EPA Summary and Assessment of ADEQ Visibility Analysis
May 9, 2013
In the supplement submitted on May 3, 2013, ADEQ provided an additional analysis of visibility trends in Arizona based on IMPROVE monitoring data for the baseline period (2000-2004) through the 2005-2009, which ADEQ terms the "progress period". EPA reviewed the data and analyses and we agree with the presentation of the observed trends of the data. The monitoring data presented in the supplemental analysis is consistent with the data that EPA obtained from the Regional Haze Planning section of the WRAP Technical Support System (TSS) website located at http://vista.cira.colostate.edu/tss/Results/HazePlanning.aspx.
Interannual variability in the emissions of certain aerosol components caused by natural, uncontrollable events such as wildfires or dust storms is often much larger than the expected change in the emissions of these species from anthropogenic sources. The large variability in size and occurrence of these events can obscure a trend in the species-specific contribution to visibility impairment as well. In order to provide a more complete explanation of the observed trends, ADEQ performed different types of analyses that are sensitive to these variations in different ways.
ADEQ's first analysis of the monitoring data was to compare total and species contributions to visibility impairment during the baseline period and the progress period. The TSS calculates the multi-year average for the two periods by averaging the annual average visibility for each of the five years, for the best and worst twenty percent days. This approach tends to average out the impact of large events, but still can be difficult to assess actual improvement between the two periods if a large event occurred more often in one period than the other. In order to demonstrate the effect of wildfires on the results of this analysis at GRCA1 and IKBA1, ADEQ replaced the annual average total extinction with an average value for some years with particularly large wildfires. This analysis was successful in demonstrating the influence of high impairment years on the 5-year average. 
In addition to providing the results of the Regional Haze method of determining trends (the difference between baseline and progress period averages) ADEQ also provided an alternative analysis that uses year to year changes to derive an overall trend through the entire data record from 2000 through 2010. The Theil method, a nonparametric regression technique, was used to characterize ten-year visibility trends at the Class I areas and determine their statistical significance. This type of method is useful since it does not make assumptions about the distribution of values or their uncertainties. This reduces the impact of outliers and large interannual variability, often seen in ambient air quality data, on the estimated trend. EPA has used these methods for characterizing air quality trends as well as in reports on visibility trends. This method tests whether the value from year to year in a dataset tends to increase or decrease consistently; it is not concerned with the magnitude of the year-to-year differences, only the significance and direction of the trend. 
The comparison of baseline and progress averages shows that visibility impairment on best and worst 20 percent days has remained constant or decreased at most sites. The worst 20 percent days at GRCA2 and IKBA1 show a slight decrease in visibility. Organic carbon is the largest contributor to extinction during the progress period at these sites, and there were large wildfires impacting GRCA2 and IKBA1 in 2009 and 2005, respectively. ADEQ replaced fire-affected data with long term averages and found in both cases that the new progress period average visibility was higher than the baseline period, also demonstrating the sensitivity of these trends to infrequent, high extinction events. The Theil method detected no significant increasing trends in total extinction from 2000 to 2009.
The analysis of trends in the species contributing to extinction yield mixed results which are presented in more detail in section 11.4.3 ADEQ's supplement. In particular, the regional haze method of comparing period averages shows increases in ammonium sulfate extinction on the worst 20 percent days at most sites, but the Theil method determined that there was no statistically significant annual increasing trend. The Regional Haze Rule method also showed increasing extinction at BALD1, SAGU1, SYCA1 and TONT1, but only the trend at BALD1 was statistically significant using the Theil method. ADEQ did further analysis of the monitoring data to evaluate if any of the trends observed were due to anthropogenic sources, rather than interannual variability in natural events.
One characteristic of the Regional Haze method in determining trends of 20 percent worst or best total extinction days is that the causes of the visibility impairment on the days with a certain level of total extinction can shift due to changing emission patterns, including interannual variability of large natural events. This can, for example, lead to worst days in one year more affected by anthropogenic emissions, while worst days in other years are dominated by events such as large wildfires. Convoluting these sources in estimating trends obscures the underlying trends in visibility due to controllable emissions, especially in the contribution of individual species. For example, if worst days in the baseline period were caused primarily by wildfire emissions, and there were fewer wildfires in the progress period, the progress period worst days are more impacted by anthropogenic sources. This can lead to misinterpretation of percent contribution since the dominant contributing species could shift from organic carbon to ammonium sulfate, without necessarily an increase in the emissions of ammonium sulfate between the two periods. 
Further analyses on ammonium sulfate trends
The analysis of the 20 percent worst ammonium sulfate days submitted by ADEQ demonstrated that there was not a significant increasing trend at any site using the Theil method. There was still an increase in the base and progress period averages, but ADEQ asserts that this can be explained by anomalously high ammonium sulfate impairment in 2005 and 2007, leading to an uncharacteristically high average for the 2005-2009 period. In fact, ADEQ notes that shifting the 2005 year to the baseline period results in decreasing differences between the two periods at all sites except TONT1. Regional data indicated that there were large contributions to visibility impairment throughout the entire four corners area. In 2005, there were anomalous number of ammonium sulfate spike events in the summer and fall, and the frequency of the events made it difficult to differentiate specific sources. In the May 3, 2013 RH supplement, ADEQ asserted that there were no anomalous elevated anthropogenic emissions in 2005 after reviewing all significant sources of SO2 in the State (including coal-fired power plants and copper smelters). Data submitted for six large SO2 sources showed that 2005 was, in fact, lower than prior years for most of the facilities.
ADEQ did investigate further the cause of the August 2007 ammonium sulfate event that largely drove the 2007 value for 20 percent worst ammonium sulfate days. Back trajectories indicate that the plume source was to the southeast of the CHIR1 monitor, very likely outside Arizona (Figure G of ADEQ May 2013 submittal). This was supported by tracking the peak in monitored ammonium sulfate extinction at IMPROVE monitors first through Texas, then New Mexico and finally into Arizona over the course of the episode (see Figures H-L of ADEQ May 2013 submittal). Removing the data from this episode in 2007, as well as the anomalous 2005 data year results in decreases in the extinction on the 20 percent most impaired ammonium sulfate days between the baseline period and the alternate progress period of 2006-2010. 
Further analyses on coarse mass trends
The analysis of the 20 percent worst coarse mass (CM) days submitted by ADEQ demonstrated that there was only a significant increasing trend at PEFO1 using the Theil method. Most sites exhibit no significant trend, with GRCA2 and SYCA1 showing a decrease in 20 percent worst CM days. ADEQ further evaluated the data from the PEFO1 monitor and found there was a large amount of interannual variability in the number of high CM days in late March to early June throughout the monitoring period. In particular, each year from 2008-2010 had an increasing number of these episodes relative to the previous year. This increase in episode frequency is responsible for the apparent increasing trend in 20% worst days between the baseline and progress periods. ADEQ chose the four highest CM days in 2010 to analyze further to understand the events contributing to elevated CM extinction. Back trajectories from these days indicate that windblown dust from area sources are more likely the source of these elevated CM impacts than the nearby point sources. More generally, there was also a lack of a pattern statewide between the observed amount of (and trend in) impairment due to coarse mass, and the proximity and size of nearby large PM10 point sources. These results along with inverse relationships between the CM observed data and precipitation patterns throughout the state support the conclusion that the CM extinction is more affected by area sources, rather than point sources.
