, 703–711.
Tanre´, D., Deroo, C., Duhaut, P., Herman, N., Morcrette, J.J., Perbos, J., Deschamps, P.Y., 1986. Simulation of the Satellite Signal in the Solar Spectrum. Technical Report, Laboratoire d’Optique Atmospherique, Universite´ des Science et Techniques de Lille, 59655 Villeneuve d’ Ascq Cedet dex, France. Tarpley, J.D., 1979. Estimating incident solar radiation at the surface from geostationary satellite data. J. Appl. Meteorol. 18, 1172–1181. Vignola, F., Harlan, P., Perez, R., Kmiecik, M., 2007. Analysis of satellite derived beam and global solar radiation data. Sol. Energy 81, 768–772. Zain-Ahmed, A., Sopian, K., Othman, M.Y.H., Sayigh, A.A.M., Surendran, P.N., 2002. Daylighting as a passive solar design strategy in tropical buildings: a case study of Malaysia. Energy Convers. Manage. 43, 1725–1736.