6 Applying Solar Resource Data to Solar Energy Projects
6.2 Data Applications for Feasibility, Engineering, and Financial Assessments
6.2.3 Examples of Mean Irradiance Estimation and Hourly Data Selection Using
Example 1 is a proposed trough plant near Harper Lake, California. Harper Lake is actually a dry lake bed with very bright salt deposits on the surface. Our goals for exploring this example are to:
• Determine the best estimate for mean DNI by month and year for a chosen site • Procure one or more years of time series DNI (and weather) data for use in time-
dependent modeling (CST plant models or electrical grid models).
To quickly assess the annual and monthly mean DNI, we use the Solar Prospector43 with satellite ground surface imagery (from Google Maps) as background. Figure 6-8 shows nine
NSRDB/SUNY grid cells in the area near Harper Lake in the Mojave Desert. The values of average DNI can be obtained from Solar Prospector. The upper value is the mean DNI from the hourly data, which is not corrected for specular reflection. The lower is from the map, which has been corrected for this artifact.
Next, we look at the mean DNI values, by month, for the 8 years of data from the 1998–2005 NSRDB. We do this for the desired location and a few nearby locations. If the map value and the hourly averaged values are different by more than 0.2 (kWh/m²/day), the grid-cell map is
corrected. In this example, Cells B1 and B2 were corrected. If the candidate site for a solar power plant is located in cell B2, the analyst could select hourly data from another cell that has
not been corrected, such as A2 or C2. This procedure will ensure that the hourly simulations (e.g., Solar Advisor Model for CSP)44 produce results that are more consistent with the mean value at the proposed site.
Figure 6-8. NSRDB/SUNY 10-km grid cells near Harper Lake, California. The upper values shown in text boxes are averaged from (uncorrected) hourly files. The lower values are averaged DNI from
corrected maps. The values in red show uncorrected time series mean values, which are substantially lower than the corrected map values. Image from NREL
In general, cells in need of correction have bright or uneven areas, especially near the center. Adjacent cells with a darker, more uniform background will have more reliable hourly DNI data. The goal is to select the correct time series to match the estimate of the mean values. The Google Map shows that the time-series data from the selected cell (B2) should not be used, because the time series produces different (lower) means. The SUNY team developed corrections to this artifact; in the near future, corrected maps will be available that avoid this problem.
Although it is not recommended, the user could choose one of the TMY2 or TMY3 data sets to act as a surrogate for the 8 years of data. If a TMY2 or TMY3 data set is proposed as a surrogate for this site, the data set should be carefully evaluated for applicability of the mean values in space and time. Figure 6-9 shows the monthly DNI values for the C2 site and the nearby
Daggett, California, TMY3, which is a higher quality, Class I, NSRDB site. In this case, the TMY3 may be a suitable surrogate for the site-specific SUNY data.
Figure 6-9. Monthly mean DNI for Harper Lake (Cell C2) and Daggett TMY3. Minimum and maximum values for cell C2 are also shown for each month. Image from NREL
Example 2 is a proposed plant site near Desert Rock, Nevada. We assume for this example that we have chosen the NSRDB/SUNY data for preliminary analysis, and we have obtained new measured data for the desired site. We show the effects on the annual DNI and GHI estimate of including measured and modeled data. Table 6-3 shows the results of using 8 years of modeled NSRDB data with 2 years of measured data (2004–2005), year 2004 of measured data only, and year 2005 of measured data only.
Table 6-3. Annual Mean Values of Global and Direct Radiation for Measured and Modeled Data at Desert Rock, Nevada
Measured Time Period (kWh/m
2/day) 2004–2005 2004 Only 2005 Only 1998–2005 Model global 5.615 5.656 5.574 5.622 Model direct 7.642 7.720 7.564 7.658 Measured global 5.703 5.799 5.607 Measured direct 7.564 7.901 7.227 MBE global –1.54% -2.46% -0.58% MBE direct 1.04% -2.28% 4.67% Adjusted direct 8-year mean 7.579 7.833 7.300 Meyer-corrected mean DNI 7.582 7.859 7.305 Meyer MBE direct 0.8% -1.8% 3.6% Meyer-adjusted 8-year mean 7.597 7.793 7.386
We adjust the average DNI from the 8-year period using the bias error from our observed data with the simple “ratio method” described above. The bias error using both years is a relatively low value of 1.04%. The bias errors from individual years are higher and do not show a
consistent pattern. The adjusted direct is the new estimate of the long-term mean DNI, and it is simply the 8-year mean DNI (7.658) times (1.0 - MBE). The method of Meyer et al. (2008) described previously can also be used advantageously here. If we assume the uncertainty is 3% for measured data and 10% for SUNY data, we can calculate the corrected means for all the months we have both measured and modeled data. If we adopt this value as our best guess for the actual DNI for the years 2004 and 2005, then our new bias error is (SUNY-Meyer)/Meyer, and our bias errors are smaller. The Meyer estimate is calculated using the following equation:
Iest = (Ime/Ume = Imo/Umo)/(1/Ume + 1/Umo) (6-1) where
• the Meyer estimate = Iest • Ime = measured value • Imo = modeled value
• Ume = measurement uncertainty (0.03) • Umo = modeled uncertainty (0.10).
Monthly mean values of GHI and DNI are shown for the Desert Rock site (see Figure 6-10). For many months, especially during 2005, the bias errors are very small for GHI and large for DNI. GHI and DNI bias errors are well correlated in 2004, but not in 2005. One interpretation is that the principal source of error during 2004 is the cloud estimation, and the principal source of error in 2005 is in the AOD.
A small error in global radiation along with a large overestimate of the DNI indicates that AOD at the site may have been much higher than the estimated AOD used in the satellite model. A diligent analyst might pursue an explanation for the higher than normal AOD and ask whether higher levels of AOD could be caused by dust storms, forest fires, or a general underestimation of the AOD. The average monthly values shown in Figure 6-10 would be helpful in pinpointing the cause of the problem.
Figure 6-10. Desert Rock annual average GHI and DNI from satellite and measurements. Mean bias error is defined as (satellite - measured)/measured by 100%. Image from NREL
The monthly values in Figure 6-10 show large shortfalls in the measured DNI in January and April 2005, indicating higher than normal AOD. After the likely cause has been determined, the analyst should then assess whether that phenomenon might be more prevalent in the future or if it is possibly a rare event.
The broadband AOD may be estimated from the new DNI measurements using a clear-sky model such as REST2 (Gueymard 2013), with supplemental data to estimate total column water vapor. These values can then be used to adjust the modeled DNI estimates; however, AOD is also highly variable from month to month and from year to year (and also on smaller timescales), so it would take several years of data to show conclusively that the mean AOD used in the satellite model needs to be adjusted at this site.
In this example (see Table 6-3), the new estimate for the 2-year data set DNI, 7.597 kWh/m2/day, is less than 1% different from the 8-year model estimate of 7.658. With only 1 year of
measurements, the errors are larger, up to 3.6%.