Appendix B: Solar Radiation Estimates Derived from Satellite Measurements
B.1 Introduction
Satellite-derived irradiance data augment ground-based measurements as satellites survey large areas and provide continuous information for long-term studies of the solar resource. Satellite data and images are used to generate time-and site-specific irradiance data and high-resolution (10 km x 10 km or smaller) maps of solar radiation. In fact, if nearby high-quality ground-based irradiance measurements are not available, the best characterization of the solar resource comes from satellite-derived irradiance data.
Satellite-derived solar resource surveys are used to characterize the long-term variability of the solar resource and climatologies areas for the best location for solar facilities, while ground-based monitoring stations are essential for accurately quantifying the solar irradiance at a specific site, measuring short-term variability of the solar resource, and providing ground truth for the satellite-derived values. However, when extrapolating the solar resource to nearby locations, satellite-derived hourly values become more accurate beyond 25 km from a ground station (Zelenka, Perez, Seals, and Renné, 1999).
Geostationary satellites are most suitable for modeling solar irradiance as they monitor the state of the atmosphere and the earth’s cloud cover with a spatial resolution near 1 km in the visible range and with a 30-minute time resolution. However, the view of the earth from these satellites limits their effectiveness at high latitudes (see Figure B.1). Polar satellites are closer to the earth’s surface and provide a variety of measurements that can be transformed into surface solar irradiance values, but they pass over a particular area only once during the day, which limits their temporal coverage at the lower latitudes.
In this appendix, the satellites used to generate the measurements for the irradiance models will be discussed, along with a brief description of the Perez, Ineichen, Moore, Kmiecik, Chain, George, and Vignola (2002) satellite irradiance model.
B.2 Geostationary Satellites
Geostationary satellites are located 35,880 km (22,300 miles) above the equator in a geosynchronous orbit. The United States has two satellites, GOES-West (135° W) and GOES-East (75° W), that cover North and South America. The European Union operates two satellites, Meteosat–9 (0°) and Meteosat–7 (57.5° E), that cover Europe, Africa, and the Middle East. The Japanese operate MTSAT (140° E), which covers Asia and Australia (Figure B.1). The curvature of the earth limits the useable images to between –66° and +66° latitude, so there is some redundancy in the satellite images and this allows some coverage even if one satellite has problems. Russia (GOMS), China (FY–2 series), and India (InSat and KALPANA) also have geosynchronous satellites that provide meteorological data and images.
FIGURE B.1 Geostationary weather satellites cover most of the globe from latitude 60°S to 60°N. This figure shows the coverage of five geostationary satellites. (Illustration courtesy of Angie Skovira of Luminale, LLC.)
Geosynchronous satellites have to be replaced periodically because they become outdated or run out of fuel for stabilization. For example, GOES-West and GOES-East are the twelfth and thirteenth satellites in the series. Historic meteorological satellite data from the United States, the European Union, and Japan are more complete and readily available than satellite data from China, India, and Russia.
B.3 Deriving Irradiance From Satellites
Many models exist to derive irradiance values from satellite data. An overview of satellite models will be presented in this section. For a more complete summary of satellite models see Chapter 4 of the NREL Best Practices Handbook (Stoffel et al., 2010). The empirical satellite models derive a cloud index (CI) from the satellite visible-channel measurements of surface radiance and use this index to modulate a clear-sky global irradiance model of the solar resource. Physical models use satellite radiance data and other atmospheric information to calculate the irradiance as it passes through the atmosphere by accounting for the radiative transfer processes. Empirical models estimate irradiance by using correlations to bypass some of the more cumbersome and time-consuming atmospheric modeling procedures. Physical models can take considerable computer time and require knowledge of the distribution of the gases, aerosols, and particles that make up the atmosphere in addition to understanding how each constituent affects the incoming irradiance. Physical models have been useful in understanding radiative transfer and were instrumental in uncovering the infrared (IR) radiative losses (thermal offset) that skewed the DHI measurements made using high-quality thermopile pyranometers (Cess, Nemesure, Dutton, DeLuisi, Potter, and Morcrette, 1993).
A good example of a physical model is one developed by Pinker and Ewing (1985) that divides the solar spectrum into 12 intervals and applies a radiative transfer model to a three-layer atmosphere. A primary input for this model is cloud optical depth. Pinker and Laszlo (1992) enhanced the model, and cloud information from the International Satellite Cloud Climatology Project (ISCCP) (Schiffer and Rossow, 1983) was used to develop irradiance data for the Surface Radiation Budget database that was created for a 2.5° × 2.5° grid (Whitlock et al., 1995). The clouds in the ISCCP climatology are separated into low, middle, and high clouds with three different optical thicknesses. Low and middle clouds are also categorized into water and ice clouds, whereas high clouds are always ice clouds. This creates 15 different cloud types for the ISCCP. The ISCCP climatology is used for cloud input for many models (Stoffel et al., 2010).
Empirical models take less computer time to run, are easier to apply, and don’t demand as detailed input information as required by the physical models. These empirical models are based on regression relationships between satellite observations and ground-based instrument measurements. The CI and regression relationships with other meteorological data are used to estimate the solar irradiance. Accurate clear-sky DNI modeling is important for all models because it is these clear-sky DNI values that are modulated by the cloud cover index. Good atmospheric turbidity values are necessary for accurate clear-sky DNI estimates.
B.3.1 GLOBAL IRRADIANCE (GHI)
Early work by Cano, Monget, Aubuisson, Guillard, Regas, and Wald (1986) is based upon the observation that shortwave (solar) atmospheric transmissivity is linearly related to the earth’s planetary albedo (Schmetz, 1989). This is a good first-order approximation. Without any clouds and with a fixed albedo, the irradiance incident on the earth’s surface is proportional to the intensity of the reflected irradiance as measured using the counts in the image pixel. The initial task in modeling is to assign the image pixel to a precise ground-based location. As the satellite wobbles in space, this limits the precision to which a pixel can be assigned. The original Perez procedure for extracting irradiance data is the basis for the following general discussion, and much of this material is extracted from Vignola and Perez (2004). A more detailed description can be found in Perez et al. (2002) As with many models, improvements are made over time, and a third-generation Perez model is being validated as of 2012.
There are two distinct steps for this process:
Pixel-to-cloud index (CI) conversion
CI to global irradiance conversion
B.3.2 PIXEL-TO-CLOUD INDEX CONVERSION
Individual pixel brightness values in the satellite image are stored as counts. The pixel counts are first normalized by the cosine of the solar zenith angle to account for the solar geometry. This normalized pixel is then gauged against the satellite pixel’s dynamic range at the chosen location to estimate a cloud index (CI). The lowest pixel count is assumed to be the ground brightness under clear sky conditions.
An additional normalization is necessary to adjust for a secondary atmospheric air mass effect and for the “hot spot” (Zelenka et al., 1999). The hot spot is cause by the sun–satellite angle and incorporates both atmospheric back-scatter brightness intensification and the fact that the ground surface becomes brighter as the sun–satellite angle diminishes due to the reduction of ground shadows seen by the satellite (e.g., Pinty and Verstraete, 1991).
The dynamic range represents the range of values a normalized pixel can assume at a given location from its lowest (darkest pixel, i.e., clearest conditions) to its highest value (brightest pixel, i.e., cloudiest conditions). A record of the normalized pixel counts at a given location is kept to track the changes in the dynamic range. While the upper bound of the range remains relatively constant except for the drift in the satellite’s calibration, the lower bound evolves over time as the local ground albedo changes (chiefly from changes in snow, moisture, and vegetation). Obtaining an accurate lower bound that represents the true ground albedo is difficult, especially during periods with long-term cloudiness. This can be especially important during the winter months when snow covers the ground and significantly reduces the dynamic range. Techniques using the IR satellite image are being considered to better distinguish between snow on the ground and clouds.
In the Perez procedure, a sliding time frame is used to determine this lowest bound. The lower bound is determined as the average of the 10 lowest pixels in the sliding time frame. Note that the secondary normalization is applied to the lower bound of the dynamic range. Subtracting the lower bound from the normalized pixel count and dividing this number by the normalized dynamic range defines the cloud index. The CI can then vary from 0 (clearest) to 1 (cloudiest).
B.3.3 CLOUD-INDEX-TO-GHI CONVERSION
In the Perez procedure GHI is determined by
where GHI is the clear-sky global irradiance per Kasten (1984). GHI. is adjustable for broadband turbidity as quantified by the Linke turbidity coefficient (Kasten, 1980) and ground elevation.
where Io is the extraterrestrial normal incident irradiance, sza is the solar zenith angle, am is the elevation-corrected air mass, TL is the Linke turbidity coefficient, and alt is the ground elevation in meters.
B.3.4 DIRECT IRRADIANCE (DNI)
DNI is obtained, as GHI, by modifying the clear-sky direct irradiance (DNI). An example of the clear-sky model (Ineichen and Perez, 2002) is
where DHIc is the minimum clear-sky diffuse irradiance given by
Unlike GHI, the direct modulating factor is not derived from CI but from global, using a GHI-to-DNI model. One quasi-physical model, DIRINT (Perez, Ineichen, Maxwell, Seals, and Zelenka, 1992), is based on the DISC model developed by Maxwell (1987) that uses a clear-sky model that is a function of the clearness index kt. In the DIRINT model, a modified clearness index is used along with atmospheric water vapor and the change in clearness index.
DNI is obtained from DNI = DNlc x DIRINT (GHI)/ DIRINT (GHIc), where DIRINT(GHI) is the DIRINT model using the input global irradiance GHI.
Other global–beam irradiance models have been used (Ineichen, 2008).
B.3.5 DIFFUSE IRRADIANCE (DHI)
The DHI is obtained by subtracting the DNI times the cosine of the solar zenith angle from the GHI. When the DNI value is very small, DHI is approximately equal to GHI and the uncertainty in the DHI value is the same as the percent uncertainty in the GHI value. When DNI is large, then the DHI value is small and the uncertainty percent in the DHI value is large compared with the uncertainty percent in the GHI value. When it is neither clear nor totally overcast, the uncertainty in the DHI value can be very large. Therefore, it is important to precisely state the conditions when specifying the uncertainty in the DHI value.
B.4 Status of Satellite Irradiance Models
Satellite irradiance models provide a good sampling of the location’s solar resource. The bias error is very small—a few percent for hourly GHI and random error of about 25% when compared with ground-based measurements. Of course, the different perspectives of ground-based and satellite views account for a good portion of this random error. In addition, the models have been adjusted to reproduce the statistical distributions in the solar resource while minimizing any bias errors.
As with any modeling effort, one has to be mindful of the accuracy of the data with which the model is derived and validated. The absolute accuracy of ground-based global irradiance is ±3%, at best, so a claim of 2% mean bias error has to take into account the measurement error.
TABLE B.1
Uncertainty in the NASA-Modeled Satellite Data for Monthly Averaged Values
Measurement | MBE (%) | RMS (%) |
|---|---|---|
GHI | –0. 0 | 10. 3 |
DHI | 7.5 | 29.3 |
DNI | –4. 1 | 22. 7 |
Note: The RMS errors are smaller for irradiance values obtained between ±60° latitude and larger for values for locations closer to the poles.
The uncertainty in the NASA-modeled 1-degree gridded data compared with high-quality BSRN data is given in Table B.1 (from Stoffel et al., 2010). While the mean bias error (MBE) appears small overall, it can vary several percent depending on the site examined. For example, the DNI MBE varies from –15.7% above 60 degrees north latitude to 2.4% below 60 degrees. It is difficult to compare ground-based BSRN-site-measured data with satellite-derived data on a 1-degree grid because of the differences in the size of the coverage.
According to Perez, Seals, Ineichen, Stewart, and Menicucci (1987), satellite-based GHI estimates are accurate to 10–12%. According to Renné, Perez, Zelenka, Whitlock, and Di Pasquale (1999) and Zelenka, Perez, Seaks, and Renné (1999), the target-specific comparison with ground-based observations will have a root mean square error (RMSE) of at least 20%; the time-specific area-wide accuracy is 10–12% on an hourly basis.
The RMSE decreases with averaging time, as expected. An hourly RMSE for one site will decrease from 20 to 25% for hourly estimates to 10 to 12% for daily estimates and down to 5 to 10% or less for monthly estimates. Mean bias errors generally range from +5% to –5%, with most studies reporting MBEs in the 2 to 3% range. Considering that the best GHI measurements have an absolute accuracy of 3%, it is necessary to be sure of what is being compared and cognizant of the uncertainties in the comparison measurements. The uncertainty in the GHI estimates increase if there is snow on the ground, the albedo in the image pixel varies considerably, or the terrain creates significant shadowed areas.
DNI and DHI have larger fractional RMSE and MBE than GHI estimates with the DNI, being up to twice the RMSE compared with the GHI estimate uncertainties and the DHI having a slightly larger RMSE than DNI.
With satellite images not centered on the hour, but at other times, the correspondence between ground-based data and data derived from satellite is not always easy to make. For example, when an image is taken at 9:15, it has to be shifted to 9:00 to be compatible with other meteorological data. The satellite-derived data in the National Solar Radiation Data Base is a weighted average of irradiance values that correspond to the hourly averaged meteorological data. This smoothing reduces some of the variability in the data values but gives an overall better representation when the irradiance data are used with other meteorological values for system performance calculations.
Satellite-derived values near sunrise and sunset have high uncertainties. This is caused by two factors: (1) the large incident angles, and (2) the fact that sometimes the satellite images are taken when the sun is below the horizon but there is some irradiance during the hour. As an illustration of the type of problem that can occur, if sunrise is 6:30 and the satellite image is taken at 6:15, there will be no irradiance recorded for the time period when there really is GHI between 6:30 and 7:00. Using an averaging method with the 7:15 value can give some value at 7:00. Of course the irradiance values are relatively small and large uncertainties do not significantly affect the usefulness of the data, but it is important to understand the limitations of the data values used.
B.5 Comments on Modeling and Measurement
As with any modeling effort, it is important to know the estimated uncertainty of the measurements used to develop and validate a model. Specifically, the systematic errors, which are part of the type B uncertainties, are different for an Eppley PSP pyranometer, a LI-COR LI-200 pyranometer, and a Kipp and Zonen CM 21 pyranometer. A model developed using one type of instrument will work best with that type of instrument if the systematic errors are not removed. Therefore, care should be taken when developing a model to use data where known systematic errors of the instruments have been removed. If they are not, then it is likely that the model will agree best with the measurements made with the instruments used to develop the model, but not necessarily the best measurements.
If an instrument is fully characterized, it is possible to remove most of the systematic errors in the measurements. Alternatively it is possible to use very accurate, and usually more expensive, instruments like absolute cavity radiometers to make the measurements. It should be possible to make significant refinement to models based on very accurate measurements. However, if one wants to develop models with data taken with lesser-quality instruments, the systematic errors associated with these less accurate instruments have to be accounted for or else they will skew the modeled result. Alternatively, if one uses measured data without correcting for systematic errors, the modeled results will be skewed. Therefore, it is important to know the instrument’s measurement performance characteristics. Important information such as temperature dependence, calibration uncertainty and frequency, and known systematic errors do affect the usefulness of the data and the accuracy of any results.
Deriving irradiance data from satellite images is dependent on the characteristics of the local albedo and also may be affected by local physical and environmental conditions. While models using satellite images are validated for a variety of sites, there is always the possibility that conditions at a given site are such that the satellite-derived data may be skewed. For example, the satellite pixel encompasses two types of surfaces, say, a salt bed and a lake. Any wobble in the pointing of the satellite will pick up more albedo from the salt bed or the lake. This wobble makes it difficult to determine the surface albedo and hence decreases the accuracy of the modeled data. Therefore, it is always useful to have some ground-based measurements to validate the satellite-derived data for the specific location under study.
Questions
What type of satellites are used to estimate solar radiation?
Briefly describe the process of obtaining global solar radiation data from satellite images.
Compare and contrast the strengths and weaknesses of solar radiation values obtained using satellite images and data from ground based stations.
References
ARM. 2002. Atmospheric radiation measurement program. Available at: http://www.arm.gov/
Broesamle, H., H. Mannstein, C. Schillings, and F. Trieb. 2001. Assessment of solar electricity potentials in North Africa based on satellite data and a geographic information system. Solar Energy 70:1–12.
BSRN. 2002. Baseline surface radiation network. Available at: http://bsrn.ethz.ch/
Cano, D., J. M. Monget, M. Aubuisson, H. Guillard, N. Regas, and L. Wald. 1986. A method for the determination of global solar radiation from meteorological satellite data. Solar Energy 37:31–39.
Cess, R. D., S. Nemesure, e.g. Dutton, J. J. DeLuisi, G. L. Potter, and J. Morcrette. 1993. The impact of clouds on shortwave radiation budget of the surface-atmosphere system: Interfacing measurements and models. Journal of Climate 6:308–316.
Ineichen, P. 2008. Comparison and validation of three global-to-beam irradiance models against ground measurements, Solar Energy 82: 501–512.
Ineichen, P. and R. Perez. 1999. Derivation of cloud index from geostationary satellites and application to the production of solar irradiance and daylight illuminance data. Theoretical and Applied Climatology 64:119–130.
Ineichen, P. and R. Perez. 2002. A new airmass independent formulation for the Linke turbidity coefficient. Solar Energy 73(3): 151–157.
Ineichen, P., R. Perez, M. Kmiecik, and D. Renne. 2000. Modeling direct irradiance from GOES visible channel using generalized cloud indices. In Proceedings of the 80th AMS Annual Meeting, Long Beach, CA.
Kasten, F. 1980. A simple parameterization of two pyrheliometric formulae for determining the Linke turbidity factor. Meteorologische Rundschau 33:124–127.
Kasten, F. 1984. Parametriesirung der Globalstahlung durch Bedeckungsgrad und Trubungsfaktor. Annalen der Meteorologie Neue Folge 20:49–50.
NOHRSC. 2002. National Operational Hydrologic Remote Sensing Center. Available at: http://www.nohrsc.nwsgov/
NSRDB. 1995. National Solar Radiation Data Base—final technical report, Volume 2, 1995. NREL/TP–463–5784.
Perez, R., P. Ineichen, E. Maxwell, R. Seals, and A. Zelenka. 1992. Dynamic global-to-direct irradiance conversion models. ASHRAE Transactions-Research Series, pp. 354–369.
Perez, R., P. Ineichen, K. Moore, M. Kmiecik, C. Chain, R. George, and F. Vignola. 2002. A new operational satellite-to-irradiance model. Solar Energy 73(5):307–317.
Perez, R., R. Seals, P. Ineichen, R. Stewart, D. Menicucci. 1987. A new simplified version of the Perez Diffuse Irradiance Model for tilted surfaces Description performance validation. Solar Energy 39:221–232.
Pinker, R. and J. Ewing. 1985. Modeling surface solar radiation: Model formulation and validation. Journal of Climate and Applied Meteorology 24: 389–401.
Pinker, R. and I. Laszlo. 1992. Modeling surface solar irradiance for satellite applications on a global scale. Journal of Applied Meteorology 31:194–211.
Pinty, B. and M. M. Verstraete. 1991. Extracting information on surface properties from bidirectional reflectance measurements. Journal of Geophysical Research 96:2865–2874.
Schmetz, J. 1989. Towards a surface radiation climatology: Retrieval of downward irradiances from satellites. Atmospheric Research 23:287–321.
Stoffel, T. , D. Renné, D. Myers, S. Wilcox, M. Sengupta, R. George, and C. Turchi. 2010. CONCENTRATING SOLAR POWER best practices handbook for the collection and use of solar resource data. NREL/TP–550–47465.
SWERA. 2002. Solar and wind resource assessment. Available at: http://www.uneptie.org/energy/act/re/fs/swera.pdf
Vignola, F. and R. Perez. 2004. Solar resource GIS data base for the Pacific Northwest using satellite data—final report. Project ID DE-FC26–00NT41011.
Vignola, F., P. Harlan, R. Perez, and M. Kmiecik. 2007. Analysis of satellite derived beam and global solar radiation data. Solar Energy 81:768–772.
Whitlock, C. H., T. P. Charlock, W. F. Staylor, R. T. Pinker, I. Laszlo, A. Ohmura, H. Gilgen, T. Konzelman, R. C. DiPasquale, C. D. Moats, S. R. LeCroy, and N. A. Ritchey. 1995. First global WCRP shortwave surface radiation budget dataset. Bulletin of the American Meteorological Society 76:905–922.
Zelenka, A., R. Perez, R. Seals, and D. Renné. 1999. Effective accuracy of satellite-derived irradiance. Theoretical and Applied Climatology 62:199–207.