An Overview of Satellite Retrieved Solar Radiation Values

An Overview of Satellite Retrieved Solar Radiation Values

The range of observation points in the NSRDB database is limited, and many potential solar power plants do not have good meteorological observation stations near them. Usually, the average solar energy resource in a certain area is relatively constant, but due to the influence of microclimate and the local weather system, the average solar energy resource will also vary greatly within a relatively short distance. For example, a solar power plant may be located fairly close to a weather observatory in the NSRDB, but separated by, or in a different direction from, a water or mountain, and therefore actually has a different solar resource.

It is impossible to build enough ground-based meteorological observatories or solar monitoring stations to cover every possible location with abundant solar energy. The main factor in the model for estimating solar irradiance is cloud cover. Meteorological satellite imagery is also capable of estimating cloud cover on a common scale compared to ground-based meteorological observatories. The solar irradiance values ​​in the NSRDB starting around 1998 were also inverted from the model using satellite images.

From 1998 to 2010, hourly solar radiation values ​​from satellite imagery were available on a 0.1° grid for the continental United States and territories through the International Climatic Data Center (NCDC) and NREL. Satellite irradiance values ​​extend the range of ground-based measurements due to the large scale of satellite surveys and the provision of continuous data for long-term solar resource studies. Satellite data and satellite imagery can be used to generate time- and location-specific irradiance data, as well as high-resolution (10km x 10km or smaller) maps of solar radiation. If nearby high-quality ground-based irradiance measurements are not available, then satellite irradiance data can accurately describe solar energy resources.

There is already evidence that satellite solar radiation data provide better estimates of hourly solar resources than inferred data from high-quality ground-based observatories if the target observation site is more than 25 km from the measured value (Zelenka et al., 1999). In addition, satellite solar resource survey results can be used to characterize the year-to-year changes in solar resources and to investigate the location of the best locations for solar installations; ground-based monitoring stations can accurately quantify solar irradiance at specific locations, measure short-term solar resources The role of variability, providing ground truth against satellite data values, is also critical.

  1. Irradiance from satellite imagery

Many models are based on irradiance values ​​from satellite data. Both models use satellite imagery to estimate surface irradiance. The physical model uses satellite imagery and other atmospheric data to calculate the irradiance of the sun through the atmosphere and accounts for the radiation-transport process. The empirical satellite model obtains the cloud extinction coefficient (Cl for short) from measurements of reflected light in the satellite visible light channel, and uses it to adjust the clear-sky level total irradiance model for solar resources. Although physical models take a lot of computational time, they are fairly accurate if the concentrations and spatial distributions of the gases, aerosols, and particles that make up the atmosphere are known, and the effect of each component on anthropogenic radiation is known of. The physical model established by Pinker and Ewing in 1985 is a good example (Pinker and Ewing, 1985). The model divides the solar spectrum into 12 spaced bands, and applies the radiation-transport model to the three-layer atmosphere, whose input is mainly the cloud optical thickness. This model was improved by Pinker and Laszlo in 1992 (Pinker and Laszlo, 1992), who developed it for the surface radiation budget dataset using cloud data from the International Satellite Cloud Climatology Programme (ISC-CP) (Schiffer and Rossow, 1983) irradiance data. The surface radiation budget dataset was created by Whitlock et al. 1995 on a 2.5° × 2.5° grid (Whilock et al., 1995). Clouds in ISCCP project climatology are classified into low, medium and high clouds with three different optical thicknesses. Low and medium clouds are also classified as water and ice clouds, while high clouds are always ice clouds, thus forming 15 different types of clouds. The climatology of the ISCCP project is used for cloud input to many models (Stoffel et al., 2010). This data was used to develop the NASASSE dataset.

Empirical models require less computation time to run, are easier to apply, and require less level of detail than physical models. Based on the regression relationship between satellite observations and ground-based instrument observations, the model estimates solar irradiance using the cloud extinction index and the regression relationship with other meteorological data. Since the cloud extinction index adjusts the clear sky value, establishing an accurate clear sky model is very important for all models. Good atmospheric turbidity values ​​are necessary for accurate clear sky estimates. Empirical models typically use average turbidity and optical thickness measurements, but tend to ignore changes in solar resources due to changes in aerosol type and concentration. This means that empirical values ​​of solar radiation based on long-term average aerosols have a limited role in determining trends in climate change.

  1. Geostationary satellites

Polar-orbiting satellites are closer to the surface and can provide a variety of measurements that can be converted to surface solar irradiance values. But since Earth-orbiting satellites only pass a specific area once a day, their temporal coverage is limited. Geostationary meteorological satellites have a spatial resolution of about 1 km in the visible range, and can monitor the atmospheric state and cloud amount with a temporal resolution of 30 minutes, so they are the most suitable for building solar irradiance models. The geostationary satellite is located in a geosynchronous orbit 35,880km (22,300mile) above the equator.

The curvature of the Earth limits the available imagery between -66° and +66° latitudes. The US GOES-West (135°W) and Goes-East (75°W) satellites cover North and South America. The EU’s Meteosat-9 (0°) and Meteosat-7 (57.5°E) satellites cover Europe, Africa and the Middle East. Japan’s MTSAT (140° east longitude) satellites cover Asia and Australia. Russia’s GOMS geostationary satellite, China’s FY-2 series of geostationary satellites and India’s InSat and KALPANA geostationary satellites also provide meteorological data and weather images. Therefore, even if one satellite fails, the data and images it should have observed will be included in the observations of other satellites, that is, the observations of different satellites will overlap.

  1. Satellite modeling irradiance model accuracy

Table 1 compares the uncertainty between the NASA modeled 1° gridded data and the high-quality Baseline Surface Radiation Observation Network (BSRN). It should be noted that it is difficult to compare satellite data on a 1° grid with ground-based BSRN site measurements due to large differences in observed areas. In general, however, the mean deviation (MBE for short) appears to be smaller. Depending on the observation location, the MBE can vary by several percentages. For example, DNI’s MBE has changed from -15.7% above 60°N to 2.4% below 60°N. Note: Both the RMSE and MBE of DVI and DHI are larger than the estimates of GHI, while the RMSE of DNI is twice that of GHI. The RMSE of DHI is a few percent higher than that of DNI.

Table 1 - Uncertainty of the monthly mean of NASA/SSE modeling satellite data
Table 1 – Uncertainty of the monthly mean of NASA/SSE modeling satellite data

The RMSE between satellite modeled inversions and ground-based measurements decreases with increasing averaging time. In terms of hour-by-hour comparisons, the RMSE of GHI was 20% to 25% compared to ground-based measurements. When the daily mean RMSE falls to 10% to 12%, the monthly mean RMSE is in the range of 5% to 10% or lower (Zelenka et al., 1999; Perez et al., July 1987; Renne et al., 1999). The latest Solar Anywhere dataset reduces the RMSE of hourly GHI data, daily GHI data, and monthly GHI data to 17%-22%, 8%-13%, and 4%-7%, respectively. Information from infrared satellite channels improves winter estimates (Hoff and Perez 2012). In general, the range of MBE is ±5%, while in most studies, the range of MBE is 2% to 3%. Table 2 compares the RMSE of the SUNY Albany satellite data and the NSRDB database METSTAT modeling and inversion data with ground-based measured data. Data were measured by Myers et al. 1989 (Myers et al., 1989), and ground-based data in Texas were excluded from comparison sites.

Table 2 - Comparison of measured satellite data and modeled data in NSRDB
Table 2 – Comparison of measured satellite data and modeled data in NSRDB
  1. NASA/SSE database

While the National Renewable Energy Laboratory is creating the NSRDB database for U.S. sites, NASA is developing the Surface Weather and Solar Energy (SSE) database for global sites. Instead of an empirical model, NASA opted for a physical model based on the work of Pinker and Ewing (Pinker and Ewing 1985). The original database was located on a 2.5 × 2.5° grid and ran between 1983 and 1993. Currently, with the improvement of the method, the size of the grid has been reduced to 1.0° × 1.0°, and the runtime of the database has also been extended from 1983 to 2005 (SSE database on March 1, 2012). NASA plans to further improve the method in the future, and shrink the grid.

Although a 1.0° grid is too large for observation point analysis, the observation points within the grid roughly follow how the solar resource varies. The Dagate and Phoenix observations and the 1.0° grids of these two observations were compared. Both NASA/SSE closely track the NSRDB data, except for years when the data obtained are replaced by data from other time periods (especially 1996-1997).

  1. Discussion on the accuracy and status of satellite data

Regarding the accurate information of atmospheric composition, the calculation results of total radiation, direct radiation and scattered radiation all have high accuracy under the condition of clear sky and no clouds. For example, the difference between calculated and measured scattered radiation is about 10 W/m², which results in thermal compensation that distorts DHI measurements from high-quality thermopile pyranometers (Cess et al., 1993). While this discrepancy is produced using radiative transfer models, there are many eye-air models that successfully calculate clear-sky irradiance without requiring detailed knowledge of the aerosol distribution within the radiative transfer model. Therefore, models that use satellite imagery to obtain irradiance data perform very well during clear-sky periods and provide sufficient aerosol input data, especially for GHI calculations. This is because many aerosols scatter light preferentially in the forward direction, and the DNI estimation errors associated with aerosol distribution biases in the calculations are compensated for by estimating the inverse effect on the DHI.

Another reason is that GHI=DNIxcos(SZA)+DHI, where SZA represents the solar zenith angle. Calculations during cloudy or partially cloudy periods are more complicated because the extent of the sky observed by satellites is larger than that seen by ground-based instruments. At the same time, the multi-layered nature of cloud cover and varying turbidity complicates the modelling, which means that in years with cloudy or partly cloudy periods, the model cannot be as accurate as the model for the clear sky period. Therefore, sunny sites and months have smaller RMSEs and possibly smaller MBEs.

As with any modeling study, the modeler must pay attention to the accuracy of the data required for modeling and validate it. Since most of the cited satellite data RMSE and MBE do not take into account the uncertainty of the measured data, the uncertainty of the orthogonality of the ground-based measurement data (RMSE or MBE) to the satellite data should be considered. Orthogonal means taking the square root of the sum of the squares of two numbers.
For example, the uncertainty of the RMSE of the satellite GHI data is 10%, while the uncertainty of the ground-based measured GHI is 5%, and the sum of the errors after quadrature is 11.2%.

Currently, satellite datasets are available on grids based on the native resolution of the images (about 1km in the US and 3km in the UK). Validating these models and demonstrating that spatial specificity does not increase with uncertainty in radiance values ​​(eg, ground cover reflectivity pointing accuracy issues or variability issues) requires more work.

Since satellite imagery is not focused on the hour, it is not easy to always match ground-based data with satellite data. For example, when an image is taken from a 09:15 o’clock position, the image data must be adjusted to match the weather data, which is usually an average of the previous hour’s data. The satellite data in the NSRDB database are transferred irradiance values, which are equivalent to hourly average meteorological data. This shift may include smoothness data (if interpolated between consecutive satellite frames to better match specific hourly conditions), reducing the variability of some calculated values. However, when calculating system performance using irradiance data and other meteorological data values, it is a better representation overall.

The satellite values ​​obtained near sunrise and sunset have high uncertainty due to the following two reasons: (1) the person’s shooting angle is large; (2) the sun is just below the horizon when the satellite image is obtained, and the sun still exists at this time. irradiated. For example, if the sunrise is at 6:30 am and the satellite imagery was acquired at 6:15, there is no record of solar irradiance at 7:00, while 6;30-7:00 is already available GHI was measured. Although the irradiance values ​​are quite small and the large uncertainties do not significantly affect the usefulness of the data, it is important to understand the limiting factors of the data values ​​used.

Read more: What factors will change energy forecasts?

Related Posts