High-altitude dust layers on Mars: Observations with the Thermal Emission Spectrometer



[1] Limb-scanning observations of Martian atmospheric dust with the Thermal Emission Spectrometer (TES) over 3 Mars years indicate two distinct altitude layers with persistent maxima in the dust mixing ratio vertical profile. The first, lower maximum in the dust distribution profile is the “high-altitude tropical dust maximum” (HATDM) centered at 20–30 km, previously detected by the Mars Climate Sounder (MCS). Through the observation period, the HATDM followed a repeatable seasonal cycle with a brief absence in early northern spring and reached its highest altitudes and largest amplitude during the dust storm season in southern spring and summer. The HATDM is likely maintained during the day by a combination of convective and topographic updrafts and then degraded at night by scavenging from water ice clouds. The second, upper maximum in the dust distribution profile, which we refer, for convenience, to as the upper dust maximum (UDM), is centered at 45–65 km and is only detected in daytime observations. We see additional evidence of its presence in the limited number of MCS aerosol opacity retrievals available at these altitudes. Comparable dust mixing ratios are nearly absent from this altitude range at night. This upper maximum is generally a northern hemisphere phenomenon, peaking in amplitude in northern summer and nearly absent from the TES observational domain during the dust storm season. We suggest topographic updrafts over Martian volcanoes, small particle size, diurnal transport associated with thermal tides, and scavenging by water ice as probable key factors in the creation of the UDM.

1 Introduction

[2] Mineral dust plays a key role in Martian atmospheric dynamics. The background haze level of dust significantly warms the atmosphere, and periodic global dust storms can radically alter the atmospheric circulation and temperature structure on time scales of weeks to months [Martin and Kieffer, 1979; Haberle et al., 1982; Smith et al., 2002]. Dust possesses a highly variable spatial and temporal distribution as a result of the complex mix of processes that govern its lofting: dust devils, dust storms, and topographic and convective updrafts [Newman et al., 2002a, 2002b; Kahre et al., 2005, 2006; Fisher et al., 2005; Michaels et al., 2006; Fuerstenau, 2006]. Once lofted into the atmosphere, complex feedback effects between the radiatively active dust and the atmospheric circulation begin to occur; these feedbacks have been modeled with limited success [Richardson and Wilson, 2002; Basu et al., 2004, 2006; Kahre et al., 2006]. Higher atmospheric dust loading has been confidently linked to a deepened circulation [Haberle et al., 1982; Wilson, 1997; Newman et al., 2002a, 2002b], enhanced thermal tides [Leovy and Zurek, 1979], and more intense polar warming [Wilson, 1997].

[3] Discrete high-altitude layers of aerosols in Mars' atmosphere have been imaged since Mariner 9 first reached orbit [Anderson and Leovy, 1978]. Observations of similar phenomena were made with the Viking orbiters [Jaquin et al., 1986], which suggested dust was reaching heights of 50 km; with the Phobos spacecraft, which indicated an enhanced layer of dust near 22 km [Korablev et al., 1993]; and with Mars Global Surveyor (MGS) [Cantor, 2007]. Radiative transfer modeling of the Viking orbiter data further hinted that the vertical profile of dust included discrete layers of enhanced dust concentrations [Clancy and Lee, 1991; Montmessin et al., 2002].

[4] More recently, the Thermal Emission Spectrometer (TES) observed the 2001 global dust storm in limb-scanning mode and observed dust lofted to heights above 60 km [Clancy et al., 2010]. The improved vertical resolution of the Mars Climate Sounder (MCS) has shown that dust mixing ratios have a peak in the tropics near 20 km altitude and that the vertical dust profile is not well represented by the customary Conrath-ν profile [Conrath, 1975; McCleese et al., 2010; Heavens et al., 2011a]. Recently published comprehensive Spectroscopy for Investigation of Characteristics of the Atmosphere of Mars (SPICAM) solar occultation observations show pervasive high-altitude aerosol layers with climatological similarities to the dust layers discussed in sections 'LDM Discussion' and 'UDM Discussion' [Määttänen et al., 2012].

[5] This work describes TES limb-scanning observations that expand the climatology of the tropical 20 km altitude dust mixing ratio peak first described in Heavens et al. [2011a] and that detect a second maximum in the vertical profile at 45–65 km. Section 2 describes the climatology of the two dust maxima as observed with TES. Section 3 discusses the possible mechanisms that could support these layers. Finally, section 4 states our conclusions.

2 TES Limb Observations

[6] TES has provided a wealth of atmospheric data that have significantly expanded understanding of Mars atmospheric dynamics and properties [Christensen et al., 2001; D. E. Smith et al., 2001; Smith et al., 2004]. Arriving in 1997, TES (and the MGS spacecraft) provided nearly 10 years (nearly 5 Mars years) of measurements in both nadir-pointing and limb-scanning operation modes. This paper uses the limb-scanning observations from Mars year (MY) 24 at Ls = 110° (March 1999) to MY 27 at Ls = 80° (August 2004). Limb scan retrieval methods and algorithms are described in more detail in the appendix and by McConnochie and Smith [2009] and McConnochie et al. [2009]. While, in nadir-pointing mode, TES retrievals provide column-integrated aerosol absorption opacity (at a wavelength of 9.3 µm) and a vertical temperature profile at a resolution of approximately 10 km, from the surface to approximately 35 km altitude [M. D. Smith et al., 2001]. Limb scans have a similar vertical resolution, but the observable domain extends to near 65 km altitude and vertical profiles of aerosols can be retrieved [Smith, 2004]. Limb scan observations below 10 km altitude are considered unreliable due to excessive influence from surface reflectance and have been omitted from the figures. The complete set of TES limb scan observation profiles was binned by 10° of solar longitude (18–19 Martian solar days (sols)), 30° in longitude, 10° in latitude, and 5 km in altitude. Typically, each bin value is the average of approximately 50 profiles. The data presented in the figures have been smoothed with a 2 × 2 horizontal bin sliding boxcar average. For individual retrievals, the standard errors of the parameters (i.e., dust and ice extinction at each of the six retrieved levels) are the 1 sigma uncertainties from the retrieval covariance matrix, multiplied by the square root of the reduced chi-square of the retrieval fit. These standard errors are then propagated through the weighted averaging process to form the binned data, using an assumption that the retrieval errors are uncorrelated from retrieval to retrieval (see the appendix for additional details). Standard fractional errors in dust mixing ratio (dust mixing ratio error divided by dust mixing ratio) are typically 0.01–0.2 of the bin values but can be much larger at the lowest mixing ratios and highest altitudes (Figure 1). Data below 10 km and above 60 km are not plotted due to decreased confidence in those values. To visually represent confidence in our results, portions of each plot that are color filled represent bins with dust extinction coefficients that are above zero to a 2 sigma confidence level. Dust mixing ratio is presented as the change in extinction optical depth with vertical coordinate, in this case, pressure, d τ/d P, and thus, the units of dust mixing ratio are expressed as mb−1. The extinction optical depth used is that at a reference wavelength of 9.3 µm. (When water ice extinction mixing is presented, a reference wavelength of 12.4 µm is used.) Many comparisons are made in this paper between TES and MCS results [e.g., Heavens et al., 2011a]. The MCS team has generally reported dust mixing ratio in the units of “density-scaled opacity” (m2/kg). Assuming hydrostatic equilibrium, this unit and our use of d τ/d P (mb−1) are equivalent with a factor of 1/g, where g is the gravitational acceleration.

Figure 1.

Zonally averaged daytime dust mixing ratio fractional error for (a) MY 26 at Ls = 0°, (b) MY 26 at Ls = 90°, (c) MY 26 at Ls = 180°, and (d) MY 26 at Ls = 270°.

[7] Section 2.1 focuses on the 15–30 km dust maximum, termed the “high-altitude tropical dust maximum” (HATDM) by Heavens et al. [2011a] and which we refer to as the “lower dust maximum” (LDM) for the remainder of this work. Section 2.2 describes the 45–65 km dust maximum, which we refer to as the “upper dust maximum” (UDM). For the purposes of this study, we define a dust mixing ratio maximum as a point in the vertical profile where the dust mixing ratio, expressed as a change in optical depth with height (pressure), is greater than or equal to 10−2 mb−1 and is at least 50% greater than an adjacent minimum.

2.1 LDM Discussion

[8] During the approximate 3 Mars year limb-scanning observation period of TES, the LDM was a persistent feature of the tropical lower atmosphere that exhibited a high degree of repeatability from year to year. For a particular time period, the variability in structure and location of the LDM between Martian years was driven by regional or global dust storms altering the basic state of the atmospheric dust field. Figure 2 shows daytime (approximately 2 P.M. local solar time) zonal averages at the equinoxes (Figures 2a and 2c) and solstices (Figures 2b and 2d) of MY 26, a representative “average” year of the observation period, and Figure 3 shows the nighttime (approximately 2 A.M. local solar time) zonal averages for the same time periods. Figure 4 shows interannual variability, over the 3 Martian years of observations, at two example seasons, early northern summer (Ls = 120°) and late southern spring (Ls = 240°). Periods outside of the dust storm season (dust storm season extends from approximately Ls = 190° to Ls = 330°) show little variability in dust mixing ratio amplitude, height, or latitude between years. The greatest interannual variability occurs in the dust storm season, seen in Figures 4d–4f. The 2001 global dust storm (hereafter GDS01) is seen in Figure 4e.

Figure 2.

Zonally averaged daytime dust mixing ratios (mb−1) for (a) MY 26 at Ls = 0°, (b) MY 26 at Ls = 90°, (c) MY 26 at Ls = 180°, and (d) MY 26 at Ls = 270°. Color-filled portions indicate nonzero dust mixing ratios at greater than 2 sigma confidence.

Figure 3.

Zonally averaged nighttime dust mixing ratios (mb−1) for the same times as in Figure 2.

Figure 4.

Zonally averaged daytime dust mixing ratios (mb−1) for (a) MY 24 at Ls = 120°, (b) MY 25 at Ls = 120°, (c) MY 26 at Ls = 120°, (d) MY 24 at Ls = 240°, (e) MY 25 at Ls = 240°, and (f) MY 26 at Ls = 240°.

[9] From a zonal average perspective, this repeatability allows us to chart the daytime movement and amplitude of the LDM through the year. At Ls = 0°, the daytime LDM is centered near 20 km altitude just south of the equator with an amplitude of 0.05–0.1 mb−1 and extends from 20°S to 20°N (Figure 2a). Moving into northern spring, the LDM collapses downward in height and weakens dramatically. By Ls = 20°–30°, the LDM is absent and does not regenerate until after Ls = 50° (not shown). SPICAM has observed a similar paucity of high-altitude aerosol layers during northern spring [Määttänen et al., 2012]. During this period, the highest dust mixing ratios in the tropics occur at or near the surface and are typically 0.01–0.04 mb−1. This corresponds to a period that has been noted for being relatively dust free and having low interannual variability [Smith, 2004]. By Ls = 90°, the daytime LDM exhibits slightly larger amplitude than the beginning of the year and again is near 20 km altitude but has shifted to the north side of the equator, following the Sun (Figure 2b). During northern summer, it slowly tracks southward while remaining at nearly the same height, returning to at or just south of the equator by Ls = 180° (northern fall equinox) (Figure 2c). For each of the 3 years observed, the LDM briefly strengthens in amplitude in northern summer near Ls = 120° (Figures 4a–4c). As the dust storm season begins and perihelion approaches, the altitude of the LDM rises to near 30 km by Ls = 210°. During GDS01 (Figure 4e), the LDM was subsumed into globally enhanced dust mixing ratios, and as reported by Clancy et al. [2010], dust reached great heights. The relatively dust-free atmosphere seen at low altitudes in Figure 4e is a retrieval artifact due to the extremely high opacities at high altitudes preventing significant near-surface radiance from reaching the detector. During the dust storm season, the LDM expands poleward, to reach from about 35°S to 35°N (Figures 2d and 4f). By late in the year, the LDM begins to erode on the northern edge as winter deepens there, the amplitude weakens, and the height lowers back toward 20 km (not shown).

[10] The nighttime LDM was 2–5 times weaker in maximum mixing ratio than the daytime LDM for most of each Martian year, with the exception of the dust storm season when the mixing ratios are diurnally comparable (Figure 3). While the daytime LDM recovers quickly following the northern spring collapse, the nighttime LDM did not fully return to the values seen earlier at Ls = 0° (Figure 3a) until at least Ls = 120°. This recovery was slowest in MY 26 and did not recover until after Ls = 180° (Figure 3c). By the early portion of the dust storm season at Ls = 210°, the nighttime LDM was similar in both amplitude and location to the daytime LDM (not shown), and this pattern was maintained until midway through southern summer, when the nighttime LDM again begins to weaken relative to the dayside. During all large-scale dust storms, the nighttime LDM is comparable to the daytime LDM.

[11] The daytime LDM is not distributed uniformly longitudinally but often peaks in two locations, near those identified by Heavens et al. [2011a] (Figure 5). Both are found on or near the equator in all seasons but do somewhat follow the Sun north and south with the seasons. The most persistent of these two longitudinal maxima has its western edge near 270°E, proximal to the eastern edge of the Tharsis plateau and the western terminus of Valles Marineris, with the core of the maximum typically near 300°E (Figures 5a–5c). The second maximum is far more variable in both location and amplitude but often begins around 50°E, over the high terrain of Syrtis Major (Figure 5). At Ls = 0°, the two maxima are near these typical locations, each with dust mixing ratio of about 0.1 mb−1 (Figure 5a). The western maximum (Syrtis Major) is larger in size for each of the northern spring equinox seasons observed by TES. However, as northern spring progresses, the western maximum completely disappears. At Ls = 30°, the eastern maximum (Tharsis/Valles Marineris) is weakened and shifted slightly northward (not shown). The western maximum begins to regenerate by Ls = 50°–60°, coincident with the time the LDM itself recovers after the northern spring absence. Following the seasonal pattern of the LDM, each longitudinal maximum reaches a brief peak in amplitude and size early in northern summer (about Ls = 120°) (not shown), before weakening briefly at equinox prior to the dust storm season. The vertical dust distributions in the equinox seasons are very similar. Understandably, the dust storm season exhibits the most interannual variability. Each of the longitudinal maxima moves south of the equator during southern spring and, as mentioned previously, rises in altitude. At Ls = 210°, the 3 Mars years show no consistent pattern. In MY 24, the pattern is similar to that of Ls = 180°: The two maxima are of comparable strength. GDS01 was raging at this time in MY 25, and the dust mixing ratios are distributed nearly uniformly with longitude. Planetwide dust opacities had reached their highest levels, and the last visible signs of active dust lifting occurred between Ls = 210° and Ls = 214° [Strausberg et al., 2005]. The eastern maximum is nearly absent at lower altitudes (around 20 km) in MY 26 but is present at higher altitudes, while the western maximum was much larger than normal. However, as the dust storm season progresses, the eastern maximum begins to dominate the LDM and often extends around the planet to include the western maximum. From Ls = 240° to Ls = 330°, dust mixing ratios were fairly uniform throughout the low latitudes in each year, but the highest mixing ratios were persistently associated with the eastern maximum (Figure 5d).

Figure 5.

Daytime dust mixing ratios (mb−1) for the 17.5–22.5 km altitude bin for the same times as in Figure 2.

[12] At all times of the year, there is no shift in longitude of the daytime dust mixing ratio maxima with altitude. The nighttime maxima also have no longitudinal shift with altitude but occur at a different longitudinal position from the daytime maxima. Outside of the dust storm season, the eastern nighttime maximum was 10–50° east of the daytime eastern maxima (compare Figures 5 and 6). This shift suggests that zonal winds transported dust downstream quickly after it was injected into the LDM, if the assumption is made that the dust is injected during the daytime. Winds of 15 m/s would be sufficient to transport dust 10° in longitude (near the equator) in 12 h, and therefore, to transport material 50°, winds of 75 m/s would be required. While 15 m/s eastward winds commonly occur at these locations and altitudes in global circulation model (GCM) simulations [Richardson et al., 2007], 75 m/s eastward winds are nearly twice the highest simulated velocities near the location of the LDM. Highest simulated eastward winds occur over Tharsis, near the eastern maxima. This disparity may be related to the somewhat coarse longitudinal binning used in analyzing the TES limb observations or longitudinally variable scavenging by water ice clouds at night (see section 4).

Figure 6.

Nighttime dust mixing ratios (mb−1) for the 17.5–22.5 km altitude bin for the same times as in Figure 2.

2.2 UDM Discussion

[13] Comparison of Figures 2 and 3 illustrates that the UDM was primarily a daytime phenomenon. Dust mixing ratios in the UDM often equaled and occasionally surpassed mixing ratios in the daytime LDM. Dust mixing ratios associated with the UDM begin to increase in the 45–50 km altitude range and reach a local (and occasionally absolute) maximum at the top of the TES limb observable domain in the 60 km altitude bin. Conceivably, dust mixing ratios could continue to increase above the observable domain, but very few MCS dust retrievals are available above 60 km to support this idea. Although mixing ratios in the UDM are high, the total dust optical depth within the layer is still low due to the rarefied atmosphere at these altitudes and thus low consequent total dust mass. The UDM's integrated infrared optical depth is 10−3–10−2, with the higher bound corresponding to 2–10% of the total average infrared dust opacity (0.1–0.5) for the Martian atmosphere.

[14] The retrieval algorithm (see the appendix) does not specifically account for the possible presence of CO2 clouds. Thus, these observations of the UDM may be partially contaminated by these clouds. However, based on the published observations of Martian mesospheric CO2 clouds by MGS [Clancy et al., 2007], Mars Odyssey [McConnochie et al., 2010], Mars Express [Montmessin et al., 2006, 2007; Määttänen et al., 2010], and Mars Reconnaissance Orbiter (MRO) [Vincendon et al., 2011], we view any contamination of dust retrievals as isolated and unlikely to substantially impact the spatially broad and temporally persistent signature of the UDM. These works find CO2 clouds to be generally above the altitude of the UDM (typically 70–80 km) and confined to specific latitude (tropical), longitude (between Tharsis and Syrtis Major), and solar longitude (near-aphelion) zones. The climatology of the UDM does not conform to those confines. Additionally, the temperatures at these levels are typically too warm to support CO2 condensates.

[15] The seasonal evolution of the UDM exhibits marked differences from that of the LDM. Like the LDM, the UDM weakens during early northern spring and lowers in altitude; however, the UDM was never completely absent during this period. The UDM quickly recovered from its less intense state in each year, however, and, by Ls = 60°, contains higher mixing ratios than at Ls = 0°. Dust mixing ratios continued to increase until approximately Ls = 120° in MY 24 and MY 25, while in MY 26, they slightly decreased (Figures 4a–4c). By Ls = 150°, the UDM began to weaken in amplitude in all 3 years, and this weakening trend continued through the dust storm season. During the dust storm season, the UDM is almost entirely absent from the observable TES limb domain (Figures 2d and 4d–4f). As mentioned in section 2.1, during the dust storm season, the LDM increases in height. Plausibly, the UDM would increase in height as well and move beyond the observable domain. Although this is not observed with MCS, this cannot be ruled out given the scarcity of MCS dust retrievals above 65 km. The SPICAM instrument has observed aerosol layers at 70 km and higher but cannot distinguish between dust, CO2 ice, and water ice [Rannou et al., 2006].

[16] The UDM exhibits a spatial pattern distinct from the LDM. It is much broader in latitude, often extending from the southern midlatitudes to the high northern latitudes, and is not narrowly confined to the equatorial region like the LDM. The longitudinal pattern is highly variable between seasons and interannually (Figure 7). Often, the UDM is nearly uniform in dust mixing ratios through longitude, and the eastern and western maxima of the LDM are not seen in the UDM. A local dust mixing ratio maximum is seen near Tharsis approximately 25% of the time. The northern summertime peak of the UDM followed similar patterns in MY 24 and MY 25. Early in MY 25, high dust mixing ratios were confined to a narrow strip along 35°N, which moved to 50°N by Ls = 30°, and then expanded to cover the entire northern hemisphere by Ls = 60°. By northern summer (e.g., Ls = 120°), the UDM extended from around 25°S to cover nearly the entire northern hemisphere in MY 24 (Figure 7a) and MY 25 (Figure 7b). Maximum mixing ratios occurred in a broad swath from the equator to 40°N in northern spring and summer of MY 26 and MY 27 (Figure 7c). The UDM spanned 40°S–50°N with the highest mixing ratios in the eastern hemisphere (longitudes ≥180°E) northern fall equinox, at Ls = 180°, in MY 24 and MY 25. In MY 26, the high mixing ratios fell in a swath along 20°S–40°S. In each year TES observed, the mixing ratios then quickly decreased at these altitudes during northern fall/southern spring (with the exception of MY 25 during GDS01 (Figure 7e)), with the maximum near or just south of the equator (Figures 7d and 7f). After Ls = 300°, dust mixing ratios in the UDM began to increase again, but the pattern of this increase was inconsistent from year to year.

Figure 7.

Daytime dust mixing ratios (mb−1) for the 57.5–62.5 km altitude bin for (a) MY 24 at Ls = 120°, (b) MY 25 at Ls = 120°, (c) MY 26 at Ls = 120°, (d) MY 24 at Ls = 210°, (e) MY 25 at Ls = 210°, and (f) MY 26 at Ls = 210°.

[17] Clancy et al. [2010] identified a pattern in high-altitude dust during GDS01 that was suggestive of a westward propagating Rossby wave, and this wave number 1 pattern is clearly seen in Figure 7e. Prior to the onset of GDS01 at MY 25 at Ls = 180°, there was substantial dust in the UDM (mixing ratios of >0.1 mb−1) centered near the equator. By Ls = 190°, the dust began to take the appearance of a wave number 1 structure, with the highest dust mixing ratios centered near 250°E at the equator. At Ls = 200°, the wave number 1 structure was dominant with the highest dust mixing ratios near 330°E. Maximum dust mixing ratios then propagated slowly westward to nearly 270°E and 20°S by Ls = 210° (Figure 7e), finally reaching 160°E and 30°S by Ls = 220°. Following this period, the wave number 1 structure rapidly fell apart and the 60 km altitude range was nearly dust free by Ls = 250°. Clancy et al. [2010] found that larger particles made up the bulk of the dust at this altitude during GDS01 and the rapid clearing following the storm is suggestive of accelerated gravitational sedimentation of larger particles.

[18] The UDM was occasionally separated from the well-mixed boundary layer dust and the LDM (if present) by a region of nearly dust-free atmosphere with dust mixing ratios at least 2 orders of magnitude less than the mixing ratios within the UDM. The UDM began to increase in amplitude immediately following aphelion, when the tropical aphelion cloud belt developed and strengthened (Figure 8). Comparing Figures 8a and 8b, it is readily apparent that dust and ice mixing ratios are strongly anticorrelated (below 20 km, some of the anticorrelation is due to a retrieval artifact wherein high dust opacity at high altitudes causes the retrieval algorithm to falsely indicate high water ice mixing ratios below that level; see the appendix for additional details) and that the UDM had its highest values in the low latitudes at the same time that the aphelion cloud belt reached its annual maximum from Ls = 105° to Ls = 125° [Smith, 2004]. Profiles exhibiting this dust-free region between the UDM and lower altitudes were most frequent in the northern middle and high latitudes, away from locations of densest clouds [see Smith, 2004, Figure 16]. Such dust profiles exhibit mixing ratio minima near 40 km. Daytime ice mixing ratio profiles in the same locations often see their highest mixing ratios at similar altitudes. Dust profiles without a dust-free region below the UDM also typically exhibited a local mixing ratio minimum in the 30–40 km range, and this minimum was typically 1–1.5 orders of magnitude less than the mixing ratios in either the LDM or UDM. Again, these profiles had dust mixing ratio minima at similar altitudes as ice mixing ratio maxima.

Figure 8.

Time versus height daytime TES limb (a) dust and (b) water ice mixing ratios (mb−1) averaged from 30°S to 30°N.

[19] On the nightside, while dust mixing ratios nearly uniformly decreased above the boundary layer outside of the dust storm season, mixing ratios in the region of the UDM decreased by nearly an order of magnitude in most times observed by TES relative to daytime values (compare Figures 2 and 3). Isolated nighttime mixing ratio maxima occurred near the equator at UDM altitudes during the first half (Ls = 0°–180°) of the years but with no clear pattern.

[20] We examined the available MCS database covering MY 28 at Ls = 110° (September 2006, the start of MRO's prime mission) through MY 31 at Ls = 20° (October 2011) which comprises over 2.3 million profiles. Approximately 40,000 daytime profiles retrieved dust above 45 km, which amount to 1.7% of the total MCS profiles. The statistics of small numbers thus prevented us from making any confident determinations about the UDM from MCS. What dust profiles are available do share some similarities with TES UDM observations with many profiles in northern summer (Ls = 90°–180°) exhibiting a local dust mixing ratio maximum above 45 km. Kass et al. [2011] reported that MCS was, in fact, seeing “detached haze layers” quite commonly (50 or more times per sol), although their predominant composition was not stated. According to Kass et al. [2011], the MCS retrieval algorithm does not work well at these high altitudes, and thus, the aerosol retrievals are typically not continued to these heights.

[21] For MCS, field-of-view (FOV) wing radiance artifacts complicate the retrieval at high altitudes and appear as high-opacity features [Kleinböhl et al., 2009]. For TES, this issue is significantly lessened due to the differences between the optics of the two instruments. Christensen et al. [2001] discuss the calibration and testing of TES in great detail and test for FOV wings and out-of-field radiance contaminating the FOV. They found that no near- or far-field out-of-field energy was detected.

3 Discussion

[22] The climatology of the LDM and UDM that TES provides allows us to constrain their source mechanism(s). The LDM is a persistent feature for the entire Martian year, with the exception of a 20–30° window of solar longitude in early northern spring prior to aphelion. Dust mixing ratios are higher in the daytime than in the nighttime. Finally, the altitude of the LDM is near 20 km for most of the year but rises to near 30 km during the dust storm season. The source mechanism(s) for the LDM must be consistent with these climatological traits. Similarly, the UDM contains the highest dust mixing ratios and spans the greatest distance in latitude during northern summer while being absent from the TES domain during the dust storm season.

[23] Heavens et al. [2011a] posited four potential mechanisms to sustain the HATDM/LDM, namely, dust storms, orographic circulations, scavenging by water ice clouds, and “pseudo-moist” convection (lofting via dust devils and enhanced boundary layer lift), and favored a combination of scavenging and pseudo-moist convection. Rafkin [2012] described “non-local deep transport,” isolated topographically enhanced updrafts that loft aerosols well above the boundary layer where they can be mixed horizontally via wind. This is analogous to the “hot towers” described in Riehl and Malkus [1958] and the pattern of vertical motion in the Earth's tropics, where much of the zonally averaged upward motion representing the rising branch of the Hadley circulation actually occurs in isolated thunderstorms and is not evenly distributed longitudinally.

[24] The 1 µm dust particles at the height of the LDM should fall at velocities of approximately 0.6 cm/s [Kahre et al., 2008] and therefore reach the surface after about 40 sols. Our 10° of solar longitude binning corresponds to 18 or 19 sols, and the persistence of the LDM over these periods implies the mechanism maintaining it operates at that frequency or faster. Otherwise, the processes of advection and gravitational sedimentation would disperse the LDM over time, and the structure would not be consistent between 10° periods of solar longitude. Additionally, the consistent diurnal variability exhibited by the LDM during TES observations indicates that the mechanism sustaining the LDM is reliably occurring nearly every day. Both pseudo-moist convection and orographic injection could plausibly meet this criterion. To conform to the observed climatology of the altitude of the LDM, the source mechanism injecting dust must reach greater altitudes during the dust storm season.

[25] The persistence of the UDM over long time periods, despite its near-total absence at night, suggests that dust is being transported above and/or below the level of the UDM at night and then returns to that level during the day or the UDM is substantially eroded during the night and constantly replenished during the day. We find the latter explanation implausible given the very broad horizontal distribution of the UDM and the steadily evolving pattern over time. At UDM altitudes (45 km), fall velocities are significantly faster than those at LDM levels, with a 1 µm particle falling at approximately 6 cm/s [Kahre et al., 2008]. Two factors lend themselves to explaining the dramatic diurnal variation in mixing ratios: transport by thermal tides and scavenging by water ice clouds.

[26] In this section, we discuss the following possible mechanisms that maintain and/or influence the UDM and LDM: pseudo-moist convection, dust storms, scavenging by water ice clouds, small particle size, orographic updrafts, and thermal tides.

3.1 Pseudo-moist Convection

[27] The TES limb data set can provide little support to (or cast little doubt on) the possibility of pseudo-moist convection sustaining the LDM. The limb-scanning resolution of about 1 scale height is effectively the entire depth of the boundary layer or more, and the retrievals are not reliable below 10 km. Therefore, the lowest-altitude “pixel” in a given retrieval likely combines emission from the boundary layer and the free atmosphere. As described in Heavens et al. [2011a], pseudo-moist convective updrafts potentially need only cover a tiny fraction of Mars' surface to loft sufficient dust to maintain the LDM. Spiga et al. [2013] recently modeled convective “rocket dust storms” that could supply dust to the LDM and would be maintained over several sols. It is still an open question whether such convective updrafts or storms occur in selected regions of the tropics or whether they occur stochastically across the low latitudes. On this point at least, the persistence of the two longitudinal maxima (on the east side of Tharsis and near Valles Marineris and again over Syrtis Major) suggests that these are the preferred source regions for LDM dust. Neither of these areas was seen as a preferential location for dust devils in several studies of dust devil tracks and images from the Mars Orbiter Camera (MOC) on board MGS [Fisher et al., 2005; Cantor et al., 2006; Whelley and Greeley, 2008]. If pseudo-moist convection is consistently supplying the LDM, the updrafts must either be smaller than MOC pixels in size (approximately 1.5 m typically for the narrow angle camera and approximately 230 m at maximum resolution for the wide angle camera [Cantor et al., 2006]), occur at times other than near 1400 local solar time (the time of MGS overpasses at the equator), or be optically thin. The first possibility is the most plausible of the three since the most comprehensive (to date) dust devil studies [Fisher et al., 2005; Cantor et al., 2006; Whelley and Greeley, 2008] focused on specific regions of the planet and/or counted dust devil tracks. However, it is also possible, though less likely in our opinion, that pseudo-moist convective updrafts have a bias toward being optically thin (as seen by spacecraft) or their occurrence frequency does not peak during times of MOC observations.

3.2 Dust Storms

[28] The comprehensive database of Mars daily global maps (MDGMs) produced from MOC images [Wang and Ingersoll, 2002] is contemporaneous with nearly the entire TES limb data set and allows us to rule out local and regional dust storms as the prime mechanism to maintain the LDM. While local dust storms occur almost daily on Mars during the TES observation period, they infrequently occur at low latitudes, typically only last 1–2 sols, and do not exhibit the consistent longitudinal distribution seen in the LDM. However, higher dust loading in the atmosphere seasonally and from regional storms appears to be a factor in raising the height of the LDM and increasing the mixing ratios at its core. For example, the LDM was abnormally large and intense near Ls = 330° in MY 26 relative to MY 24 and MY 25. From Ls = 312° to Ls = 317° of that year, two large “flushing” dust events [Wang et al., 2003] occurred where dust from Acidalia Planitia and Chryse Planitia moved southward, crossed the equator, and spread horizontally throughout the entire southern hemisphere (Figure 9 shows the MDGM from that period). Figure 10 shows the (combined night and day zonally averaged) LDM evolution from Ls = 300° to Ls = 0° of MY 26.

Figure 9.

MDGM from MY 26 at Ls = 316.97° showing two large “flushing” dust storms. Dust from the first storm is spreading out in the southern hemisphere (green arrows), while additional dust is flowing southward out of Chryse Planitia and Acidalia Planitia (blue arrow).

Figure 10.

(a–f) Evolution of zonally averaged dust mixing ratios (mb−1) and the LDM in response to two large dust storms from MY 26 at Ls = 300° to MY 27 at Ls = 350°.

[29] Again, using the MDGM database, we can quickly discount dust storms as the primary source of dust for the UDM. Not only are regional and global dust storms rare or absent, respectively, for northern spring and summer when the UDM reaches its maximum amplitude and areal coverage but the UDM is considerably weaker during the seasons when those storms are most frequent. Dust from the GDS01 did reach these altitudes and even higher [Clancy et al., 2010], but this pattern was unique to MY 25. The UDM strengthened in late MY 26, contemporaneous with the two “flushing” dust storms seen in Figure 9. A similar, but weaker, dust event occurred during MY 25 at Ls = 315°, also when dust mixing ratios began to increase in the UDM. However, the highest dust mixing ratios at that time at 45–65 km occur on nearly the exact opposite side of the planet from the Chryse Planitia storm track. Additionally, in all 3 Mars years observed, dust mixing ratios begin to increase in the UDM after Ls = 300°, although the spatial pattern of that increase varied from year to year.

3.3 Water Ice Clouds

[30] Scavenging by water ice is playing a role in the dynamics of the LDM but not in the way suggested by Heavens et al. [2011a]. They suggested (and challenged) that scavenging of dust by water ice during the night could deplete the low altitudes in dust, leaving the LDM behind. The comprehensive observations of the LDM by TES show that the daytime dust mixing ratios are not uniform from the LDM to at least the 10 km altitude level, which implies that nighttime scavenging of dust is not the mechanism that produces the relatively dust-free layer below the LDM. However, the significant diurnal variability in the LDM, similar to what Heavens et al. [2011a] predicted (a factor of 3, versus a factor of 2–5 observed with TES), implies that scavenging by ice is primarily responsible for the weakening of the LDM during the night. It cannot be determined with our data whether this scavenging is occurring through increased sedimentation of dust particles following their coating with ice or whether ice is coating dust particles during the night (and hence appearing as ice to TES observations) and then subliming during the day. Throughout the Martian years observed by TES, the daytime LDM occurred around 10–20 km lower in altitude than the daytime water ice clouds (compare Figures 2 and 11). During the first half of the year at nighttime, water ice mixing ratios peak at the same altitudes as the LDM (compare Figures 3 and 12). This dramatic vertical motion of water ice clouds is caused by the vertical propagation of temperature inversions associated with the migrating diurnal tide [Lee et al., 2009; Hinson and Wilson, 2004]. Clouds preferentially form at altitudes of minimum temperature during the night and then sublimate at those levels during the day as the temperature inversion moves downward and the column warms [see Hinson and Wilson, 2004, Figure 12]. This tidal signature can be seen clearly in Figures 12a and 12c (compare to Lee et al. [2009, Figure 12]). During the dust storm season, the nighttime clouds formed at higher altitudes (Figure 12d), above the LDM, and the nighttime LDM was comparable in magnitude to the daytime LDM. By midway through southern summer (near Ls = 320°), higher water ice mixing ratios return to the location of the LDM and the nighttime LDM begins to weaken again relative to the daytime. The exception to this was during the two large flushing dust storms in late MY 26. Water ice clouds were absent from the tropics during all large dust storms viewed by TES limb scans. This link to the height of water ice clouds through the year was suggested by Heavens et al. [2011b], but they were not able to distinguish true LDM variability from retrieval biases during periods of high ice cloud opacity.

Figure 11.

Zonally averaged daytime water ice mixing ratios (mb−1) for the same times as in Figure 2.

Figure 12.

Zonally averaged nighttime water ice mixing ratios (mb−1) for the same times as in Figure 2.

[31] High water ice mixing ratios were nearly absent from UDM altitudes during daytime throughout the TES limb observation period. The exception was during the dust storm season each year, when higher mixing ratios did occur above 50 km, but typically, the daytime peak water ice mixing ratios occurred near 40 km altitude. At nighttime (Figure 12), the propagation of the migrating diurnal thermal tide decreased temperatures at UDM altitudes, and high water ice mixing ratios were found at these heights almost year-round. The longitudinal pattern of the UDM was not strongly correlated to the nightly water ice mixing ratio pattern, as the water ice clouds were tightly coupled to high terrain (Tharsis and Elysium Mons in particular) and the polar caps, but there were still widespread ice mixing ratios of 10−2 mb−1 at UDM height levels for most of the TES observation period. Much like the LDM, scavenging by water ice appears to be a factor in the diurnal variation in the dust mixing ratios at 45–65 km altitude. Of the two most likely scenarios associated with water ice scavenging, water ice particles increasing sedimentation of dust and a dust transport mechanism resupplying each layer at high frequency or water ice particles coating dust particles at night and then subliming during the day, we favor the latter for its simplicity but cannot exclude the first possibility.

3.4 Small Particle Size

[32] Small particle size is an intriguing possibility. Any distribution in particle sizes in the population of airborne dust would naturally differentiate through gravitational sedimentation, leaving the smaller particles at highest altitudes. During dust storms, intensified vertical mixing prevents this differentiation, and larger particles (1.5 µm) have been seen at UDM altitudes [Clancy et al., 2010]. Radiative transfer calculations required to perform TES retrievals assume a constant particle size, and an effective radius of 1.5 µm was used, consistent with Wolff et al. [2006] and Clancy et al. [2003]. Several other studies using the SPICAM instrument have indicated submicron dust particles [Montmessin et al., 2006; Fedorova et al., 2009], possibly down to 10–100 nm above 20 km [Rannou et al., 2006]. If particles in the UDM are smaller than the 1.5 µm assumption, the true optical depth would be larger than the calculated optical depth [see Wolff et al., 2006, Figure 12]. Again, using the fall velocities in Kahre et al. [2008], 0.1 µm particles at 10 Pa would take as much as 2 sols to fall 1 km (0.6 cm/s). These fall speeds would also logarithmically decrease at increasing pressures. The 10 nm sized particles would fall even more slowly. These already slow fall speeds are at least partially compensated by upward motion at and below the level of the UDM. The zonally averaged vertical velocity, w, and the residual mean vertical velocity, w*, from the previously discussed MarsWRF simulation are both positive and of order 0–0.01 m/s from 20 to 60 km in the low latitudes throughout the year (not shown). Taken together, small particle sizes and weak, but persistent, upward motion could maintain the UDM nearly indefinitely (ignoring factors like water ice scavenging) once submicron dust reached that altitude.

3.5 Orographic Updrafts

[33] Injection of dust and water vapor to high altitudes through the intense updrafts on the flanks of Mars' impressive volcanoes is not a novel idea. Rafkin et al. [2002] utilized a mesoscale model to simulate an “aster” cloud at the top of Arsia Mons and found that intense updrafts associated with this caldera-topping dust storm could inject dust well above the planetary boundary layer and that dust and aerosols would spread horizontally on short time scales after reaching altitude. Similarly, Michaels et al. [2006] studied volcanic updrafts and found that water and dust reached altitudes as high as the UDM. Cushing et al. [2005] observed a 700 m tall dust devil inside the caldera of Arsia Mons, at a height of more than 16 km above the datum.

[34] In support of our analysis on these mechanisms, we analyzed specific fields (e.g., vertical wind) of an archived global atmospheric simulation employing the MarsWRF GCM. MarsWRF is the Mars-specific utilization of the PlanetWRF GCM system first described by Richardson et al. [2007]. PlanetWRF is the expansion and adaption of the National Center for Atmospheric Research terrestrial Weather Research and Forecasting (WRF) mesoscale model to a global circulation model with varying planet-specific input parameters. Additional details relating to MarsWRF, including the standard parameter configuration used in the simulation results described here, can be found in Toigo et al. [2012]. The results analyzed and discussed here utilize the 2° × 2° horizontal resolution simulation described in Toigo et al. [2012]. Using this simulation, we examined the updrafts associated with the five tallest volcanoes on Mars: Olympus Mons, Arsia Mons, Pavonis Mons, Ascraeus Mons, and Elysium Mons.

[35] The seasonal pattern of updraft height and velocities is similar at all five volcanoes. During local summer, the mean updraft velocities are highest and the maximum height of upward vertical motion is quite consistent sol to sol. In local winter, the mean updraft velocities are weaker, but the maximum heights of upward motion vary dramatically. Upward motion reaches above the level of the LDM for the four Tharsis volcanoes year-round. At Elysium Mons, the updrafts in northern winter are nearly nonexistent. In summer, the updrafts on the Tharsis volcanoes reach 30–50 km altitude above the datum, but in winter, they reach as high as 80 km. The average vertical velocities for the Tharsis volcanoes are 3–6 m/s in summer and 1–3 m/s in winter (Figure 13).

Figure 13.

(Left) Maximum height of positive vertical motion for Olympus Mons (solid line) and Arsia Mons (dashed line) and (right) mean vertical velocity in the layer from the surface to the maximum height of positive velocity.

[36] The updraft heights and velocities around the volcanoes are more than sufficient to transport aerosols to UDM and LDM heights at all times of the year. If topographic updrafts are supplying and maintaining these two dust layers, as appears possible, there is still some disparity with the seasonality of the two layers and the seasonality of the updrafts. During early northern spring when the LDM collapses, there are still sufficient updrafts in the GCM simulations to reach the LDM and water ice clouds are well above the daytime altitude of the LDM. Perhaps, this represents the seasonal minimum of pseudo-moist convection due to the weak insolation at this near-aphelion time of year, but the LDM actually recovered during aphelion each of the 3 years TES observed. The seasonal pattern of the UDM conforms to that of the updrafts more closely. Four of the five tallest volcanoes are in the northern hemisphere (if barely for Pavonis Mons), and the updraft velocities are weaker in northern winter, when the UDM is weakest; however, the updrafts also reach UDM altitudes more frequently at this time of year.

3.6 Thermal Tides

[37] Long-distance vertical and horizontal transport of dust by the migrating diurnal tide has previously been documented and found dramatic diurnal variation of dust mixing ratios during GDS01 from TES limb observations via upward and poleward transport of dust during the day and downward/equatorward transport at night. Similarly, the terrestrial migrating diurnal tide has been found to transport atomic oxygen and other species in the mesosphere [Ward, 1999; Zhu et al., 2007]. The vertical velocity varies diurnally in a manner that clearly indicates the influence of the diurnal tide in our MarsWRF simulation. The zonally averaged sign of the velocity varies in the vertical with a wavelength of 30–40 km, near the theoretical vertical wavelength of the diurnal tide. There is also downward phase propagation in the vertical velocity from day to night. However, in a global sense, the vertical wind field at UDM altitudes is chaotic, with intense updrafts adjacent to strong downdrafts and signatures of the diurnal tide or Hadley circulation only becoming present when the field is zonally averaged. In a zonally averaged sense, there is no large-scale vertical motion that could produce the 10–20 km altitude change to evacuate dust from the UDM during the night. In fact, the zonally averaged vertical wind is more positive during the night in the 45–65 km range. The meridional wind at the altitude of the UDM in MarsWRF generally matches the sense of the diurnal changes in the UDM with the strongest meridional winds producing horizontal convergence (at net wind speeds of greater than 100 m/s) near the equator during the night and horizontal divergence during the day. These strongest convergence and divergence occur over the western hemisphere, and given the nature of the diurnal tide, winds are in the opposite sense in the opposing hemisphere but of weaker magnitude. Comparing Figures 2 and 3, the tidally driven meridional motion matches the TES observations of a meridionally broad UDM and then a narrow equatorially confined remnant at night, but there is no sign of widespread vertical transport of the UDM to lower altitudes or above the observable TES domain.

3.7 Summary

[38] Within the limitations imposed by available observations, none of these proposed mechanisms can solely explain the behavior and structures in the LDM and UDM observed with TES. We are unable to make a confident statement on the role of pseudo-moist convection in providing dust to the LDM but have placed additional constraints on its occurrence, particularly in regard to the longitudinal distribution of dust mixing ratio maxima. We confidently exclude dust storms as the primary mechanism supplying dust to both layers, but dust storms do influence those layers on short time scales (e.g., Figure 10). Scavenging by water ice clouds appears to play a significant role in the dynamics of both layers, particularly in regard to the diurnal variability and seasonal cycles of both layers seen by TES. Small particle size is something worthy of additional study, but we make no conclusions about the size of particles in either layer given the use of 1.5 µm dust particle size in the retrieval algorithm. The persistence of a longitudinal dust mixing ratio maximum near and downwind of Tharsis in the LDM provides circumstantial support to the idea that orographically forced updrafts supply dust to both layers. Our GCM results also suggest that these updrafts are sufficient to transport dust to LDM and UDM altitudes; however, the seasonal cycle of simulated updrafts does not precisely match that seen in dust mixing ratios. Lastly, GCM-simulated winds indicate that meridional transport by the migrating diurnal tide could explain the meridionally broad structure of the UDM at daytime and the narrow equatorially confined remnant at night (in concert with water ice cloud scavenging). Vertical winds forced by the tide in a global sense are insufficient in the GCM to produce a significant change in altitude of the UDM.

4 Conclusions

[39] Analysis of retrievals of dust mixing ratios from TES limb scan observations has revealed the presence of two discrete layers of enhanced dust mixing ratios throughout the TES limb-scanning observation period: the lower dust maximum (LDM) at 20–30 km altitude (also seen in MCS dust retrievals [Heavens et al., 2011a] and termed the high-altitude tropical dust maximum therein) and the upper dust maximum (UDM) at 45–65 km. Both of these follow distinct season cycles with amplitude minima in early northern spring and late northern fall and maxima in northern fall and northern summer, respectively. In at least isolated cases, MCS has also observed the UDM, but MCS retrievals above 45 km are infrequent.

[40] Nocturnal scavenging by water ice clouds plays a key dynamical role in the control of each dust layer, and thus, the constant presence of these dust layers implies their constant resupply, likely by a combination of orographically forced updrafts and pseudo-moist convection for the LDM and orographically forced updrafts for the UDM. While dust storms can enhance and influence both layers, analysis of their seasonal behavior does not correlate with the seasonal pattern of dust storm activity, implying that dust storms are not the primary mechanism to supply dust to these layers. Meridional transport of dust by the migrating diurnal tide also appears to be a factor in the UDM.

Appendix A: TES Limb Sounding Retrieval Algorithm

[41] Using TES limb sounding observed radiance from the Planetary Data System (PDS), the aerosol retrieval algorithm proceeds by solving for the mixing ratios (expressed as the change in optical depth at a reference wavelength per unit pressure drop) of dust and water ice aerosols at six altitudes in the Martian atmosphere between 10 and 60 km above the areoid. To do so, it (1) compares a quasi-spherical plane-parallel source function radiative transfer model with TES absolute radiance between 240 and 1200 cm−1 excluding the CO2 absorption band region from 590 to 750 cm−1, (2) performs a change of basis to diagonalize the background error covariance matrix, and (3) iterates using Levenberg-Marquardt optimization to minimize the χ2 residuals. Note that the retrieval does not explicitly account for the possibility of CO2 ice.

A1. Forward Model

[42] For the aerosol retrieval, we use a discrete ordinates thermal infrared radiative transfer model that incorporates both multiple scattering and gaseous absorption. We have used the gaseous absorption capability to verify that gaseous absorption has no significant effect on the model for the chosen aerosol retrieval spectral ranges, and so for aerosol retrievals, we switch off the gas absorption in the model for computational efficiency. The model solves the discrete ordinates problem for the source function at all levels in plane-parallel coordinates and then handles the inherently spherical geometry of the limb observations by integrating the radiance along the curved path traced by the TES pointing vectors in the plane-parallel coordinate system. The mixing ratios at the six solution levels are interpolated onto a much finer grid of model levels with a spacing of 0.4 in log pressure coordinates or one seventh of the surface pressure, whichever is a smaller interval.

[43] For the aerosol optical properties, we assume spherical particles and use (complex) indices of refraction from Warren [1984] for water ice and from Wolff et al. [2006] for Martian dust. We assume fixed particle size distributions, using the modified gamma distribution [Deirmendjian, 1964] functional form (consistent with Wolff and Clancy [2003]) with effective radius (reff) of 2.0 µm and effective variance (veff) of 0.1 (consistent with Clancy et al. [2003] and Wolff and Clancy [2003]) for water ice and reff = 1.5 µm and veff = 0.3 (consistent with, e.g., Lemmon et al. [2004] and Wolff et al. [2006]) for dust. The true particle size distributions are known to vary with location and season [Clancy et al., 2003] and with altitude [Federova, 2009]. Fixed distributions were chosen for the present analysis in order to minimize retrieval complexity.

A2. Temperature Profile

[44] The temperatures at each model level are estimated from the TES radiance spectra performed prior to the aerosol retrieval, with geometric altitudes being obtained from the hydrostatic relation using the same surface pressures assumed by the nadir temperature retrievals and the surface altitudes at the locations of the nadir retrieval. The temperature retrievals associated with each limb sounding set (a limb sounding set is defined below) are a combination of nadir-sounding-derived profiles, which control lower levels, and limb-sounding-derived profiles, which control upper levels, and use the nadir-derived profiles as the initial guess and constraint. The weight given to the limb-sounding portion of the algorithm phases in from near zero below the 80 Pa pressure level to near 100% above the 10 Pa level. The nadir profile used for each limb set is actually a spatial average of all nadir retrievals from the same orbit that fall within 2° of latitude along the orbit track of the central tangent latitude of the limb pointings.

[45] Both the nadir and limb sounding portions of the temperature retrievals are as described in previous work [Conrath et al., 2000; M. D. Smith et al., 2001], except that we have improved the limb sounding portion of the temperature retrieval by incorporating three-point triangle function smoothing of the spectra (as discussed below), accurate instrument background measurements, and most importantly accurate background noise covariance matrices. This more accurate background and background noise information for the temperature retrieval is obtained in a manner identical to that used for the aerosols retrievals, which will be described below.

A3. Other Inputs

[46] The surface pressures, surface altitudes, and surface temperatures associated with the nadir retrievals are all averaged in the same way as the temperature profiles to form their counterparts for limb sounding for both the limb aerosol retrievals and the limb sounding temperature retrievals. The assumed surface pressure, as described in Conrath et al. [2000], is derived from the Mars Orbiter Laser Altimeter (MOLA) topography and the annual cycle of atmospheric pressure as observed by Viking. The surface temperatures used are the “SPECTRAL_SURFACE_TEMPERATURE” values from the TES Planetary Data System data set. They are not the “CO2 continuum temperature” that is calculated as part of the nadir temperature retrieval.

A4. TES Limb Sounding Data Set

[47] Limb sounding aerosol retrievals are performed on each set of limb scan spectra in the TES data set. Limb sets are identified in the PDS archive by examining the spacecraft clock (“SCLK” in the Planetary Data System TES data set) value of all observations with recorded emission angles of 90°. Each limb set is separated from the next limb set by searching for gaps of more than 20 s in this sequence of SCLK values. TES limb sets are found at intervals of roughly 10° latitude along the orbit track, with the limb set locations offset by 5° on adjacent tracks. For the complete details of the TES observing strategy, instrument design, and instrument performance, see Christensen et al. [2001] and the NASA Planetary Data System (PDS) TES data archive.

[48] The TES spectra are first smoothed by convolving them with a three-data-point triangular smoothing function (the function is {0.25, 0.5, 0.25}) and then subsampled at an interval of 20 cm−1 within the 240–590 and 750–1200 cm−1 working ranges used for the aerosol retrieval. The 20 cm−1 sampling is used regardless of the resolution mode of the original TES spectra. The three-point convolution removes Nyquist frequency noise that is commonly present in TES spectra due to the interferogram being distorted by the finite response time of the TES electronics [Christensen, 2006]. This Nyquist frequency noise occurs at the Nyquist frequency of whichever resolution mode is used. TES supported two resolution modes. Roughly one third of the limb sets, mainly in the second Martian year of TES mapping operations, were acquired with the spectrometer in “double” (5 cm−1 sampling) resolution mode. The rest use “single” (10 cm−1 sampling) resolution mode.

[49] The radiative transfer model is compared to those spectra in a TES limb set which have pointing vectors with tangent heights less than 85 km above the local surface and for which less than 2% of the vertical spatial response function intersects the Martian surface at any point. We use the MOLA Mars topography data set [Smith et al., 2001] at 4 pixels-per-degree resolution to check for surface intersections at all points along the observations vectors. The spatial response of the TES instrument is described in Christensen et al. [2001]. We consider the full 2-D spatial response functions of the six TES spectral detectors as well as the rotation of those response functions by the TES pointing mirror to calculate the 1-D spatial response function in the vertical direction. For all of the TES detectors, the 1-D vertical response function projected onto the limb has an ~10 km full width at half maximum but has broad and asymmetrical wings. To calculate the emergent radiance integrated over this complicated spatial response function, we first calculate the radiance of emergent rays tangent to each of the model levels plus a set of surface-intersecting emergent rays sufficient to span the surface-intersecting wings of the response functions. The modeled integrated radiance for a detector is then the average of the discrete radiances weighted by the 1-D vertical response function. Because the radiative transfer model cannot account for surface slopes or for foreground and background surfaces in the field of view, the calculated radiance contribution of surface-intersecting rays is inherently imprecise, which leads to the above-mentioned restriction that only spectra which have 2% or less of their integrated weighting function supplied by rays which could intersect the surface are used. Since far-field stray light is plausibly a concern for measurements of the faint high-altitude aerosols that are discussed in this paper, it is important to note Christensen et al. [2001] report no detectable far-field stray light in preflight testing and that the absence of such stray light was verified after launch with Earth flyby stray light tests.

A5. TES Background Error and Correlated Noise

[50] TES spectra are affected by a substantial slowly time-varying background error and by a significant amount of correlated background noise. These effects are clearly visible in the “spacelook” (i.e., targeted at zero-radiance empty space) spectra that precede every limb set. In order to compensate for the background error and background correlated noise in a limb set, we identify all spacelooks with the same resolution mode and pointing direction (forward or aft) as that limb set from within a 200-orbit period centered on the orbit of the retrieval. We apply the same sampling and smoothing as is used for the limb spectra to each spacelook, and then for each detector, we calculate the average radiance, the covariance matrix, and the eigenvectors and eigenvalues of the covariance matrix, using only spacelooks for which all six detectors have tangent heights greater than 80 km and less than 200 km. We have experimented with raising the 80 km lower bound for spacelooks to 90 or 100 km and found no significant effects on the average radiance or the covariance matrix for the aerosol retrievals. For the temperature retrievals, we use a 90 km lower bound and of course the sampling and smoothing appropriate to the temperature retrieval.

A6. Inversion Algorithm

[51] The average radiances for each detector are estimates of the slowly varying background radiance error and are therefore simply subtracted from the limb spectra. For the limb sounding temperature retrievals, the retrieval algorithm can accommodate a nondiagonal covariance matrix [see Conrath et al., 2000], and so in our revised limb temperature retrieval, we simply plug in the updated spacelook-derived covariance matrix. For the aerosol retrievals, the error covariance eigenvectors are used to transform the forward model and the data to a basis that diagonalizes the error covariance matrix prior to computing the χ2 statistic used by the Levenberg-Marquardt optimization. In the new basis, the signal-to-noise ratio for each of the three data points in each spectrum with the largest noise standard deviation is checked, and if it is found to be less than two, that data point is ignored in computing the χ2 statistic. Once the low signal data points have been checked and eliminated as necessary, in order to account for the fact that the forward model is an imperfect representation of the true radiative transfer physics, we add in a second source of uncertainty equal to a fixed percentage of the data spectral radiance in the original basis. We find that adding a fixed fractional uncertainty of ~10% to ~15% produces retrievals with reduced χ2 values that, over the full data set, average close to unity, and so we have adopted 10% as the fixed fractional error.

A7. Uncertainties

[52] The Levenberg-Marquardt optimization routine halts when subsequent iterations change all of the aerosol parameters by less than 0.1%. To produce our reported estimates of the aerosol parameter uncertainties for each retrieval, we treat the forward model and the relative uncertainties that we assigned to all of the data points as a priori assumptions and therefore multiply the formal uncertainties given by the final calculated Jacobian by the square root of the reduced χ2 of the fit. As a practical matter, this means that poor model fits will have large uncertainties and thus receive little weight when aggregating large numbers of retrieved profiles.

A8. Spurious Results at High Column Opacities

[53] We observe one prominent retrieval artifact that is sometimes statistically significant according to our reported uncertainties but which is clearly spurious. When the total column opacity is very high, and the retrieval algorithm therefore has little available information at lower levels, it very often reports very large water ice opacities at the lowest levels during daytime over warm surfaces in dusty conditions. This result is of course physically implausible. An empirical work around this problem is to filter out any portion of a retrieved profile for which the overlying dust column opacity (at 9.3 µm) exceeds 0.23 or the overlying ice column opacity (at 12.4 µm) exceeds 0.3.


[54] This work was funded by a Johns Hopkins University Applied Physics Laboratory graduate student fellowship. We thank Mike Wolff and an anonymous reviewer for their helpful comments which have significantly improved this work.