Proc. 29th IEEE PVSC, New Orleans, May 19-24, 2002 pp. 927-930. USING A NATURAL AM1.5G SPECTRUM TO HELP DEFINE AN AM1.5D SPECTRUM APPROPRIATE FOR CPV PURPOSES 1 1 1,2 1 David Faiman , Arnon Karnieli , Nurit Ninari , Vladimir Melnichak and Sareet Jacob 1 3 1,3 Department of Solar Energy and Environmental Physics 2 Wyler Department of Dryland Agriculture The Jacob Blaustein Institute for Desert Research Ben-Gurion University of the Negev Sede Boqer Campus, 84990 Israel Department of Chemistry, Brown University, Providence, RI 02912, USA ABSTRACT Spectral measurements were performed of direct beam and global solar irradiance at normal incidence. The measurements were taken at Sede Boqer, during noontime, on a selection of cloudless days during the period December 2001 – March 2002. For those days having a low level of diffuse radiation, the ratios of direct beam to global spectra were found to follow a previously suggested, semi-empirical, 2-parameter formula based on Rayleigh scattering. For one of the days, with an unusually high level of diffuse radiation, this formula did not provide an adequate fit to the data. In order to provide an adequate fit it was found necessary to effect an ad hoc modification to the algebraic form of the original formula. INTRODUCTION Because of their sensitivity to temperature, illumination intensity and light spectrum, PV cells are characterized under specified standard values of these external parameters. So-called Standard Test Conditions (STC) o specify: a cell temperature of 25 C, an illumination -2 intensity of 1000 W m , and a computer-generated spectrum referred to as AM1.5 [1]. The latter was devised so that it approximates the intensity of sunlight that would be received on a tilted plane surface, on a clear day, with the sun at a zenith angle of arcsec 1.5, and for a model atmosphere containing specified concentrations of water vapor, aerosol, etc. [2]. Recently, owing to increased interest in PV cells that are designed to operate under high concentration levels of solar energy (so-called CPV cells), it has been deemed desirable to distinguish between global AM1.5G and direct AM1.5D reference spectra. The motivation behind this need is the realization that if the global clear sky intensity -2 is, say, 1000 W m , the direct beam fraction would be -2 substantially smaller, typically 800 W m . Hence, the use -2 of a 1000 W m reference intensity for characterization purposes would tend to under-rate a CPV cell relative to a non-concentrator PV cell. Furthermore, given that it might be desirable to adopt a lower level of illumination as a test standard for CPV cells, it is natural to inquire what spectrum should be used for this purpose. It will be realized that this entire argument is mainly cosmetic because the performance of a cell, whether PV or CPV, under any set of standard operating conditions is by no means a universal index of its relative excellence. For example, if one cell has a higher efficiency than a second cell under one set of external conditions, it is entirely possible for the second cell to have a higher efficiency than the first under a different set of conditions. What one really needs to know, for comparison purposes, is how the current-voltage characteristics of the two cells depend on changes in temperature and light quality over the entire range of their operating conditions. Nevertheless, assuming it is agreed that a standard AM1.5D reference spectrum is truly needed, an immediate question arises as to how it should be generated. One way might be to adopt the same AM1.5 spectrum for both types of cell characterization. Another way, and one that has served as a basis for recommendation as an international standard [3], is to use a uniform set of assumptions concerning the atmospheric constituents in order to computer-generate a pair of standard AM1.5G and AM1.5D spectra. In the present paper we examine a third possible approach. It takes advantage of the known similarity [4] of natural clear-sky, noontime spectra at Sede Boqer (Lat. = o o 30.9 N, Lon. = 34.8 E, Alt. = 470 m) to the international standard AM1.5G spectrum [1]. METHODOLOGY Our method is to measure pairs of clear-sky, noontime, solar spectra, both normal global and direct beam, and compare their relative shapes. This study was started on the winter solstice day of the year 2001. At that time we discovered that the ratio of the beam/global spectra could be well approximated by a simple 2parameter formula based on a model in which Rayleigh Proc. 29th IEEE PVSC, New Orleans, May 19-24, 2002 pp. 927-930. 4 B/G = 1 / [ a + (b/λ) ] (1) where a and b are constants. The present paper extends our previous work by studying additional pairs of spectra measured on sample days during the first three months of 2002. We shall examine the extent to which our previously proposed 2parameter formula remains valid, and the month-to-month variation in parameter values observed for those situations in which the simple model holds. EXPERIMENTAL METHOD Solar spectra were measured at Sede Boqer at noontime on sample cloudless days. For this purpose a Li-Cor L-1800 spectroradiometer, having a wavelength range 300 - 1100 nm and a measurement interval of 2 nm, was employed. Global spectra were measured using the Teflon dome that is attached to the body of the instrument, with the latter tilted at normal incidence to the incoming beam radiation. Direct beam spectra were measured by adding a collimating pipe with a ratio of 10:1 for its length to internal diameter. For overall calibration purposes, Eppley PSP and NIP full-spectrum radiometers were employed, the calibration of which can be traced to the IPC-IX international comparison. It should be emphasized that although our spectroradiometer had recently been re-calibrated by its manufacturer, its absolute calibration is of little concern to us as our main interest is in the ratio of pairs of measurements made with this instrument. Because each spectral scan lasts approximately 20 s (when scanning the range 300 nm < λ < 1100 nm at 2 nm intervals), there is an intrinsic degree of uncertainty in the definition of the beam/global ratio. Our experimental method therefore consisted of making, as rapidly as possible, a series of alternate global and direct beam spectral scans, during a half-hour time period starting approximately 15 min before solar noon. In this manner, a given direct beam scan could be compared with the global scans that preceded and followed it, in order to test for atmospheric stability. Spectral scans were made on the dates: December 21, 2001; January 17, 2002; February 28, 2002 and March 24, 2002. RESULTS All spectral scans were made on basically cloudless days. Table 1 lists the average values of the normal global and direct beam radiation measured while the spectral scans were in progress. Also shown is the reading from a nearby shadow band pyranometer (uncorrected for shading ring losses), and some other relevant geometrical parameters of interest. Table 1: Solar irradiance and geometrical factors during the spectral scans discussed here. [Horiz. Diffuse = uncorrected shadow band reading, Air Mass = pressurecorrected value] Date Normal Global -2 [Wm ] 1083±9 1097±7 1095±9 1105±6 21.12.01 17.1.02 28.2.02 24.3.02 Direct Beam -2 [Wm ] 930±12 937±9 762±16 917±9 Horiz. Diffuse -2 [Wm ] 85±3 94±3 229±6 150±2 Zenith Angle [Deg.] 54.3 51.7 39.5 30.2 Air Mass 1.62 1.52 1.22 1.09 One observes from Table 1 that 28.2.02 was an unusually hazy day. For this reason its spectral scans will be discussed separately from those taken on the other dates. If, for the purpose of this discussion, we label each day’s scans, G1, B1, G2, B2, ….. Bn-1, Gn, then all pairs of ratios B1/G1, B1/G2, B2/G2, … Bn-1/Gn, were first graphed and examined visually. The graphs fell into two classes: Class I graphs increased “almost” monotonically with wavelength, whereas Class II graphs had varying amounts of waviness superimposed on a monotonically increasing shape similar to that of the Class I graphs. More careful examination of the Class II graphs revealed that, in addition to slight bumps and dips at fixed wavelengths (which correlate with known water vapor absorption bands in the atmosphere, and which can also be discerned in the Class I graphs), there were other, often much stronger, dips at arbitrary wavelengths. Fig.1 shows a typical Class II data set. 0.8 Ratio of spectral runs B3 and G4 Sede Boqer, 28.2.02 0.7 Beam/Global Ratio scattering is the dominant factor. If B and G represent simultaneously measured direct beam and normal global solar spectra, respectively, and λ is the wavelength, then the suggested relationship is [5]: 0.6 B3/G4 0.5 0.4 0.3 200 400 600 800 1000 1200 Wavelength [nm] Figure 1: A typical set of Class II data (No attempt was made to fit such sets) The waviness that characterizes Class II data sets was almost certainly brought about by intensity variations in irradiance (probably caused by invisible passing clouds) while a particular scan was in progress. For this reason all Class II data were discarded as being of no physical Proc. 29th IEEE PVSC, New Orleans, May 19-24, 2002 pp. 927-930. interest. The Class I data sets were then earmarked for more detailed study. In the case of data from the December, January and March scans, fits were made to eq.(1), using the method of least squares. In all cases, extremely good fits were 2 obtained (i.e. low values of χ ). Fig. 2 shows a typical example of a Class I data set and the corresponding fitted curve. Ratio of spectral runs B8 and G8 Sede Boqer,28.2.02 0.7 Beam/Global Ratio Results for the days with low haze levels 0.75 B8/G8 0.65 Fit(1.40,266) 0.6 0.55 Model: y = 1/[a + (b/lambda)^4] a = 1.403, b = 265.9 nm, chi sq. = 0.00013 0.5 0.45 200 400 600 800 1000 1200 Wavelength [nm] 1 Ratio of spectral runs B5 and G5 Sede Boqer,17.1.02 Figure 3: A typical data set for the day 28.2.02,with superimposed least-squares fit to eq.(1) B5/G5 0.8 Fit(1.09,321) For this reason, higher powers of the inverse wavelength term were explored. The fifth power was found to be an improvement but it was still too “round at the elbow”. However, with higher powers, good fits could be obtained. Specifically, by adopting a sixth-power expression of the form: 0.7 0.6 Model: y = 1/[a + (b/lambda)^4] a = 1.089, b = 320.7 nm, chi sq.= 0.00002 0.5 0.4 6 B/G = 1 / [ a + (c/λ) ] 0.3 200 400 600 800 1000 Wavelength [nm] Figure 2: A typical Class I data set with superimposed least-squares fit to eq.(1) Table 2 lists the resulting mean values of the parameters a and b, together with their standard deviations among all Class I data fits for each date in question. Table 2: Mean (and 1 σ) values of the parameters resulting from fitting eq.(1) to all Class I data sets for noontime spectral scans taken at Sede Boqer on clear days during the period December 2001 – March 2002. Date 21.12.01 17.01.02 24.03.02 Class I data sets 4 5 3 Parameter a 1.082 ± 0.004 1.093 ± 0.003 1.132 ± 0.004 (2) 1200 Parameter b [nm] 305.9 ± 1.1 321.5 ± 2.5 282.0 ± 1.2 Results for the day with high haze level where a and c are constants, adequate fits could be obtained. Fig.4 shows a typical sixth-power fit (using the same data set shown in Fig.3). 0.75 0.7 Beam/Global Ratio Beam/Global Ratio 0.9 Ratio of spectral runs B8 and G8 Sede Boqer, 28.2.02 B8/G8 0.65 Fit6(1.42,292) 0.6 0.55 Model: y = 1/[a + (c/lambda)^6] a = 1.419, c = 291.9 nm, chi sq. = 0.00002 0.5 0.45 0.4 200 400 600 800 1000 1200 Wavelength [nm] Figure 4: Same data as Fig.3 but with superimposed least-squares fit to eq.(2) DISCUSSION AND CONCLUSIONS For the Class I data sets taken on 28.2.02, it was 2 found that only poor fits (χ values were typically 5x higher than those for the clear-day fits) were possible to the algebraic form given in eq.(1). Fig.3 shows the best fit that was obtained from among all data sets for this day. 4 Qualitatively one observes that the (b/λ) term does not allow a steep enough initial rise with λ, which is a characteristic of all data sets measured on this day. If we compare the present results, obtained from a study of selected days during 4 consecutive months, with those from our published study [5] of a single day (21.12.01), a number of clarifications become apparent. First, for “clear” days, the originally suggested, semiempirical, formula, Eq(1), remains valid as a good description of the ratio of direct-to-global spectra. Furthermore, the month-to-month changes in the parameters a and b are relatively small, as was to be expected from our previous theoretical discussion [5]. Proc. 29th IEEE PVSC, New Orleans, May 19-24, 2002 pp. 927-930. However, because the changes are relatively small, and because, as we have seen here, the diffuse fraction can have a qualitatively large effect, our data sample is still too small to enable us to determine whether any underlying seasonal trend exists in the variation of these parameters. Because a seasonal trend was to be expected for parameter a [5], this study will continue. The interesting new result is the qualitatively different shape that results when high haze levels are present. A theoretical explanation for this observation is rendered difficult by the fact that what we measure as “direct beam” irradiance becomes less and less well-defined as the haze level increases. On clear days, the length-toaperture ratio of 10:1 that characterizes the construction of standard pyrheliometers represents approximately 10 solar diameters. In such situations the pyrheliometer records essentially all of the direct beam radiation and a negligibly small amount of spurious diffuse radiation. However, as the haze level increases, so too does the sun’s circumsolar halo. This results in an increasing ambiguity as to how much of the pyrheliometer reading is true direct beam radiation. For this reason, eq.(2) should be regarded as a purely empirical formula, with no theoretical basis at the present time. Its only value lies in the fact that it provides a simple 2-parameter description of the direct-beam spectrum (relative to the global spectrum) on very hazy days. Such a description may be useful for improving the understanding of CPV cells operating under high levels of solar concentration on such days. In any event, the possibility of characterizing outdoor spectral conditions by two easily-measurable parameters would lend further quantification to the testing of CPV cells and modules. ACKNOWLEDGMENTS This research was performed with partial funding from the Israel Ministry of National Infrastructures. One of the authors [DF] is indebted to Tom Stoffel for operating our cavity radiometer during the IPC-IX inter-comparison, 25.9-13.10.2000, Davos, Switzerland, to Daryl Myers for some penetrating insights upon reading an early version of this study, and to Indra Karki for assistance with data processing. REFERENCES [1] International Standard: Photovoltaic devices, Part 3; Measurement principles for terrestrial photovoltaic (PV) solar devices with reference spectral irradiance data, CEI/IEC 904-3, 1989. [2] R.E. Bird and R.L. Hulstrom, Review, evaluation, and improvement of direct irradiance models, ASME J. of Solar Energy Eng. 1981; 103: 182-192. [3] D. Myers, K. Emery and C. Gueymard. Proposed reference spectral irradiation standards to improve photovoltaic concentrating system design and performance evaluation, (Poster 3 P1.8 at this conference). [4] D. Berman and time-dependence of with and without environment, Solar 1997; 45: 401-412. D. Faiman, EVA browning and the I-V curve parameters on PV modules mirror-enhancement in a desert Energy Materials and Solar Cells [5] D. Faiman, A. Karnieli and S. Jacob, Concerning the relationship between clear-sky, global and direct beam, solar spectra, Progress in Photovoltaics (accepted for publication).