Interactive comment on “Toward enhanced capability for detecting and predicting dust events in the Western United States: the Arizona Case Study”

in the revealed intensiﬁed dust activity in the western US during the past decades determine this. I of evidence of the authors’ and important. presentation

North America contributes to a small proportion of the world's total dust emissions, ranging from < 0.1 to ∼ 5 % as reported in previous studies (Miller et al., 2004a, b;Tanaka and Chiba, 2006;Zender et al., 2003;Ginoux et al., 2004;Ravi et al., 2011), and the important emitters include the four major deserts in the western US, i.e., the Great Basin, Mojave, Sonoran, and Chihuahuan deserts.Dust storms in the western US usually last for 2-21 h, due to various mechanisms (Lei and Wang, 2014).Surface and satellite observations, along with modeling analysis, have provided evidence that the western US is not only affected by local dust emissions but is also susceptible to dust transported from overseas (e.g., Van Curen and Cahill, 2002;Fischer et al., 2009;Uno et al., 2009;Fairlie et al., 2007;Chin et al., 2007;Eguchi et al., 2009;Stith et al., 2009;Dunlea et al., 2009;Liu et al., 2009b).Using a global transport model, Fairlie et al. (2007) reported that dust from overseas contributed to < 30 % of the total dust in the southwestern US, to > 80 % of the total dust in the northwestern US in spring 2001, and that these non-US contributions were much larger than in other seasons.Recent dust observations have revealed rapid intensification of dust storm activity in the western US (e.g., Brahney et al., 2013), despite the weaker dust activity in many non-US regions (e.g., Mahowald et al., 2007;Zhu et al., 2008;Shao et al., 2013).This increasing trend enhances the concerns about their various impacts and even the possibility of another Dust Bowl, as that which occurred in the 1930s due to severe drought conditions and inappropriate farming methods (Lee and Gill, 2015; http://www.livinghistoryfarm.org/farminginthe30s/water_02.html; http://www.ncdc.noaa.gov/paleo/drought/drght_history.html) and led to significantly negative agricultural and ecological impacts in the western/central US.
Surface and satellite observations have been used to study dust trends and variability, as well as for model evaluation (e.g., Tong et al., 2012;Appel et al., 2013;Torres et al., 2002;Ginoux and Torres, 2003;Draxler et al., 2010;Vukovic et al., 2014;Mahler et al., 2006;Raman and Arellano, 2013;Morain et al., 2010).Surface observations used in many of these studies are sparsely and/or infrequently sampled, and there is delay for obtaining some of these data sets which prevents timely updates on the observed dust records.The capability of satellite aerosol optical depth products to capture the dust events depends on various factors such as sensor characteristics, cloud conditions, surface reflectance and dust mineralogy (e.g., Baddock et al., 2009).There is still a lack of comprehensively developed observational dust records with broad spatial coverage up to the very recent years, and accurately simulating dust aerosols is challenging.Therefore, it is important to extend the temporal changes of observed dust activity to recent years using diverse observations.These various observations can assist in evaluating the chemical transport model skills especially during dust events.Furthermore, better understanding the linkages between the temporal changes of dust observations and the observed surface/weather conditions can be beneficial for advancing the dust emission modeling skills via improving the meteorology and dust source input data, as well as for projecting future dust activity under the changing climate.
Several studies found that dust events can be accompanied by stratospheric intrusions in multiple regions of the world (e.g., Pan and Randel, 2006;Yasunari et al., 2007;Yasunari and Yamazaki, 2009;Reddy and Pierce, 2012).Recently, substantial attention has been called on the influences of stratospheric ozone intrusions on western US surface/nearsurface ozone variability (e.g., Lin et al., 2012;Langford et al., 2014).Observations and modeling tools are useful for identifying the periods when dust events are associated with stratospheric intrusions, as well as to assess the impact of elevated surface/near-surface ozone and PM (particulate matter) concentrations on public health and the environment during such events.
This study develops decadal dust records in the state of Arizona using multiple in situ and satellite observation data sets, and relates the diurnal and inter-annual variability of observed dust activity to the observed surface conditions (e.g., land cover type and vegetation conditions) and weather patterns (e.g., wind and soil moisture; Sects.3.1-3.3).We also analyze observations and model simulations during a recent strong dust event in the western US accompanied by a stratospheric ozone intrusion.The modeling analyses include the US National Air Quality Forecasting Capability (NAQFC) 12 km Community Multi-scale Air Quality (CMAQ) regional model base and sensitivity simulations (Sect.3.4).In the analyses, we discuss the usefulness and limitations of different observations for identifying potential exceptional events and for model evaluation.We also suggest future directions of integrating observations into regional dust emission modeling in the western US for further improvement of the air quality forecasts.

Drought indicators
Three data sets were analyzed to interpret the observed interannual variability of the drought conditions from 2005 to 2013 in Arizona, an important dust source and receptor region in the western US.They are the normalized difference vegetation index (NDVI) from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument on the NASA Aqua satellite, a European soil moisture data set that merged both passive and active satellite sensor data, and the Palmer drought severity index (PDSI).NDVI is the most commonly used vegetation index, calculated using the reflected visible and near-infrared light by vegetation (Scheftic et al., 2014;Brown et al., 2006).Smaller NDVI values refer to less vegetated areas, which may have a high potential of emitting dust (D.Kim et al., 2013;Vukovic et al., 2014).NDVI has been used for monitoring land cover changes and indicating drought (Tucker and Choudhury, 1987;Karnieli et al., 2010;Wan et al., 2004), and it has been found to be correlated with meteorologically based drought indexes such as the standardized precipitation index (Ji and Peters, 2003).In this study we used the monthly mean 1 km MODIS NDVI product Collection 5, which temporally aggregated the 16-day 1 km MODIS NDVI using a weighted average.Following the users' guide instructions (http://vip.arizona.edu/documents/MODIS/MODIS_VI_UsersGuide_01_2012.pdf),only the data flagged as good quality were used.To avoid the known effects from the degradation of the Terra MODIS sensor (e.g., Wang et al., 2012), only the NDVI data from the Aqua MODIS (MYD13A3) were used.
Soil moisture has also been used for drought monitoring and several studies have found that satellite and modeled soil moisture is related to dust outbreaks in Asian countries (Liu et al., 2004;Y. Kim et al., 2013;Kim and Choi, 2015).This study used a multi-sensor satellite soil moisture product from the European Space Agency (ESA) within the soil moisture Climate Change Initiative (CCI) project that merged all available passive and active products and preserved the original dynamics of these remote sensing observations.The data are produced daily on a 0.25 • × 0.25 • horizontal resolution grid.Long-term soil moisture changes in the US based on the CCI soil moisture product contributed to the US National Climate Assessment report (Melillo et al., 2014, pp.72-73, last accessed November 2015).
Monthly PDSI data, calculated from temperature and precipitation (Palmer, 1965;Alley, 1984), are widely used for identifying long-term and abnormal moisture deficiency or excess.Studies have found that PDSI is moderately or significantly correlated (r = 0.5 to 0.7) with observed soil moisture content within the top 1 m depth during warm-season months in various regions (Dai et al., 2004).In this study, we analyzed the inter-annual variability of PDSI in two NOAA climate regions in Arizona (Karl and Koss, 1984; http://www.ncdc.noaa.gov/monitoring-references/maps/us-climate-regions.php).Drought conditions are defined with negative PDSI values (e.g., −2 is moderate drought, −3 is severe drought, and −4 is extreme drought), and positive PDSI values indicate wet conditions.

Specification of dust sources using satellite (MODIS) land cover and NDVI products
The dust productive areas depend on surface conditions such as land cover types and vegetation conditions, and therefore are temporally variable.Several studies specified dynamic dust source regions using either or both satellite land cover types and NDVI products (e.g., Vukovic et al., 2014;Yin et al., 2007;D. Kim et al., 2013).In this study, to explore the inter-annual variability of dust sources in the western US and its influences on the dust activity, we specified the dust sources following the methods in Vukovic et al. (2014) Friedl et al., 2010) and its 17-category International Geosphere Biosphere Programme (IGBP) land cover classification scheme (defined at: https://lpdaac.usgs.gov/dataset_discovery/modis/modis_products_table/mcd12q1).Then, for each month and each of the three erodible land cover types, dust source areas were determined based on the monthly mean Aqua MODIS NDVI values (introduced in Sect.2.1) and the following criteria.

Aerosol observations
Both remote sensing and in situ aerosol observations were used to explore the dust aerosol distributions in Arizona.We first demonstrate the large-scale spatial distributions of aerosols using satellite aerosol products and discuss their diurnal (e.g., late morning vs. early afternoon times) and interannual variability link to the weather and surface conditions.We mainly focus on spring and summer time periods when dust activity is generally strong in Arizona, as found by Ginoux et al. (2012a) for the 2003-2009 period.In situ observations at Arizona surface monitoring sites were then analyzed, focusing on their temporal variability in the populated Phoenix urban area (i.e., with a population of ∼ 1.5 million).Finally, we identify dust events in Phoenix using hourly surface observations and discuss the time of occurrence of these identified dust events.

MODIS deep blue aerosol optical depth (AOD) and dust optical depth (DOD)
We extracted scenes dominated by dust aerosols from the MODIS level 2 deep blue aerosol product Collection 6 (Hsu et al., 2013(Hsu et al., ) during 2005(Hsu et al., -2013.This product includes the values of AOD and single scattering albedo (SSA) at 412, 470, 550, and 670 nm, as well as the Ångström exponent between 412 and 470 nm.It is recommended for identifying both dust sources and plumes at high spatial resolution (e.g., Baddock et al., 2009).The Collection 6 deep blue data were created using the enhanced deep blue algorithm (from the previous Collection 5.1), with improved surface reflectance determination, aerosol model selection, and cloud screening schemes.Also, the deep blue data from Terra MODIS have been extended beyond 2007 using suitable calibration corrections (Hsu et al., 2013).Compared with the Aerosol Robotic Network (AERONET) AOD data, the Collection 6 deep blue AOD data from Aqua MODIS show a ∼ 0.03 change in bias through the decade, with overall negative biases in 2005-2007and 2011, and positive biases in 2009, 2010, and 2012(Sayer et al., 2013)).
The very good (quality assurance flag = 3, as recommended by Shi et al., 2013, andSayer et al., 2013) MODIS deep blue AOD data from Terra and Aqua were selected and gridded on 0.1 • × 0.1 • horizontal resolution for each day.The DOD values were then determined by screening the 550 nm AOD data based on three criteria to represent dustdominated scenes: (1) an Ångström exponent within 0-0.5, which selects the particles in large sizes; (2) SSA at 412 nm, < 0.95, which selects the absorbing aerosols and efficiently eliminates the sea-salt-dominated scenes; and (3) a positive difference of SSA between 412 and 670 nm, due to the specific optical property of dust by which there is a sharp increase of absorption from red to deep blue (Ginoux et al., 2012a;Hsu et al., 2013).

Particulate matter (PM) measurements from the surface Interagency Monitoring of Protected Visual Environments (IMPROVE) sites
Most IMPROVE surface sites are located in rural regions, many of which are in the national parks to measure background pollution levels.Here, we analyzed the temporal variability of observed particulate matter mass PM10 (i.e., < 10 µm in diameter) along with the fine (i.e., < 2.5 µm in diameter) soil particles at the Phoenix site (PHOE1, 33.5038 • N, 112.0958 • W) within the IMPROVE network during 2005-2013.These fine soil data are computed based on five (Al, Si, Ca, Fe, and Ti) soil-derived trace metals in their assumed oxidized form measured at the IMPROVE site (Malm et al., 2004).Daily mean IMPROVE data are available every 3 days, and there is an approximate delay of 1 year for obtaining these data.

Air Quality System (AQS) and AirNow PM and trace gas measurements
In general the US Environmental Protection Agency (EPA) AQS sites are designed to monitor air quality in populated urban or suburban areas.In this study the AQS hourly PM10 and PM2.5 data from 2005 to September 2013 and the AirNow from September to December 2013 at the Phoenix JLG supersite (co-located with the IMPROVE PHOE1 site, AQS site no.040139997) were analyzed to study the temporal variability of dust events on hourly temporal resolution.In the case study on the dusty year of December 2006-November 2007, AQS trace gas measurements (i.e., carbon monoxide, CO, and oxides of nitrogen, NO x ) were used as tracers of anthropogenic or biomass burning sources to evaluate the dust events that are identified based on the hourly PM observations.The AQS data qualifier codes were also examined, which provided us with clues of the event types (e.g., high winds and long-range transport of PM from non-US regions).

Other satellite aerosol products
The achieved NOAA Hazard Mapping System (HMS) text product narratively describes the observed smoke and dust events based on images of multiple satellites.It qualitatively indicates the dust's locations and the intensity, which in this study supports the analysis during a recent strong event we selected for the case study in Sect.3.4.We also used the dust score data from the Atmospheric Infrared Sensor (AIRS) instrument on board the Aqua satellite to qualitatively represent the presence of atmospheric dust during this recent event.The Aqua satellite has ascending overpass times in the early afternoon (∼ 13:30 LT, local time).

Observed wind speed and direction
As atmospheric dust concentrations depend on the wind fields (e.g., Kavouras et al., 2007;Ravi et al., 2011;Csavina et al., 2014), we used the observed hourly surface wind speed and direction in December 2006-November 2007 at the Phoenix Encanto site (33.4792• N, 112.0964 • W, within the Arizona meteorological network; AZMET) together with the hourly AQS PM observations to identify the dust events.Phoenix Encanto is the closest site to the Phoenix JLG supersite within the AZMET that had meteorological observations available during this period.

Backward air mass trajectory analysis
Backward air mass trajectories were computed to locate the sources of dust aerosols observed at the Phoenix JLG site during the identified dust events in December 2006-November 2007.These trajectories were calculated using NOAA's Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model, version 4 (Draxler and Rolph, 2015;Stein et al., 2015).The accuracy of the trajectories depends on the resolution of the wind data (Draxler and Hess, 1998), and we calculated these trajectories based on the 3-hourly North America Regional Reanalysis (NARR) data (Mesinger et al., 2006) on 32 km horizontal resolution and nine vertical levels below 800 hPa.NARR is the finest meteorology HYSPLIT can currently run with for studying this year, as the horizontally finer (12 km) North American Mesoscale Forecast System (NAM; Janjic, 2003;Janjic et al., 2004) wind fields are only available for HYSPLIT calculations for the time after May 2007.These trajectories were initiated at 500 m above Phoenix's ground level at identified dust periods and were computed for 24 h.The HYSPLITindicated air mass origins during the Phoenix dust events will be discussed together with the MODIS land cover product (details in Sect.2.2).

Chemical transport model base and sensitivity simulations
The US NAQFC 12 km CMAQ (Byun and Schere, 2006;Chai et al., 2013;Pan et al., 2014) model simulations were used to depict the PM distributions during a recent strong dust event in the western US that was accompanied by a stratospheric ozone intrusion.Dust emissions for NAQFC's CMAQ simulations were calculated by the FENGSHA dust emission model based on a modified Owen's equation, which is a function of wind speed, soil moisture, soil texture and erodible land use types (Tong et al., 2015).Both the FENG-SHA and CMAQ model calculations were driven by meteorological fields from the NAM model, which is known to usually have positive biases in temperature, moisture, and wind speed in the continental US (e.g., McQueen et al., 2015a, b).The CMAQ base simulation was evaluated against surface observations at the AirNow and IMPROVE sites, and we focused on PM2.5 concentrations as it is one of the standard NAQFC products.To quantify the impact of west-

Ozone and carbon monoxide products from AIRS
The level 3 daytime ozone and carbon monoxide (CO) profiles (AIRX3STD version 6, gridded in 1 • × 1 • horizontal resolution) from the AIRS instrument were used to help identify the stratospheric intrusion during a recent dust event in Sect.3.4.AIRS ozone is sensitive to the altitudes near the tropopause, with positive biases over ozonesondes in the upper troposphere (e.g., Bian et al., 2007).Due to its broad spatial coverage and the capability of reproducing the dynamical variability of ozone near the tropopause, AIRS ozone has been used in a number of studies on stratospheric intrusions (e.g., Lin et al., 2012;Pan and Randel, 2006;Pan et al., 2007).AIRS CO, which is most sensitive to 300-600 hPa (Warner et al., 2007), can distinguish stratospheric intrusions from long-range transported pollution when used together with its ozone product.

Decadal drought indicators, dust sources and satellite DOD in Arizona
We first review the spatial and inter-annual variability of the drought conditions during 2005-2013 in Arizona, in the dusty seasons (i.e., spring and summer, from March to August), based on satellite NDVI (Fig. 1a) and soil moisture (Fig. 1b) products.These observations show that southwestern and south central Arizona, a region close to the Sonoran Desert, is overall drier, with less greenness, than the rest of the state.Most of these dry regions fall into two NOAA climate divisions (i.e., south central including the Maricopa and Pinal counties and southwest including the La Paz and Yuma counties).The mean PDSI values in spring and summer in these two climate divisions were calculated (Fig. 1c), indicating moderate to severe dry conditions under warm weather in these regions in the past decade, except for 2005 (extremely wet), 2008 (near neutral), and 2010 (moderately wet).The PDSI values were then correlated with the anomalies of satellite NDVI and soil moisture, defined as the ratio of the annual mean value over the multi-year mean value.In general, Fig. 1c shows that the PDSI-indicated drought conditions are consistent with those based on the satellite NDVI and soil moisture products: i.e., with correlation coefficients (r) of PDSI vs. NDVI anomaly and PDSI vs. soil moisture anomaly of 0.96 and 0.84, respectively.Gridded MODIS DOD maps are shown in Fig. 2a, b for each year's dusty season during 2005-2013 and they were related to the satellite-based weather and vegetation conditions (Fig. 1c).To exclude the locations occasionally affected by long-range transported dust aerosols, data are shown only for the grids in which DOD data are available on > 5 % of the total number of days in each year, defined as "areas of dust impact".In all maps, high DOD values (> 0.2) are seen in the dry southwest and south central climate divisions.Aqua MODIS observed higher DOD than in Terra MODIS DOD by 4-19 % (∼ 11 % on average).Assuming Terra and Aqua MODIS DOD have similar quality in this region, this indicates higher dust in the early afternoon than in the late morning.Inter-annual variability is also seen from these DOD maps over large spatial scales, with smaller "areas of dust impact" and DOD values in these areas in the wetter years (e.g., 2005 and 2010).The differences among the annualmean DOD values are often much larger than those of the MODIS AOD biases reported by Sayer et al. (2013).The correlation coefficients between the anomalies of Aqua MODIS DOD and the three drought indicators (NDVI, soil moisture, and PDSI) in the past decade are −0.82,−0.58, and −0.79, respectively.The anomalies of Terra DOD show similar correlations with these three drought indicators.Such anticorrelations suggest the importance of drought monitoring to the interpretation and prediction of dust activity.In particular, it is noted that satellites can provide soil moisture measure-ments of much broader spatial coverage than the surface sites (e.g., there is only one site, Walnut Gulch, in Arizona within the Soil Climate Analysis Network), and drought monitoring can be better assisted by newer satellite soil moisture observations, such as those from NASA's newly launched Soil Moisture Active Passive (SMAP).
The correlations between dust activity and drought conditions can be partially attributed to the dependency of dust source regions as well as the threshold wind velocity (i.e., the minimum wind velocity required to initiate soil erosion; Ravi et al., 2011, and the references therein) on the surface conditions in the western US. Figure 3 shows the MODISderived annual-mean dust source regions during the dusty season in 2005-2013 over several land use types (maps of the dust sources from three land use types are shown for selected wet and dry years in Fig. S1 in the Supplement).In most years, barren contributed the most (> 50 %) and cropland contributed the least (< 5 %) to the dust source regions, qualitatively consistent with the findings by Ginoux et al. (2012a) and Nordstrom and Hotta (2004).In general, larger dust source regions are found in drier years, with the strongest inter-annual variability from the open shrubland category.As an important nonerodible roughness element, the variable vegetation also modified the threshold wind velocity for the soil erosion.These findings suggest that dust emission modeling can be improved by using satellite land products, instead of those based on static land data.Similar land products of smaller footprints from newer satellite instruments, such as those from the Visible Infrared Imaging Radiometer Suite (VIIRS) instrument launched in 2011, can also be considered.In addition, soil moisture affects dust activity by modifying the threshold wind velocity, dependent on the soil type.Therefore, dust emission modeling can also benefit from careful evaluation and improvement of the soil moisture inputs using surface and satellite soil moisture measurements.

Decadal surface in situ PM measurements in Phoenix
We then analyze the long-term surface PM measurements at the AQS and IMPROVE monitoring sites in the Phoenix area.The time series of PM10 from AQS/AirNow and IM-PROVE sites in Phoenix are shown in Fig. 4a during 2005-2013 in their original temporal resolution.It is shown that the 24 h mean IMPROVE PM10 data missed the extreme values (e.g., > 150 µg m −3 ) that were captured by the hourly AQS/AirNow observations at this location.The 9-year mean PM10 concentration at the AQS site (31.6 µg m −3 ) is slightly higher than at the IMPROVE site (28.2 µg m −3 ) due to the different sampling frequency and methods.Another advantage of AQS/AirNow observations over those at the IM-PROVE sites is that they are timely made available.IM-PROVE fine soil particles demonstrate the similar temporal variability to IMPROVE PM10 with a correlation coefficient of ∼ 0.8.To explore the inter-annual variability of PM10 in dust seasons (spring-summer) at this site, we calculated the anomalies for each variable in each year (Fig. 4b).Similar to the results from satellite observations, the inter-annual variability of surface PM observations are anticorrelated with regional soil wetness and vegetation cover.Inconsistency exists among the anomalies of these three variables, due to different sampling methods and densities and because the particle size distributions depend on soil wetness (Li and Zhang, 2014).Due to the different observation methods, uncertainties, and sampling strategies (spatial and temporal), the anomalies of surface PM concentrations are more consistent with (i.e., whether > 1 or < 1) those of the MODIS DOD only in several significantly wet or dry years (i.e., 2005, 2007, 2010, 2011).

Phoenix dust events in 2007 identified by hourly surface observations
We take the dry and dusty year of December 2006-November 2007 (Fig. 4b) as an example to introduce a novel approach of identifying dust events using hourly observations.We first calculated the seasonal averages of PM10 and wind speed in Phoenix based on the AQS PM10 and AZMET wind speed observations.It is shown that in this year dominant westerly and easterly winds in spring and summer carried much PM10 to Phoenix (Fig. S2), whereas most PM10 in autumn and winter came from the north and east.Hourly mean wind speed is highly correlated with the hourly maximum wind speed (r = 0.93, slope = ∼ 0.5), and stronger winds were observed during spring and summer (Fig. S3).Two steps followed to identify the individual dust events.In the first step, any period in which PM10 and wind speed exceeded the seasonal mean values for no shorter than 2 h (the lower end of dust storm duration in the western US reported by Lei and Wang, 2014) was defined as a dusty period.The second step screened the dust events selected in the first step using their median values of PM10 (55 µg m −3 ) and PM2.5 / PM10 (∼ 0.2) as lower and upper thresholds and therefore relied on data availability of both PM2.5 and PM10.After these two steps of selection, 29 high dust periods are found as denoted in Fig. S4 on 7, 10, and 27 December; 27 March; 8, 11 (twice), 12 (twice), 16,18,and 20 April;19,28,and 30 July;[13][14]19,20,24,and 25 August;4,5,7,15,and 19 September;5,13,and 16 October; and 15 November.Around 76 % of these events lasted for no longer than 5 h, consistent with the findings by Lei and Wang (2014) that the majority of the exceptional dust storms in Arizona during 2003-2012 lasted 2-5 h mainly due to meso-or small-scale weather systems (e.g., thunderstorms, convections along dry lines, and gusty winds caused by high pressure systems).Hourly PM10 during these high dust periods ranged from 57 to 8540 µg m −3 , with the PM2.5 / PM10 ratio between ∼ 0.07 and ∼ 0.2, and PM2.5 was highly correlated with PM10 during these peri- ods (r > 0.9).In April-May 2007, the Pacific Dust Experiment (PACDEX) was carried out to study dust emission and transport from Asia (Stith et al., 2009).The University of Iowa STEM chemical transport model tracer calculations (http://data.eol.ucar.edu/codiac/dss/id=96.013)estimated dust to be ∼ 2 µg m −3 on average (and not exceeding 10 µg m −3 during transport events) at ∼ 5.3 km altitude in Arizona during this period, which can serve as the upper limit of extra-regional dust impacts on the surface PM concentrations.During our identified dust events, PM10 concentrations were much higher than this magnitude and therefore were mainly due to the impact from local dust emissions.The identified high dust periods were validated using the hourly AQS trace gas observations.Figure S5 includes the scatterplots of AQS CO and NO x over the PM10 measurements at the Phoenix JLG AQS site.Two distinct slopes are shown in both scatterplots, representing the times mainly affected by anthropogenic/biomass burning sources and dust.PM10 values during most of the identified dust events fall into the slope ends in these scatterplots.Using the PM2.5 / PM10 ratio as an additional constraint (as suggested in Tong et al., 2012;Lei and Wang, 2014) in the second step of selection excluded some less strong events interfered by anthropogenic/biomass burning emission sources but possibly also some real dust events.After the second step of selection, higher-than-median CO or NO x values were observed at only ∼ 10 % of the identified dust times.In addition, AQS qualifier codes provide useful information for interpreting the event types, e.g., the IJ and RJ flags (https:// aqs.epa.gov/aqsweb/codes/data/QualifierCodes.html)inform that 19-20 July was a high wind event.
Independent IMPROVE and satellite observations can also assist in validating these identified dust events.IMPROVE observations were only available on ∼ 29 % of these identified dusty days (7 and 10 December; 12 and 18 April; 13, 19, and 25 August; 15 September), and they were more likely to be able to indicate exceptionally strong and long-lasting events due to the 24 h sampling duration.Tong et al. (2012) reported two strong dust storm events at the PHOE1 IM-PROVE site (∼ 12 April; ∼ 20 July) using total PM concentrations and its speciation, both of which were also captured by our method.In addition, ∼ 48 % of the days impacted by strong blowing dust were possibly captured by MODIS (i.e., dust events occurred between 09:00 and 15:00 LT: 10 December; 27 March;11,12,16,18,and 20 April;19 and 30 July;7 and 15 September;15 November).To further demonstrate the advantages of using frequently sampled observations for capturing dust events, we plotted the time of occur-  rence of these AQS-/AZMET-based dust periods in Phoenix for this year (Fig. 5a).Dust events occurred more frequently during Aqua overpass times than during the Terra overpasses, consistent with the findings from Fig. 2. Most of these dust events occurred between 15:00 and 21:00 LT, when winds were stronger (also in Fig. 5a) and the soil was drier (derived by looking at NAM soil moisture at the top soil layer in recent years, not shown), rather than at MODIS overpass times from late morning to early afternoon times.A similar longterm diurnal variability of dust event occurrence has been found in Utah based on analyzing the weather code (Hahnenberger and Nicoll, 2012).Therefore, current polar orbiting satellites are unable to observe all dust events, and the hourly sampling frequency of the future geostationary satellites can help better capture dust events together with the surface monitoring network.Such conclusions were also drawn by Schepanski et al. (2012) for the African dust source regions.
We classified PM mass by wind direction observed at the Phoenix AZMET site, which indicates the dominant westerly/southwesterly winds at the Phoenix high dust times.Furthermore, based on the NARR meteorology, HYSPLIT air mass trajectories were originated from 500 m above the ground level (a.g.l.) of Phoenix at the identified dust periods to locate the origins of Phoenix dust episodes and indicate the regional transport patterns.The endpoints of these HYSPLIT backtrajectories are overlaid on the MODIS land classification map (Fig. 5b), showing that most of the transported dust particles were at the shrublands or deserts (primarily Sonoran, also Chihuahuan) 0-12 h before arriving in urban Phoenix areas below ∼ 900 hPa.This is consistent with the finding from Figs. 3 and S1, in which barren and sparsely vegetated open shrubland are the major contributors to the dust producing areas in 2007.

Case study of a recent strong dust event accompanied by a stratospheric ozone intrusion
Multiple satellites identified a recent dust event (10-11 May 2014) in the western US: as described by NOAA's HMS text product (http://www.ssd.noaa.gov/PS/FIRE/DATA/SMOKE/2014/2014E111659.html; http://www.ssd.noaa.gov/PS/FIRE/DATA/SMOKE/2014/2014E120143.html),dust was originated in southern California.It swept across northern Baja California and Arizona, and then entered New Mexico after a cold-frontal boundary and impacted Texas, Oklahoma and Kansas.We evaluated the current NAQFC PM2.5 (a standard NAQFC air quality modeling product) forecasting skill during this event and assessed the impact of dust emission on the regional air quality based on model sensitivity analysis.The NAQFC 12 km CMAQ base simulation produced 24 h mean PM2.5 of over 50 µg m −3 in western Arizona and > 15 µg m −3 in southwestern Arizona on 11 May 2014 (Fig. 6a).A sensitivity analysis using the base and no-dust simulations indicates that over 50 µg m −3 of hourly PM2.5 during this event were contributed from dust emissions in populated urban regions in Arizona (such as Phoenix in the Maricopa county and Tucson in the Pima county) and, on average, dust contributed to > 70 % of the total PM2.5 in most Arizona grid cells (Fig. 6b).
The modeled PM2.5 was evaluated mainly for the Maricopa and Pima counties in Arizona where both IMPROVE and AirNow observations were available during this event.Time series of observed and modeled PM2.5 are shown in Fig. 6c and d.AirNow observations indicate daily maxima to be over 100 µg m −3 in Maricopa (at ∼ 08:00 LT) and over 50 µg m −3 in Pima (at ∼ 14:00 LT), with PM2.5 / PM10 ratios at the dust times below 0.2 (not shown).Both the model and observations show significant temporal variability (standard deviations), indicating the advantages of the AirNow data for capturing the extremely high PM concentrations during the dust events.The model was fairly well correlated with the observations (with median/high correlation coefficients of 0.7-0.9;Table 2).CMAQ underpredicted the daily maxima in Maricopa by a factor of ∼ 2 with a 2 h lag, while it slightly overpredicted them in Pima with the right timing.PM was measured at more AirNow sites than at IMPROVE sites in both counties on this day.The observed 24 h mean concentration at the AirNow sites was lower than at the IM-PROVE sites in Maricopa, but those in Pima were close.This can be mainly due to the different sampling areas that AirNow and IMPROVE networks cover.The model underpredicted the 24 h mean values in both counties, with more significant negative biases in Maricopa than in Pima.
This dust event was accompanied by a stratospheric ozone intrusion, as shown from a RAQMS model simulation that assimilated ozone columns from the Ozone Monitoring Instrument and ozone profiles from the Microwave Limb Sounder, as well as the AIRS satellite products (Figs. 7,S6).Descending dry air containing rich ozone enhanced the surface ozone concentrations in eastern Arizona and New Mexico at late morning and early afternoon times, when dust was strongly impacting the same locations.Observed surface ozone at the Petrified Forest National Park in eastern Arizona (AQS/AirNow site no.040170119) at this time exceeded 65 ppbv.However, the current NAQFC CMAQ modeling system is unable to capture the exceptionally high ozone during stratospheric intrusion episodes, as the CMAQ lateral boundary conditions were downscaled from the monthly mean GEOS-Chem simulation in 2006 and no upper boundary conditions were used.

Conclusions and suggestions
We developed dust records in Arizona for 2005-2013 using multiple observation data sets, including the MODIS level 2 deep blue aerosol product and in situ measurements at the surface AQS and IMPROVE sites in Phoenix.Both satellite and surface aerosol observations were anticorrelated with three drought indicators (i.e., NDVI, soil moisture, and PDSI).Dust events were stronger and more frequent in the afternoon times than in the morning due to stronger winds and drier soil; in addition, the Sonoran and Chihuahuan deserts were important dust source regions during the identified dust events in Phoenix.These findings suggest a potential for use of satellite soil moisture and land products to interpret and predict dust activity.We also emphasized the importance of using hourly observations for the better representation of dust events, and we expect that the hourly geostationary satellite observations will in future complement the current surface PM and meteorological observations, especially considering their broader spatial coverage.Continued development of products from the polar-orbiting satellites is also important in that they can provide higher-spatialresolution observations from each swath due to their lower orbit level.Future efforts should also be devoted to better characterizing and attributing the observed dust, by integrating additional satellite measurements (such as ammonia as shown in Ginoux et al., 2012b) and in situ measurements of trace gases and aerosol compositions.
In a case study, we evaluated the capability of the current NAQFC CMAQ modeling system to capture the magnitude of aerosol concentrations and their temporal variability during a recent dust event.Sensitivity simulations from this modeling system assessed the impact of this dust event on western US air quality, and showed that dust contributed to > 70 % of the total PM2.5 in Arizona, on average.Satellite weather and land products are currently being integrated into dust emission modeling for future improvement of NAQFC's PM forecasting skill.Finally, we showed that this recent dust event was accompanied by a stratospheric ozone intrusion, and we emphasized the importance of representing both PM and ozone well under such conditions.
The Supplement related to this article is available online at doi:10.5194/acp-15-12595-2015-supplement.
Acknowledgements.This study was supported in part by a NASA ROSES grant (NNX13AO45G).We thank Janae Csavina and William Sprigg for their constructive comments on an earlier version of the manuscript.We thank the useful information from the NASA Air Quality Applied Science Teams, SMAP early adaptor working teams, and the NAQFC and HYSPLIT groups at NOAA ARL.We also acknowledge the open access to the surface and satellite observations used (the sources of data are included in the main text).The views, opinions, and findings contained in this paper are those of the author(s) and should not be construed as an official National Oceanic and Atmospheric Administration or U.S. Government position, policy, or decision.

Figure 1 .
Figure 1.Inter-annual variability of drought indicators in dust seasons: (a) MODIS NDVI on 0.1 • × 0.1 • horizontal resolution and (b) ESA multi-sensor soil moisture (SM) product on 0.25 • × 0.25 • resolution are shown on selected moderate-to-severe dry and wet years.The text in the upper left corner of each panel indicates the year of the data.(c) Time series of PDSI and the anomalies (i.e., the annual mean value over the multi-year mean value) of satellite SM, NDVI and Aqua MODIS DOD.The anomalies of satellite data were calculated using data within the box defined in (a).The inset panel in (c) shows the NOAA climate divisions, and PDSI values in the southwest (region 5) and south central (region 6) regions were used in the time series plot.

Figure 2 .
Figure 2. (a) DOD maps (in 0.1 • × 0.1 • horizontal resolution) in dust seasons from Terra MODIS during 2005-2013.Data are plotted only for the grids in which DOD data are available > 5 % of the total number of days in each year (defined as "areas of dust impact").The purple star in the upper left panel of (a) indicates the location of Phoenix.(b) Same as (a), but for Aqua MODIS.

Figure 3 .
Figure 3. MODIS-derived dust sources over the western US (from the MODIS tile grid horizontal 8/vertical 5, defined in Fig. S1) and in the southwestern US (lower, defined as the region within the box in Fig. 1a), during the dust seasons in 2005-2013.The absolute source areas for three types of land cover are shown in the left column, and the contributions (%) from individual land cover types to the total source areas are shown in the right column.

Figure 4 .
Figure 4. Time series of surface PM data at AQS and IMPROVE sites in Phoenix.These observations are shown in their original temporal resolution in (a), and their anomalies in each year's dust season are shown in (b), along with the Aqua MODIS DOD anomalies (i.e., the annual mean value over the multi-year mean value).

Figure 5 .
Figure 5. (a) Frequency of identified dust storms in Phoenix in 2007 as a function of the time of occurrence.Hourly mean wind speed (∼ half of the hourly maximum, with correlation coefficient of ∼ 0.93) during these dust storms is shown in red dots, and the inset panel shows the frequencies of PM10 within various concentration intervals by wind direction during these dust storms.(b) Hourly HYSPLIT endpoints colored by four time intervals, overlaid on a 500 m MODIS land cover type image.The MODIS land cover types mentioned in the text and their corresponding numbers are barren or sparsely vegetated: 16; urban and built-up: 13; open shrublands: 7; cropland: 12; and cropland/native vegetation: 14 (source: https://lpdaac.usgs.gov/dataset_discovery/modis/modis_products_table/mcd12q1).

Table 1 .
Data used in this study * .

Table 2 .
Evaluation of NAQFC CMAQ PM2.5 predictions during a recent dust storm event on 11 May 2014.County in Arizona Site typeNo. of sites Observed PM2.5 * Modeled PM2.5 * Units in µg m −3 ; mean ± standard deviation during this 24 h period shown for the AirNow results. *