1. Introduction
The intensity and distribution of solar radiation at the surface determines the growth of vegetation and affects surface temperatures. After entering the earth’s atmosphere, solar radiation is scattered and absorbed by clouds, aerosols, and atmospheric molecules before reaching the surface. Therefore, changes in surface incident solar radiation G depend on changes in clouds and atmospheric aerosol loading (Augustine et al. 2005; Wang et al. 2012c; Wild 2009), both of which have large uncertainties (Trenberth et al. 2007; Wang et al. 2009, 2012b).
There are two ways to measure G. The first is to directly measure it with (a) a single pyranometer. The construction of such instrument [i.e., Eppley Precision Spectral Pyranometer (PSP)] remains unchanged since the 1950s. The second is to measure diffuse irradiance using a shaded pyranometer with (b) a solar tracker and (c) direct normal irradiance using a pyrheliometer and then to calculate G by summing the direct and diffuse radiation measurements using Eq. (1). Pyranometers with a low thermal offset (such as Eppley 8-48 here; photo downloaded from http://www.srrb.noaa.gov/isis/index.html) have been deployed since the early 2000s to measure diffuse irradiance.
Citation: Journal of Climate 26, 12; 10.1175/JCLI-D-12-00482.1
Do these two methods provide similar estimates of the long-term trend of G? Do differences between earlier versus later measurement methods affect the reported trends? Because the responsivities of pyranometers may drift over time, the data can show a false dimming trend if the instruments have not been regularly and properly calibrated (Wild 2009). In 1979, the World Meteorological Organization (WMO) recommended that a world radiometric reference (WRR; Fröhlich 1977) be used to help ensure a worldwide homogeneity of G measurements (Gueymard and Myers 2008). Such regularly calibrated pyranometers would be expected to provide more accurate estimates of the long-term trends of G. All data used for this analysis are from radiometers that have been regularly calibrated during the analysis period by reference standards that are traceable to the WRR.
2. Data and processing method
As a backup to the direct and diffuse measurements and to provide an independent measure of G to assess whether a solar tracker is operating correctly, single pyranometers have also been deployed to directly measure G at state-of-the-art networks, such as the BSRN (Ohmura et al. 1998); the Atmospheric Radiation Measurement Program (ARM), which operates the southern Great Plains site (Ackerman and Stokes 2003); the Surface Radiation (SURFRAD) Network (Augustine et al. 2000); and the Integrated Surface Irradiance Study (ISIS) network (Fig. 2). The availability of data from well-calibrated instruments at these modern stations from the two independent measurement methods has made a comparison between them feasible. We collected 17 years of G data (1995–2011) measured by the two methods at nearly 50 stations (see Table 1, Fig. 2) to test whether these two measurement methods provide similar long-term trends.
Map of the sites where high-quality surface solar irradiance measurements were made (green dots, SURFRAD sites; blue stars, ISIS sites; purple square, the domain of the ARM southern Great Plains sites; red squares, BSRN sites excluding ARM, SURFRAD, and ISIS).
Citation: Journal of Climate 26, 12; 10.1175/JCLI-D-12-00482.1
Summary of basic information about the sites. Acronyms in the Instruments used columns for Eppley are the normal incidence pyroheliometer (NIP) and precision spectral pyranometer (PSP).
The National Oceanic and Atmospheric Administration (NOAA) operates the SURFRAD and ISIS networks. Data with 1-min (or 3-min) temporal resolution of surface incident solar irradiances are available (http://www.srrb.noaa.gov/surfrad/ and http://www.srrb.noaa.gov/isis/index.html). The ARM Program is supported by the U.S. Department of Energy’s Office of Science (information about ARM is available at http://www.arm.gov/ and 1-min data are available at http://www.archive.arm.gov). Data with 1- or 3-min are available from the BSRN archive (http://www.bsrn.awi.de/).
Each BSRN site has its own independent principle investigator and is sponsored by its host country (see Table 1). Both SURFRAD and ARM submit data to the BSRN archive. We found that most of the nighttime BSRN data were subjectively set to zero or within ±2 W m−2. Because global pyranometer data that have not been adjusted are necessary for this study (i.e., Fig. 6, described in greater detail below), we only analyzed original data that was downloaded from the SURFRAD and ARM websites.
Before 2001, diffuse solar radiation at the SURFRAD sites was measured by a shaded pyranometer with a solid black detector, and afterward, it was measured with a shaded black and white detector (Eppley model 8-48; see section 3 for detailed information). To homogenize the data, the diffuse measurements before 2001 have been corrected using a method similar to Dutton et al. (2001). We only used data from the ARM and ISIS sites after 2002 (see Table 1), when all daytime diffuse measurements were made with an Eppley 8-48.
The surface incident solar radiation data at 1- or 3-min temporal resolution were downloaded. For quality control (or to minimize biases caused by missing data, for example), a month (or a day or a half hour) of measurements were used only when data were available for more than half of the month (or day or half hour). Half-hourly, daily, and monthly means were calculated. The long-term trends that we report were derived from the monthly means. The impact of the averaging method on our results is expected to be very small as this study focuses on the differences between measurement methods, and data are processed using exactly the same method for all comparisons.
3. Results
We first compared monthly averages of the two independent measurements of G. At monthly time scales, the two measurements agree very well (Fig. 3), with an average standard deviation of 3.6 W m−2 (or 2.0%) and a mean difference of 0.4 W m−2 (or 0.2%) (Fig. 4). The correlation coefficient between the monthly averages of the two measurements of G is larger than 0.99 for each site. The uncertainties shown here are within the specifications of the radiometers; therefore, we believe that all radiometers involved have been well calibrated and maintained.
Scatterplots of the two measurements of G: (x axis) measurements of a single pyranometer and (y axis) diffuse radiation measured by a shaded pyranometer plus direct radiation measured by a pyrheliometer at six SURFRAD sites. Consult Table 1 for expansion of Site ID acronyms. Comparison of other networks is similar (see Fig. 4).
Citation: Journal of Climate 26, 12; 10.1175/JCLI-D-12-00482.1
Comparison of (a) standard deviation, (b) average difference, (c) relative standard deviation, and relative average difference among monthly averages of G by a single pyranometer and those calculated from diffuse radiation measured by a shaded pyranometer and direct radiation measured by a pyrheliometer. From 1995 to 2011, the agreement between these two measurements is very good at the monthly scale, with a correlation coefficient higher than 0.99 at each site. On average, the standard deviation of the difference between the two measurements is 3.6 W m−2 (2.0%) and the mean difference is 0.4 W m−2 (0.2%). The relative values were calculated by normalizing the values with their site-averaged G.
Citation: Journal of Climate 26, 12; 10.1175/JCLI-D-12-00482.1
However, we found that the monthly difference between the two independent measurements (G measured by a single pyranometer minus that calculated from the sum of the direct and diffuse components) increases linearly with monthly G (Fig. 5). This difference may be caused by factors associated with traditional outdoor pyranometer calibration methods and/or thermal offsets that affect single black detector pyranometers. In the traditional calibration method, a single responsivity value is set at a solar zenith angle (SZA) of about 45°, even though the responsivity varies with SZA. Use of the 45° responsivity is acceptable for the diffuse measurement or global solar measurements under overcast skies. However, for times when the sun illuminates the pyranometer sensor, it would be better to use a responsivity function, which would give more accurate results across the range of SZAs (Ji et al. 2011).
Scatterplots of monthly differences between the two measurements of G [G measured by a single pyranometer minus that calculated from the sum of the direct and diffuse components, as in Eq. (1)] at six SURFRAD sites. Consult Table 1 for expansion of Site ID acronyms. Comparison of other networks is similar.
Citation: Journal of Climate 26, 12; 10.1175/JCLI-D-12-00482.1
Figure 6 illustrates the thermal offset effect of the single pyranometer by comparing the nighttime values of an Eppley 8-48 (green points), which has little-to-no thermal offset, with those of the single pyranometer with a black detector (red points). In addition, it shows that a thermal offset of opposite sign is present during the day by showing differences between daytime measurements of G made from the single black detector pyranometer and those from the summation method for which the diffuse measurements have little or no thermal offset (blue points).
Time series of monthly thermal offset defined as the differences between the two measurement methods during the daytime (blue dots), nighttime thermal offset of a solid black thermopile pyranometer (red dots, measured by SpectroSun model SR-75), and nighttime thermal offset of a shaded black and white pyranometer (green dots, measured by Eppley 8-48) at SURFRAD sites. Diffuse irradiance has been measured using an Eppley 8-48 with a black and white thermopile detector since 2002, and the data from 1996 to 2001 were corrected for the thermal offset by the data provider.
Citation: Journal of Climate 26, 12; 10.1175/JCLI-D-12-00482.1
Figure 6 also shows that thermal offsets during the night are negative by up to several watts per square meter in monthly averages of G measured by a single solid black detector pyranometer. These biases are introduced when the detector responds to differences in internal surface temperatures (Bush et al. 2000; Ji and Tsay 2000) caused by the protective domes cooling to a cold sky and, subsequently, the detector cooling to the cold dome. A specially designed pyranometer for diffuse radiation measurement (such as a black and white thermopile pyranometer Eppley 8-48; see Table 1) exhibits a vastly reduced (nearly zero) thermal offset (see the green points in Fig. 6).
The positive differences between the two methods in the daytime data during periods of high sun (peaks in the blue curves of Fig. 6) are partly because the single pyranometer sensor’s surface is heated more during those periods by greater incident solar irradiance (Cess et al. 2000; Ji 2007). Note that the nighttime offsets (red points in Fig. 6) are greatest (more negative) during high sun periods because warmer nighttime temperatures maintain warmer sensors that cool more efficiently under the cold clear sky. Here, we use the difference between the two measurements to quantify the error of a single pyranometer because the diffuse plus direct component sum has a negligible thermal offset effect (Cess et al. 2000) (see Fig. 6).
This effect of the summation method is likely negligible because 1) the pyrheliometer, which is used for measuring direct solar radiation, is less prone to a thermal offset effect than a pyranometer; 2) the lower intensity of the diffuse radiation should cause a lesser enhancement of the thermal offset because the direct beam is not heating the sensor surface; and 3) the summation method should be insensitive to the cosine response of a single pyranometer. The use of a specially designed black and white thermopile pyranometer for the diffuse measurement could further mitigate the thermal offset of the component sum. However, the black and white pyranometer will not replace other types of pyranometers for directly measuring G because, when exposed to the direct solar beam, a black and white pyranometer exhibits a spurious azimuthal response because of its alternating pie-segmented design (Augustine et al. 2005).
Notice that the difference between the two measurements of G has a linear tendency with G (Fig. 5). This indicates that the magnitude of G is the determining factor of the error of a single pyranometer during the day. Previous studies generally used data collected during a short time period and found that the thermal offset effect during the day is greater under clear sky conditions (Cess et al. 2000; Dutton et al. 2001; Ji 2007). This is consistent with our results, as clouds are the determining factor of G and G is larger under clear sky conditions. Also, the cosine error introduced by applying the responsivity at 45° SZA to pyranometer and pyrheliometer measurements is more effective under clear skies than under cloudy conditions.
Could the thermal offset and cosine response errors introduce long-term spurious trends in the single pyranometer measurements of G? Figure 6 shows that the monthly differences of G as measured by the two methods have become more negative with time from 1995 to present, and Fig. 5 shows that the differences increase monotonically with absolute values of G at monthly time scales. Linear trends of G derived from single pyranometer measurements from 1995 to 2011 are 2–4 W m−2 decade−1 less than the trends of G derived from the summation method (Figs. 7a–c). This underestimation of the long-term trend from single pyranometer measurements occurred at most sites during the study period, no matter whether dimming or brightening had occurred (see Table 2). Previous studies using the summation method also reported a higher rate of brightening (Long et al. 2009; Wild et al. 2005) than studies based on a single pyranometer measurement (i.e., Table 2 of Wild 2009).
(a) Histogram of the trend in differences derived from the two measurements of G [G measured by a single pyranometer minus that calculated from a summation method using Eq. (1)] at all the stations from 1995 to 2011. (b) As in (a), but for the stations where trend in differences pass an α = 0.05 confidence test, (c) As in (a), but for the stations where both the trend in differences and the trend of G pass an α = 0.05 confidence test. The numbers in the box in each panel indicate averages of the trend in differences and their standard deviation.
Citation: Journal of Climate 26, 12; 10.1175/JCLI-D-12-00482.1
The averaged trend of G from the two measurement methods and their differences over the four networks (W m−2 decade−1). Data as in Fig. 7a. Figure 2 shows a map of the sites of the four networks.
We investigated the relationship between trend differences derived from the two measurement methods and trends of G. We first averaged the data in Fig. 7a over the four networks, and the results are shown in Table 2. Generally, long-term trends in G derived from single pyranometer measurements are less than trends of G derived from the component sum of direct plus diffuse. The magnitudes of the underestimation are larger at the ARM and ISIS networks than those at the BSRN and SURFRAD networks (Table 2). During the study periods, G has brightened at the BSRN and SURFRAD networks while G has dimmed at the ARM and ISIS networks. SURFRAD, ARM, and ISIS are all located in the United States, but the SURFRAD data used here extends back further in time than ARM and ISIS data. The strong brightening in SURFRAD data primarily occurred before 2000 (Long et al. 2009), and if only data after 2000 were considered, these three networks (SURFRAD, ARM, and ISIS) would provide similar trends of G.
We further plotted the differences in trends as derived from the two measurement methods against the average trend of G at stations where both trends pass an α = 0.05 confidence test (Fig. 8). A linear least squares fit to that data has a slope of 0.13. That slope is qualitatively consistent with the network averages shown in Table 2.
A scatterplot of the trend differences derived from the two-measurement method as a function of the averaged trend of G at stations where both trends pass an α = 0.05 confidence test. Each star represents a site. The data are the same as Fig. 7c.
Citation: Journal of Climate 26, 12; 10.1175/JCLI-D-12-00482.1
4. Conclusions and discussion
Since the late 1950s, G has been widely observed with a pyranometer that has a measurement uncertainty of 5%, which is uncomfortably large for climate change study. For a better estimate of G (i.e., with a uncertainty of 2%) (Gueymard and Myers 2008), WCRP BSRN suggested that its diffuse and direct components be measured separately with a shaded pyranometer and a pyrheliometer, respectively. A dimming trend can be found in the dataset before the early 1990s, when single pyranometers were used to measure G (Gilgen et al. 1998). For most of the 1990s and beyond, time series of G, based on the summation method [i.e., Eq. (1)] (Long et al. 2009; Wild et al. 2005), showed a higher rate of brightening relative to simultaneous data from a single pyranometer (Wild 2009).
We collected 17 years of G data (1995–2011) as measured by the two methods from nearly 50 stations to test whether these two measurement methods provide similar long-term trends. We found that, although the data from the two measurement methods agree very well on a monthly time scale, the long-term trend from 1995 to 2011 determined by a single pyranometer is generally ~2–4 W m−2 decade−1 less than that from the summation method, regardless of whether dimming or brightening has occurred. The differences in the trends of G derived from these two types of measurement methods vary with the rate of brightening, and they are partly caused by the thermal offset effect of the solid black thermopile pyranometers used to directly measure G and partly from application of a constant responsivity measured at 45° SZA.
The thermal offset effect of a solid black thermopile pyranometer is well known and is manifested by negative values at night when G should be zero. This study shows that the thermal offset of the solid black thermopile pyranometer is even stronger positive during daytime. This offset cannot be removed by traditional calibration methods (Gueymard and Myers 2008; Ji et al. 2011) and is included in most datasets of G measured by a pyranometer. Previous studies have shown that this thermal offset may contribute to the so called “cloud absorption anomalies” (Arking 1996; Cess et al. 1995; Li et al. 1995; Pilewskie and Valero 1995; Stephens and Tsay 1990); that is, cloud absorption calculated by atmospheric radiation transfer models was less than that derived from ground observations (Philipona 2002). This study shows that this thermal offset during daytime and the cosine-response-related error of applying a single responsivity valid at 45° SZA to all measurements may introduce a spurious trend in G that is measured by a single pyranometer.
The dependence of trends of G on measurement methods uncovered here has an important implication for the widely reported global dimming and brightening based on datasets collected by different measurement methods; that is, the dimming indicated by solid black thermopile pyranometer studies might have been less by 2–4 W m−2 if measured with current summation methods. Thermal offset effects and application of a constant responsivity measured at 45° SZA lessen the difference in positive trends and amplify the difference of negative trends. Besides these two factors, we have not uncovered any further significant sources of bias in the thermopile measurements. Thus, although perhaps plausible, we have not been able to conclusively establish that the estimations of G by a single pyranometer are an underestimate, because we cannot show that the trends with the newer instruments are not overestimates.
Acknowledgments
Kaicun Wang is funded by the National Basic Research Program of China (2012CB955302) and the National Natural Science Foundation of China (41175126). We thank Drs. Ellsworth G. Dutton, Charles N. Long, and Kevin E. Trenberth for their helpful comments. We thank the three anonymous reviewers for their insightful and thorough comments, which substantially improved our paper. This paper was edited by Robert Wood.
REFERENCES
Ackerman, T. P., and G. M. Stokes, 2003: The Atmospheric Radiation Measurement Program. Phys. Today, 56, 38–44.
Arking, A., 1996: Absorption of solar energy in the atmosphere: Discrepancy between model and observations. Science, 273, 779–782.
Augustine, J. A., J. J. DeLuisi, and C. N. Long, 2000: SURFRAD—A national surface radiation budget network for atmospheric research. Bull. Amer. Meteor. Soc., 81, 2341–2357.
Augustine, J. A., G. B. Hodges, C. R. Cornwall, J. J. Michalsky, and C. I. Medina, 2005: An update on SURFRAD—The GCOS Surface Radiation budget network for the continental United States. J. Atmos. Oceanic Technol., 22, 1460–1472.
Bush, B. C., F. P. J. Valero, A. S. Simpson, and L. Bignone, 2000: Characterization of thermal effects in pyranometers: A data correction algorithm for improved measurement of surface insolation. J. Atmos. Oceanic Technol., 17, 165–175.
Cess, R. D., and Coauthors, 1995: Absorption of solar radiation by clouds: Observations versus models. Science, 267, 496–499.
Cess, R. D., T. T. Qian, and M. G. Sun, 2000: Consistency tests applied to the measurement of total, direct, and diffuse shortwave radiation at the surface. J. Geophys. Res., 105 (D20), 24 881–24 887.
Dutton, E. G., J. J. Michalsky, T. Stoffel, B. W. Forgan, J. Hickey, D. W. Nelson, T. L. Alberta, and I. Reda, 2001: Measurement of broadband diffuse solar irradiance using current commercial instrumentation with a correction for thermal offset errors. J. Atmos. Oceanic Technol., 18, 297–314.
Fröhlich, C., 1977: World radiometric reference. Commission for instruments and methods of observation (CIMO) 7th session, WMO/CIMO Final Rep. WMO 490, 97–100.
Gilgen, H., M. Wild, and A. Ohmura, 1998: Means and trends of shortwave irradiance at the surface estimated from global energy balance archive data. J. Climate, 11, 2042–2061.
Gueymard, C. A., and D. R. Myers, 2008: Solar radiation measurement: Progress in radiometry for improved modeling solar radiation at the Earth’s surface. Modeling Solar Radiation at the Earth’s Surface: Recent Advances, V. Badescu, Ed., Springer, 1–27.
Ji, Q., 2007: A method to correct the thermal dome effect of pyranometers in selected historical solar irradiance measurements. J. Atmos. Oceanic Technol., 24, 529–536.
Ji, Q., and S. C. Tsay, 2000: On the dome effect of Eppley pyrgeometers and pyranometers. Geophys. Res. Lett., 27, 971–974.
Ji, Q., S. C. Tsay, K. M. Lau, R. A. Hansell, J. J. Butler, and J. W. Cooper, 2011: A novel nonintrusive method to resolve the thermal dome effect of pyranometers: Radiometric calibration and implications. J. Geophys. Res., 116, D24105, doi:10.1029/2011JD016466.
Li, Z., H. W. Barker, and L. Moreau, 1995: The variable effect of clouds on atmospheric absorption of solar radiation. Nature, 376, 486–490.
Long, C. N., E. G. Dutton, J. A. Augustine, W. Wiscombe, M. Wild, S. A. McFarlane, and C. J. Flynn, 2009: Significant decadal brightening of downwelling shortwave in the continental United States. J. Geophys. Res., 114, D00D06, doi:10.1029/2008JD011263.
Ohmura, A., and Coauthors, 1998: Baseline Surface Radiation Network (BSRN/WCRP): New precision radiometry for climate research. Bull. Amer. Meteor. Soc., 79, 2115–2136.
Philipona, R., 2002: Underestimation of solar global and diffuse radiation measured at Earth’s surface. J. Geophys. Res., 107, 4654, doi:10.1029/2002JD002396.
Pilewskie, P., and F. P. J. Valero, 1995: Direct observations of excess solar radiation by clouds. Science, 267, 1626–1629.
Stephens, G. L., and S.-C. Tsay, 1990: On the cloud absorption anomaly. Quart. J. Roy. Meteor. Soc., 116, 671–704.
Trenberth, K. E., and Coauthors, 2007: Observations: Surface and atmospheric climate change. Climate Change 2007: The Physical Science Basis, S. Solomon et al., Eds., Cambridge University Press, 235–336.
Wang, K., R. E. Dickinson, and S. Liang, 2009: Clear sky visibility has decreased over land globally from 1973 to 2007. Science, 323, 1468–1470.
Wang, K., J. Augustine, and R. E. Dickinson, 2012a: Critical assessment of surface incident solar radiation observations collected by SURFRAD, USCRN and AmeriFlux Networks from 1995 to 2011. J. Geophys. Res., 117, D23105, doi:10.1029/2012JD017945.
Wang, K., R. E. Dickinson, L. Su, and K. E. Trenberth, 2012b: Contrasting trends of mass and optical properties of aerosols over the Northern Hemisphere from 1992 to 2011. Atmos. Chem. Phys., 12, 9387–9398.
Wang, K., R. E. Dickinson, M. Wild, and S. Liang, 2012c: Atmospheric impacts on climatic variability of surface incident solar radiation. Atmos. Chem. Phys., 12, 9581–9592.
Wild, M., 2009: Global dimming and brightening: A review. J. Geophys. Res., 114, D00D16, doi:10.1029/2008JD011470.
Wild, M., and Coauthors, 2005: From dimming to brightening: Decadal changes in solar radiation at Earth’s surface. Science, 308, 847–850.
