Assessment of the local windblown component of dust in the western United States



[1] We estimated the contributions of windblown dust from nearby area sources to dust concentrations at Class I areas in the western United States including Alaska and Hawaii. The approach utilized multivariate linear regression of dust concentrations against categorized wind conditions (wind direction and speed) for all 2001–2003 data for 70 Interagency Monitoring of Protected Visual Environments sites. Statistically significant associations between dust concentrations and at least one of the wind variables were found at 41 sites with correlation coefficients as high as 0.97. At some sites, primarily in New Mexico and Texas, windblown dust from nearby sources accounted for up to 3 μg m−3 over the 2001–2003 period. In addition, the impact of local windblown dust sources during the 20% worst visibility days when dust was the major component of visibility reduction (worst dust days) was examined. A total of 608 worst dust days were identified for 2001–2003, mostly at Class I areas in southwestern states during spring and summer with 24-h average dust concentrations as high as 153 μg m−3. Windblown dust from local sources was present with statistical confidence on many of the worst dust days at sites in New Mexico, Utah, Colorado, southern Texas, and Death Valley in California. A smaller percentage of worst dust days were associated with local windblown dust in Arizona and other sites in southern California, suggesting either nonwindblown or distant sources of dust. The methods discussed can serve as a useful, semiquantitative tool for identifying sites where local wind conditions affect dust concentrations.

1. Introduction

[2] Mineral dust is an important component of atmospheric aerosol on a global scale that influences the terrestrial radiation balance, atmospheric chemistry, air quality, and visibility. Erosion by wind in arid regions of the world is the primary pathway for the release of dust particles into the atmosphere [Tegen and Fung, 1995; Prospero, 1999; Nordstrom and Hotta, 2004]. Globally, windblown dust emission estimates, which vary from 1018 to 2150 Tg yr−1, are largely responsible for the 17–36 Tg of dust present in the atmosphere [Penner et al., 2001; Ginoux et al., 2001; Zender et al., 2003a, 2004]. In the Northern Hemisphere, the most conspicuous area sources for dust are located in North Africa (∼ 980 Tg yr−1) and Asia (415 Tg yr−1) including the Arabian Peninsula. Patagonia in Argentina (35 Tg yr−1) and deserts in Australia (37 Tg yr−1) are the major sources of windblown dust in the Southern Hemisphere [Ginoux et al., 2001; Prospero et al., 2002; Zender et al., 2003a]. Dust emissions from the North American continent are comparatively modest with estimates for the arid southwestern United States (including Texas and the Great Plains) and Mexico being in the vicinity of 8 Tg yr−1 [Zender et al., 2003a].

[3] The major sources of dust in the western USA include windblown emissions [Okin and Gillette, 2001; Reynolds et al., 2001; Breshears and Allen, 2002; Whicker et al., 2002], agricultural activities [Zobeck, 1991; Nordstrom and Hotta, 2004], travel on paved and unpaved roads [Etyemezian et al., 2003], mining and storage operations, and regional transport or long-range transport from Mexico, Canada, and Asia [Bates et al., 2004; Darmenova et al., 2005; VanCuren, 2003]. Dust originating from the African continent exerts a minimal influence on the western USA compared with dust transported from Asia [Prospero, 1999]. Source regions of windblown dust in the western USA most likely include large areas of shrub land and grass land [Okin and Gillette, 2001; Reynolds et al., 2001; Breshears and Allen, 2002; Whicker et al., 2002], unpaved roads, and agricultural land, especially in arid areas that are susceptible to soil wind erosion such as the southern high plains in Texas and southern California [Hagen and Woodruff, 1973; Clausnitzer and Singer, 1996; Etyemezian et al., 2003].

[4] Several factors can influence the extent of soil wind erosion and dust emission into the atmosphere. They include the inherent susceptibility of the soil to aeolian transport processes, the frequency, magnitude, and direction of wind, moisture content/snow cover, the presence/absence of nonerodible elements (for example, gravel, vegetation), aggregate formation, surface crusting, level of disturbance, and the supply and availability of sand-sized particles in the sediment [Gillette, 1999]. Emissions of windblown dust have been modeled to occur only after a minimum threshold value for wind speed is reached [Lyles and Krauss, 1971; Gillette, 1999]. Commonly, wind erosion and dust emission are assumed to be related to the portion of the wind speed that is in excess of this threshold. This results in noncontinuous emissions during episodic, high-wind events and zero emissions when wind speed is below the threshold [Belnap and Gillette, 1998]. Estimates of threshold friction velocity for bare and seriously disturbed soils demonstrate greater erodibility relative to undisturbed areas or surfaces with nonerodible roughness elements [Gillette et al., 1980; Raupach et al., 1993; Gillette and Chen, 2001].

[5] Dust aerosol constitutes more than 20% of ambient PM10 mass (suspended particles with aerodynamic diameter smaller than 10 μm) in many Class I National Parks and Wilderness Areas of the western USA [Malm et al., 1994, 2000a, 2000b, 2004; Malm and Sisler, 2000]. These “Class I”-designated areas are afforded visibility protection by the Clean Air Act amendments passed by the US congress in 1977. This protection is embodied in the “Regional Haze Rule” [US Environmental Protection Agency, 1999] which requires reduction of haze on the 20% worst visibility days for Class I areas and the prevention of further visibility degradation on the 20% best days. “Natural background” conditions are to be met by 2065. The Interagency Monitoring of Protected Visual Environments (IMPROVE) network utilizes the measured concentrations of aerosol constituents to numerically infer or reconstruct light extinction (bext), a metric of visibility, through a simple linear combination [equation (1)] of the individual light extinction contributions from aerosols consisting of sulfate (SO42−), nitrate (NO3), organic carbon (OMC), elemental carbon (LAC), fine soil (FS), and coarse mass (CM) [Malm et al., 2000a; Xu et al., 2006].

equation image

2. Approach

[6] In this paper, an empirical, semiquantitative approach for estimating the contributions of local windblown dust to ambient PM10 is presented. Aerosol concentrations over the period 2001–2003 from IMPROVE monitors in the western USA are combined with data from surface meteorological networks to extract relationships between surface wind speed and wind direction and windblown dust generated in the vicinity of the IMPROVE monitoring site. A multivariate linear regression analysis method is used to (1) determine and evaluate the relationships between wind conditions and dust concentrations at an IMPROVE site; (2) estimate the contribution of windblown dust generated in the vicinity of the site, and; (3) assess the spatial and seasonal importance of windblown dust emissions for visibility impairment. Owing to their regulatory importance under the Regional Haze Rule, we examined 20% worst visibility days when dust was the primary cause of visibility extinction in some depth. To our knowledge, this type of ambient data analysis has not been reported in the literature previously and can be used to complement existing modeling methods which rely on either measurement or modeling of windblown dust emissions at the source. This analysis is part of a larger effort to enumerate the major causes and frequencies of occurrence of dust-resultant haze in relatively remote regions of the western USA. Kavouras et al. [2005] have utilized the methods described here to ascribe elevated dust at IMPROVE sites to transcontinental transport, locally generated windblown dust, or windblown dust transported from regions upwind of the monitoring site.

[7] The analysis began with 70 selected sites from the IMPROVE network located in the Western Regional Air Partnership (WRAP) domain and the state of Texas (Figure 1). This preliminary list of sites was based on the availability of aerosol data over the 2001–2003 period as well as the potential to obtain surface meteorological data from nearby monitoring networks. Filters collected by the IMPROVE network, are analyzed for PM10 and PM2.5 (particulate matter with aerodynamic diameter less than 10 and 2.5 μm, respectively) mass concentration as well as for metals, sulfate, nitrate, organic and elemental carbon content in the PM2.5 size fraction. In this paper, we adopt an operational definition of dust where it is assumed equal to the sum of FS and CM as measured by IMPROVE monitors. CM is defined as the difference between PM10 and PM2.5, and FS is calculated from a linear equation [equation (2)] based on the measured concentrations of five metals (Al, Si, Ca, Fe, and Ti). The assumptions behind equation (2) are explained by Malm et al. [1994, 2000a, 2000b].

equation image

Note that the definition of dust adopted for this study presumes that particles larger than 2.5 μm in aerodynamic diameter are primarily associated with geologically derived material. This assumption is never fully true and in some cases may be quite erroneous. For example, at the coastal site of Point Reyes, California, the CM concentration on 18 June 2001 is reported to be in excess of 100 μg m−3 while FS is quite low (< 1 μg m−3). It is improbable that there would be such a high concentration of crustal CM at a coastal site, especially in the absence of a commensurately large amount of FS. Possible errors associated with this assumption are discussed in the context of the present study in a later section.

Figure 1.

Map showing the locations of IMPROVE sites in the western United States including Alaska. The open circles indicate sites with statistically significant association between observed dust concentration and at least one of the wind variables, and the solid triangles represent sites with no associations between dust and wind conditions for the 2001–2003 period.

[8] For a specific site, the subset of days when dust, through the combined effect of FS and CM [equations (1) and (2)], is the largest contributor to reconstructed visibility extinction on a 20% worst visibility day are referred to as “worst dust days”. Worst visibility days and worst dust days are determined on a site-by-site and year-by-year basis. Consequently, the number of worst dust days at a site does not necessarily correlate directly with the absolute concentration of dust at that site.

3 Dust Aerosol Data

3.1 Annual Average Dust Concentrations in the Western United States

[9] All particle mass and elemental concentrations used in this study were retrieved from the Visibility Information Exchange Web System Column A of Table 1 shows the mean and standard deviation (σ) of 24-h dust concentrations measured on all sample days for the 70 IMPROVE and IMPROVE protocol sites in the WRAP domain in addition to two sites in Texas over the 2001–2003 period (Figure 1). Among IMPROVE sites, mean dust concentrations ranged from 1.4 (σ = 1.5) μg m−3 at Denali National Park in Alaska to 16.8 (σ = 14.5) μg m−3 at Saguaro West in Arizona (Table 1). Noting that there were variations of dust levels among sites within a state and that the density of IMPROVE sites varies among states, in general, the highest dust concentrations were in Texas (Geom. Mean = 12.0 μg m−3), Arizona (Geom. Mean = 10.0 μg m−3), New Mexico (Geom. Mean = 6.5 μg m−3), California (Geom. Mean = 6.5 μg m−3), and North and South Dakota (Geom. Mean = 6.7 and 5.6 μg m−3, respectively). This spatial pattern of dust concentrations is consistent with spatial characteristics of soil erodibility in the USA and with previous studies [Gillette, 1999; Prospero et al., 2002; Zender et al., 2003b]. The mean dust concentration at the Phoenix Arizona site (28.0 μg m−3) was comparatively much higher, but that site is not strictly part of the IMPROVE network (though sampler and analysis follow the same protocol) and is most likely influenced by emissions of dust from local urban sources (for example, construction, motor vehicle travel on paved and unpaved roads).

Table 1. Dust Concentrations and Calculated LWD for the 2001–2003 Period for 70 IMPROVE Sites in the Western USAa
24-h Dust Concentration for All Available Data From 2001–2003 n, Mean (σ)Correlation Coefficient (R) Dust Versus Wind [equation (3)]Mean LWD for Days When LWD − 2E > 0 n, Mean (σ)Mean LWD for All Days. (LWD Assumed 0 if LWD − 2E < 0) Mean (σ)
  • a

    Format: Number of days (n), mean, and standard deviation (σ). Standard deviation represents variation of concentration (or LWD) over the n days and not regression error.

Alaska (Geom. Mean = 2.3 μg m−3)
Denali333, 1.4 (1.5)
Simeonof228, 4.4 (3.5)0.5919, 2.2 (2.3)0.2 (0.9)
Trapper Creek271, 1.7 (1.6 )
Arizona (Geom. Mean = 10.0 μg m−3)
Chiricahua357, 8.5 (10.0)0.384, 26.9 (16.3)0.4 (3.5)
Grand Canyon325, 3.4 (3.7)
Hillside267, 6.7 (6.5)0.7255, 3.0 (4.4)0.7 (2.5)
Ike’s Backbone330, 6.6 (6 1)0.3021, 3.2 (5.4)0.7 (2.5)
Mount Baldy314, 4.0 (4.1)0.5571, 2.1 (1.9)0.2 (0.9)
Queen Valley314, 13.4 (11.8)0.5339, 5.5 (4.6)0.8 (2.6)
Phoenix297, 28.0 (14.9)
Saguaro East313, 10.0 (9.7)0.1120, 9.9 (12.9)1.0 (5.0)
Saguaro West258, 16.8 (14.5)0.2424, 4.6 (3.4)0.3 (1.4)
Sierra Ancha302, 6.6 (7.5)0.0819, 4.9 (3.1)0.3 (1.4)
Tonto328, 8.7 (9.7)0.184, 13.1 (8.7)0.2 (1.7)
California (Geom. Mean = 6.5 μg m−3)
Aqua Tibia299, 10.7 (7.9)
Bliss310, 2.6 (3.1)0.3272, 0.7 (0.8)0.3 (0.6)
Death Valley334, 9.8 (9.2)0.49113, 6.0 (6.2)2.1 (4.7)
Dome Lands307, 9.3 (9.9)
Hoover273, 4.3 (8.7)0.0272, 1.9 (4.6)0.9 (2.9)
Joshua Tree335, 8.3 (8.1)1.003, 39.2 (27.1)0.4 (4.3)
Lava Beds336, 2.9 (2.9)0.027, 3.2 (1.7)0.1 (0.5)
Lassen Volcanic337, 2.5 (2.9)
Pinnacles342, 6.3 (3.8)
Point Reyes249, 8.5 (8.3)
San Gabriel234, 8.5 (23.7)
San Gorgonio331, 6.7 (5.8) 
Sequoia307, 10.5 (8.4)
Trinity300, 2.8 (4.4)
Yosemite333, 4.2 (3.2)
Colorado (Geom. Mean = 5.0 μg m−3)
Great Sand Dunes339, 7.3 (8.1)0.6246, 8.8 (7.9)1.2 (4.2)
Mesa Verde314, 6.9 (9.9)0.708, 34.4 (22.0)1.1 (7.0)
Rocky Mt.357, 4.6 (4.6)
Weminuche347, 3.5 (3.6)0.6535, 4.7 (4.4)0.9 (2.7)
White River331, 3.0 (4.0)0.3635, 4.3 (2.8)0.5 (1.6)
Hawaii (Geom. Mean = 1.6 μg m−3)
Hawaii Volc.341, 1.6 (1.2)
Idaho (Geom. Mean = 3.1 μg m−3)    
Craters Moon319, 4.0 (4.3)0.36156, 1.6 (1.9)0.8 (1.6)
Sawtooth314, 2.2 (2.5)0.815, 2.7 (1.6)0.3 (1.0)
Montana (Geom. Mean = 5.0 μg m−3)
Glacier298, 5.4 (7.8)
Monture336, 3.4 (4.8)
Medicine Lake327, 6.3 (6.0)0.3035, 3.1 (3.9)0.4 (1.8)
UL Bend330, 4.6 (5.1)0.544, 10.2 (5.7)0.1 (1.4)
North Dakota (Geom. Mean = 6.7 μg m−3)
Lostwood355, 6.7 (6.1)0.3328, 3.5 (3.6)0.3 (1.4)
Th. Roosevelt310, 6.8 (5.9)0.6644, 3.7 (3.6)0.7 (2.1)
New Mexico (Geom. Mean = 6.5 μg m−3)
Bandelier336, 4.0 (3.8)0.6553, 3.6 (4.4)0.6 (2.2)
Bos. del Apache309, 7.8 (8.0)0.977, 26 (24.8)0.6 (5.3)
Gila320, 3.5 (3.8)0.6117, 3.0 (2.7)0.2 (1.0)
Salt Creek340, 13.0 (12.8)0.67101, 7.8 (12.3)3.0 (8.6)
San Pedro292, 3.2 (3.7)0.8915, 7.3 (7.0)0.4 (2.3)
White Mt219, 7.6 (10.2)0.53128, 5.1 (6.8)3.0 (5.8)
Nevada (Geom. Mean = 4.0 μg m−3)
Great Basin337, 4.0 (6.1)
Oregon (Geom. Mean = 3.2 μg m−3)
Kaliompsis348, 2.9 (4.2)0.1332, 1.7 (1.1)0.2 (0.6)
Mount Hood344, 2.0 (3.1)
Starkey358, 5.4 (8.6)
Three Sisters354, 2.6 (3.2)
South Dakota (Geom. Mean = 5.6 μg m−3)
Badlands352, 6.7 (6.0)
Wind Cave327, 4.5 (3.9)
Texas (Geom. Mean = 12.0 μg m−3)
Big Bend216, 9.3 (10.1)0.5041, 10.5 (10.4)1.5 (5.3)
Guadalupe Mt228, 14.6 (14.8)0.37155. 6.6 (5.7)2.7 (4.9)
Utah (Geom. Mean = 4.9 μg m−3)
Bryce Canyon305, 4.1 (5.0)0.8082, 2.6 (6.0)0.7 (3.3)
Canyonlands325, 5.0 (5.7)0.7626, 8.1 (8.1)0.7 (3.3)
Zion343, 5.5 (5.8)0.9212, 11.8 (10.8)0.5 (3.2)
Washington (Geom. Mean = 4.6 μg m−3)
Columbia RG306, 7.5 (9.3)0.6270, 12.4 (7.6)6.1 (8.2)
Mt. Rainier329, 2.7 (2.9)
Pasayten335, 2.1 (2.9)0.603, 6.3 (6.0)0.1 (0.8)
Puget Sound282, 6.6 (4.0)0.315, 3.8 (2.4)0.1 (0.7)
Snoqualamie352, 1.6 (1.7)
Spokane214, 10.0 (10.1)0.577, 13.3 (7.5)0.6 (3.2)
Wyoming (Geom. Mean = 2.8 μg m−3)
Brooklyn Lake331, 2.3 (2.3)
North Absaroka301, 3.6 (3.0)
Yellowstone320, 2.5 (3.6)

3.2 Dust Concentrations During Worst Dust Days

[10] Column A of Table 2 shows average dust concentrations but only for worst dust days. Overall, more than one fifth (608 sample days at 70 IMPROVE sites) of 20% worst visibility days were classified as worst dust days [dust was the largest contributor to bext according to equation (1)]. Average dust concentrations during the worst dust days ranged from 9.5 μg m−3 (σ = 2.5 μg m−3) at Denali, Alaska, to 153.2 μg m−3 (σ = 168.4 μg m−3) at San Gabriel, California. In general, dust concentrations on worst dust days at a site were from 3 to 10 times the average concentrations over the entire 2001–2003 period at that same site, emphasizing that dust can have a large impact on visibility through episodic events. By state, there were 204 worst dust days in Arizona (31.9% of worst visibility days at 11 sites), 86 in Colorado (25.7% of worst visibility days at 5 sites), and 79 in New Mexico (22.3% of worst visibility days at 6 sites). In contrast, the number of worst dust days and the fraction of worst visibility days for which they accounted for were much less in South Dakota (9 days at two sites), Washington (5 days at five sites), Alaska (5 days at three sites), and North Dakota (2 days at two sites). Interestingly, there were only 72 worst dust days (6% of worst visibility days at 15 sites) in California, though there were large site-to-site variations. For example, at Death Valley California, the number of worst dust days was 24 and represented two thirds of all worst visibility days at the site. In contrast, worst dust days represented a small fraction of 20% worst visibility days for sites located south of Los Angeles (for example, San Gorgonio, Joshua Tree, Agua Tibia; Table 2), and there were no worst dust days at Lava Beds in Northern California. Overall, these data emphasize the significance of dust in the southwestern states of Arizona, New Mexico, Colorado, and Utah, as well as portions of southern California.

Table 2. Dust Concentrations and Calculated LWD for Worst Dust Days in the 2001–2003 Perioda
24-h Dust Concentration for 2001–2003 Worst Dust Days n, Mean (σ)Mean LWD on Worst Dust Days When LWD − 2E > 0 n, Mean (σ)Mean LWD for All Worst Dust Days (LWD Assumed 0 if LWD − 2E < 0) Mean (σ)Group
  • a

    Format: Number of days (n), mean, and standard deviation (σ). Standard deviation represents variation of concentration (or LWD) over the n days and not regression error.

Denali2, 9.5 (2.5) 
Simeonof1, 20.3 
Trapper Creek2, 13.2 (3.6) 
Chiricahua30, 32.6 (18.9)4, 26.9 (16.3)4.1 (11.4)1
Grand Canyon10, 15.1 (7.6) 
Hillside19, 22.4 (10.5)7, 5.4 (5.0)2.1 (4.0)2
Ike’s Backbone20, 22.0 (10.6)2, 2.9 (3.1)0.3 (0.9)2
Mount Baldy7, 19.3 (12.4)2, 6.0 (4.5)2.0 (3.7) 
Queen Valley23, 39.3 (26.2)2, 10.8 (4.1)1.3 (3.7)2
Phoenix10, 65.7 (30.9) 
Saguaro East24, 31.7 (22.2)5, 4.8 (4.6)1.1 (2.9)2
Saguaro West28, 43.6 (27.9)6, 16.8 (22.2)4.4 (13.0)2
Sierra Ancha13, 29.4 (20.9)1, 2.50.2 (0.7)2
Tonto20, 34.5 (23.4)3, 8.7 (0.1)1.3 (3.2)2
Aqua Tibia3, 59.5 (24.9) 
Bliss3, 24.6 (11.3)2, 1.3 (0.1)0.8 (0.7) 
Death Valley24, 30.7 (14.4)17, (10.1) 13.17.1 (11.9)1
Dome Lands12, 36.7 (23.4) 
Hoover7, 39.2 (39.9)2, 2.9 (3.1)1.0 (2.1) 
Joshua Tree6, 50.9 (23.0)1, 70.511.7 (28.8) 
Lava Beds1, 33.31, 3.73.7 
Lassen Volcanic5, 18.0 (5.4) 
Point Reyes1, 101.0 
San Gabriel3, 153.2 (168.4) 
San Gorgonio2, 32.7 (4.9) 
Sequoia2, 30.2 (16.2) 
Trinity2, 44.2 (28.5) 
Yosemite1, 17.9 
Great Sand Dunes30, 25.8 (14.4)12, 16.9 (8.5)6.7 (9.9)1
Mesa Verde27, 30.9 (19.7)6, 34.5(23.3)10.9 (20.6)1
Rocky Mt.7, 22.7 (8.4) 
Weminuche14, 14.1 (4.5)10, 8.6 (5.7)7.2 (6.1)1
White River9, 14.3 (6.0)5, 5.9 (3.9)3.3 (4.2) 
Craters Moon11, 17.5 (4.9)8, 5.2 (6.2)3.8 (5.7) 
Sawtooth2, 16.7 (4.4)1, 4.92.5 
Glacier3, 33.0 (13.5) 
Monture7, 18.2 (5.0) 
Medicine Lake3, 31.6 (12.5)2, 1.3 (0.4)0.6 (0.4) 
UL Bend6, 23.2 (7.8)2, 13.6 (6.6)4.5 (7.6) 
North Dakota
Th. Roosevelt2, 34.0 (10.5) 
New Mexico
Bandelier13, 17.0 (6.3)8, 8.1 (7.6)5.0 (7.1)1
Bos. del Apache15, 30.4 (19.0)5, 31.6( 27.9)10.5 (21.5)1
Gila7, 21.4 (7.8)2, 5.8 (6.6)1.9 (4.2) 
Salt Creek22, 45.3 (21.5)20, 21.4 (19.6)20.4 (19.7)1
San Pedro12, 15.9 (7.1)4, 13.9 (11.7)4.6 (9.2)1
White Mt10, 43.1 (23.2)8, 15.1 (10.6)12.1 (11.3) 
Great Basin16, 18.8 (21.6) 
Mount Hood3, 19.1 (6.0) 
Starkey8, 42.7 (17.2) 
Three Sisters2, 24.2 (9.3) 
South Dakota
Badlands8, 27.9 (7.8) 
Wind Cave1, 18.4 
Big Bend7, 46.7 (12.8)5, 18.1 (9.7)12.9 (11.8) 
Guadalupe Mt23, 45.1 (17.3)12, 9.1 (7.3)6.7 (9.9)1
Bryce Canyon9, 21.2 (11.4)5, 9.9 (18.1)5.5 (13.8) 
Canyonlands15, 20.9 (10.6)6, 13.4 (10.4)5.7 (9.4)1
Zion17, 21.4 (10.9)3, 19.6 (10.3)3.7 (8.8)1
Columbia RG2, 74.5 (7.8)2, 25.0 (4.2)25.0 (4.2) 
Pasayten2, 23.9 (10.0) 
Spokane2, 40.6 (19.8) 
Brooklyn Lake4, 11.9 (3.4) 
North Absaroka4, 12.5 (2.7) 
Yellowstone6, 16.6 (5.2) 

[11] At some sites, aerosol extinction is primarily a result of nondust aerosol components such as sulfate, nitrate, or elemental or organic carbon. This is especially true for sites located close to urban areas where the influence of these aerosol constituents is more pronounced. This is evidenced by the rather high dust concentrations at sites with comparatively few worst dust days. For example, the average dust concentration was 153.2 μg m−3 during three worst dust days at San Gabriel, California, 101.0 μg m−3 for only one worst dust day at Point Reyes, California, and 74.5 μg m−3 for two worst dust days at the Columbia River Gorge in Washington. As previously noted for Point Reyes, other sources of coarse aerosol, such as sea spray, may also contribute to high apparent dust concentrations, especially at coastal sites.

[12] In contrast, sites located in areas with limited anthropogenic influence exhibited much lower average dust concentrations during comparatively more frequent worst dust days. For example, the average dust concentration during 14 worst dust days was 14.1 μg m−3 at Weminuche, Colorado, and 15.9 μg m−3 during the 12 worst dust days at San Pedro Parks, New Mexico.

[13] Figure 2 shows the seasonal distribution of worst dust days and the corresponding mean dust concentration for each state. For most states, a majority of worst dust days occurred in the spring (March to May) and summer (June to August), with fewer worst dust days in the fall (September to November) and a very few cases during winter (December to February). Although limited in number, the worst dust days that occurred during winter were associated with very high dust concentrations (from ∼30 to 150 μg m−3), while the concentrations of dust on worst dust days occurring in the three other seasons were roughly comparable. Northern states (and Texas) exhibited an opposite seasonal profile, with higher dust concentrations measured during spring and summer worst dust days (Figure 2b).

Figure 2.

Seasonal variation of worst dust days at IMPROVE sites by state (a) distribution of worst dust days by season and (b) mean dust concentration during worst dust days.

4. Estimation of Locally Generated Windblown Dust

[14] The premise behind this analysis is that at IMPROVE monitors where there is the potential for wind to entrain a significant amount of dust from local sources, a positive correlation should be present between wind speed at/near the monitor and the measured dust concentration. We present the methodology and results of applying it to the 70 preliminary IMPROVE sites for all 2001–2003 sample days as well as for the worst dust days at each site. Interpretation of these results and the potential errors associated with applying this method are discussed subsequently.

4.1. Meteorological Data

[15] Hourly wind direction, wind speed (WS), and precipitation (if available) data were obtained from meteorological sites located at or near each of the 70 IMPROVE sites. Meteorological sites were part of the Clean Air Status and Trends Network (CASTNET), Remote Automatic Weather Stations, Arizona Department of Environmental Quality, sites in NOAA’s Integrated Surface Hourly database, or networks operated by NASA or the National Park Service. The appropriateness of using a specific meteorological station to represent a specific IMPROVE site was determined using a combination of objective and subjective methods. Initially, all meteorological sites within a 90-km radius and an elevation within 500 m of the IMPROVE site were considered. The IMPROVE site and meteorological stations that satisfied these criteria (along with data completeness and length of record criteria) were overlaid on a topographic map. The most appropriate surface meteorological site for each IMPROVE monitor was chosen so as to minimize distance, elevation difference, and intervening landform features (topography). In some cases, it was not possible to find a reasonable match for IMPROVE monitors and those were not included in the list of 70. For the 70 IMPROVE monitors on the preliminary list, 56 were located within a 25-km distance, and 59 were within a 200-m elevation difference of a meteorological station. The greatest distance between an IMPROVE monitor and a meteorological station was 83 km (Great Sand Dunes, Colorado), and the greatest difference in elevation was 410 m (Hillside, Arizona). The effect of a meteorological station not adequately reflecting conditions at an IMPROVE site is discussed below.

[16] Hourly wind speed data (reported in integer miles per hour at most meteorological monitoring networks) were binned into one of three categories: WS2, 6.2–8.9 m s−1 (corresponding to 14–20 miles h−1); WS3 > 8.9–11.6 m s−1 (corresponding to > 20–26 miles h−1); and WS4 > 11.6 m s−1 (corresponding to > 26 miles h−1). These wind speed bins were further divided into wind direction bins, each representing 90° centered about one of the four cardinal directions. Since aerosol data from the IMPROVE network represent 24-h average values, wind data were aggregated over the same period. Thus, the wind conditions on a specific sample day (midnight to midnight local time) were represented by 12 bins (three wind speed × four wind direction bins). Hourly wind speeds less than 6.2 m s−1 (corresponding to < 14 miles h−1) were not included in the regression since windblown dust emissions are not likely to occur at those low wind speeds [Lyles and Krauss, 1971]. Gillette et al. [1982; 1996] reported that emissions are nonlinear with respect to wind speed and are proportional to u(u2uf2),where u and uf are the measured wind speed and threshold wind speed, respectively. Therefore ideally, wind speed and wind direction bins would be as narrow as possible to fully represent the nonlinearity of emissions. However, the work reported here relies heavily on statistical regression, and if bins are made too narrow (for example, 0.44 m s−1 increments instead of 2.7 m s−1), then the number of independent variables would be quite large and obtaining statistically significant regression results would not be possible. The discretization scheme used here was based on trial and error and was intended as a compromise between precisely reflecting wind conditions while minimizing the number of independent variables.

4.2. Multivariate Linear Regression Analysis

[17] Multivariate linear regression analysis assumes the ability to predict the value of a dependent variable from the values of n independent variables based on the existence of a correlation between the dependent and independent variables. The equation used in this study to describe the relationship between measured dust mass on a given sample day at a given site (ym, in μg m−3, dependent variable) and wind condition variables (x1, x2,……., x12, independent variables) was:

equation image

where b1, b2,……., b12 are the regression coefficients of the wind condition variables (three wind speed bins × four wind direction bins), and α is the intercept. For a given day, xi represents the number of occurrences of wind condition i (maximum 24 since wind data are hourly). The intercept is interpreted here to be the average dust concentration not associated with local wind resuspension. Thus, it may represent dust from transport of windblown dust from distant sources or local and long range transport of mechanically suspended dust (agricultural, road dust).

[18] Best fit coefficients in equation (3) were determined using all available data for the 2001–2003 period except for days when precipitation had occurred on that day or the prior day. In many cases, owing to the prevalence of specific wind directions at a site, the effective number of independent variables was much smaller [that is, some of the xi in equation (3) were equal to 0 for a majority of the days].

[19] Least squares variable screening methods, including stepwise forward and backward procedures, were employed to objectively determine which variables were significant using 0.15 significance level t value criteria. When a variable did not meet these significance criteria, the corresponding value of b was set to 0. The local windblown dust at a specific site for day j (LWDj) and the associated error (Ej) were calculated as:

equation image


equation image

where e1, e2, …., ek are the standard errors of b1, b2, ..., bk regression coefficients when the value of b was significant for a specific site and zero when the value of b was not significant.

[20] A set of external statistical tools were applied to evaluate the accuracy, robustness, and sensitivity of regression models. Significance of bi was determined by rejecting the null hypothesis that bi = 0 at the 0.15 level. The variance infiltration factor (VIF) which is related to a partial correlation coefficient for each independent variable was applied to detect multicollinearity for the individual regression coefficients. Independent variables with VIF values higher than 10 were rejected. The accuracy and the adequacy of the linear model was evaluated by estimating the F value (the ratio of explained variation divided by the model’s degrees of freedom) and the adjusted multiple coefficient of determination for the specific model.

[21] The intercept α was omitted from equations (4) and (5) since it is assumed that α represents dust concentrations that are not a result of the influence of wind erosion in the vicinity of the site. LWDj was calculated for all site days when meteorological data were available. However, when LWDj values were associated with high levels of uncertainty (LWDj − 2Ej ≤ 0), it was assumed that the local windblown dust for that site-day was zero (i.e., not significant in a statistical sense).

4.3. Estimates of Local Windblown Dust for the 2001–2003 Period

[22] Of the 70 sites considered, there were 41 where a statistically significant (p value < 0.15) relationship was found between measured dust concentrations and at least one of the twelve wind conditions (empty circles in Figure 1), mostly located in southern states. The associations between dust concentrations and wind conditions appeared to be quite robust (R > 0.50; Table 1) for many sites located in New Mexico, Colorado, Utah, and Texas, and several sites in northern states. Twenty-nine sites showed no associations (p value > 0.15) between surface wind conditions and the dust concentrations measured at the IMPROVE site for any of the twelve combinations of wind speed and wind direction (filled triangles in Figure 1). Table 1 (Column B) shows the average values and standard deviations of LWD for only when LWD − 2E ≥ 0. Column C shows the averages for all sample days in the 2001–2003 period where if LWD − 2E ≤ 0 for a sample day, LWD was set to 0 for that day. Overall, for the 2001–2003 period, aeolian dust from sources located near the site contributed from 0.2 μg m−3 at Gila Wilderness up to 3.0 μg m−3 at Salt Creek and White Mountain Wilderness Areas in southern New Mexico, and as high as 6.1 μg m−3 at Columbia River Gorge (Table 1). Note that at the Columbia River Gorge site, a significant fraction of the CM may be associated with nonwindblown dust sources such as wildfires. Thus, the LWD for this site may overestimate actual windblown dust emissions from the vicinity of the site.

[23] The majority of sites that exhibited a statistical relationship between dust concentration and wind conditions were located in Arizona, New Mexico, southern California, Utah, and Colorado, although significant variations in LWD were observed among sites within the same state. For example, the Death Valley site which is located in an open area with sparse vegetation showed rather good correlation with wind conditions (R = 0.49) while low correlation coefficients were computed (from 0.02 up to 0.32) for sites located west of the Sierra Nevada Mountains. Similar differences were observed in Arizona with correlation coefficients ranging from 0.08 to 0.72. These discrepancies may be explained by the proximity of the sites to agricultural fields and urban areas. For example, Los Angeles is upwind of the southern California sites, and the Tucson metropolitan area is in the vicinity of the Saguaro sites. For those sites, regression-based LWD contributed only up to 1 μg m−3 during the 2001–2003 period (Table 1). Perhaps this is due to nonwindblown sources such as construction activities, travel on unpaved roads, agricultural activities, or transport from far upwind. Death Valley was heavily influenced by windblown dust from sources located in the vicinity of the site, representing approximately 2 μg m−3 during the 2001–2003 period.

4.4. LWD on Worst Dust Days

[24] Table 2 (Column B) shows calculated LWD values for worst dust days only. Sites with more than 12 worst dust days were examined in greater detail. A total of 19 sites located in Arizona, California, Colorado, New Mexico, and Texas are included (Figure 1). The 19 sites fell into two categories (Table 2, Column D) based on mean LWD contribution (assuming LWD = 0 for days with LWD − 2E ≤ 0). Group 1 (Zion and Canyonlands in Utah, Death Valley in California, Great Sand Dunes, Mesa Verde, and Weminuche in Colorado, and San Pedro Parks, Bandelier, Bosque del Apache, and Salt Creek in New Mexico and to a lesser extent, Chiricahua in Arizona and Guadalupe Mt. in Texas), is characterized by sites where the LWD contribution to dust concentrations on worst dust days (Column C divided by Column A in Table 2) was greater than 10%. The polar plots showing the orientation and the relative importance of regression coefficients for each of group 1 sites are depicted in Figure 3. The high (> 0.25) standardized coefficients of regression for specific wind directions for those sites (Figure 3) are indicators of the likely direction of the source areas and required wind speeds necessary for local windblown contributions and provide additional evidence of the different types of influence of local windblown dust sources, for example, strong contribution only from area sources located north or south of Salt Creek (Figure 3f) versus moderate contributions from sources located all around Guadalupe Mt. (Figure 3l).

Figure 3.

Polar plots of standardized regression coefficients obtained from multivariate linear regression analysis of observed dust concentrations and wind conditions for IMPROVE sites with more than 12 worst dust days over the 2001–2003 period. WS2, WS3, and WS4 correspond to wind speed bins of 6.2–8.9, 8.9–11.6, and higher than 11.6 m s−1, respectively. Standardized coefficients provide an indication of the contribution of individual coefficients to LWD.

[25] Group 2 is comprised of sites that are located in central Arizona and includes Hillside, Ike’s Backbone, Queen Valley, Saguaro(both East and West), and Tonto National Monument. In group 2, LWD estimates accounted for a rather small fraction (<10%) of dust concentrations on worst dust days (from 0.2 μg m−3 of measured dust at Sierra Ancha, to 4.4 μg m−3 at Saguaro West) (Table 2). Other nonwindblown dust sources including agricultural activities, road dust, construction activities, or mining operations may be responsible for high dust concentrations at those sites. Alternatively, it is possible that high dust concentrations at those sites are caused by transport of wind entrained dust from far upwind.

[26] Figure 4 shows the times series plot of measured dust concentrations in each Group 1 site, as compared with local windblown dust estimates (where LWD − 2E >0) (denoted with black squares) throughout the 2001–2003 period including worst dust days (denoted with asterisks). For Death Valley, source regions located northwest and southwest of the site appeared to affect dust concentrations when wind speeds were moderate (WS2, 6.2–8.9 m s−1) throughout the 2001–2003 period (Figures 3a and 4a). Very high wind speeds (WS4, >11.6 m s−1) were also significant (in a statistical sense) contributors to dust concentrations at Death Valley on worst dust days. However, owing to their infrequent occurrence, dust from those high wind speeds was not as important as from the more frequent moderate wind speeds (Note that normalized regression coefficients are nonzero but small for WS4 compared with WS2). On a seasonal basis, windblown dust from nearby sources was more important in winter and to a lesser extent during spring, while most of the episodes in summer and fall appear to be associated with other types of sources. A similar trend with sources that are susceptible to high wind speeds (WS4) was observed for Salt Creek and Guadalupe Mt. sites. Dust emissions during winter and early spring from area sources located north and south of Salt Creek contributed more than 50% to dust concentrations on worst dust days but had a lower contribution in summer and fall (Figures 3f and 4f). Local windblown dust from sources around Guadalupe Mt. appeared to contribute to dust concentrations frequently, though they represented a relatively small percentage of total measured dust on worst dust days (Figures 3l and 4l).

Figure 4.

Observed (black line) dust concentration and computed local windblown dust (black squares) during the 2001–2003 period in twelve IMPROVE sites in southwestern USA. Worst dust days are denoted with an asterisk.

[27] For Canyonlands (Utah), Bandelier (New Mexico), Great Sand Dunes, and Weminuche (Colorado), the local windblown dust contribution was quite sporadic but present at significant levels on most of the worst dust days (Figures 4b, 4d, 4h, and 4j). Source areas located northwest and south/southwest of those sites were mostly susceptible to moderate-to-high wind speeds (Figures 3b, 3d, 3h, and 3j). With the exception of Weminuche, which showed significant contribution of local windblown dust in both spring and summer, the rest of the sites followed a similar seasonal trend to that observed for the Death Valley and Salt Creek sites. Windblown dust from local sources at Zion (Utah), Bosque del Apache (New Mexico), Mesa Verde (Colorado), and Chiricahua (Arizona) was responsible for a few worst dust days (Figures 4c, 4e, 4i, and 4k). Each site exhibited a unique seasonal profile, probably because of the type of dust sources. For example, for Bosque del Apache and Chiricahua, most of the windblown dust events were observed during winter and spring, similar to other sites in that area. In contrast, local sources were quite active during summer and fall at Zion and Mesa Verde which may be related to windblown dust emissions from nearby disturbed agricultural lands after harvesting and tilling [Nordstrom and Hotta, 2004].

[28] Overall, these observations suggest that there exists a large range of conditions that can cause worst dust days to occur during different times of the year and local windblown dust to account for varying amounts of dust on those days. Efforts are underway to improve characterization of causes of seasonal variation on a site-by-site basis as well as numerating plausible dust creating activities/conditions other than locally generated windblown emissions [Kavouras et al., 2005].

4.5. Limitations of Methodology

[29] It is important to remember that the methodology presented is based on the ability to extract a statistical relationship between wind conditions and aerosol concentrations. There are several important assumptions and limitations that apply to the results of this study. First, contrary to our assumption, coarse mass will not be composed entirely of geological material. The high CM concentration at the coastal site of Point Reyes, California, was given as one example of CM likely resulting from another source. In recent work, Malm et al. [2006] have examined the composition of coarse particles (2.5–10 μm) at 9 IMPROVE sites, 4 of which overlap with the 70 sites considered here. At Sequoia and San Gorgonio in California, and at Grand Canyon in Arizona, those authors found that on an annual basis, only about 70–80% of the coarse particle mass was associated with soil, indicating that a substantial fraction (20–30%) was not. At Mount Rainier in Washington, only 32% of the coarse mass was associated with soil, and 57% was associated with organic material. Without detailed chemical information for all IMPROVE monitors in this study, it is difficult to determine errors associated with assuming that all coarse mass is associated with soil. In general, the assumption will lead to an overestimate of windblown soil dust. We note parenthetically that neither Point Reyes nor Mount Rainier exhibited a correlation between wind speed and dust-defined as CM + FS.

[30] Second, since in most cases, surface meteorological data were obtained from stations that were not in the immediate vicinity of an IMPROVE site (with the exception of some CASTNET stations), it is difficult to ascertain how well the meteorological data represent the IMPROVE monitor. If wind conditions are not adequately represented by the chosen meteorological station, then the correlation between wind and dust concentration would appear to be weaker than it really is or else absent altogether. This results in an underestimate of locally generated windblown dust. Related to this point, by applying somewhat strict significance criteria to the regression coefficients in equation (3) and in the application of equation (4), on average, the calculated LWD will underestimate the actual contribution of local windblown dust to ambient concentrations. In this context, the LWD should be thought of as the amount of dust that can be associated with elevated wind (with some statistical confidence) based on the available meteorological data.

[31] Another aspect of our approach that is likely to result in somewhat underestimated values of LWD is that the regression coefficients in equation (3) were calculated with the underlying assumption that the susceptibility of erodible surfaces in the vicinity of an IMPROVE site to wind erosion is invariant with time. In reality, under certain conditions, wind erodibility can vary temporally. Higher threshold velocities for entrainment can result from seasonal precipitation, stabilizing range, and shrub lands [Stout, 2003; Gillette et al., 1980, 1982; Zender et al., 2003a, 2003b]. At other locations such as Owens Lake and the Salton Sea in California, hard salt crusts may form during dry summer and fall conditions protecting the soil surface from wind erosion [Houser and Nickling, 2001; Gillette et al., 2004; Etyemezian et al., 2006]. The net result of not allowing for seasonal differences in wind erodibility would be to weaken the confidence in the coefficients of equation (3) causing rejection by the significance criteria and resulting in underestimates of LWD. As the IMPROVE data set continues to grow in terms of temporal coverage, this problem may be avoided in the future by running the regression between wind conditions and dust concentrations by season.

[32] Third, synoptic scale weather systems may affect wind conditions over a large region. Thus, elevated winds at an IMPROVE site (or at the meteorological station representing the IMPROVE site) may be correlated with high dust concentrations, but the source of the dust may be windblown emissions from far upwind. This would lead to an overestimate of the influence of wind on local dust emissions and LWD values may be unrealistically high. It is worth noting that the relatively strict significance criteria placed on the regression coefficients in equation (3) would require that if this type of error was to occur, the wind conditions at the IMPROVE site and those at the “real” source of windblown dust would have to be very highly correlated. Otherwise, periods of high wind and high dust would be counteracted by periods of high wind in the absence of dust, and the regression would return statistically insignificant correlations. Here again, the issue may be sidestepped somewhat by defining LWD as the amount of dust that can be attributed to elevated wind with some confidence based on the available meteorological data. However, this does highlight one important weakness of this method, namely that the distance scales associated with an operationally defined “local” source (i.e., one that emits dust when winds are high) are not well bounded and that additional information is required on a site-by-site basis to determine what “local” encompasses. Sources of such information may include on-site photographs or measurements, satellite images of specific wind events, and/or soil stability maps.

5. Conclusions

[33] We analyzed aerosol data from Class I areas in the western United States in combination with local meteorological conditions (wind speed, direction, and precipitation) to determine the relationships between dust concentrations and wind conditions. Dust concentrations showed significant variations, presumably because of spatial and temporal differences in windblown dust source strengths as well as source types. Spatial concentration gradients, with southern and western locales exhibiting higher dust concentrations, corroborate intuition as well as previous work. Approximately one-third of 20% worst-case visibility days in the western USA during 2001–2003 were due to high dust contributions to aerosol extinction coefficient, indicating the importance of dust in the implementation of the Clean Air Act and Regional Haze Rule. The timing of occurrence of worst dust days and the magnitudes of concentrations on those days varied substantially among sites.

[34] Forty-one IMPROVE sites, most of them located in southwestern states, showed statistically significant correlations between dust concentration and specific wind conditions, providing information on wind direction and magnitude required to activate wind erosion at sources local to the sites. Sites with more than 12 worst dust days during the 2001–2003 period were further analyzed. The comparison of measured dust and estimates of local windblown dust suggested that for the sites located in New Mexico, Colorado, Utah as well as Death Valley (California) and Guadalupe Mt. (Texas), the impact of dust sources near the site was important and responsible for most of the worst dust days. In contrast, dust originating from other sources appeared to dominate worst dust days at the remaining sites which were mostly located in Arizona and California. Despite the spatial variations among the sites, a rather strong seasonal trend was observed for local windblown dust with higher contributions during winter and spring.

[35] The method used here to infer dust concentration because of local wind erosion relies on extracting a statistical relationship between wind conditions and the sum of fine soil and coarse particle mass concentrations. Thus, the method may provide overestimated values at locations where coarse particles are not overwhelmingly composed of crustal (geologic) material. On the other hand, at sites where local windblown dust is a frequent and dominant source, the method may underestimate the local contribution since statistically weak relationships between wind and dust concentration are not used to estimate the locally generated windblown dust contribution. Another shortcoming is that the distances to “local” sources are not well defined. Rather, a source is defined as local if a statistical relationship can be found between dust concentrations and elevated wind. Overall, our method provides an empirical, semiquantitative procedure for determining the contribution of local wind blown sources to dust concentrations at Class I areas in the USA. This information is important for understanding the causes of reduced visibility in national parks and evaluating the relative magnitude of impacts from local/regional and transcontinental dust sources. However, the method provides only one tool for assessing relationships between wind conditions and dust concentrations at a site. Efforts that utilize the work presented here, in conjunction with several other independent tools, are underway to help provide a more complete picture of the sources of dust at remote locations in the western USA.


[36] This research was supported by Western Air Regional Partnership. We thank Lee Alter, Tom Moore, and the members of Dust Emissions Forum for useful comments on data analysis.