Source reconciliation of atmospheric dust causing visibility impairment in Class I areas of the western United States

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 114,, doi: /2008jd009923, 2009 Source reconciliation of atmospheric dust causing visibility impairment in Class I areas of the western United States Ilias G. Kavouras, 1 Vicken Etyemezian, 1 David W. DuBois, 1 Jin Xu, 1,2 and Marc Pitchford 3 Received 5 February 2008; revised 19 August 2008; accepted 12 November 2008; published 27 January [1] Aerosol data from the Interagency Monitoring of Protected Visual Environments (IMPROVE) network, air mass backward trajectories, land use maps, soil characteristics maps, diagnostic ratios of elemental composition, and multivariate linear regression were utilized as part of a semiquantitative analysis. The purpose of the analysis was to determine the types of dust-causing events that contribute to low visibility at a given site when the sum of extinction from coarse mass (CM) and fine soil (FS) was larger than any other aerosol component and the reconstructed aerosol extinction coefficient was among the 20% highest (calculated on a calendar year basis) for that site. For these worst dust days, the above tools were used to ascribe the cause of low visibility to one of the following types of events: (1) transcontinental transport of dust originating from Asia; (2) windblown dust events from sources located nearby the site and; (3) transport of windblown dust from sources upwind of the site. Depending on the weight of evidence, a low or high level of confidence was associated with the assignment of one of these three events. Absence of convincing evidence resulted in ascribing the worst dust day to undetermined events. Of the 610 worst dust days over the period, 51% were associated with one of the three event types with high confidence and an additional 30% were accounted for with low confidence. Of the 496 worst dust days associated with an event (either low or high confidence), Asian dust was the assigned event on 55 days (for ), locally generated windblown dust on 201 days, and transport from upwind source areas susceptible to wind erosion on 240 days. Events associated with windblown episodes from source areas in the United States and Mexico exhibited the highest dust concentrations. Asian dust events were associated with lower dust concentrations and a larger FS-to-CM ratio. Some variations between Asian dust and continental North American dust were observed in organic matter (OMC), black carbon (LAC), and nitrate (NO 3 ) content. None of the tools used in this study was adequate for identifying events associated with mechanically released dust by anthropogenic activities including, agriculture, construction and motor vehicle travel on paved and unpaved roads. Some of the worst dust days may have been caused by these types of activities, especially in central Arizona and northern and Southern California, where the fraction of undetermined events was higher than in other regions within the western United States. All in all, the methods and results of this study can help improve the performance of large-scale dust emission models and provide insight into the distribution of the types of events that cause dust resultant haze in relatively remote areas of the western United States. Citation: Kavouras, I. G., V. Etyemezian, D. W. DuBois, J. Xu, and M. Pitchford (2009), Source reconciliation of atmospheric dust causing visibility impairment in Class I areas of the western United States, J. Geophys. Res., 114,, doi: /2008jd Division of Atmospheric Sciences, Desert Research Institute, Las Vegas, Nevada, USA. 2 Also at California Air Resources Board, Sacramento, California, USA. 3 Air Resources Laboratory, National Oceanic and Atmospheric Administration, Las Vegas, Nevada, USA. Copyright 2009 by the American Geophysical Union /09/2008JD Introduction [2] Mineral dust constitutes one of the major components of ambient PM 10 (suspended particles with aerodynamic diameter smaller than 10 mm) in Class I National Parks and Wilderness Areas of the western United States [Malm et al., 1994, 2000a, 2000b, 2004; Malm and Sisler, 2000]. Soil mass concentrations account for more than 5% of fine 1of18

2 (smaller than 2.5 mm aerodynamic diameter) and 50% of coarse (between 2.5 and 10 mm) particle mass in the United States and as much as 20% and 90% for IMPROVE sites located in the southwest United States, respectively during [Malm et al., 2007]. Sources of dust from open sources by wind erosion or mechanical abrasion can vary greatly in spatial scale, time, location, and causes of emission. Dust emissions from travel on paved and unpaved roads are related to traffic characteristics and the quality of the road [Etyemezian et al., 2003]. Windblown dust emissions generally occur over larger spatial scales and the magnitude of dust emissions during these events can eclipse the comparatively smaller, but more regular road dust emissions [Okin and Gillette, 2001; Reynolds et al., 2001; Breshears and Allen, 2002; Whicker et al., 2002; Mahowald et al., 2007]. On a transcontinental scale, dust from regional wind storms in the Saharan and Chinese/Mongolian deserts can be transported across continents and impact areas in North America [Prospero, 1999; VanCuren and Cahill, 2002; Darmenova et al., 2005]. [3] The Clean Air Act Amendments mandated visibility protection in Class I National Parks and wilderness areas of the United States. The regional haze rule (RHR) requires the reduction of haze in Class I areas on the 20% worst visibility days ( worst days hereafter), prevention of visibility degradation on the 20% best days and natural background conditions to be met by 2065 [U.S. Environmental Protection Agency (EPA), 1999]. The IMPROVE network monitors the concentrations of aerosol constituents in a large number of Class I areas to evaluate the importance of aerosol components and their sources on reduced visibility as is defined by the aerosol extinction coefficient (b ext )[Malm et al., 1994, 2000a; Pitchford et al., 2007]. To address the goals of the RHR, States and Federal Land Managers embarked on multistate regional planning efforts to coordinate technical analyses and development of state and tribal implementation plans for improving visibility on the haziest days in Class I areas. In this frame, the Western Regional Air Partnership (WRAP) launched a large data analysis and modeling effort to determine the major causes of reduced visibility in Class I areas located in the states of Montana, North and South Dakota, Colorado, New Mexico, Arizona, Nevada, Idaho, Washington, Oregon, California, Alaska, and Hawaii and two additional sites in Texas. [4] The study presented here aims to identify the types of events that result in high concentrations of dust on the haziest days. The specific objectives of this study were: (1) to develop a methodology for identifying the possible influence of long-range transport and windblown dust sources on dust concentrations at IMPROVE sites located within the boundaries of the WRAP domain and (2) to determine the most important type of dust event that was responsible for high dust concentration during the period when dust was the dominant aerosol component of impaired visibility and the reconstructed aerosol extinction coefficient fell within the 20% worst values for daily visibility. For a given IMPROVE site, this subset of 20% worst visibility days (where dust is the largest contributor to haze) is referred to hereafter as Worst Dust Days. This paper builds on an earlier effort by Kavouras et al. [2007] to estimate the fractional contribution of locally generated windblown dust to the dust concentration measured at a site. [5] The approach relies on a number of qualitative and semiquantitative indicators including multivariate regression analysis, diagnostic ratios of chemical constituents, air mass backward trajectories, and land use and soil properties databases. These indicators were overlaid and integrated within a GIS-based mapping tool. This tool was used to individually examine the indicators for every worst dust day at the WRAP IMPROVE sites during the period. Using this brute force approach, a set of criteria was established to help differentiate between worst dust days caused by the following types of events: (1) transcontinental transport of dust from Asia; (2) locally generated windblown dust or; (3) transport of windblown dust from upwind source areas (excludes the possibility of locally generated windblown dust). Depending on the body of evidence to support the assignment of the worst dust day to one of the three event types mentioned above, a confidence rating of weak, moderate, or strong was assigned to the final assessment of the cause of each worst dust day. In many cases, the worst dust day could not be attributed to any of the above three event types with the available information. [6] The utility of this somewhat heuristic, rather laborintensive approach is twofold. First, unlike aerosols caused by other anthropogenic activities such as coal combustion or motor vehicle traffic, sources of dust are diffuse and geographically widespread. Furthermore, regional dust does not contain specific chemical markers (using current widespread techniques for chemical characterization) associated with specific sources. That is, in general, the relative abundances of common crustal elements of the shallow soil sediment that is susceptible to wind erosion are comparable for different primary dust sources geographic areas and lithologic substrates in western United States [Reheis et al., 2002; Labban et al., 2004; Reynolds et al., 2006a, 2006b]. This makes it difficult to perform source attribution using chemical profiles, though some authors have been able to extract unique signatures for cases involving transcontinental transport [VanCuren, 2003]. Large-scale dust emission models [Zender et al., 2003] and satellite observations of dust clouds [Prospero et al., 2002] can help identify the areal extent of large regional-scale events. However, those techniques cannot be used to confidently distinguish if dust measured at the receptor site was a combination of several sources (local or regional) or just a single source (either local or regional) primarily because of uncertainties in describing the conditions for dust mobilization for area sources at IMPROVE sites in the southwest United States. The methods described here allow for retrospective analysis of the types of sources that contributed to high dust concentrations in the western United States. Second, by examining the impact of certain types of dust events on multiple sites in an area, it is possible to begin to identify regional regimes for dust emissions. This information is very useful for planning agencies that must show progress toward complying with the Regional Haze Rule, determining how much of the visibility impairment owing to dust is amenable to mitigation, and underscoring gaps in knowledge with regard to high dust concentrations that cannot be associated with specific events. Examination of 2of18

3 the results of this study using the present methodology for different periods of time and geographic locations can in turn be used to fine tune the method and perhaps yield an enriched, multifaceted approach to identifying and quantifying the magnitude of different types of dust events. 2. Methodology [7] The worst dust days at seventy IMPROVE sites located in the WRAP domain as well as Big Bend and Guadalupe Mt in Texas were included in this analysis. The tools developed for examining the potential contribution of each type of dust event on a given worst dust day, the criteria used to determine the type of event that caused the worst dust days, and the level of the confidence for the event assigned to each worst dust day are described below IMPROVE Aerosol Chemistry [8] The IMPROVE network consists of automated samplers that collect aerosol samples on filters for a 24 h period on a 1 in 3 day sampling basis. Filter samples are analyzed for concentrations of ions, metals, organic carbon and elemental carbon. The reconstructed aerosol extinction for each 24-h IMPROVE sample at a given site is calculated as the sum of extinction owing to sulfate (SO 2 4 ), nitrate (NO 3 ), elemental carbon (LAC), organic carbon (OMC), fine soil (FS), and coarse mass (CM) aerosols [Malm et al., 1994, 2000a]. An updated version of the equation that relates the aerosol extinction to the size distribution of sulfate, nitrate and organic carbon, changes in Rayleigh scattering for different elevation and ambient temperatures and contribution for sea salt and NO 2, with no changes for fine soil and coarse mass has been suggested by Pitchford et al. [2007]. However, that information was not available at the time of data analysis for this study, and we utilized a previous version of the IMPROVE equation [Malm et al., 1994, 2000] to determine the aerosol extinction and the 20% worst visibility days for each site. Comparison between the two algorithms showed that the modified algorithm does not alter the characteristic of worst visibility days and has a minor effect on identifying which worst visibility days were also worst dust days for sites in the western United States. For this study, the dust concentration was defined as the sum of the fine soil (FS) and coarse mass (CM) components of aerosol. CM is obtained by subtracting the mass of particles collected on a filter with a PM 2.5 size selective inlet from the mass collected on a filter with a PM 10 size selective inlet. FS is defined operationally using the measured concentrations of metal elements found on the filter with the PM 2.5 size selective inlet as ½FSŠ ¼ 2:2½AlŠþ 2:49½SiŠþ 1:63½CaŠþ 2:42½FeŠþ 1:94½TiŠ ð1þ and applying corrections for MgO, Na 2 O, water and carbonate. A sample day is labeled as a worst dust day if the calculated aerosol extinction is among the 20% highest values for that site and calendar year and if the combined contribution of CM and FS to aerosol extinction on that day was greater than the contribution of any of the other components. Since we draw on the earlier work of Kavouras et al. [2007] for the present analysis, it is worth mentioning that the definitions utilized here are consistent with ones employed by those authors Data Sources [9] Fine particle mass, chemical composition, and coarse mass (CM) were obtained from the Visibility Information Exchange Web System ( Hourly meteorological data were retrieved from a number of networks and data sources that included the Clean Air Status and Trends Network (CASTNet), Remote Automated Weather Stations (RAWS), Arizona Department of Environmental Quality (AZDEQ), sites in NOAA s Integrated Surface Hourly database (ISH), and networks operated by NASA and the National Park Service (NPS). In some cases, supplemental PM 10 concentration records, satellite remote sensing maps and weather maps were obtained from the U.S. Environmental Protection Agency Air Quality System, level 2 and level 3 aerosol optical depth from MODIS instruments on the Terra/Aqua satellites ( nascom.nasa.gov/index.html), and Unisys Weather ( weather.unisys.com/index.html), respectively. State Soil Geographic Database (STATSGO) soil survey data created by USDA Natural Resources Conservation Service were downloaded for the entire study region Wind Erosion Potential Maps [10] We used the Wind Erosion Group (WEG) index from the STATSGO database to screen the regions that may be susceptible to wind erosion and potentially contribute to dust measured at nearby IMPROVE sites [Wolock, 1997]. The WEG index ranges from 1, representing the most erodible soil types, to 8 for the least erodible soil types. These 8 categories were reduced to three categories: WEG index between 1 and 3 for soil textures that were most likely to emit large quantities of dust, WEG index between 4 and 6 for soil textures with moderate wind erodibility, and WEG index between 7 and 8 for soil textures that were the least likely to be subject to wind erosion. The STATSGO WEG index only provides information on the susceptibility of the soil to windblown dust emission on the basis of soil textural properties. However, the actual potential for wind erosion is determined to a large extent by the land cover and the degree of surface disturbance. By sheltering the soil surface from direct exposure to the shearing action of wind, thick vegetation or other types of surface roughness can mitigate wind erosion even if the underlying soil is highly susceptible on the basis of its textural properties [Raupach et al., 1993; King et al., 2005]. To account for this effect, data from the STATSGO database were combined with information on land cover/use obtained from the National Land Cover Characterization 2001 database. The Wind Erosion Equation (WEQ) and, more recently, the Wind Erosion Prediction Systems (WEPS) have been extensively used to assess the annual soil loss by wind from agricultural fields using soil data from STATSGO and SSURGO databases, respectively; but both models are applicable to a field, or at most, a few adjacent fields and for a single crop (see L. J. Hagen, WEPS technical documentation BETA Release 95 08, 1996; available at wepsoverview.pdf). [11] The NLCD 2001 provides delineation of land use/ cover data using twenty-nine (29) categories with a 30 m by 3of18

4 30 m resolution [Homet et al., 2004]. These 29 categories were condensed into three major categories, representing areas: (1) with soil disturbance caused primarily by anthropogenic activities (human influence; includes residential/ commercial, mines and quarries, and agricultural activities); (2) that are forested and therefore very unlikely to be significant sources of windblown dust (includes forests and wetlands categories) and; (3) that are susceptible to wind erosion (includes shrubland, grasslands/herbaceous and bare rock/sand/clay categories). Shrublands and grasslands are prevalent in the desert southwest and can represent significant source areas for windblown dust especially during prolonged dry periods. When used as a map background, the nine resultant combinations of soil erodibility potential and land cover allowed for rapid assessment of the likelihood that high winds would result in dust emissions. For example, if high winds were observed to occur (e.g., through air mass backward trajectories discussed below) over forested lands, it is not reasonable to expect that there would be significant windblown dust emissions regardless of the inherent soil erodibility. On the other hand, if high winds were observed to occur over shrublands that are also characterized by a soil in the high-erodibility group (WEG between 1 and 3), then there would be substantial potential for windblown dust Elemental Diagnostic Ratios [12] A diagnostic index, the Asian Dust Score (ADS), was developed as a measure of the confidence that the ratios of certain elemental concentrations within a given aerosol sample were consistent with those expected for dust with an Asian source. The well-recorded Asian dust episode on 19 April 1998 that was detected at seventeen of the IMPROVE monitoring sites in western United States ten days later (29 April 1998), was used as the reference sample [Husar et al., 2001; Tratt et al., 2001]. For that episode, the values of Al/Si, K/Fe and Al/Ca were 0.52 ± 0.06, 0.59 ± 0.07 and 2.1 ± 0.3, respectively, which were significantly different from the typical values observed in western United States (with the averages of 0.35, 0.70 and 1.5, respectively, during ). The Asian Dust Score (ADS) was calculated as follows: ADS ¼ Q 3 i¼1 1! ð2þ 100 R ij R i;ref Eij pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi R ij Eij 2 þ s2 i;ref where i is equal to1 for Al/Si, 2 for K/Fe and 3 for Al/Ca; R i,ref and s i,ref are the reference ratio value and standard deviation calculated from the 19 April 1998 dust event, and; R ij and E ij are the ratio and propagated uncertainty of component X to component Y at a specific j-day. The ADS was designed to increase when ratios on j-day approached those of the reference sample ratios, and decrease with increasing measurement uncertainty. Thus, the ADS provides a numeric embodiment of the confidence that the elemental ratios in a given aerosol sample are close to those in the reference event, but does not provide any estimate of how much of the IMPROVE sample collected for j-day comprised Asian dust. These ratio values were comparable to those computed for soil samples from Gobi [Nishikawa and Kanamori, 1991] and Taklamakan deserts [Suzuki et al., 1993] and aerosol samples in Beijing [Wang et al., 1982] and Japan [Nishikawa and Kanamori, 1991] Multivariate Regression Analysis [13] Regression analysis of dust concentrations against wind conditions was utilized to estimate the concentration of dust associated with emission from area sources in the proximity of each site. The methodology and results of the regression analysis are presented in detail elsewhere [Kavouras et al., 2007]. Briefly, where available, a surface meteorological station from one of the networks described in section 2.2 was associated with each IMPROVE site included in this analysis. Hourly wind direction and speed data for the entire period were categorized into one of twelve wind variables (W ij ) (4 groups for wind direction 3 groups for wind speed). Each wind direction group represented 90 centered about one of the four cardinal directions, while wind speed groups corresponded to miles h 1 ( m s 1 ), miles h 1 ( m s 1 ) and >26 miles h 1 (>11.6 m s 1 ). Categorical wind data were aggregated over the 24-h period to facilitate comparison with daily aerosol data. Conditions corresponding to low winds (0 14 miles h 1, <6.2 m s 1 ) were excluded from the regression because windblown dust generally does not occur at those low wind speeds [Lyles and Krauss, 1971]. The regression analysis included all sampling days within the 3-year period for which aerosol and meteorological data were available and there was no precipitation during the day or the previous day [Kavouras et al., 2007]. The reconstructed Local Windblown Dust (LWD j ) at a specific site for day j,, was calculated as follows LWD j ¼ X b i W ij where b i are the regression coefficients of statistically significant (at p-value < 0.15) wind variables (W ij ). While the intercept was included in the regression analysis, it was omitted from equation 2 because it does not represent dust generated by winds at the site [Kavouras et al., 2007]. When the total uncertainty (the sum of the products of regression coefficient standard errors and the wind variables) of computed LWD j values was high (i.e., LWD j 2 E j 0), the local windblown dust for that site-day was set to zero [Kavouras et al., 2007]. A discussion of how the LWD parameter can be interpreted and the confidence in its value is deferred to a later section. The ratio of LWD to total measured dust (TMD = CM + FS) was used as an indicator of the fraction of dust generated from local sources owing to wind erosion Air Masses Backward Trajectories [14] The origin and the path of the air mass to the IMPROVE site was tracked using NOAA s HYSPLIT trajectory model with input from either the Eta Data Assimilation System (EDAS) and/or the hemispheric FNL meteorological fields [Draxler and Hess, 1997; Xu et al., 2006]. Air mass trajectories going back in time for two, eight and fifteen days were generated for all sites considered in this analysis every three hours at a starting height of 500 m above ground level. The model calculated the location of the air mass every hour. Thus, the distance traveled between two consecutive air mass locations was ð3þ 4of18

5 Figure 1. Analysis maps including overlays of tools used for (a) 16 April 2001 and (b) 9 May 2003 worst dust days. See section 2.7 for an explanation of symbols. related to the trajectory speed which was used as a surrogate for surface wind speed at that location. To facilitate comparison with the meteorological data, trajectory speed data were also divided into three categories, lower than 14 miles h 1 (<6.2 m s 1 ), between 14 and 20 miles h 1 ( m s 1 ) and greater than 20 miles h 1. (>8.9 m s 1 ) Source Attribution Criteria [15] The tools discussed previously were integrated into a master geographic information system (GIS) mapping utility (as shown in Figure 1) that was used to inspect every worst dust day at every site included in the analysis and 5of18

6 applying a set of criteria to determine the type of event that likely caused the worst dust day. [16] 1. The background of the maps was the combination of land use (black color for human-influenced land use, green color for forests and wetlands, and orange color for native lands) and wind erosion index (three shades of WEG categories with lighter colors corresponding to more erodible soils). [17] 2. An open yellow square was used as an indicator of the ADS (small size < 750; medium size: and large size > 1500). [18] 3. The LWD/TMD ratio at the site was represented as a filled red circle (small size < 0.25; medium size: and large size > 0.50). A large filled green circle was used for LWD/TMD ratio equal to zero. [19] 4. Three back trajectories (with trajectory points coded for wind speed; light yellow: 0 14 miles h 1 ; yellow: miles h 1 and brown > 20 miles h 1 ) corresponding to start times of 0800 local time (LT) (triangle), 1400 LT (circle), and 2000 LT (square) (Central Standard Time for all sites were superimposed). [20] 5. IMPROVE sites for which a worst dust day was not recorded were represented by a small sized filled white circle. If precipitation occurred on the sample day, an open blue inverted triangle would be assigned. Sites where a measurement was not available on that day were plotted as a blue X. [21] The primary aim of the data analysis was to use a weight of evidence approach to determine which type of event likely caused a worst dust day at a given site. There were three possible categories of events: (1) transcontinental transport of Asian dust, (2) locally generated windblown dust, and (3) windblown dust transported from upwind sources. Because the tools used for this analysis constitute a mix of semiquantitative indexes and qualitative observations, in addition to assigning an event type to a worst dust day, we also assigned a confidence level to the result. Depending on the weight of evidence, event types were considered to cause the worst dust day with weak, moderate, or strong levels of confidence. In the absence of sufficient evidence for an assignment to one of the three event types with even a weak level of confidence, a catchall undetermined event category was used Asian Dust [22] Analysis of intercontinental transport of dust from Asia on April 2001 [DeBell et al., 2004; Darmenova et al., 2005] in conjunction with the calculated ADS values, the following guidelines were adopted: ADS values lower than 750 (which corresponds to the median ADS for all worst dust days) indicated that Asian dust influence was unlikely; ADS values between 750 and 1500 suggested a possible Asian dust signature and; ADS higher than 1500 indicated a strong Asian dust signature. For days with ADS > 750 at a given site, ADS values at surrounding IMPROVE sites were also examined since a dispersed plume traveling from Asia would encompass a large area. In addition, for worst dust days where transport from Asia was a plausible explanation of the measured dust concentrations, we obtained corroborating evidence including 8- and 15-day backward trajectories, satellite images, aerosol observations, and model results from U.S. Naval Research Laboratory ( nrlmry.navy.mil/aerosol/) showing the burst of Asian dust and transport over the Pacific Ocean and Alaska. [23] The level of the confidence that long-range transport of dust from Asia was the predominant cause of a worst dust day was determined as follows: [24] 1. Strong confidence. ADS values higher than 1500 for more than one site in the WRAP domain in combination with evidence from satellite images and aerosol measurements of the developments of a dust storm in Asia days prior to the worst dust day and air mass backward trajectories originating from Asia and traveling over the Pacific Ocean. [25] 2. Moderate confidence. For cases in which the estimation of ADS was not possible at one site, but Asian dust appeared to be the causing event of reduced visibility in several surrounding sites with strong confidence, a moderate degree of confidence was assigned. A moderate degree of confidence was also allocated to cases of high ADS (>1500) at the site of interest as well as surrounding sites, but without verification of dust storms in Asian deserts. [26] 3. Weak confidence. A weak degree of confidence was given to cases in which either only one site exhibited high ADS values or multiple sites exhibited an elevated ADS score, but backward trajectories did not provide strong indications of passage over the Pacific Ocean Local Windblown Dust [27] The LWD/TMD ratio was used as an indicator of the potential impact of locally generated windblown dust. The susceptibility of the soil surface to wind erosion (on the basis of combination of WEG erodibility index and land use) near the IMPROVE site of interest was examined in conjunction with backward trajectories. The level of confidence that the worst dust day was caused in large part by windblown dust in the vicinity of the site was determined as follows: [28] 1. Strong confidence. LWD/TMD values higher than 0.50 at the site and high-speed (>20 miles/hour) trajectories over terrain with moderate or greater wind erodibility provided evidence of the impact of windblown dust from nearby sources at a high level of confidence. [29] 2. Moderate confidence. A moderate level of confidence was assigned to cases with LWD/TMD between 0.25 and 0.50 and high-speed trajectories over terrain with moderate or greater wind erodibility. The same level of confidence was also given for cases with no meteorological data available on the day of interest but that meet all of the following conditions: (1) regression analysis showed a statistically significant relationship between wind conditions and dust concentrations [Kavouras et al., 2007]; (2) backward trajectories are consistent with high wind speed over terrain with moderate or greater wind erodibility and; (3) locally generated windblown dust appeared to be the major cause of the worst dust day at surrounding sites with a high level of confidence. [30] 3. Weak confidence. For cases in which the LWD/ TMD ratio was between 0.00 and 0.25, a weak degree of confidence was assigned. [31] It is worth noting that while the LWD provides a numerical value, it represents a fairly blunt instrument intended to probe the physical relationship between high 6of18

7 winds and dust emissions at the site. Detailed discussion of the potential for these types of errors is provided by Kavouras et al. [2007]. For example, the adjective locally does not refer to any specific radial distance from the site. Rather, locally generated windblown dust is defined operationally as the amount of dust that can be attributed to high winds (measured at a specific meteorological station) on the basis of a regression analysis. Thus, there may be cases where high winds at a site are not the causative agent of dust, but are merely correlated with high winds at locations upwind where windblown dust sources may exist. Likewise, because the surface meteorological stations are not exactly collocated with the IMPROVE sites, and in some cases may be quite far, the absence of a statistical relationship between wind conditions at a meteorological station and the IMPROVE site does not necessarily preclude the possibility that high winds at the IMPROVE site would cause local dust emissions Transport From Upwind Sources [32] The passage of an air mass trajectory at relatively high speed (>20 miles h 1 ) over soil textures/land covers susceptible to wind erosion within one day of arrival at a site was used as an indicator of the possible contribution of sources upwind of the IMPROVE site exhibiting a worst dust day. Assigning a worst dust day to transport from an upwind event required that there was no indication of locally generated windblown dust at the site. That is, the transport from upwind event would not be used if there was evidence of locally generated windblown dust. The level of the confidence that dust originated from upwind sources was related to persistence of flow over moderately to highly erodible soils; beginning at least 15 h for strong confidence level, at least 8 h for moderate confidence, or at least 3 h for weak confidence. In addition to these guidelines, causes of dust at surrounding IMPROVE sites were also examined since it is expected that the scale of weather systems associated with upwind dust sources would be regional, therefore impacting multiple sites. 3. Results and Discussion [33] The rules for selecting which event type caused a worst dust day and the level of confidence to assign to that assessment were discussed in detail above. It is important to note several points regarding this type of analysis. First, while they are consistent with our understanding of the physics and chemistry of dust and dust emission, the rules we used are essentially arbitrary. They were constructed through an iterative process where a set of rules was applied to the entire data set, the results were examined and compared to known events that caused worst dust days, and then the rules were reformulated to better match the known events. Second, a different set of tools was used for different event types. Therefore, the levels of confidence assigned to the three event types are not necessarily comparable. For example, the rules for assigning a strong degree of confidence to an Asian transport event may be stricter (in some absolute reference frame) than those for assigning a strong degree of confidence to a locally generated windblown dust event. Third, the analysis was intended to determine the type of event that likely caused a worst dust day at a given site when sufficient evidence was available to do so. In other words, our aim was not to maximize the number of worst dust days that could be explained by one of the three event types considered. Thus, in general, the method was designed to be conservative overall. When in doubt, the catchall category, undetermined event, was invoked. Fourth, the three event categories do not occupy tidy, mutually exclusive domains. Wind storms that cause dust emissions are frequently regional in scale. Thus, even if a worst dust day was attributed to locally generated windblown dust, this does not preclude the possibility (perhaps likelihood) that some of the dust measured at the site was transported from an upwind source area. Evidence of locally generated windblown dust received a higher priority than evidence of transport from upwind sources which in turn received higher priority than evidence of Asian transport Identification of Types of Dust Events [34] Figures 1 and 2 show maps for 16 April 2001 (Figure 1a), 9 May 2003 (Figure 1b), 14 May 2002 (Figure 2a), and 7 June 2002 (Figure 2b). These dates correspond to worst dust days for a number of IMPROVE sites. The shading of the background and the symbols used in these maps were described earlier (section 2.7). Supporting evidence for assigning events that caused the worst dust days is provided in Figure 3 and includes 8-day composite level 3 aerosol optical depth generated by Terra satellites for the period of 7 14 April 2001 (Figure 3a), surface weather map on 9 May UTC (Figure 3b), and 24-h PM 10 mass concentrations measured in El Paso, Texas, on May 2002 (Figure 3c). These four days are discussed in detail below and serve as examples of how the methods described were applied to assign an event type to each worst dust day along with a level of confidence in that assignment. Owing to the size of the data tables, the event assignments for every worst dust day and associated maps are included in the auxiliary material. 1 In addition, a web-based tool was developed to allow users to query the maps generated for the 610 worst dust days at 70 sites over the period ( Using this tool, the event type and associated confidence level can be queried by date range, site, event type, state, and/or the time of year. [35] Twenty-nine (29) sites in the western United States exhibited worst dust days on 16 April 2001 (Figure 1a). Dust concentrations ranged from 12.8 mg m 3 at Sierra Ancha in Arizona to 51.6 mg m 3 at Death Valley in California. Asian Dust Scores (ADS) were substantially higher than 1500 ( ,740; open, large yellow squares in Figure 1a) with the highest values at sites located in the Pacific Northwest and California. These values indicated a strong Asian signature, which is in agreement with previous studies suggesting the occurrence of several Asian dust storms in March April 2001 with dust clouds reaching as far as the eastern United States [DeBell et al., 2004; Darmenova et al., 2005; Heald et al., 2006]. More specifically, the passage of two low-pressure fronts over the Gobi and Taklamakan Deserts on 6 9 April 2001 initiated the mobilization of dust up to altitudes of 8 9 km. The windblown dust caused a severe air pollution episode in 1 Auxiliary materials are available at ftp://ftp.agu.org/apend/jd/ 2008jd of18

8 Figure 2. Analysis maps including overlays of tools used for (a) 14 May 2002 and (b) 7 June 2002 worst dust days. See section 2.7 for an explanation of symbols. Gansu Province, Beijing, other parts of China, the Korean peninsula, and Japan [Xu et al., 2004]. The resultant dust cloud arrived in Alaska and northern America on April 2001 (Figure 3a). This body of evidence, combined with the high ADS values resulted in the assignment of transcontinental transport from Asia as the event that caused worst dust days at most of the sites in Figure 1a. Although calculated ADS values for some of the sites (Bliss and Lava Beds in California, Ike s Backbone in Arizona, Mesa Verde in Colorado, and Brooklyn Lake in Wyoming) were between 8of18

9 Figure 3. (a) Eight-day composite level 3 aerosol optical depth generated by Terra satellites for the period 7 14 April 2001 (see (b) surface weather map on 9 May 2003 (see and (c) 24-h PM 10 mass concentrations measured in El Paso, Texas, in May 2002 (U.S. Environmental Protection Agency Air Quality System). 750 and 1500 (midsized yellow square), the confidence level in Asian transport as the event causing the worst dust days at those sites was still moderate to strong, owing to the overwhelming evidence of widespread Asian dust influence. [36] Two sites (located on the southeast end of WRAP domain), Salt Creek Wilderness in New Mexico and Guadalupe Mt in Texas, showed only a moderate ADS. For Salt Creek, local windblown dust represented 74% of total measured dust (large filled red circle in Figure 1a; LWD/ TMD = 0.74). This was supported by air mass trajectories moving at fairly high speed over erodible terrain (smallsized filled yellow or dark brown circles/squares/triangles in 9of18

10 Figure 1a). Therefore, this worst dust day at Salt Creek was attributed to locally generated windblown dust with a strong degree of confidence. At Guadalupe Mountain in Texas, surface meteorological data indicated that winds at the site were not of sufficient force to cause windblown dust and back-trajectories indicated that high winds were possible over portions of Mexico. However, those high winds represent activity that was more than 24 h prior to the worst dust day. Therefore, using the guidelines mentioned previously, the cause of dust at this site was completely undetermined for this worst dust day. Note that other sites in the south and southeastern part of the domain (e.g., Big Bend in Texas, Saguaro East/West in Arizona, White Mountain in New Mexico) did not experience reduced visibility caused by dust on this day (as indicated by the small, open circles on Figure 1a). [37] Dust concentrations on 9 May 2003 (Figure 1b), resulted in worst dust days at Great Sand Dunes and Mesa Verde, and Weminuche Wilderness in Colorado and San Pedro National Park and Bosque del Apache Wilderness in New Mexico. As the medium and large-sized red dots on the map indicate, these sites showed a strong correlation between dust concentrations and local high-speed surface winds from the south and southwest [Kavouras et al., 2007]. The estimated LWD accounted for more than 67% of measured dust at all these sites (LWD/TMD > 0.67). Back-trajectories indicated high-speed air masses for periods of 2 12 h prior to arrival of the air mass at the site. This was consistent with surface weather maps showing the presence of a front the Four Corners region and moderate winds in central Arizona in New Mexico (Figure 3b). For these five sites, the local windblown dust was the event assigned to the worst dust days with a strong level of confidence. [38] Fourteen (14) sites located Arizona, New Mexico and Colorado encountered low visibility caused by dust on 14 May 2002 (Figure 2a). For all sites, the ADS was low (<1000), suggesting a negligible contribution of Asian dust. The LWD/TMD ratio values for Tonto in Arizona and Weminuche in Colorado were 0.26 and 0.27, respectively and back-trajectories indicated that there were high-speed winds in the vicinity of these sites. Thus, for these two sites the cause of the worst dust day was assigned to locally generated windblown dust with a moderate degree of confidence. [39] Trajectory analysis for the remaining sites with worst dust days showed moderate-to-high-speed trajectories over areas with moderate to high erodibility in Paso del Norte region, west/northwest Texas and Great Plains. The LWD calculated for these sites was not statistically different from zero. This indicated that there was significant dust contribution from sources located in regions upwind of the sites. This was further supported by the variation of aerosol concentrations (Figure 3c) in El Paso (near the westernmost point of Texas), which showed a large spike in PM 10 mass concentrations on 13 May 2002, about h away in terms of transport time from the sites with worst dust days (consecutive back-trajectory dots represent 1 h of transport). As a result, a strong level of confidence was assigned to the upwind transport category for these sites. [40] Figure 2b shows a typical example of a worst dust day (7 June 2002) at several sites (only for sites in Arizona) for which none of the tools developed in this study provided indicative information of the origin of dust. Eleven (11) sites located in east Arizona, west/central New Mexico and in Wyoming exhibited low visibility. The highest dust concentration was measured at Bosque del Apache in New Mexico (51.9 mg m 3 ); the ADS value was For the other sites, dust concentrations on 7 June 2002 ranged from 10.1 in Grand Canyon to 23.0 mg m 3 in Chiricahua. The ADS and LWD/TMD ratio values were insignificant and trajectory analysis showed no strong evidence of transport from upwind sources. Analysis of air quality data from CASTNET and U.S. EPA AQS networks did not show any significant differences for that day as compared to the remaining days of June As a result, these worst dust days were all classified as having undetermined events. It is worthy to note that (1) each IMPROVE sampling day in June was associated with a worst dust day in one or more sites in Arizona and (2) the vast majority of worst dust days that occurred in summer months were caused by undetermined events. It is possible that these worst dust days are caused by sources where the mechanical suspension is the main mechanism of dust emission. This could include anthropogenic activities such as driving on unpaved roads, off-road recreational activities, and some agricultural operations (e.g., tilling). Although, these types of sources are generally confined in time and space, their contribution to dust levels may be important under certain weather condition such as local (low wind speed) circulation favored by high-pressure systems over Arizona and New Mexico in early June. In warm climates, resuspension of dust can also occur by convective vortices (dust devils). Renno et al. [2004] estimated that dust fluxes in dust devils are on the order of 1 g/m2s, which is one order of magnitude higher than the vertical fluxes during dust storms. However, dust devils tend to occur over fairly small areal extents. Dust particles can also be released into the atmosphere because of convective flows that result in the development of short-lived wind gusts. This phenomenon is more pronounced during summer as heating of the surface is more intense [Cakmur et al., 2004]. In any case, the present analysis does not allow for determining the individual or combined effects of these types of activities or natural events and we present these explanations here only as hypotheses that require verification Patterns of Events Causing Worst Dust Days [41] For the 70 IMPROVE sites included in this study, a total of 610 of the % worst-case visibility days were caused by elevated dust concentrations. A detailed analysis of the characteristics and the spatiotemporal patterns of worst dust days in western United States is presented elsewhere [Kavouras et al., 2007]. The number of worst dust days assigned to one of the three types of dust events at strong and moderate confidence levels is depicted in Figure 4. Also shown in Figure 4 are cases where the event was identified at weak confidence. The tools described above provided sufficient information to assign 496 days (81% of all worst dust days) to one of the three types of dust events with some confidence. The majority of those cases (313 worst dust days, about 51% of total worst dust days) were attributed to one of the three event types with a confidence level of moderate or strong. For the purposes of discussion, events assigned at a moderate or strong confi- 10 of 18

11 Figure 4. Number of worst dust days attributed to one of the three types of dust events at a confidence level of high (includes strong and moderate categories) and low (corresponds to weak category). dence level are classified as high, while weak confidence levels are shown as low in Tables 1 3 and Figure 6. [42] Aeolian dust from windblown sources, either local to the site or from areas upwind was responsible for 441 worst dust days (72% of all worst dust days with local windblown dust responsible for 201 and upwind transport responsible for 240 worst dust days, respectively). Most of these events (265) were assigned to worst dust days with a fairly high level of confidence (Figure 4). Long-range transport of dust released into the air over Asian deserts was responsible for reduced visibility during 55 worst dust days over the period, in most cases with high level of confidence. This represented about 10% of all worst dust days. However, Asian dust influence was not determined for any of the IMPROVE samples in 2003 owing to laboratory issues related to the analysis of Al, a critical component for the calculation of Asian Dust Score. Thus, the influence of Asian dust is likely understated, though it is unknown to what degree. By state, assigned worst dust days accounted for more than 80% in Alaska (4 of 5 worst dust days), Colorado (73 of 87), Idaho (11 of 13), Montana (16 of 19), New Mexico (73 of 79), South Dakota (9 of 9), Texas (27 of 30), Utah (36 of 41), Washington (6 of 6) and Wyoming (14 of 14). A smaller fraction worst dust days was accounted for in California, especially sites in southern California (47 of 73 worst dust days), Arizona (155 of 203 worst dust days), Nevada (12 of 16 worst dust days), Oregon (11 of 14 worst dust days), and North Dakota (1 of 2 worst dust days). [43] The IMPROVE sites were grouped into ten geographic clusters (nine on continental United States and one for Alaska) based roughly on the number of worst dust days, and the distribution of event types including undetermined events. Table 1 shows the sites associated with each cluster and the number of worst dust days associated with each event at high and low levels of confidence. The ten clusters and the percentage of worst dust days per type of dust event are shown in Figure 5. Death Valley, comprising its own cluster, was the only cluster in the southwest part of the WRAP domain for which locally generated windblown dust explained the majority of the worst dust days (17 of 24). Five worst dust days were associated with transport from source areas located in northern Nevada. A large playa is within view of the IMPROVE site and westerly winds appear to cause windblown dust to be suspended from the playa surface [Kavouras et al., 2007]. [44] The southeast region cluster, encompassing sites in Texas and most of the sites in New Mexico and southern Colorado, demonstrated a similar pattern, with local windblown dust being the likely event that caused about half of worst dust days. Transport from upwind sources located either in northern Mexico (southerly winds) or north/central Texas and the Great Plains (northerly/northeasterly winds) was associated with low visibility for 33% of the worst dust days. The northern (14 sites located in Idaho, Montana, Wyoming, North Dakota, South Dakota, and the Rocky Mountain site in Colorado) and central (4 sites in Utah and Nevada) region clusters showed comparable distributions of events causing worst dust days with local windblown dust representing 30 33%, upwind transport representing 33 37%, Asian influence representing 19 21%. [45] Sites in Arizona and California were primarily contained in four clusters. For the northern and Southern California clusters, the events leading to more than 40% of worst dust days were entirely undetermined. Local or regional area sources of windblown dust were responsible for about one-third (7% local windblown dust; 25% upwind transport) of worst dust days at those sites. There are a number of possible reasons for this and we provide a partial list here. Sources of coarse particles in California can include sea spray and epicuticular waxes from plants in northern California. These plant materials contain primarily organic compounds, but get incorporated in the calculation of dust concentration via the coarse mass fraction, which is not chemically speciated at IMPROVE sites. However, Malm et al. [2007] showed that particulate organic matter in Mt. Rainier in Oregon accounted for about 60% of coarse particle mass but less than 25% at the Sequoia and San Gorgonio sites in Southern California. Soil dust was a major constituent of coarse particles, at least in Southern California. Considering the large number of worst dust days with undetermined causes, this suggests that nonwindblown sources may have a significant impact in the region. These can include agricultural activities, construction, and travel on paved and unpaved roads. The usual harvesting dates for potatoes and wheat (durum and winter) in Arizona are between 15 May and 1 July, while agricultural activities during the same period in California, include planting of corn, beans and sugar beets and harvesting of barley, hay, sugar beets and wheat [U.S. Department of Agriculture (USDA), 1997]. [46] Sites in the central and north Arizona clusters were also characterized by a large number of undetermined worst dust days. This was more pronounced for the central Arizona cluster, where sites are located in the vicinity of the Phoenix and Tucson urban areas and along the Interstate 10 highway, an area characterized by extensive agricultural facilities and rapid expansion. These areas and related activities may have a direct impact on the measured dust concentrations at IMPROVE sites. It is also possible that the same activities may disturb soil surfaces and increase the potential for local and regional wind erosion. Local windblown dust accounted for 25 worst dust days (representing about 23% of total worst dust days in the central Arizona 11 of 18