Using the UV index forecasts to independently validate and intercompare surface ultraviolet radiation observations in the United States.

C.S. Long

• Introduction
• Overview
• Characterization Factors
• Overall Results
• Sample Comparison of Surface Observatons and UV Index Forecasts
• Multiple Comparisons at one Local
• One on One Comparisons
• Conclusions

FIGURES

• Figure 1
• Figure 2
• Figure 3
• Figure 4
• Figure 5
• Figure 6
• Figure 7
• Figure 8
• Figure 9

Introduction

The NOAA/National Weather Service (NWS) has been issuing a next day UV Index forecast for locations within the United States since the summer of 1994. The UV Index forecast incorporates satellite observations of the earth's ozone field, a radiative transfer model, and atmospheric parameters from a numerical weather forecast model. As with all other NWS forecast products, the UV Index must be validated. At nearly the same time, several U.S. governmental agencies have established surface ultraviolet radiation observation networks. The UV Index forecasts are validated against these observations.

In addition to providing validation information to the NWS, this process is unique in that it combines all of the different networks' observations and compares them against an independent source. In the past, this has lead to uncovering calibration and characterization problems with specific instruments and an entire brand of instrument. Subsequently, these calibration and characterization issues have been addressed and correction values are made available to the user community.

Overview

As the NWS strives to improve the accuracy of the UV Index, it remains an excellent independent source of comparison for surface UV observations. The UV Index is a dose rate of erythemally weighted UV irradiances. The U.S. National Weather Service (NWS) produces a daily forecast of the UV Index for the following day valid during the local solar noon hour for nearly 800 U.S. cities(Long, et al., 1996). Total ozone forecasts based upon observations from NOAA satellites, cloud forecasts from NCEP numerical models provide the necessary input for the daily UV Index forecasts. No surface UV observation information is used in the generation of the daily forecasts.

There are no less than seven networks taking surface UV radiation observations operated by various governmental agencies. Each network has its own specific purpose. Which directs the locations of the surface sites. For this exercise, three of the larger networks were chosen: USDA, NOAA's SURFRAD, and EPA's UVnet.

The USDA and the SURFRAD networks use the Yankee UV-B broadband instrument to make their surface observations. The EPA uses the spectral Brewer Mk IV. More information about these instruments can be found at the networks home pages.

Figure 1 shows the location of the three networks' sites within the contiguous 48 states. Note that there are several locals where more than one network is located. This offers the chance for inter-comparison of the observations.

Figure 2 shows the locations of nearly 800 cities where the UV Index forecasts are issued. Stations closest to the surface observation sites are used for this comparison. Some parings of UV observation site and UV forecast city have small distances between them. While others, especially for the sites in National Parks, are much further away.

Characterization Factors

In previous comparisons surface observations made by the Yankee UV-B instrument using the manufacturer's conversion factors produced values with a high bias by as much as 10%.

The NOAA Central Ultraviolet Calibration Facility characterized this instrument and determined conversion factors that varied with both solar zenith angle and total ozone amount(see Figure 3). Once applied, agreement between the Yankee instrument observations and the UV Index forecasts increased tremendously. Therefore, whenever using surface observation data, one must be certain that the correct scaling and calibration factors have been used. Usually, 'readme' files accompany the data sets, but a phone call to the network manager reassures that the data is used properly.

Figure 3

Overall Results

Differences between the observed and forecast UV Index values at all the sites for each of the three networks have been determined. Histograms of these differences for each of the networks are presented in Figure 4.

Figure 4

Note the high number of occurrences for the -1 to +1 UV Index differences for the USDA and the SURFRAD. The great number of these occurrences implies that there is great consistency between the observed and the forecast UV Index values throughout the two networks. However, the EPA histogram results imply a greater inconsistency between observed and forecast UV Index values among its network sites. The high consistency for the USDA and SURFRAD sites imply that similar differences would be expected any where the UV Index was forecast.

The histograms also show one weakness of the UV Index forecasts. The relatively disproportionate number of differences of negative values (-3 to -5) indicate that many times the UV Index forecasts are much greater than the observations. As will be shown below, this occurs when overcast conditions occur. The forecasts allows too much UV radiation to transmit through the clouds.

Also, the SURFRAD histograms exhibit a positive bias. This is interesting since the same type of instruments are used as the USDA which exhibits no bias at all.

Sample Comparison of Surface Observatons and UV Index Forecasts

A good example of how well the UV Index forecasts and observations agree is shown in Figure 5.

Figure 5

This figure shows how the forecasts and observations are compared "by eye". Besides the observations and the UV Index forecasts, the "clear sky" UV Index forecasts are also shown. This provides a indicator of the maximum values at each particular site. The only factors affecting this value is the solar zenith angle and the total ozone amount. Note that the day-to-day variation of total ozone does introduce variations in the clear sky amounts. Note also, the number of occasions when overcast conditions occur and the large differences between observed and forecast UV Index amounts. As explained above, the large differences that occur when overcast conditions exist show up on the negative side of the histograms.

Multiple Comparisons at one Local

As stated earlier there are several locals in the U.S. which have more than one network taking observations. This allows for intercomparisons between the observations and can also be used to exhibit the validity of the UV Index forecast as a unbiased comparison standard.

In Figure 6, the observations from the EPA Brewer at the National Institute for Standards and Technology (NIST) in the northern suburbs of Washington D.C. show a large difference from the UV Index forecasts. These large differences could raise concern for the validity of the UV Index forecast.
However, in the eastern suburbs, in Beltsville, MD is a USDA site.

Figure 7 shows much better agreement. Knowing that similar agreement at the other USDA sites implie that the UV Index forecasts are reliable. As a result, the data from the EPA Brewer is then questionalbe. This same situation occurred at the Table Moutain site outside of Boulder, CO. This time data from a Brewer exhibited a very low bias against the UV Index forecasts. A co-located SURFRAD instrument had a much lower bias. This suggests that this Brewer's data is questionable. This does not mean that data from all the other Brewers is questionable. Other sites compared very well with UV Index forecasts.

One on One Comparisons

During the time period of these intercomparisons, differences normalized to remove the annual cycle, should remain constant. When they do not is another indication of a calibration shift in the instrument.

Figure 8 shows the observations and UV Index forecasts for Davis, CA. Davis characteristically has very clear skies throughout the May to August time period.

Visual inspections of the time series suggests a change in the differences of the obserations and forecast values around day 190. Ratios of the normaled observations and forecasts are shown in Figure 9. Ratios are centered near the zero line to about day 160. The ratios following consistenly lower implying a calibration shift.

Figure 9

Conclusions

The consistency of differences between the UV Index forecasts and observations from two surface UV networks implies the viability of the UV Index forecasts at other locals. This was shown were comparisons against observations made by one instrument had a consistent bias, but comparisons against another instrument in the same local had nearly zero bias. Thus, observations from sites with consistently large biases from the UV Index forecasts would be under suspect.

The UV Index forecasts can also be use to detect unexplained changes in the observation record during the course of the time period. These attributes allow a network data manager or a single observer to use the UV Index forecasts as a first indicator of problems with an instrument.