4.0 Contents of the Data Base
The data set described in this document is the best set of measurements possible with the personnel, equipment, methods, and calibration procedures available to us. We found that making continuous measurements with complex, sensitive equipment, and accounting for all sources of error in order to specify total measurement uncertainty were not trivial tasks.
The estimated total uncertainty (95% confidence interval) for each of the spectroradiometers and each of the measurement modes is shown in Figure 5-1. The upper and lower spectral uncertainty limits are symmetrical, except when global-tilt or global-horizontal spectra were measured using the integrating sphere, and when a strong, direct-beam component was present (Kt > 60%). We added an 8% maximum bias to the lower uncertainty limits for these cases based on results of experiments in which we mapped the integrating sphere response under bright sun with the image of the sun in different locations on the inside of the sphere. During these experiments, we observed variations of up to -8% compared to the calibration point.
The high uncertainty in the absolute value of the spectral measurements less than 0.45 Ám was caused by low spectral irradiance for the calibration lamps, which resulted in a low signal-to-noise ratio for the spectroradiometer calibrations (Figure 5-2). In addition, we experienced a change (i.e., loss of response) in the ultraviolet (UV) response of the instruments when using the integrating spheres.
Total uncertainty is a combination of an estimated total spectral bias error and total spectral random error. The final estimated total uncertainty for a 95% confidence interval is calculated as the root-sum-square of the bias (B) error, plus two times the random (R) error:
To evaluate instrument-to-instrument differences or variability (also called between-instrument precision), simultaneous spectral solar radiation scans were made with each instrument in each configuration. The standard deviation of these measurements, multiplied by three, was converted to a percentage of the mean. This percentage indicates the expected instrument-to-instrument variability, or precision, for this set of instruments. Figure 5-3 shows the results obtained for one of these comparisons using the integrating spheres. Similar results occur with the other configurations. The between-instrument precision above 0.4 Ám is close to 5%.
Temperature control of the spectroradiometers with silicon detectors was important. Figure 5-4 shows the change in the response of the instruments, especially in the near-infrared, when operated at different control tempera-tures. Uncertainty of 1° C in the set point results in about a 2% uncertainty in the near- infrared measurements. On hot days, we found that the temperature of the detectors (controllers) sometimes exceeded the set point, and we experienced runaway temperatures. This resulted in amplifier millivolt drift errors and noise in some of the spectra. We achieved better temperature control by shielding the spectroradiometers with Mylar sheets supported by foam offsets to allow air flow.
Measurement uncertainty for the broadband and meteorological data is based on the calibration histories provided by the measurement sites and characteristics of the instruments. This information is given in Tables 5-1, 5-2 and 5-3 the values are included on the data-base tape.
In these tables, the measurement uncertainty for a 95% confidence interval was estimated by root-sum-squaring the estimated bias errors and by doubling the random errors, in the calibration process and measurements [15,16]. Radiometers used at PG&E in 1987 were post-calibrated at SERI, allowing estimates of the random calibration error component; the radiometers used in 1988 were calibrated elsewhere so nominal estimated errors were used in the uncertainty analysis. Instruments used at SERI were pre-and post-calibrated at SERI. Instruments used at FSEC were calibrated by FSEC, and nominal values for random and bias errors for these types of radiometers were used in the uncertainty analysis. Calibration information for the meteorological instruments is included in these tables, but no measurement uncertainty analysis was performed.
We urge users of the data base to refer to the daily field notes for special circumstances recorded by the operator that may affect data uncertainty. These notes include references to the spectroradiometer temperature controller, solar-tracking accuracy, and weather conditions (such as rain). Included in the daily notes on the data- base tape are observations made in quality-control processing, as well as references to the existence and quality of photographs or slides.
Quantitative post-processing, quality-control information is also included on the data-base tape, and a complete description is given in the appendix of this report. Redundant measurements (such as before-scan and after-scan broadband solar radiation, direct-normal, and global-horizontal measurements with both thermopile and silicon detectors) and comparisons of integrated spectral solar radiation with broadband solar radiation were used in the quality- control procedures. Users are encouraged to perform their own quality-control processing on the data to meet their particular requirements.
Table of Contents