IMPROVE optical data include flags to indicate whether the data are valid, suspect, invalid, or probably valid but likely to have been influenced by weather. The standard criteria used for flagging a transmissometer observation as possibly weather-influenced include RH>90%, Bext greater than a site-specific threshold ranging from 194 Mm-1 at Petrified Forest to 4419 at Shenandoah, hourly change in Bext > 10 Mm-1 (except 60 Mm-1 at Shenandoah), or hourly Bext uncertainty > 18 Mm-1. For the nephelometers, the criteria for weather-influenced include RH>90%, Bsp > 5000 Mm-1, hourly change in Bsp >50 Mm-1, or hourly ratio of standard deviation to mean > 10 Mm-1.
For this analysis, only data flagged as valid and not weather-influenced were used. After this filtering of the data, there were a few extreme outlying values still remaining in the data set. For example, at Great Smoky Mountains, all weather-filtered Bext values (nearly 33,000 observations) were less than 510 Mm-1 except for a handful of observations all on a single day when the reported Bext was in excess of 4000 Mm-1. These few outliers, though they may be valid, were then flagged as probably weather-influenced also, by re-filtering the data using lower maximum thresholds. The new thresholds based on graphical observation of the time-lines of either Bext or Bsp at each site, ranged from 150 Mm-1 at Yellowstone to 2000 Mm-1 at Shenandoah.
As an aid to understanding the diurnal and seasonal patterns in the data, the variable, f(RH), was calculated for each hour of filtered optical data. F(RH) is the ratio of the light scattering by hygroscopic particles at a given RH to the scattering that would be expected if the particles were dry (Tang, 1996). Figure 2 shows the values of f(RH) vs. RH for ammonium sulfate particles with a median diameter of 0.3 microns and geometric standard deviation of 1.5. Similar curves can be generated for other species and for different size distributions. Scattering by sulfates is a large fraction of the light extinction at most IMPROVE sites (Malm, 1992, 1994; Malm et al., 2000; Sisler et al., 1993, 1996) so these f(RH) values were used for this analysis, primarily to illustrate how changes in RH can have a large influence on changes in scattering at some sites, while much less influence at other sites that have lower concentrations of hygroscopic aerosols and/or lower mean RH.
The means of the filtered optical and meteorological data were calculated by season and hour of day and plotted for each site. Because the optical data and f(RH) are usually log-normally distributed, geometric means were used for these values. Arithmetic means were used for RH and temperature, which are more normally distributed. Means of RH, temperature, and f(RH) were plotted on the same scale for all sites to facilitate comparisons between sites. The magnitudes of the mean values of Bext and Bsp are quite different between sites, so in order to see the diurnal and seasonal patterns of these values, the scale had to be allowed to vary by location.
Four sites, Crater Lake National Park, Edwin B. Forsythe National Wildlife Refuge Brigantine Division, Great Gulf Wilderness Area, and Lye Brook Wilderness Area were not analyzed because there were too few data points. All remaining sites had at least 7800 hourly weather-filtered observations, most had at least 10,000, and several long-running sites in the western United States had more than 70,000 observations.
Weather-filtering of the data has the advantage of removing
hours that are likely to have been influenced by precipitation, fog, clouds in
the transmissometer path, or even events such as interference by wildlife. However, the disadvantage of weather-filtering
is that the resulting dataset is biased.
In the western United States where average relative humidities are
fairly low, there is no apparent adverse consequence to weather-filtering for
this analysis. At sites in the eastern
United States, where average RH is higher, more hours were weather-filtered at
night than during mid-day when the mean RH was lower. Thus, the same analysis without weather- filtering would probably
show greater differences between day and night at these sites.