The scientific literature is replete with satellite-based methodologies for lofted dust detection and characterization. Ultraviolet techniques (e.g., Herman et al., 1997; Torres et al., 1998, 2007) take advantage of differences (spectral- and angular-dependent) in backscattered radiances for absorbing and scattering aerosol species with respect to a molecular atmosphere. Moving into the visible light and near-infrared wavelengths, algorithms take advantage of the preferential absorption of blue light to enhance contrast over bright surfaces and vis-à-vis meteorological clouds (e.g., Miller et al., 2003; Hsu et al., 2004, 2013; Qu et al., 2006). Combinations of visible, shortwave-infrared, and thermal infrared are used to attain contrast between dust over water and over land, respectively (e.g., Shenk and Curran, 1974; Ackerman, 1989; Tanré and Legrand, 1991; Legrand et al., 2001; Hansell et al., 2007; Cho et al., 2013). Recent advancements have enabled the use of O2 A and B absorption bands and their nearby continuum channels to derive dust and smoke layer height over dark ocean and vegetated land surfaces (Xu et al., 2017, 2019).

The most commonly implemented techniques for satellite-based dust detection involve the reststrahlen band of silica (or quartz), a common and often significant constituent of mineral dusts found worldwide (Di Biagio et al., 2017) caused by the Si–O vibrational bending mode in the 8–10 µm range (Peterson and Weinman, 1969; Salisbury et al., 1987; Wald et al., 1998). This optical feature for mineral dusts results in elevated values of the real (scattering) and imaginary (absorption) parts of the complex index of refraction for silicates, producing commensurately higher values of extinction than surrounding wavelengths.

Passive radiometer narrowband channels positioned in the 8–12 µm atmospheric window (a spectral region where the gaseous atmosphere is largely transparent), used for imaging and describing the properties of meteorological clouds and the surface, can take advantage of the silica spectral fingerprint for dust detection. When used in tandem, the ∼10 and ∼12 µm narrowband spectral channels are often referred to as the split window, so called because although this part of the infrared spectrum is an atmospheric window, water vapor absorption is not entirely negligible; namely it is slightly stronger at ∼12 µm than at ∼10 µm. As such, this atmospheric spectral window is “split” between channels residing in relatively clean and more absorbing portions of it. Hereafter, we will refer to the brightness temperature difference (BTD) between these two wavelengths, i.e., T(10 µm)–T(12 µm) as the split-window BTD, or SWBTD.

Sensitivity of the SWBTD to atmospheric water vapor is well established, and it has long been incorporated as corrections to retrievals of sea surface temperatures (e.g., McMillin, 1975), land surface temperature (e.g., Wan and Dozier, 1996), and characterization of the lower atmosphere moisture itself (Lindsey et al., 2014). The enhanced absorption of water vapor at 12 µm produces a slightly positive SWBTD for conditions of a warm surface (skin) temperature and adiabatic lapse rate – a signal that is of opposite sign to that produced by mineral dust. Thus, the negative SWBTD signal produced by lofted dust may be counterbalanced by the positive SWBTD produced by water vapor in certain situations, especially when the dust is embedded within or residing below the “masking” moist layer. The masking effect of water vapor on the negative SWBTD dust signal establishes a working hypothesis for the missing dust plume of our southern Arabian Peninsula case study.

Moving away from the reststrahlen band, the strength of dust extinction decreases gradually from 10 to 12 µm with some variability in magnitude related to composition and shape (Hansell Jr. et al., 2011). This tendency typically results in a negative SWBTD that can become significant (e.g., values between 0 and -6 K) when the underlying surface is warmer than the lofted dust layer. Assuming other factors (e.g., dust layer height) are equal, the dust signal is most pronounced during the daytime over deserts, when the skin temperature is much higher than the air temperature and the lapse rate of the lower atmosphere is close to dry adiabatic. The optical properties of meteorological clouds differ from those of mineral dust, making the SWBTD test useful for isolating lofted dust layers in a cloudy scene. Optically thin cirrus clouds typically exhibit a relatively large positive SWBTD – opposite to the dust signal (e.g., Strabala et al., 1994). Optically thick clouds tend to show near-zero to slightly positive values of the SWBTD. However, intervening clouds (above the dust layer) will obscure and mask out the SWBTD dust signal, limiting the applicability of this passive detection satellite technique to cases of cloud-free line of sight from the satellite to the dust layer.

The European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT; https://www.eumetsat.int/website/home/Data/Products/Atmosphere/index.html, last access: 9 September 2019) demonstrates a red–green–blue (RGB) dust enhancement technique (EUMETSAT Dust RGB; e.g., Lensky and Rosenfeld, 2008) that takes advantage of the SWBTD dust signature. The appeal of the infrared-based techniques is their 24 h utility, although the strong dependencies on the atmospheric profile and the diurnal pattern of surface temperature can make for varying performance and ambiguity. Over deserts, the surface possesses a similar mineralogy and radiometric behavior to the locally lofted dust, producing SWBTD false alarms. Over ocean surfaces, where the temperature of the water may be close to or cooler than the lofted dust layer, the SWBTD dust signal is inherently weaker and does not become significant unless the dust layer itself is optically thick. These challenges are readily apparent in the EUMETSAT RGBs but are overcome in part by the application of these algorithms to geostationary data, in which case motion helps analysts to differentiate between actual dust and false alarms locked to the surface structures.

Work-arounds to the surface signal ambiguity problem have been attempted (e.g., Legrand et al., 2001; Tramutoli, 2005, 2007). Miller et al. (2017) employ a front-end cloud mask with a priori information on the clear-sky background surface emissivity to identify and suppress the undesired enhancement of land surfaces, via a Dynamic Enhancement Background Reduction Algorithm (DEBRA). However, DEBRA's ability to detect lofted dust through improved signal to noise nevertheless faces the same physical limitations as the EUMETSAT Dust RGB product, as it enlists the same spectral bands (and leverages the same physics) to identify dust layers.

In both SWBTD-based dust detection techniques, we hypothesized that the effects of water vapor absorption in the lower atmosphere would significantly impact performance. Although dust characteristically is associated with arid environments and inherently lower column water vapor, cases do arise when dust is lofted within or transported into anomalously moist (with respect to regional climatology) air mass environments. The effects of SWBTD dust signal suppression by water vapor in terms of moisture amount, dust amount, temperature profile, and positioning of the dust layer within the moist profile have not been examined systematically in the literature. We begin to explore some of these questions here.

The following is a synopsis of the case study based on analysis of numerical modeling and surface station observations. The Navy Global Environmental Model (NAVGEM; Hogan et al., 2014) and the Weather Research and Forecasting model coupled with Chemistry (WRF–Chem; Grell et al., 2005) model were used to interpret aspects of the synoptic-scale and mesoscale conditions for this case. WRF–Chem was coupled to a Naval Research Laboratory (NRL) dust source database for southwest Asia (Walker et al., 2009) to examine dust lofting and transport. Figure 2 shows a NAVGEM analysis of the 500 hPa geopotential height field and the relative vorticity during the initial stages of the event. Figure 3 shows WRF–Chem simulations for the column integrated dust mass for selected times across the 3–4 August period leading up to the satellite imagery shown in Fig. 1. Both dust plumes seen in satellite imagery were captured by this coupled model system. This case study was also simulated using the Regional Atmospheric Modeling System (RAMS; Cotton et al., 2003; Saleeby and van den Heever, 2013), and differences in the amounts of dust lofted between the WRF and RAMS simulations are detailed by Saleeby et al. (2019).

For 3 August 2016 at 00:00 Z, NAVGEM analysis (Fig. 2) showed a broad 588 dm geopotential height upper-level ridge over northeast Africa and the Arabian Peninsula. At this time, a southward-moving shortwave over southern Iraq started to traverse around this high-pressure ridge and descend into the central Saudi Peninsula. Between 05:00 and 06:00 Z on 3 August, this shortwave intensified the winds from 700 hPa down to the surface, mobilizing dust from southern Iraq across the Ad Dahna desert to the Rub' al Khālī over the eastern edge of the Saudi plateau. This dust was transported within a very dry air mass (TPW < 20 mm; discussed in Sect. 3.4). As the shortwave continued to move around the ridge and southwestward on 4 August, this lofted dust was readily observable over the Saudi plateau in various dust-enhanced satellite imagery products.

Between 12:00 and 18:00 Z on 3 August, both model and surface observations showed a mesoscale low-pressure system forming over the SG east of Qatar. The alignment of the Al Hajar mountains in northern Oman and the Zagros Mountains in southwestern Iran aided in the blocking and redirection of the surface winds, reinforcing the cyclonic flow around the surface low in the SG. Strong southerly winds formed in response to this surface low, mobilizing dust from 3 August 18:00 to 4 August 09:00 Z over an area extending from Oman to the coast of the United Arab Emirates (UAE).

During this period, surface stations from the Oman coast to the UAE coast reported southerly winds at 2.5–10.5 m s-1 (5–20 kn) and were accompanied by observations of dust in suspension, dust raised by winds at the time of observation, dust devils, and slight-to-moderate dust storms. Evidence of dust lofting and pooling along a surface front formed by this southerly surge, as captured by WRF–Chem coupled to the NRL dust source database, is noted in Fig. 3b (animation provided as S1 in the Supplement). Convection initiated in this region may have produced haboobs (Miller et al., 2008) that contributed as a secondary source to the model's total dust loading. The correct placement and timing of such convection is a major challenge for numerical modeling and thus poses a source of uncertainty in regional dust forecasting. Dust from these various sources comprised the eastern plume, which was transported into the SG on 4 August (Fig. 1).

Surface observations collected across the southeastern portion of the Arabian Peninsula between 09:00 and 12:00 Z on 4 August show that the surface winds weakened across the region over this period, although dust storms and blowing dust conditions prevailed. By 18:00 Z, nearly all stations across the region reported dust in suspension, with only one station still reporting an active (i.e., blowing) dust storm. Although the surface dust lofting event had ended by this time, satellite imagery showed copious dust aloft, being transported northward. This dust remained in suspension over the region for the following 2 d before being dispersed by the synoptic-scale flow or settling out.

Figure 4 shows true color, SWBTD, and DEBRA dust-enhanced imagery for a zoomed-in portion of Fig. 1 as observed by the Visible Infrared Imaging Radiometer Suite (VIIRS) instrument on the Suomi National Polar-orbiting Partnership (Suomi NPP) satellite. In Fig. 4a, the tan coloration of the plume in the SG (as noted by label “2” in the panels of Fig. 1) is characteristic of lofted dust. Although not as obvious over the bright desert interior due to poor brightness contrast, low color contrast, and relatively low surface variability, the true color imagery shows this plume originates from inland portions of eastern UAE, flanking the Al Hajar mountain range in northern Oman. The preferential absorption of blue-wavelength light by iron-bearing minerals, responsible for the perceived tan coloration, was also keyed on by visible-based dark-target dust enhancements from the National Aeronautics and Space Administration (NASA) Aqua satellite's Moderate Resolution Imaging Spectroradiometer (MODIS) (e.g., Miller, 2003; not shown), confirming the composition of the plume as silicate dust as opposed to meteorological clouds, biomass smoke, or pollution. This dust would also be expected to produce a distinct negative SWBTD signal and thus be readily enhanced by DEBRA, but enhancement was not achieved in this case. Instead, the SWBTD signal (Fig. 4b) in this region is near zero or slightly positive, resulting in DEBRA missing the dust plume (i.e., not enhanced as yellow in Fig. 4c) almost entirely.

DEBRA includes a cloud mask preprocessing step, which enlists the SWBTD as a “restoral test” for pixels erroneously flagged as cloud instead of optically thick dust. It was further confirmed, by temporarily turning it off in the processing, that the cloud mask itself was not contributing to the missing dust in the original DEBRA algorithm. As the EUMETSAT Dust RGB algorithm (Fig. 1b), which also misses the eastern dust entirely, does not enlist a cloud mask, this exercise was done simply as a sanity check before pursuing alternative explanations for the missing dust.

Geostationary MSG-8 visible-band imagery (time-resolved at 15 min, provided as S2 in the Supplement) provides insight into the additional factors contributing to the eastern dust plume. The imagery shows morning-time convection initiating offshore of Oman, between the coastal cities of Lakabi (18.23∘ N, 56.55∘ E) and Ras Madrakah (18.98∘ N, 57.79∘ E) on 3 August 2016 at 04:00 UTC (∼ 08:00 local time). By 08:15 UTC, a well-defined boundary marked by leading-edge cumulus formed and began moving inland in a northwesterly direction. Such features are indicators of a cold pool outflow and gust front associated with evaporative cooling, which is common to convection occurring in dry lower atmospheric environments. This outflow may have produced additional dust lofting via the haboob mechanism (e.g., Miller et al., 2008). The gust front appears to superimpose with the daytime inland-moving sea breeze front along coastal Oman. The presence of dust was inferred by high visible reflectance and reduced land surface texture behind the front, although standard dust-enhancement products (SWBTD) gave no indication of lofted dust.

By 13:00 UTC (17:00 local time) on 3 August 2016, the coastal Oman dust-laden front had just crossed the border between Oman and southern Saudi Arabia at 20∘ N, 55∘ E. Infrared signatures of clouds oriented along the front were lost by 18:00 UTC. By this time, the front had come under the influence of the southerly flow (the principal source of dust lofting in the missing plume; described in Sect. 3.1), taking a more northward track along the Saudi–Omani border, moving toward the southern border of the UAE. Both visible (Fig. 1a) and true color (Fig. 4a) imagery from the following day show a dust plume which consists of both local and secondary regionally lofted sources as described, extending through central UAE and nosing offshore into the SG.

The well-detected and missed dust plumes, as observed in juxtaposition on 4 August 2016, contained very different spectral properties across the infrared window, in which the SWBTD operates. Figure 5 shows information from Aqua MODIS and Atmospheric Infrared Sounder (AIRS) instruments, collected at 09:15 UTC. Figure 5a shows the domain of interest via MODIS true color imagery and denotes locations of selected AIRS spectral plots shown in Fig. 5d. Locations 1–2 correspond to offshore and onshore locations within the SWBTD-missed dust plume, and location 3 is within part of the SWBTD-detected dust plume. Figure 5b shows MODIS retrievals of aerosol optical depth (AOD) via the visible-based “Deep Blue” algorithm (Hsu et al., 2004, 2013). Figure 5c shows a field of 10.35–12.3 µm (monochromatic wavelengths; not band averaged) SWBTD from AIRS, where negative values shown in blue correspond to dust detections.

AIRS spectra (and associated SWBTD values) at the selected dust plume locations noted in Fig. 5a are shown in Fig. 5d. The 1650 km wide swath width of AIRS (narrower than the 2330 km swath of MODIS) accounts for the differences in the width of data gaps between adjacent satellite overpasses. While the AIRS swath gap on this day corresponded to a significant portion of the well-detected dust plume, a portion of the western swath intersects the southwestern edge of the same storm (near location 3) and thus is relevant to the comparison of dust signals between the two air masses. Whereas the spectra (Fig. 5d) associated with location 3 produce a strongly negative SWBTD, indicative of lofted dust, locations 1 and 2 produce positive SWBTD values – explaining the inability of the EUMETSAT and DEBRA methods to infer the presence of dust at those locations.

Passive imaging radiometers, while providing useful information on the horizontal distribution and column-integrated properties of dust storms, offer very limited information in terms of vertical structure. Active optical sensors such as lidar are very useful in this regard, but terrestrial systems are few and far between. Even less common are spaceborne lidars, which extend coverage to the global scale but typically provide only a curtain slice of the atmospheric column, owing to their non-scanning configuration. The resultant poor spatial sampling, combined with relegation to polar-orbiting satellites that offer infrequent revisits of a given location, reduces the chances of leveraging spaceborne lidar observations for a localized case study. At the time of this case study, two NASA cloud–aerosol lidar systems were operational – the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP; Hunt et al., 2009) on the NASA Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) satellite (Winker et al., 2009) and the NASA Cloud-Aerosol Transport System (CATS; McGill et al., 2015) on the International Space Station (ISS).

By what can best be described as a stroke of incredible serendipity, both the ISS and CALIPSO satellites carrying these lidar systems crossed over the same region of the UAE within minutes of each other (nearly a simultaneous nadir overpass) on the evening between 3 and 4 August 2016. Moreover, the conjunction occurred at the very time that the “missing” dust plume was forming and being transported northward by various mechanisms described in Sect. 3.1 and 3.2. Traveling from NW to SE on the ISS, CATS crossed over the UAE on 3 August at ∼ 22:08 UTC. Roughly 15 min later, at ∼ 22:23 UTC, CALIOP crossed the same area on its NE-to-SW descending node track. This conjunction of active systems provided a golden opportunity to assess the vertical structure of the missing dust plume at a critical time in its development.

Figures 6 and 7 show the satellite ground tracks and corresponding lidar profiles of total attenuated backscatter for CATS and CALIOP, respectively. Differences seen in the magnitudes of total attenuated backscatter are due primarily to the different wavelengths (1064 nm for CATS and 532 nm for CALIOP) shown in these figures. Both profiles indicate a deep layer (0–5 km above mean sea level, AMSL) of suspended aerosol (identified by CALIPSO algorithms, not shown, as likely dust) common to the region during this time of year (e.g., Nabavi et al., 2016). Embedded within this background are more significant backscatter features tied to local, recently lofted and optically thicker dust. Meteorological clouds, especially those in the middle and lower atmospheres composed of mixed or liquid phase droplets, act as very strong attenuators of lidar energy. Attenuation from these clouds, also present within the CATS and CALIOP profiles, accounts for the occasionally seen vertically oriented dark stripes or bands in the lower atmosphere of both lidar profiles.

A key item of interest in these lidar profiles is the labeled “Frontal Feature”, seen in both CATS and CALIOP observations over the southeastern UAE. This feature arcs from the surface up to ∼2–3 km. The diffuse character of its lower and upper boundaries, combined with its significant backscatter (entering the range of the color scaling typically relegated to cloud signals), is indicative of an optically thick dust layer. In the CALIOP data of Fig. 7, the same arc feature seen in the CATS data of Fig. 6 is present but appears less continuous, due to the intermittent beam attenuation by the overriding mid-level clouds encountered along the NE approach of CALIPSO but avoided by the W-NW approach of the ISS.

The timing and location of the frontal feature are consistent with the WRF–Chem and RAMS model simulations and surface station observations of a southeasterly flow surge associated with the forming low pressure over the SG (Sect. 3.1). The southbound components of both CATS and CALIOP would have encountered the leading edge of this frontal structure. The effects that a frontal surge would have on dust already in suspension, in terms of the dust pooling and riding up and atop the surface front, is consistent with the lidar observations. Thesa effects are identified in the model simulations: in Fig. 3b, the area labeled “F” shows the bowing of suspended dust along the southerly surge and also as northbound streamline plumes characteristic of freshly lofted dust forming in its wake. From these dual lidar observations, we infer that the missing dust plume has significant mass loading in the 0–3 km layer (∼1000 to 700 hPa) of the atmospheric column.

We turn our attention now to the topic of atmospheric moisture, which is hypothesized to have played a governing role in the performance (or lack thereof) of the infrared-based dust detection techniques. Figure 8 shows WRF–Chem simulations, corresponding to Fig. 3, of the 24 h period leading up to the 4 August 2016 event depicted in Fig. 1 (animation provided as S3 in the Supplement). The simulations show a very dry (low TPW) air mass in cyclonic rotation (the shortwave discussed in Sect. 3.1) descending through central Saudi Arabia, which is associated with the well-detected dust plume. Close inspection of Figs. 1, 4 and 5 reveals that this dry air mass correlates positively with the well-detected western dust plume. Meanwhile, a tongue of high TPW extends across Oman and northward through the eastern UAE. Strong southerly winds entrained relatively moist air (TPW > 45 mm) from the Arabian Sea maritime air mass. The frontal surge that occurred over the evening hours of 3–4 August is manifest in Fig. 8b (compare to Fig. 3b) as an enhanced gradient of moisture. Cross-referencing once again to observations in Figs. 1, 4, and 5, this moist air mass envelopes the region of the missing eastern dust plume.

Properties of the dry and moist atmospheric air masses were also characterized directly by independent satellite retrievals. The National Oceanic and Atmospheric Administration (NOAA) Unique Combined Atmospheric Processing System (NUCAPS; Gambacorta and Barnet, 2013), run operationally since 2013, uses cloud-cleared radiances and an iterative regularized least squares minimization algorithm to produce vertical profiles of temperature, water vapor, and trace gases from microwave and infrared radiances. The retrieval enlists input from the Cross-track Infrared Sounder (CrIS) and the Advanced Technology Microwave Sounder (ATMS) on NOAA's Suomi NPP and NOAA-20 satellites. NUCAPS provides 30 retrievals at 100 levels between 1100 and 0.16 hPa across a 2200 km swath. Owing to the projection of the cross-track scanning sensor footprints, spatial resolution ranges from ∼50 km at nadir to ∼70km×134 km at the scan edge. Nalli et al. (2016) demonstrate that NUCAPS water vapor profiles compare favorably to radiosonde data and to numerical weather prediction models in terms of moisture magnitude and gradients in a variety of atmospheric flows.

Figure 9 shows NUCAPS retrievals for the Suomi NPP pass over the southern Arabian Peninsula on 4 August 2016 at 09:21 UTC (same overpass as shown in Fig. 4). A plan view of low- to mid-troposphere (surface-to-500 hPa) integrated water vapor (IWV; Fig. 9a), a sub-column of the TPW, shows a tongue of moisture extending from coastal Oman to the SG. The structure of this moist air mass, drawn northward by the surface low over the SG as discussed in Sect. 3.1, is consistent with the WRF–Chem analysis shown in Fig. 8. Selected NUCAPS temperature and dew point temperature profiles for two locations within portions of the dry and moist air mass dust plumes are shown in Fig. 9b and c. NUCAPS data show large differences in low- and mid-tropospheric IWV (∼15–20 mm), especially at mid-tropospheric levels around 500 hPa where the difference in dew point depression between locations 1 (dry) and 2 (moist) is ∼35 ∘C.

These diverse model and observational datasets provide a comprehensive understanding of the meteorological conditions associated with the case study. Further, they motivate a more in-depth analysis, based on RTM simulations and in consideration of varying dust optical properties, of the potential impacts of atmospheric moisture differences on the SWBTD dust signal. This analysis is aimed at determining the necessary conditions for producing the missing dust outcome in the case study, while looking toward comprehensive methods of mitigating the water vapor masking effect on infrared detection techniques in general.

Differences in SWBTD among various dust plumes may arise from water vapor masking, different optical properties, optical depth, or more likely a combination of all factors. The composition of mineral dust varies as a function of location, yielding different optical properties (and commensurate SWBTD signal strengths) for a given mass loading and size distribution. Thus, for the current case study it is important to take an inventory of variations due to a range of refractive indices (RIs) characteristic of dust sources over the Arabian Peninsula and to consider to what extent these variations might account for the disparate behaviors of the infrared (IR) dust detection performance in the two dust plumes.

Several sets of RI for varying dust types characteristic of the Arabian Peninsula reside in the literature. For this analysis, we enlisted recommendations from (i) the quartz-dominated (99 % quartz and 1 % hematite) species from the Aerosol Refractive Index Archive (ARIA; http://eodg.atm.ox.ac.uk/ARIA/, last access: 9 September 2019), (ii) the Optical Properties of Aerosols and Clouds database (OPAC; Hess et al., 1998), and (iii) Saudi Arabian dust properties from Di Biagio et al. (2017; hereafter DB17). The values of these RI databases, along with spectral optical properties derived from Mie theory for 2.4 µm effective radii (as defined by Hansen and Travis, 1974) dust at wavelengths used in constructing the SWBTD, are provided in Table 1. Mass extinction coefficients (kext; m-2 kg-1) are multiplied by the dust loading (mass mixing ratio), the density of air, and the geometric thickness of the dust-laden model layers to compute the dust optical thickness. The dust optical thicknesses of each layer are then combined to yield the total column dust AOD. The single-scattering albedo describes the fraction of the extinction that is due to scattering processes (vs. absorption), and the asymmetry parameter provides a metric for directional scattering (1.0 denotes complete forward scatter, -1.0 denotes complete backscatter, and 0.0 denotes isotropic, or directional-independent, scatter).

The spherical particle approximation of Mie theory was assumed in computing the dust optical properties. Using simulated and measured spectra, Klüser et al. (2016) found significant variance of infrared dust optical properties for spherical vs. aspherical (specific “habits” of needles and disks) and varying mineralogy and assumptions. Specifically, while the general structure of extinction is similar among the permutations, they find shifts to peak extinction and moderate variations to the structure of single-scattering albedo, phase function, and its associated asymmetry parameter. Mixtures of mineral components, randomly oriented irregular particle shapes which may skew the bulk optical properties toward those of oblate spheroids (as in the case of ice clouds; Baran et al., 2005), and the phase function smoothing effects of multiple scatter in optically thick dust media considered here may reduce the magnitude of disparities implied by the idealized simulations. Suffice to say that Mie theory, or any fixed assumption on dust optical properties, comes attached with uncertainties (Meng et al., 2010). In what remains a highly under-constrained problem, the simulation results presented hereafter should be interpreted as representative of bulk dust properties that are, to first order, consistent with the satellite-observed SWBTD behavior.

With these caveats in mind, an effective dust radius of 2.4 µm was used in the idealized RTM calculations, corresponding to the large particle radius mode of an averaged dataset collected at the Kuwait University AERONET site during 2016 (Fig. 10). Overplotted in Fig. 10 are the averaged WRF–Chem model binned size distributions for atmospheric columns having total summed dust volume concentrations (dust loading) greater than 0.6 µm-3 µm-2 for the current case study. These data show that the model assumptions made here were consistent with the AERONET-observed large particle radius mode. This representativeness is important for simulations in Sect. 4.3.

Table 2 shows example values of AOD computed at the SWBTD component wavelengths for three different total column dust loadings (40, 80, and 186 in units of columnar volume concentration; µgdust kgair-1) and the three sets of assumed RI (Table 1) based on Mie theory for a dust effective radius of 2.4 µm. At 10.35 µm, the real part of the RI (which dominates the extinction via scattering) is much lower for DB17 than the OPAC and ARIA databases, resulting in an AOD that is about a factor of 2 smaller. At 12.3 µm, the ARIA and DB17 RIs produce similar AODs, while OPAC is significantly larger – the difference in this case being OPAC's larger imaginary RI component (translating to enhanced absorption). Despite these differences, the proportional relationship of extinction between the two wavelengths maintains and AOD (10.35 µm) > AOD (12.3 µm) among the various composition and loading assumptions. Hence, the sign of the SWBTD for mineral dust for extinction alone remains negative for all three databases. However, the magnitude of the SWBTD for a given RI database will vary as a function of different temperature, moisture, and dust profile scenarios.

Corresponding values of SWBTD from the three dust RI databases were calculated using the hybrid Eddington RTM of Deeter and Evans (1998), following its implementation by Grasso and Greenwald (2004). Figure 11 shows results for idealized single dust layers having optical properties corresponding to ARIA, with an effective particle radius of 2.4 µm, residing in a dry adiabatic temperature profile with surface temperature of 300 K. A 2 km thick dust layer of variable AOD (referenced at 10.35 µm) shown (from clear sky to optically thick; AOD > 10) was raised through the atmospheric column, and the column moisture was increased linearly, which accounts for the change in TPW along the x axis. The AOD values of 0.254, 0.507, and 1.183 are a function of the dust loading values of 40, 80, and 186 (µg(dust) kg(dry air)-1), respectively. The y axis of Fig. 11 shows magnitude of the computed SWBTD. Included is the clear-sky reference (AOD = 0), which shows a trend toward positive SWBTD as TPW increases, owing to larger 12.3 µm vapor absorption.

The effect of increasing TPW on the structure of the dust signal in Fig. 11 follows this same positive trend, with little variation in dust layer height for low-AOD layers. However, as the dust layer AOD increases, two principal effects are noted. The first effect is the increasing spread of SWBTD values, for a given dust AOD and TPW value, among a family of dust layer heights. The lower altitude layers are less negative than the more elevated layers, as the former reside below a deeper column of atmospheric moisture and thus experience greater suppression of the negative SWBTD dust signal. The second effect is, for a given dust AOD and for increasing TPW, the divergence of SWBTD for the family of dust layer heights. This behavior appears for values of AOD > 0.5 and is most prominent for high values of AOD. Here, very little impact of water vapor is seen for dust layers in the middle to upper atmosphere, while strong, non-linear impacts occur for the low-altitude dust. The spread of SWBTD within a family of dust layer heights over the range of TPW shown varies from a few degrees to 10∘, with values of 5–7 K for an optically thicker (AOD 0.5 to 2.0; for example see Fig. 5b) dust plume in the lower atmosphere.

Figure 12 compares the RTM analyses for ARIA, OPAC, and DB17 RI databases assuming the same effective dust particle radius, loading, and atmospheric profile assumptions as in Fig. 11. The expected sensitivity of the SWBTD signal to the dust layer height is evident in all three databases, with higher-altitude dust layers invariably producing stronger (i.e., more negative) SWBTD dust signals. Also, the higher water vapor concentrations of the moist profiles (TPW > 30 mm) result in weaker (i.e., less negative, or in some cases positive) SWBTD dust signals compared to the dry (TPW < 20 mm) profiles. The role of AOD is also evident: since the RI of DB17 yields a much lower 10.35 µm AOD, its associated SWBTD dust signals were weaker (less negative) than the other two databases. The ARIA database, based on a nearly pure quartz composition, produced the largest SWBTD dust signals. Interestingly, DB17 exhibits a greater spread among the dust layer altitudes for higher values of AOD (Fig. 12h, i). The more scattering nature of DB17 extinction at 12.30 µm compared to the other databases (ωo in Table 1) may produce a radiometrically cooler dust layer temperature than more absorbing dust layers – a signal becomes increasingly apparent as the influence of the overlying water vapor decreases as the dust layer's altitude is increased.

Despite the differences in magnitude of signal among the three representative dust types considered, the trends in SWBTD associated with dust loading, atmospheric profiles, and the heights of the dust layers were similar in the dry atmosphere. Thus, whereas variation in the dust mineralogy cannot be neglected when considering the selection of SWBTD thresholds, it alone cannot explain the lack of a negative SWBTD dust signal as was observed in the 4 August 2016 missing plume case, particularly for the optically thick component of the missing plume residing over inland portions of the UAE. Thus, the role of water vapor must be considered for these IR-based dust detection techniques.

One practical way to illustrate the impact of water vapor on SWBTD-based dust detection under more realistic (non-idealized) conditions is via radiative transfer simulations of the SWBTD conducted on the fully configurable environmental state of a forecast model. Specifically, we can examine the differential signal for dust as would be observed by a satellite for spatially varying lofted dust within an atmosphere with and without water vapor. A WRF–Chem forecast was used for this exercise, valid for 12:00 UTC on 4 August 2016. Whereas the model does not capture the exact details of the moisture and lofted dust distributions for this case, it is sufficiently representative to illustrate the impacts of water vapor on SWBTD-based dust detection. Figure 10 demonstrates the representativeness of the model's dust distribution for this case. For the radiative transfer calculations, again using the hybrid Eddington model of Deeter and Evans (1998) and following its implementation by Grasso and Greenwald (2004), we used dust concentration and spatial distribution as determined by WRF–Chem and assigned dust optical properties based on the ARIA database described in Sect. 4.2.

The results of these simulations are shown in Fig. 13. Figure 13a shows the distribution of IWV in shading, and the location of dust plumes, in terms of the 10.35 µm AOD, as contours. Dust in the 1.4 µm WRF–Chem model size bin was used, as it provided the closest match to the DEBRA-based satellite observations of the dust distribution for this case. The two significant dust plumes are evident in the contours of Fig. 13a, although the model's moist air mass dust plume is displaced to the south and west of where it was observed (this shift is of no consequence for the illustration of water vapor impacts). Figure 13b shows in shading the SWBTD dust signal for a “dry” (i.e., water vapor mixing ratios were set to vanishingly small values of 1.0×10-8g kg-1 at all levels) atmosphere. The differential effect of moisture on this signal, shown as contours in Fig. 13b, is defined as SWBTD (moist) minus SWBTD (dry). In this case, “moist” pertains to the original distribution of atmospheric moisture in the model. Positive values of this difference show how column moisture is making the SWBTD less negative and thus weakening the observable dust signal that is used for dust enhancement by algorithms such as DEBRA.

The impacts to DEBRA can be evaluated directly, by producing a suite of synthetic observations and running these through the DEBRA algorithm. Based on this approach, Fig. 13c shows how DEBRA would perform in a completely dry environment and under the assumption that the background is completely characterized. Here, the region of the non-zero dust detection confidence factor is widespread and is indicative of the pervasive dust suspension in this model run. Figure 13d shows the effects of DEBRA dust detection suppression by the inclusion of atmospheric moisture for the exact same model distribution of lofted dust. While parts of the two significant dust plumes remain enhanced in the full-atmosphere simulation, the preferential suppression of the dust signal particularly in moist regions is very evident. This synthetic DEBRA performance is consistent with the idealized single-column simulations of Sect. 4.2, and its structure is anticipated by the location of significant positive-valued contours in Fig. 13b. These simulated results are also consistent with the observed disparity in IR-based dust detection for the dry and moist environment dust plumes.