IRRADIANCE VARIABILITY AND ITS DEPENDENCE ON AEROSOLS

Transcription

1 Proc. SolarPACES Conf., Granada, Spain, 211 IRRADIANCE VARIABILITY AND ITS DEPENDENCE ON AEROSOLS Christian A. Gueymard 1 1 Ph.D., President, Solar Consulting Services, P.O. Box 392, Colebrook, NH 3576, USA. Chris@SolarConsultingServices.com Abstract The aerosol optical depth (AOD) is known to be a critical input for radiation modeling purposes, and partially determines the accuracy of modeled direct normal irradiance (DNI). This contribution examines to what extent time variations in AOD also determine the observed variability in DNI, particularly at the daily and longer time scales. Two measures of variability are introduced: the Aerosol Variability Index (AVI) characterizes the magnitude of the variability in AOD over specific periods, from daily to yearly, whereas the Aerosol Sensitivity Index (ASI) relates the magnitude of relative variations in irradiance to absolute variations in AOD. AOD measurements at 8 Aeronet sites over the sun belt are used to obtain clear-sky irradiances with the REST2 radiative model, as well as determinations of ASI and AVI. Large geographic variations exist in AVI, whose largest values are found over western Sahara. The variation of ASI follows a different pattern because it decreases when AOD increases. The variability in global irradiance is much lower than that in DNI. On a long-term basis, the normal aerosol-induced variability in DNI is less than ±5% at most sites, but some areas might experience a much larger variability, comparable to that created by large volcanic eruptions. The latest such events predate most current modeled DNI datasets, making resource assessments potentially too optimistic for bankability if based on such limited data series alone. Keywords: Aerosols, DNI, irradiance, variability, Sahara, REST2. 1. Introduction Many CSP projects currently exist in the world, over an area commonly referred to as the sun belt, which has a good resource in direct normal irradiance (DNI). Recent studies [1, 2] have shown that the accuracy of existing modeled DNI datasets or maps was not always satisfactory. Moreover, it has been found [1] that a large part of these uncertainties could be explained by inaccurate aerosol data used to model DNI. More specifically, the aerosol optical depth (AOD) is known to be a critical input for radiation modeling purposes, and thus its accuracy partly conditions the quality of DNI predictions. In general, cloudiness is the main factor affecting the magnitude of DNI and its time or space variations. Under cloudless skies, however, AOD becomes the driving factor. Over the sun belt, where low cloudiness prevails most of the time, it is logical to assume that the variability in DNI is the result of variations in both cloudiness and AOD. Among the numerous questions this statement implies, this contribution examines more specifically to what extent time variations in AOD determine the observed variability in DNI at various time scales. Existing studies (e.g., [3]) have shown that the interannual variability in DNI is much larger than that in global horizontal irradiance (GHI). Although the latter is not what matters in CSP applications, most existing radiative models obtain DNI from GHI through the use of some simple empirical function. Moreover, GHI data are usually more prevalent, cover longer periods, and have less missing values. It is therefore desirable to examine the relationship between the variability in both DNI and GHI as a function of that in AOD. 2. DNI variability vs. bankability In their preliminary steps, world evaluations of CSP potential (e.g., [4]), as well as solar resource assessment studies and CSP plant designs, are usually based on average DNI information, such as monthly or annual values. The most common way of characterizing the solar potential of any given site is through its long-term mean DNI value. For bankable financial projections, however, detailed knowledge of the inherent seasonal and interannual variability in the resource is of critical importance. Whereas various design options are usually centered around average or typical solar resource data, bankable production scenarios must be based on a statistical analysis of long time series, spanning at least 1 years [5]. But even a 1-year period might not suffice because solar radiation is easily affected by natural climate cycles or infrequent events, which increase variability. A large interannual variability in DNI means a larger risk of insufficient revenue during a bad year, which threatens bankability from the standpoint of financial institutions.

2 One important aspect of the long-term variability in DNI is its dependence on volcanic aerosols. Large eruptions inject considerable amounts of aerosols in the stratosphere, whose effect on turbidity and DNI can be felt for a long time. This has been well demonstrated after the eruptions of El Chichon in 1982 and Pinatubo in 1991 [6-8]. No such large eruption has (fortunately) taken place during the last 2 years, but there is no guarantee that this fortunate situation will continue over the next 2 3 years, i.e., during the life of a solar power plant. Moreover, it must be borne in mind that none of the current commercial high-resolution DNI datasets includes these extreme volcanic years, since their starting dates are all in the mid- or late-199s. This means that the true long-term variability in DNI cannot be fully assessed with these sources of modeled data alone, which may be a non-negligible risk factor as far as bankability is concerned, unless additional historical data series are considered. This issue is demonstrated here with a case study, based on long time series of radiation measurements at Golden, Colorado. All irradiance components have been monitored there by the National Renewable Energy Laboratory since July 1981 (with a 16-month interruption in ). The annual-average daily values of DNI and GHI are calculated for each complete year, and compared to their respective long-term average (5.61 kwh/m 2 for DNI and 4.68 kwh/m 2 for GHI) over the period , which is typical of the period that satellite-derived datasets currently cover. The relative difference between each annual average and these long-term average values defines an annual anomaly in percent, which is shown in Fig. 1. It is obvious that this recent 13-year period has little interannual variability compared to the years before. The variability before 1998 was mostly caused by strong volcanic activity. In 1992, Pinatubo s eruption resulted in a significant loss of 26% in Golden s DNI resource. From a resource assessment standpoint, the period appears optimistic in both magnitude and variability, compared to a truly climatological period of 3 years, such as The variability in DNI also appears significantly larger than that in GHI. Similar results are obtained at other sites with long irradiance records, such as Sede Boker, Israel (Fig. 2). From measurements over the period , the daily-average DNI (6.67 kwh/m 2 ) and GHI (5.76 kwh/m 2 ) have been calculated, and are again used as reference to obtain annual anomalies as in Fig Golden, CO Anomaly (%) average Measured DNI GHI Fig. 1. Percent anomaly of measured annual DNI and GHI at Golden, CO compared to their average; arrows indicate major volcanic eruptions (El Chichon and Pinatubo). 3. Aerosol and irradiance data AOD happens to be highly variable in both space and time, as well as difficult to measure. Three main sources of AOD data currently exist: ground measurements using sunphotometers (from, e.g., NASA s Aeronet network), retrievals from satellite-based spectral radiance observations, and calculations based on chemical transport models. The latter two sources still lack the absolute accuracy of ground-based measurements. Efforts are underway (e.g., [9, 1]) to improve the quality of global AOD datasets, using both satellite-based and model-based gridded data, and constraining them by the ground-truth provided by the higher-quality sunphotometer data. To remove as many uncertainties as possible, only aerosol data from ground sunphotometers have been considered for this study. Similarly, to avoid the possibly large and time-variable uncertainties in modeled DNI data, which were documented in previous studies [1, 3], only measured irradiance data from high-quality stations are used here. However, extremely few coincident series of AOD, GHI and DNI data exist for long periods, thus currently limiting the spatial and temporal scope of this kind of study. (Aeronet has started its operations in 1993, making its data representative of the post-pinatubo era only, and only a few of its instruments are collocated with radiometric stations.)

3 1 Sede Boker, Israel Anomaly (%) average Measured DNI GHI DNI sensitivity to aerosols Fig. 2. Same as Fig. 1, but for Sede Boker, Israel. The sensitivity of DNI to AOD can be analyzed in various ways (see, e.g., [1, 11, 12]). For the present study, it is assumed that the broadband DNI, E bn, can be calculated from E bn = S E sc T a ΠT i (1) where S is the sun-earth distance correction factor, E sc is the solar constant, T a is the broadband aerosol transmittance, and ΠT i is the product of transmittances resulting from all other extinction processes. By analogy with the dependence on AOD of the spectral aerosol transmittance [13], and by taking advantage of the simplification offered by Ångström s Law, T a can be conveniently expressed as T a = exp(-mß λ e α ) (2) where m is the air mass, ß is the AOD at 1 µm (also known as the Ångström turbidity coefficient), λ e is the effective wavelength (µm), and α is the wavelength exponent. This basic equation has been used to construct various broadband radiative models [14-16]. From Eqs. (1) and (2), the relative dependence of DNI on AOD can be expressed as ΔE bn /E bn = γ Δß (3) α where the Aerosol Sensitivity Index (ASI), γ, represents the average value of the product mλ e over the period of interest. For λ e and α, typical values are.7.9 µm and.5 1.5, respectively. Assuming values of 1 2 for m, it is found that γ is typically in the range per unit ß, i.e., 1 35% for Δß =.1. The simplified derivation above assumes that neither λ e nor α is related to ß, which is not perfectly true. To obtain more precise sensitivity results, a more detailed derivation, involving higher-order terms, would have to be considered [12]. (See also Section 6 for a different analysis, aimed at daily results.) The important fact conveyed here is that an absolute variation (Δß) in AOD translates into a relative variation in DNI that is (approximately) proportional to it, and can be significant. 5. Aerosol variability at various time scales Rapid and intense fluctuations in DNI are typically caused by scattered thick clouds. Such dramatic effects do not normally occur with aerosols, thus their variability at time scales less than a day is not considered here. In contrast, aerosols do vary a lot at time scales from daily to interannual. In all what follows, only normal aerosol variations are considered, i.e., excluding major volcanic eruptions. Measured AOD data from 8 stations that are part of NASA s Aeronet world network are used here. All these stations are in the sun belt (Fig. 3), and have accumulated over 3 days of Level 2 AOD data, during as many clear or partly clear days. Similarly to the analysis in Section 2, the variability in AOD is defined as an anomaly, i.e., the difference between the mean daily (Section 5.1), mean monthly (Section 5.2), or mean annual (Section 5.3) AOD and the corresponding mean monthly or annual long-term AOD (or climatology ). In the present context, long term refers to the complete period of record. In all cases, the ß coefficient is derived from multiwavelength AOD data using the conventional linearized Ångström Law fitting approach [1].

4 Fig. 3. Aeronet sites (8) used in this study, arranged in three classes depending on the magnitude of their daily Aerosol Variability Index: low (blue triangles), moderate (brown diamonds), and high (red stars). The background resource map is the mean annual DNI according to NASA-SSE calculations. 5.1 Daily variability The range of calculated mean daily anomalies of the observed ß is found to vary widely. To help the quantitative analysis below, an annual-average Aerosol Variability Index (AVId) is defined as AVId =!!!"!!!!!!!!!!!!! /! (4) where!! is the AOD at 1 µm during day i of month M,!! is the long-term daily average for that month, and N is the number of days. Figure 4a shows an example of a site with very low daily aerosol variability: Boulder, Colorado, which is in the vicinity of Golden (Fig. 1), and thus with a similar aerosol regime. At the other extreme, Agoufou, Mali, experiences considerable variability, owing to intense Saharan activity in particular (Fig. 4b). For Boulder, AVId =.162, which is close to the dataset s absolute minimum of.124 observed here. At Agoufou, AVId =.244, which is close the absolute maximum of For easier classification, the observed AVId values are divided into three different ranges of variability: low ( AVId <.4; 28 stations), moderate (.4 AVId <.9; 25 stations), and large (AVId.9; 27 stations). These categories are shown in Fig. 3 with different color and shape codes to emphasize geographic differences. Interestingly, the stations with the highest AVId are all located in northwestern Africa, in a region where AOD is typically always large, and most affected by Saharan dust or smoke. These findings are not coincidental at all, since AVId appears well correlated (Fig. 5) with the long-term mean annual ß, <ß>, according to AVId =.48 <ß> (R =.963). (5) This shows that the larger the annual AOD, the more the aerosol load is caused by strong but irregular phenomena, which could be expected. 5.2 Monthly variability The same kind of analysis is repeated here to obtain the AVI for monthly-average values (AVIm). Equation (4) can still be used, but now defining!! as the mean value of ß during a specific month, rather than during a specific day. As could be expected, AVIm is lower than AVId, but roughly proportional and well correlated to it: AVIm =.444 AVId (R =.941). The geographic distribution of AVIm is similar to that of AVId in Fig. 2. (6)

5 Ångström's! Daily Anomaly (a) Boulder, Colorado Ångström's! Daily Anomaly Agoufou, Mali (b) Fig. 4. (a): Calculated daily anomaly of ß at Boulder, CO; (b): Same as (a), but for Agoufou, Mali..3 Data from 8 Aeronet sites Daily Aerosol Variability Index Annual variability Fig. 5. Relationship between AVI d and mean annual ß at 8 Aeronet sites. In theory, a similar analysis as above could be done to obtain the AVI for annual values, AVI a. This is difficult, however, because most AOD measurement records span a few years only, and very rarely more than 1. Furthermore, there are missing months during any year at most Aeronet stations for various reasons, such as offsite instrument recalibration, replacement, or failure. Because of seasonal variations in AOD, calculating the annual average with a few missing months (or even just one) can adversely bias it. An in-depth analysis is therefore not attempted here. However, based on a limited data subset of 1 Aeronet sites where at least 7 continuous and non-overlapping 12-month periods could be defined, it is found that AVI a is equal to.444 times AVI m. This means that the attenuation of AVI caused by aggregating days into months (as described by Eq. (6)) is conserved when aggregating months into years, and moreover that the annual variability is about 5 times less than the daily variability (with a lot of scatter, though). 6. Aerosol-induced DNI and GHI variability Mean Annual! To evaluate the aerosol-induced variability in DNI and GHI independently from that induced by cloudiness, it is necessary to associate clear-sky DNI and GHI predictions with known AOD conditions. This can be done with a broadband radiative model. Among all current models of this type, REST2 [15] is known for its unsurpassed performance [17, 18], and has thus been selected here. The main atmospheric inputs to the model are α, ß and precipitable water, which are provided here by the high-quality Level 2 daily-average data from Aeronet. Daily ozone data from satellite observations, estimated (constant) surface albedo, and default values for the aerosol single-scattering albedo are also used as inputs to REST2. Modeled values of daily clear-sky DNI and GHI are thus obtained. An example of results for Sede Boker over the period 2 29 is shown in Fig. 6. This site is characterized by significant aerosol variability (AVI d =.65). The relative anomalies for the daily DNI and GHI are calculated in a similar way as the daily AOD anomaly.

6 Daily DNI Anomaly (%) Predicted Clear-Sky DNI Sede Boker, Israel Ångström! 1! Anomaly Fig. 6. Observed daily ß anomaly at Sede Boker, and related daily anomaly in modeled clear-sky DNI. These calculations show that the daily anomalies in ß (up to 1.2) induce significant opposite anomalies in daily GHI (up to 28%) and even much larger anomalies in daily DNI (up to 87%). The same irradiance predictions can also be used to evaluate the daily value of ASI (defined in Section 4). It is obtained here as the ratio between the daily relative anomaly in DNI (or GHI) and the daily anomaly in ß. It is found that ASI varies on a monthly basis, and also depends on the daily-average value of ß, which is consistent with previous findings [12]. The seasonal variation of the mean monthly averages of the daily ASI for both DNI and GHI is shown in Fig. 7 for three sites with different aerosol regimes: Sede Boker, Agoufou and Solar Village, Saudi Arabia. Although AVI d at Agoufou is 3.7 times higher than at Sede Boker and 1.9 times higher than at Solar Village, the monthly ASI at Agoufou is less that that at the two other sites because their monthly-average ß is much lower, as shown in the lower part of Fig. 7. Aerosol Sensitivity Index (%) Monthly ASI GHI Agoufou Sede Boker Solar Village DNI Ångström's! Monthly Aerosol Optical Depth Month Agoufou Sede Boker Solar Village Fig. 7. Mean monthly Aerosol Sensitivity Index (top) and observed ß (bottom) at three Aeronet sites. It is obvious from the analysis in Section 5 that, due to annual compensations in natural aerosol variations, the mean annual variability in AOD is considerably reduced compared to its daily variability. But can this remaining annual variability be still of concern with respect to its effect on DNI? It is currently difficult to address this question thoroughly, considering the aforementioned dearth of long-term aerosol data and of collocated DNI and AOD measurements. Nevertheless, a preliminary analysis is offered here, using the subset of 1 Aeronet stations defined in Section 5.3. Despite the small sample size, a large range of AVI values is

7 represented here, from very low AVI d (.151 at Sevilleta, NM or.162 at Boulder, CO), to an extremely high AVI d (.244) at Banizoumbou, Niger. The effect of time averaging on AVI, and the corresponding year-specific anomalies in ß, are shown in Fig. 8 for all 1 stations. For instance, the maximum annual ß anomalies recorded at Boulder are found to be small (less than ±.1), which is consistent with Fig. 4a. Using the discussion in Section 4 and the data in Fig. 7, a typical interannual variability of about ±1 3% in clearsky DNI is estimated, which is consistent with the findings in Fig. 1 considering that a significant part of the observed all-sky interannual variability in DNI over the period is likely caused by variations in cloudiness. The same conclusion applies to Sede Boker, where the maximum annual ß anomalies are about double of that for Boulder, yielding possible interannual variations of about ±2 5% in clear-sky DNI. Larger aerosol-induced interannual changes in mean annual DNI can be expected at sites more directly impacted by dust storms, such as Solar Village and Banizoumbou. At such sites, the maximum annual ß anomaly is about ±.1. This means that there might be occasional years when DNI anomalies can reach an estimated ±1% to ±3%, hence comparable to the Pinatubo effect with the difference that these anomalies may be either positive or negative, rather than just negative in the case of volcanic eruptions. Interestingly, the absolute values of the maximum ß anomaly at all 1 sites are found roughly equal to their AVI m. If this relationship can be proven of general validity, Eq. (6) could be used to estimate the expected AVI m, and then the maximum interannual variability in ß, whenever AVI d is known. Since AVI d can be calculated with enough precision from only one or two years of AOD data at sunny sites, it is apparently possible to use short-term AOD measurements as a proxy for their interannual (long-term) effects on DNI and GHI. Alternatively, Eq. (5) could be used to derive AVI d at any site from a good AOD climatology..15 Ångström's!: Annual anomalies.1 Annual Anomaly Alta Floresta, Brazil Banizoumbou, Niger Boulder, Colorado El Arenosillo, Spain FORTH Crete, Greece Mongu, Zambia Sede Boker, Israel Sevilleta, New Mexico Skukuza, South Africa Solar Village, Saudi Arabia -.1 Mean Aerosol Variability Index Alta Floresta Banizoumbou Boulder El Arenosillo FORTH Crete Mongu Sede Boker Sevilleta Mean AVI Daily Monthly Annual Skukuza Solar Village Fig. 8. Annual ß anomaly, and corresponding AVI (daily, monthly and annual) at 1 Aeronet stations. 7. Conclusion Measured AOD data series at 8 Aeronet stations located in the sun belt have been used to study the variability in AOD due to normal variations in aerosol regime, i.e., excluding large volcanic eruptions. The AOD variability is characterized here by the Aerosol Variability Index (AVI), which is calculated from Ångström s ß coefficient at various time scales: daily, monthly and annual. The daily AVI is found directly proportional to the mean annual ß, which means that areas where DNI is strongly attenuated by aerosols are also those where variability is largest. For instance, the daily AVI is low over clean regions (e.g., Australia, southwest USA, or southern Africa), but conversely is very high over hazy regions such as northwestern Africa, with many intermediate situations. Daily AOD measurements are used as inputs to the REST2 radiative model to predict the corresponding clear-sky DNI and GHI. The sensitivity of irradiance to changes in AOD can then be evaluated via the Aerosol Sensitivity Index (ASI), which is a function of ß. DNI is much more sensitive to changes in AOD than GHI, as could be expected. The interannual variability in AOD is found generally low, resulting in low annual DNI variability, typically within ±5% or less from the long-term average. Exceptions Site

8 do occur over areas strongly impacted by desert dust or smoke, where the induced effect on the interannual variability in DNI is estimated to reach ±1 to ±3%, which is comparable to strong volcanic effects. To correctly reproduce the significant daily variations in DNI over such areas, irradiance predictions with radiative models should rely on daily AOD data rather than on the more usual monthly-average data. In all areas not affected by such high-aod conditions, it is concluded that the main cause for the occurrence of bad years over the last 3 years is, by far, volcanic activity, which may result in 2 3% drops in annual DNI. Such strong effects are not present in existing satellite-derived solar resource datasets, since they are generally post-pinatubo era only. This may have profound implications when assessing financial risks and bankability if the likelihood of future volcanic events is ignored. Preliminary results finally suggest that the interannual variability in AOD (and thus in DNI) can be estimated from the daily AVI, which can be evaluated either from only 1 2 years of AOD measurement or from a good climatological value of the annual ß. Acknowledgments The Aeronet staff and participants are thanked for their successful effort in establishing and maintaining the various sites whose data were advantageously used in this investigation. The author also wishes to thank Dr. David Faiman for sharing the Sede Boker radiation data References 1. Gueymard C.A., Uncertainties in modeled direct irradiance around the Sahara as affected by aerosols: Are current datasets of bankable quality? J. Sol. Energ-T. ASME: in press, Suri M., et al. Comparison of direct normal irradiation maps for Europe. Proc. SolarPACES Conf. Berlin, Germany, Gueymard C.A. and Wilcox S.M., Assessment of spatial and temporal variability in the US solar resource from radiometric measurements and predictions from models using ground-based or satellite data. Solar Energy 85: , Trieb F., et al. Global potential of Concentrating Solar Power. Proc. SolarPACES Conf. Berlin, Germany, Bender G., et al. The road to bankability: Improving assessments for more accurate financial planning. Proc. Solar 211 Conf. Raleigh, NC, American Solar Energy Soc., Hoecker W.H., et al., Variation of direct beam solar radiation in the United States due to the El Chichon debris cloud. B. Am. Meteorol. Soc. 66: 14-19, Michalsky J.J., et al., Degradation of solar concentrator performance in the aftermath of Mt Pinatubo. Solar Energy 52: , Molineaux B. and Ineichen P., Impact of Pinatubo aerosols on the seasonal trends of global, direct and diffuse irradiance in two northern mid-latitude sites. Solar Energy 58: 91-11, Gueymard C.A. and George R. Gridded aerosol data for improved direct normal irradiance modeling: The case of India. Proc. Solar 211 Conf. Raleigh, NC, American Solar Energy Soc., Kinne S., et al., An AeroCom initial assessment optical properties in aerosol component modules of global models. Atmos. Chem. Phys. 6: , Gueymard C.A., Direct solar transmittance and irradiance predictions with broadband models. Pt 2: Validation with high-quality measurements. Solar Energy 74: , Gueymard C.A., Importance of atmospheric turbidity and associated uncertainties in solar radiation and luminous efficacy modelling. Energy 3: , Gueymard C.A., Parameterized transmittance model for direct beam and circumsolar spectral irradiance. Solar Energy 71: , Gueymard C.A., A two-band model for the calculation of clear sky solar irradiance, illuminance, and photosynthetically active radiation at the Earth's surface. Solar Energy 43: , Gueymard C.A., REST2: High performance solar radiation model for cloudless-sky irradiance, illuminance and photosynthetically active radiation Validation with a benchmark dataset. Solar Energy 82: , Yang K., et al., A hybrid model for estimating global solar radiation. Solar Energy 7: 13-22, Gueymard C.A. and Myers D.R., Validation and ranking methodologies for solar radiation models, in Modeling Solar Radiation at the Earth Surface, V. Badescu, Editor, Springer, Gueymard C.A., Clear-sky irradiance predictions for solar resource mapping and large-scale applications: Improved validation methodology and detailed performance analysis of 18 broadband radiative models. Solar Energy: in press, 211.