Journal of Geophysical Research: Atmospheres

Fifteen-year aerosol optical depth climatology for Salt Lake City

Authors

  • Joseph Michalsky,

    Corresponding author
    1. Earth System Research Laboratory, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA
    • Corresponding author: J. Michalsky, Earth System Research Laboratory, National Oceanic and Atmospheric Administration, 325 Broadway R/GMD, Boulder, CO 80305, USA. (joseph.michalsky@noaa.gov)

    Search for more papers by this author
  • Brock LeBaron

    1. Department of Environmental Quality, State of Utah, Salt Lake City, Utah, USA
    Search for more papers by this author

Abstract

[1] Aerosol optical depth (AOD) and its wavelength dependence have been measured for the past 15 years in the Salt Lake City metropolitan area using a multifilter rotating shadowband radiometer. The instrument has not experienced a major hardware failure. It has been continuously field calibrated for extraterrestrial responses in its five aerosol channels. The instrument's cosine response was measured in 1996 and again in 2012. In our analysis of this 15 year data set, linear interpolation of these two cosine responses was used to approximate the angular response between the two characterizations. The Salt Lake City aerosol burden increased through the mid-2000s, but has dropped to its lowest level of the record since that time despite a population increase of approximately 25%. Annually, the aerosol burden is highest in midspring and midsummer with relatively coarse aerosols during the spring peak and fine aerosols during the summer peak. There is no indication of a diurnal cycle in AOD. There is a significant, but low, correlation between PM2.5 and 500 nm AOD, and a slightly lower correlation between PM10 and 500 nm AOD. The correlations between the surface-based measurements and total column AOD explain only 13% and 9% of the variance, respectively. Measurements are continuing to track future trends.

1 Introduction

[2] NASA's Goddard Institute for Space Studies in New York City started the Solar Irradiance Research Network (SIRN) with 14 sites with the intention of eventually equipping each state throughout the United States with automated radiometers for measuring aerosol optical depth (AOD) (B. Carlson, private communication, 2012). One of the sites selected was Salt Lake City. The radiometer that has operated there since March 1997 is the multifilter rotating shadowband radiometer (MFRSR), which was developed by Harrison et al. [1994] and is marketed by Yankee Environmental Systems, Inc.

[3] Although funding for this program ceased after the first 14 sites were installed, a few dedicated individuals continue to operate their SIRN MFRSRs, including the one in Salt Lake City. This paper demonstrates our attempt to automate the analysis of 15 years of SIRN MFRSR data and extract an understanding of how the aerosol column has evolved in the Salt Lake City valley. These AOD measurements are also compared to in situ surface-based aerosol measurements made by Utah's Department of Environmental Quality. Features of the daily, annual, and long-term behavior will be noted and explanations offered.

[4] A few AOD climatological records for the United States have been published since 2001. AERONET is based at Goddard Space Flight Center in Greenbelt, Maryland. The instrument used for aerosol measurements in this worldwide network is the CIMEL Sun-tracking radiometer [Holben et al., 1998]. Holben et al. [2001] published climatological data for nine of the international AERONET sites. Eck et al. [2010] recently added climatological aerosol information from three heavily polluted AERONET sites. Michalsky et al. [2001] discussed rural aerosols at northeastern U.S. sites where the aerosol optical depth was measured with the MFRSR [Harrison et al., 1994]. Augustine et al. [2008] recently published AOD records based on MFRSR measurements for the SURFRAD network, which are located in seven rural locations from the southwestern to the northeastern U.S.

[5] Salt Lake City is surrounded on three sides by mountain ranges and on the northwest by the Great Salt Lake. Michalsky et al. [2010] performed a similar analysis of MFRSR data collected in a rural area of northern Oklahoma over a similar length of time. Many of the procedures developed for that paper are applicable to this analysis of AOD in an intermountain urban area. According to the 2010 U.S. Census, the Salt Lake City-West Valley City Metropolitan Statistical Area consisting of Salt Lake, Tooele, and Summit counties had a population of 1,124,197 (http://www.census.gov/geo/www/2010census/centerpop2010/county/countycenters.html). This was a gain of 16% from the 2000 census population. Assuming that this growth rate has persisted, we can estimate that there has been a 25% increase in the Salt Lake City metropolitan area population over the 15 year period of MFRSR measurements. The current location, and the site with the most data, is at latitude 40.7117°N, longitude 111.9608°W, and altitude 1295 m. The initial site was at 40.7558°N, 111.9042°W about 7 km northeast of the current location and within a meter or two of the same altitude. The MFRSR was moved to its current site in November 1999.

2 Calibration and Characterization of the MFRSR

[6] The MFRSR is described in detail in Harrison et al. [1994]. It is our experience that if the filters within the detector head are firmly positioned so that their angle with respect to incoming sunlight does not change, their spectral profiles will remain constant. However, the overall transmission of the filters may degrade with time. Because the MFRSR is a Sun photometer, it is possible to calibrate its five aerosol channels using Langley plots. The Langley plot is based on attenuated radiation in a narrow spectral interval according to the equation

display math(1)

[7] where I and I0 are the irradiances at the surface and top of the atmosphere, respectively, τ is the total optical depth, and m is the optical air mass relative to the path in the zenith direction. By taking the natural logarithm of the equation and plotting ln(I) versus m, the intercept ln(I0) can be determined and I0 can be used to evaluate τ instantaneously from any measurement of I. This only succeeds if the intercept is robustly determined from a sample of Langley plots, and the corresponding I0’s are examined as to how well they tend toward a central value. Our method for doing this in the current analysis is explained in Michalsky et al. [2001].

[8] The robustness of this statistical sampling method for establishing an MFRSR calibration is demonstrated by comparing channel calibrations for the same MFRSR at Mauna Loa, HI, and Boulder, CO, with minimal time lapse between Mauna Loa and Boulder data. This example does not use the MFRSR from Salt Lake City; however, any instrument can be used for this demonstration including traditional, narrow field-of-view Sun pointing radiometers. At Mauna Loa, 17 clean Langley I0 calibrations were acquired between 3 and 18 November 2010. Shortly following that, 20 clean Langley I0 calibrations were acquired between 30 November 2010 and 11 February 2011 in Boulder. Figure 1 shows boxplots of normalized I0 calibrations for each of the aerosol filters at both locales, and it illustrates several points: (1) the medians agree at both sites to within 0.6% with four of the five agreeing to within 0.2%; (2) the spread of the Boulder data is greater than that from Mauna Loa data as expected for Boulder's less pristine conditions; and (3) local measurements can indeed be used for calibrations, but take longer and must be performed continuously to capture changes in the instrument calibrations with time.

Figure 1.

Top of the atmosphere response at Mauna Loa Observatory (mlo, smaller boxplots) and at Davis Skaggs Research Center in Boulder, Colorado (dsrc, larger same-colored boxplots) for the five aerosol channels of the MFRSR. These are normalized to the Mauna Loa median values for comparison. All Boulder median values are within 0.006 of the Mauna Loa calibration factors.

[9] Traditionally, ln(I0) would be given in spectral irradiance units, but ln(V0)’s (the voltages measured by the MFRSR) can be used equally well in determining AOD thus precluding the need for absolute calibration of the MFRSR channels in spectral irradiance units. That is because in equation ((1)) optical depth is determined from the ratio I/I0, where the calibration constant to convert voltages to spectral irradiances cancels in the ratio. Figure 2 is a plot of the time varying 500 nm V0’s that were used to calculate the AODs that emerge from this study. The 1695 V0’s are plotted as black points in the figure. In our analysis 20 neighboring points are combined to produce a robust estimate of V0 as described in Michalsky et al. [2001]; this is repeated by advancing one point at a time through the time series of all 1695 points. The red points in Figure 2 show these results. The V0 changes are still somewhat discontinuous, therefore a lowess smoother through the red points is used to produce a continuous and physically realistic estimate of the variability, which is represented by the blue line. Results for only one filter are shown in Figure 2; however, two changes common to all of the aerosol channels are (1) an initial rapid decrease in the sensitivity in the first few years and (2) an annual variability that is correlated with temperature. The steep initial falloff was common to all filters and was likely soiling of the porous Spectralon diffuser. Spectralon is sintered PTFE (Teflon) that is used as the fore optic in MFRSRs through which all radiation scatters and passes to the filtered detectors below. Four of the five aerosol channels experienced transmission falloffs in later years, but at different times and to different degrees. These changes are likely caused by filter degradation. Calibration values of four of the filters continued to slowly decline, but at different rates, and one filter's calibration was stable after the initial decline. Our estimate is that the V0’s determined in this manner have a 95% uncertainty of about one percent [Michalsky et al., 2001].

Figure 2.

The falloff in sensitivity of the 500 nm channel of the MFRSR over 15 years; the black dots are 1695 Langley plot slopes; the red dots are robust estimates based on a moving window of 20 of the individual Langley plot slopes; and the blue smoother reduces the scatter in the red points and is used as the calibration of the channel at any point in 15 year time period.

[10] A characteristic of the Salt Lake City MFRSR that changed significantly over the 15 years was its angular, or cosine response. Even if the diffuser was not damaged during that period, the soiled Spectralon diffuser caused a cosine response change. Figure 3 illustrates the change in cosine response over the 15 year span. The plots are normalized at 0° incidence angle. Each point of the plot for a particular filter is divided by the measured response at 0° and by the cosine of the angle of incidence from due south at the horizon to due north at the horizon through the zenith. The upper responses (solid lines) are based on factory measurements in June 1996. The lower responses (dashed lines) were measured in Boulder in February 2012. Because there is no knowledge of the rate of change from the initial response to the final one, we assumed a linear change in time and used a weighted cosine response for the cosine corrections of each year's data. For example, the 2000 cosine response was assumed to have a cosine response that was 80% factory-measured and 20% Boulder-measured. The west-to-east cosine responses were handled in a similar way. The uncertainty caused by the cosine response changes is estimated in the final section.

Figure 3.

The south to north angular response of the MFRSR normalized at 0° incidence; the upper curves (solid lines) are from the original characterization in 1996 and the lower curves (dashed lines) were measured in 2012.

3 Aerosol Optical Depth and Size Analysis

[11] In Michalsky et al. [2010], most of the analyzed data from Oklahoma were acquired at 20 s intervals. Because of the small data storage capacity available for the MFRSR in the mid-1990s, the Salt Lake City MFRSR recorded 5 min averages of 15 s samples. These 5 min data records required changes from the cloud screening that was performed in the Oklahoma study. The fundamental cloud-screening process for the Oklahoma study was to look for stability of the 20 s AOD samples over 10 min intervals. For the Salt Lake City data, five adjacent samples were examined at a time, so the AOD had to be stable for 25 min time spans to pass the cloud screen test. This implies that time windows shorter than 25 min may have cloud-free AOD, but were screened as cloud contaminated. The initial screening required that none of the adjacent (5 min separation) points differed by more than 0.02 and the difference over the 25 min period was less than 0.03. A second screening of the surviving points was similar with adjacent points not to exceed 0.1 times the estimated AOD or the change over 25 min not to exceed 0.2*AOD.

[12] Figure 4 illustrates a typical day of measurements with some points passing the cloud screening algorithm (red points) and others not (black points). In the first part of the day all of the AODs exceeded 0.3 and failed the cloud screening. Those are not shown to better display the AOD data that did pass the cloud screen tests. This figure illustrates the point about time windows shorter than 25 min having valid AOD data. As one can see, the two groupings of three consecutive points around x = 106.7, circled in red for emphasis, are likely cloud free, but they do not pass the automated 25 min stability requirement. Most of the points that did not pass cloud screening, however, are clearly influenced by clouds. The clouds that affected the data in Figure 4 are probably optically thin cirrus, because the computed optical depths added only a small additional attenuation to an estimated mean AOD of around 0.06. The AOD data that passed the cloud screening have stable values, and the method used, as judged by these results, is slightly conservative.

Figure 4.

An example of the cloud screening results using the technique described in the paper; AOD must be stable during a 25 min window to within set criteria as the window slides one point at a time over the 5 min samples. Red points were deemed true AOD, and black points were considered cloud contaminated. Points early in the day were unstable and out of this range.

[13] In this analysis, daily averages of AOD were analyzed. A daily-average AOD was calculated, even though there may have been very few cloud-free points in the day. The fewest number of points used to calculate a daily average was six. Analyzing variably sampled daily AOD averages in time series was considered a less biased way to estimate true seasonal behavior because limiting the analysis to days with a relatively high number of AOD data would produce a bias toward clean, clear days. Daily AOD averages were computed for 57.6% of the days during the 15 year period. There were six gaps that exceeded 23 days and one large gap of 58 days. Figure 5 is a plot of daily averages for 15 years of measurements for the 500 nm channel. Less than 4.2% of the daily 500 nm AODs exceeded 0.25; therefore, 0.25 was used as the upper limit for the plot for better resolution at the predominantly lower AODs. There are 3170 daily averages (black points). The red line is a lowess smoother that does a linear least square fit to the points using three months of data at a time, but with more weight for the time-centered points in the moving 3 month window. After the initial fit, points closer to the fit in AOD receive more weight so that the influence of the outlying AODs is diminished in the second fit. Note that the beginning and end of the lowess fit using this short 3 month window should be ignored because the endpoint behavior is highly influenced by a few points near the boundaries of the data record.

Figure 5.

Black dots are 3170 daily averages of AOD at 500 nm. The red line is a lowess fit with a 3 month window that shows little indication of a repeating annual pattern. The green line is a lowess fit with a 5 year window that indicates a small increase that peaks in 2007–2008 followed by the lowest AODs of the entire 15 year record. The blue line is a least square fit to a cubic polynomial that indicates a similar behavior. It is higher than the green lowess fit because it does not resist the outliers as does the lowess fit.

[14] Although some years have a tendency toward higher summertime AODs and lower wintertime AODs, a regular annual pattern is difficult to discern in most years. This lack of a routine annual pattern is in clear contrast to rural Oklahoma data [Michalsky et al., 2010] that showed consistently high summer AODs and low AODs in the winter [see Michalsky et al., 2010, Figure 8]. Augustine et al. [2008] showed the same annual tendencies for seven other rural stations in the U.S. In contrast, for much of the Salt Lake City record, low and high AODs are scattered throughout the year.

[15] The green lowess fit has a 5 year moving window to capture the longer-term tendencies of the AODs. The green lowess fit for the 500 nm filter of Figure 5 indicates a gradual increase in the first several years, a shallow dip in the 2004–2005 timeframe, a maximum for the whole record in 2007–2008, and a decrease to low AODs in the last years of the record. The blue dashed curve is a cubic equation fit to the data without the linear term. Cubic fits with the linear term and without the quadratic term produced similar fits with similar high significance for the coefficients of either fit. These least squares fits do not de-weight outliers as is the case for the lowess fit, and, therefore, they are higher than the lowess fits over the duration of the 15 year record. The cubic fit also does not show the shallow decrease around 2004–2005, but does appear similar in most ways to the lowess fit with a low beginning, a peak in the 2006–2007 time frame, and a dip to the lowest values at the end of the record.

[16] Figure 6 contains boxplot annual summaries of the 500 nm AOD data. The first boxplot contains only March through December 1997 data, and the last boxplot for 2012 contains only a little over one month of data. The medians appear as horizontal black lines in the boxes and the red dots, are the mean values for the year, which are not normally plotted as part of a boxplot. The vertical extent of the boxes contain 50% of the data (the interquartile range) and the whiskers on the top and bottom of the box extend to the nearest points that are not beyond 1.5 times the interquartile range. Points beyond this are plotted individually. The annual boxplots follow a pattern similar to the lowess fit in Figure 5. The red line is a linear fit to 14 annual averages and its slope, while slightly negative, is not statistically significant.

Figure 6.

Boxplots of the yearly data indicate interannual variability. The red dots are the annual means. The red line is a linear least-squares fit to the 14 years that have data all year long indicating no clear long-term linear trend.

[17] Ångström [1929] developed this simple expression for estimating the spectral dependence of aerosol attenuation

display math(2)

[18] The Ångström exponent α can be computed by taking the natural logarithm of equation ((2)) and performing a linear least squares fit of ln(τ) as a function of ln(λ). The absolute value of the slope of this fit is the Ångström exponent. α is inversely proportional to the predominant particle size in the particle number distribution. Values around 0 represent very large particles, values around 4 represent molecular sized particles, and typical continental aerosols have a value of about 1.3 [Ångström, 1924]. While there is more information about the size distribution of the aerosols by using all five wavelengths of AOD, in this paper we use just two points at 500 and 870 nm to estimate α to first order to focus on seasonal and longer-term size changes using

display math(3)

[19] The black points in Figure 7 are α’s calculated from daily AOD values using equation ((3)). The red line is the lowess fit to the daily values using a 6 month window, and the blue line is a lowess fit with a 3 year window. The red line suggests that most often the smallest sized aerosols occur in the summer (largest α), but the decrease in size during the summer is not as consistent as it is for Oklahoma aerosols [Michalsky et al., 2010], or for other U.S. stations [Augustine et al., 2008; Holben et al., 2001]. In Oklahoma the largest aerosols occur in early winter, but in Salt Lake City the largest aerosols (smallest α) appear much more predictably in mid spring. From the blue 3 year lowess fit it is clear that the early tendency was for aerosols to become larger until late 2000 when they hit their peak size (lowest α). Beginning in 2000 they decreased in size until early 2006 when they became stable or even grew slightly through the end of 2009. From 2010 to the end of the record, the Salt Lake City aerosols decreased in size to the 15 year minimum.

Figure 7.

The Ångström exponents for the 3170 daily averages are shown as black dots. The red line is a lowess fit with a 6 month window that indicates no repeating annual variation. The blue line is a lowess fit with a 3 year window showing a change in particle size that increases and then slowly, but not steadily, decreases to the current value (larger Ångström exponent implies smaller size).

[20] In Figure 8 all daily averaged 500 nm AODs and Ångström exponents from the 15 years of data are composited on a one-year time series to illustrate the mean annual pattern. Boxplots of 500 nm AOD are shown for each month in Figure 8a. Clearly there is a not a large variation in median AOD from month to month. April/May, July/August, and December have the highest medians. January and June are slightly lower and late winter (February/March) and fall (September/October/November) have the lowest AODs. The corresponding Ångström exponents are shown in Figure 8b. Recall that smaller Ångström exponents indicate larger aerosols; therefore, the high April values of AOD are associated with larger aerosols. On the other hand, the high values of AOD in midsummer are caused by small particles. An interesting increase in small aerosols occurs in February without an increase in AOD.

Figure 8.

(a) All of the daily averages are overlaid on one year. Monthly boxplots are made for 500 nm AOD. The boxplots indicate that the highest 500 nm AODs are mid spring to late summer and in December with lower AODs in late winter and all of the fall months. (b) Ångström exponent boxplots have peaks in February and August and a minimum in April. The high 500 nm AOD values in April correspond to a minimum in alpha indicating that large particles are causing that peak. In summer the AOD peak is associated with a peak in small particles. Small particles dominate the February data, but there is no corresponding peak in the aerosol column.

[21] The diurnal variability of the 500 nm AOD is examined in Figure 9a for the years 2007 through 2011. Boxplots are displayed for every local hour that has AOD measurements during every month of the year in Salt Lake City. The medians never differ by more than 0.006 optical depths throughout the day. The values are marginally higher in the morning and lower during the last 2 h shown in the plot, but these are not statistically different from the others. In Figure 9b the Ångström exponents for each of the same hours are plotted showing even less variability than the AOD. The green lines bracket values of 1.3 and 1.4, which are very close to the typical continental aerosol value of 1.3 ± 0.5 originally suggested by Ångström [1924].

Figure 9.

(a) Hours of the day that have some Sun every day of the year have almost the same median AOD at 500 nm. This includes the last full five years of data in the record. (b) Similarly, the size of the particles does not depend on the hour of the day. The green lines bracket Ångström exponents of 1.3 and 1.4 that compare favorably with Ångström's suggested value for continental aerosols.

4 Discussion

[22] Salt Lake City experiences seasonal aerosol events that are consistent from year to year. During the winter months the Salt Lake valley is frequently capped by intense temperature inversions creating a cold pool situation where surface air pollutants can be trapped for days or weeks. Much (70% or higher) of the fine aerosol mass consists of ammonium nitrate that is formed secondarily in the atmosphere and is submicrometer in diameter. During the spring, wind events transport crustal dust (coarse aerosol) from the western desert area of Utah into the valley. In middle to late summer, smoke (fine aerosol) from wild fires and prescribed fires regularly impact the Salt Lake valley. While the storm and hydrological cycles of individual years can influence this pattern, the long-term average validates the seasonal 500 nm AODs and α’s seen in Figure 8. Note the heights of the boxes in December and January in Figure 8a. These larger ranges result from extremely clear and extremely polluted events associated with cleansing fronts and intense inversions that occur during these winter months. Fine and coarse particle measurements at the surface, discussed in the following, have their peak values in these two months.

[23] The AOD results should be robust because the filter profiles are stable and the field calibration methodology has been established as accurate (see Figure 1). The largest uncertainty in the determination of AOD is associated with the true change in cosine response with time versus the consistent application of the factory-determined cosine response throughout the 15 year period. In Figure 10 the AOD at 500 nm for a clear day near the middle of the record is plotted. The red points represent the AOD as calculated for the data reported in this paper, that is, we used V0’s based on a linearly interpolated cosine response with time-proportioned weights for the original factory cosine response and for that measured in 2012. The green AODs were derived using the original cosine response and the blue AODs used the cosine response measured in 2012. If we assume that the green and blue points are the extreme values to expect, and the red points (from interpolated cosine responses) are the most probable, then a triangular probability distribution for the cosine responses seems a reasonable choice for calculating the uncertainty associated with the cosine response. The range of the differences from the red points over the solar zenith angles in Figure 10 (18°–80°) is 0.015 to 0.025. The 95% uncertainty for a triangular distribution is this range divided by inline image and multiplied by the coverage factor of 2 [Taylor and Kuyatt, 1994], or in this case 0.012 to 0.020 optical depths depending on the solar zenith angle. The effect of the uncertain cosine response, therefore, exceeds the uncertainty associated with our Langley calibration method for the V0’s discussed in section 2 and all other sources of uncertainty. We estimate the overall 95% uncertainty in AOD at between ±0.015 and ±0.025.

Figure 10.

The range of AOD one gets using the original cosine response (green), the latest cosine response (blue), and the cosine response interpolated for the day in the middle of the record (red; labeled “both”) when the AOD measurements were actually made. These data are used to estimate an uncertainty associated with the cosine response in the years between cosine response measurements.

[24] Concentrations of particulate matter equal to or smaller than 2.5 µm (PM2.5) and 10 µm (PM10) in aerodynamic diameter were collected at the Utah Division of Air Quality's Hawthorne monitoring site. That site is located at 1675 South 600 East in Salt Lake City, Utah, longitude 111.8730°W, latitude 40.7340°N, and elevation 1308 m. The particulate data is collected along with a number of other air quality and meteorological parameters and is sited to represent local population exposure. The site is situated less than 8 km from, and is approximately 13 m higher than, the Air Monitoring Center where the MFRSR data were collected. The small distance between these two sites is not expected to affect the comparison between the daily averaged AOD and the PM data sets.

[25] Fine PM tends to be well mixed especially during regional temperature inversion, fire smoke, and dust events, and there are no significant local sources. One might expect AOD measurements to correlate well with both of these routine air quality particle measurements, however, AOD is extinction measured through a column from the top of the atmosphere to the surface and PM2.5/10 data are local particle mass concentration measurements made near the surface that sample the boundary layer aerosols. If the boundary layer is thin, or if there are aerosol layers above it, the boundary layer contribution to the total column may be minor. Both PM measurements have been made since 1999. In 13 years of PM10/2.5 measurements there were 1997 days that had both PM10/2.5 and AOD measurements. The PM measurements are integrated over a 24 h period and expressed in units of µg m–3. The AOD measurements are the daily averaged AODs for all data that passed the cloud screening process. Correlation tests among all wavelengths of AOD and PM2.5 and PM10 measurements using Pearson's product-moment correlation find the highest correlation between the AOD at 500 nm and PM2.5. The correlation coefficient r is 0.367 with a 95% confidence interval that spans 0.328 and 0.404. The square of the correlation coefficient suggests that only a little over 13% of the variance between those two variables is explained. The highest correlation between PM10 and AOD is again at 500 nm, but only 9% of the variance is explained with a correlation coefficient of 0.301 and a 95% uncertainty range of 0.260 and 0.340. These results suggest that for the unique geography and meteorology of the Salt Lake City valley that column measurement of AOD does not relate well to the air quality measurement PM2.5 that is used most frequently for regulating particle pollution.

[26] We began this data analysis during spring of 2012 and, consequently, only included data through the winter of 2012. The summer of 2012 was marred by numerous forest and brush fires in the western U.S. that could interrupt the downward trend in AOD. We speculate that the downward trend in AOD may be associated with the economic downturn since 2008. Another possibility is that as the auto and truck inventory becomes younger with more modern pollution controls that the aerosol burden is decreasing. The authors intend to extend the Salt Lake City AOD record to better understand AOD behavior and its causes.

Acknowledgments

[27] The authors would like to thank Ellsworth Dutton and John Augustine for advice in regard to several aspects of this paper. Kimberly Kreykes in the Utah Department of Environmental Quality supplied the PM2.5 and PM10 data. The MFRSR instrument was originally purchased with a grant from the National Aeronautics and Space Administration's Goddard Institute for Space Studies.

Ancillary