Monitoring Atmospheric Dust Spring Activity at High Southern Latitudes on Mars using OMEGA
MMonitoring Atmospheric Dust Spring Activity at HighSouthern Latitudes on Mars using OMEGA
S. Douté Institut de Planétologie et d’Astrophysique de Grenoble (IPAG), France([email protected] Phone: +33 4 76 51 41 71 Fax +33 4 76 51 41 46)
Abstract
This article presents a monitoring of the atmospheric dust in the south polarregion during spring of martian year 27. Our goal is to contribute to identify-ing the regions where the dust concentration in the atmosphere shows specifictemporal patterns, for instance high, variable, and on the rise due to lifting ortransport mechanisms. This identification is performed in relation with the sea-sonal ice regression. Based on a phenomenological examination of the previousresults, hypothesis regarding the origin of aerosol activity of the southern polarregion are proposed. This is of paramount importance since local dust stormsgenerated in this region sometimes grow to global proportions. The imagingspectrometer OMEGA on board Mars Express has acquired the most compre-hensive set of observations to date in the near-infrared (0.93-5.1 microns) of thesouthern high latitudes of Mars from mid-winter solstice (Ls=110°, December2004) to the end of the recession at Ls=320° (November 2005). We use two com-plementary methods in order to retrieve the optical depth of the atmosphericdust at a reference wavelength of one micron. The methods are independentlyoperated for pixels showing mineral surfaces on the one hand and the seasonalcap on the other hand. They are applied on a time series of OMEGA imagesacquired between L S =220° and L S =280° . As a result the aerosol optical depth(AOD) is mapped and binned at a spatial resolution of 1.0°.pixel -1 and witha mean period of AOD sampling ranging from less than two sols for latitudeshigher than 80°S to approximately six sols at latitudes in the interval 65-75°S. Preprint submitted to Elsevier 30th October 2018 a r X i v : . [ a s t r o - ph . E P ] N ov e then generate and interpret time series of orthographic mosaics depictingthe spatio-temporal distribution of the seasonal mean values, the variance andthe local time dependence of the AOD. In particular we suspect that two mech-anisms play a major role for lifting and transporting efficiently mineral particlesand create dust events or storms: (i) nighttime katabatic winds at locationswhere a favourable combination of frozen terrains and topography exists (ii)large scale ( ≈ S ≈ S ≈ Keywords:
Mars; South Pole; Atmosphere; Dust; OMEGA.
Introduction
The southern high latitudes of Mars are of great interest in spring and sum-mer because of their role in the dust cycle. Local dust storms generated in thisregion sometimes develop to global storms, and a prominent dust collar encirclesthe polar cap. Several experiments aboard orbiters have recently contributed toelaborate and refine this picture.The TES and MOC instruments of Mars Global Surveyor have provided aregular, but Sun-synchronous, record of dust activity in the south polar region.Toigo et al. (2002); Imamura and Ito (2011) produced global maps of dustdistribution by integrating TES individual 9 µm optical depth measurementsaveraged over a 5°-Ls period (respectively 10°-Ls) and binned in 5x5° boxes(respectively 5x10°, latitude, longitude). Both sets of maps depict very dis-tinct space and time patterns of activity around the polar cap edge for the firstcommon Martian Year (MY) 24. These seasonal trends are sometimes in con-2radiction. For the following Martian year 25 and 26, Imamura and Ito (2011)reported a great stability of the dust opacity disturbance compared to MY 24.Thermal mapping of dust by TES was originally limited to regions where thetemperature and thus the emitted signal are sufficiently high, thus precludingthe monitoring of the seasonal polar cap itself. Nevertheless Horne and Smith(2009) modified the standard TES aerosol retrieval algorithm to retrieve at-mospheric dust and ice optical depth values for each daytime spectrum in theTES database with a surface temperature below 210 K. As a result maps of theseasonal and spatial variation of dust and water ice optical depth activity overboth poles are presented, averaged over a 2°-Ls period and binned in 2x2° boxesfrom late MY24 to early MY27. For the southern high latitudes the greatestobserved dust activity each year takes place above the growing seasonal capfrom late summer to the beginning of winter. At other seasons dust opacity isin general much lower but some interannual variability, e.g. the beginnings ofMY 25 global storm, blurs this pattern.Following the early work of James et al. (2001) that already noted correlationof storm event locations with the receding southern polar cap, Color MOC wideangle images were mosaicked together by Toigo et al. (2002) to produce dailyglobal maps. Such snapshots show very dynamic dust activity near the edge ofthe retreating south seasonal ice cap throughout mid and late southern spring,then a decline going to midsummer. Visible MOC snapshots are limited in timecoverage and do not provide quantitative values of dust opacity.The imaging spectrometer OMEGA aboard Mars Express allowed to over-come some limitations of the previous experiments since it acquired a compre-hensive set of global observations in the near-infrared (0.93-5.1 microns) of thesouthern high latitudes of Mars in spring and summer. A detailed study ofthe contribution of water ice aerosols to the OMEGA dataset is provided byLangevin et al. (2007). This study is based on the water ice absorption bandsat 1.5, 2, and 3 µm. In 2005 (MY 27) from mid-spring to mid summer mostOMEGA observations are nearly free of water ice either as aerosols or on thesurface of the southern seasonal cap. Vincendon et al. (2008) performed the3apping of the optical depth of dust aerosols above areas of the south polarcap constituted of pure CO ice as a function of Ls for dates when the contribu-tion of water ice aerosols can be neglected. The average trend of the temporalevolution is a low optical depth between Ls = 180° and Ls = 250° ( τ (2.6 µm) =0.1–0.2), an increase of atmospheric dust activity observed between Ls = 250°and Ls = 270° ( τ (2.6 µm) = 0.3–0.6), and then a decrease up to Ls = 310°. Vin-cendon et al. (2008) observed rapid time variations which are specific to a givenlocation in conjunction to large spatial variations of the optical depth observedover scales of a few tens of kilometres.Monitoring of dust activity in the high southern latitudes by the previous ex-periments was accompanied by an important effort in modelling and simulationin order to interpret the observations in terms of processes. The results of Gen-eral Circulation Models (GCM) suggest that non convective wind stress liftingproduces the peak in the atmospheric dust opacity during southern spring andsummer and that convective (dust devil) lifting is responsible for the backgroundopacity during other seasons (Basu et al., 2004; Kahre et al., 2006). However thecoarse spatial resolution achieved by GCM limits our understanding, fosteringspecific simulations conducted at mesoscales. The main picture that emergesfrom the latter studies is that flows capable of lifting dust from the surface canbe achieved by a variety of conditions, the most likely being cap edge thermalcontrasts (Toigo et al., 2002) but also topography (Siili et al., 1999). Regionalor synoptic baroclinic instabilities as well as vertical convection in the bound-ary layer could also play a role (Imamura and Ito, 2011). These conditionsas well as dust loading itself in the atmosphere can interfere constructively ordestructively.The previous compilation of observations and simulations show that someuncertainties and opened questions remain regarding dust activity in the highsouthern latitudes. First Toigo et al. (2002); Imamura and Ito (2011) indicatedifferent area where the mean atmospheric dust loading is well above backgroundlevels for the same MY 24. Such discrepancy entails an uncertainty on thelocation of the main source regions as a function of time. Second the relative4mportance of the expected mechanisms for dust lifting at local and regionalscales has not yet been clearly established. Third, dust activity around andinside the seasonal cap has only been reconstructed conjointly by Horne andSmith (2009) although with a very coarse spatial resolution and only as a meanof cross-validating their two retrieval techniques. A more spatially detailedand integrated monitoring could be of paramount importance to investigatethe atmospheric dynamics across the cap edge. Finally the main frequency ofdust cloud generation and the time they take to dissipate are also apparentlyinconsistent when examining the results of Toigo et al. (2002) and Imamura andIto (2011): daily as opposed to every 10-20 sols; a few hours as opposed to 10sols. Could that be reconciled?In this paper we bring some new insights about dust activity in the southernpolar region by monitoring the dust both inside and around the seasonal capbased on the OMEGA dataset acquired during MY 27. At the same time, specialattention is paid to the exact characteristics of the cap edge based on the work ofSchmidt et al. (2009). The mapping of the optical depth of atmospheric dust inthe near infrared above mineral surfaces is made possible by the development of anew method that is proposed in Douté et al. (2013) and that is shortly describedin Section 1. Allied to the complementary method by Vincendon et al. (2008),it is applied to analyse the time series of OMEGA observations thus producinghundred of opacity maps. The latter are integrated into a common geographicalgrid and processed by a special data procedure so as to generate a time series ofmosaics. The mosaics depict the seasonal dust loading as well as the day-to-dayvariability and local time dependence of the dust optical depth according to solarlongitude (Section 2). The mosaics are fully described and examined in Section3. As a result a synthetic view of dust activity in the south polar atmosphere inmid spring to early summer is established and discussed in Section 4. Finally,in Section 5, the main points of our study are summarised.5 . Methods for retrieving the optical depth In summary (see Douté et al. (2013) for more details) the first method thatwe operate is based on a parametrisation bringing in the mean effective opticalpath length of photons through the atmosphere composed of particles and gas.The effective path length determines, with local altimetry and the meteorologicalsituation, the absorption band depth of gaseous CO . In the following we assumethat the top-of-atmosphere (TOA) reflectance factor R k measured by OMEGAis: R k ( θ i , θ e , φ e ) ≈ T kgaz ( h, lat, long ) (cid:15) ( θ i ,θ e ,φ e ,τ k aer ,H scale ,A ksurf ) R ksurf + aer ( θ i , θ e , φ e ) where the quantity R ksurf + aer is the reflectance factor that would be measured inthe absence of atmospheric gases. The reflectance factor is defined as the ratioof the radiance coming from the planet by the radiance that would come froma idealised lambertian surface observed under the same geometrical conditions(illumination and viewing). The parametrisation can be expressed as follows.The gas contribute to the signal as a simple multiplicative transmission filterwhich is the aerosol free vertical transmission T kgaz ( h, lat, long ) scaled by themean effective optical path length (cid:15) . The transmission T kgaz is calculated ab-initio using a Line-By-Line radiative transfer model fed by the compositional andthermal profiles given by the European Mars Climate Database (Forget et al.,1999, 2006) (MY24 dust Scenario, solar averaged conditions, no perturbationsadded to the mean values) for a given date, location and altitude of Mars.Exponent (cid:15) depends on (i) acquisition geometry ( θ i , θ e , φ e )(ii) type, abundance,and vertical scale height H scale of the particles (iii) surface Lambertian albedo A ksurf . All the previous parameters are assumed to be known from ancillary dataor previous studies, except the dust aerosol integrated abundance and surfacealbedo. Information about the first quantity can be reduced to one value ofaerosol optical depth (AOD) at a reference wavelength of 1 µ m τ k aer (channel k )if the intrinsic optical properties of the particles are known.6actor (cid:15) can be further decomposed into two terms such that: (cid:15) ( θ i , θ e , φ e , τ k aer , H scale , A ksurf ) = ψ ( ν ) β ( θ i , θ e , φ e , τ k aer , H scale , A ksurf ) On the one hand factor ψ ( ν ) allows a quick and simplified calculation ofthe free gaseous transmission along the geometrical pathlength of acquisition ν = θ i ) + θ e ) knowing the vertical transmission T kgaz . On the otherhand β ( θ i , θ e , φ e , τ k aer , H scale , A ksurf ) is a precious new observable that expresseshow the aerosols influence the pathlength. It can be tabulated by performingradiative transfer theoretical calculations or experimentally estimated for eachspectro-pixel of an OMEGA image. When factor β is tabulated, the singlescattering albedo, optical depth spectral shape, and phase function retrieved inthe near-infrared by Vincendon et al. (2008) are used since these properties arerelevant for the phase angle range spanned by the data set of nadir OMEGAobservations that we consider. In addition after a series of tests described inDouté et al. (2013), the dust scale height H scale is fixed at a value of 11km inagreement with Vincendon et al. (2008). Experimental estimation of factor β for a given pixel means evaluating the intensity of the 2 µ m absorption band ofgaseous CO . Practically this is only possible for surfaces spectrally dominatedby minerals or water ice, even though the procedure can be extended for spectrashowing saturated 2 µ m CO ice absorption band but with some remainingradiance coming from the surface such as in the outer part of the seasonal cap(Douté et al., 2013). For that purpose we first need the observed spectrumand, secondly, the corresponding transmission spectrum T kgaz ( h, lat, long ) ψ ( ν ) computed ab-initio. Combining the estimation of β with the reflectance factordeep into the 2 µ m band (channel k ), we get, by iterative inversion of thetabulated function β ( θ i , θ e , φ e , τ k aer , H scale , A k surf ) , the AOD τ k aer and the surfaceLambertian albedo A k surf . Validation of the proposed method shows that it isreliable if two conditions are fulfilled: (i) the observation conditions provide largeincidence or/and emergence angles (ii) the aerosols are vertically well mixed inthe atmosphere. As for the first condition, experiments conducted on OMEGA7adir looking observations with incidence angles higher than ≈ ≈ ice. The second method by Vincendon et al. (2008) is restricted to area whereCO deposits are not contaminated by dust and water, i.e. above most placesof the seasonal cap except the cryptic sector and close to the sublimation frontwhere sub-pixel spatial mixing of ice-covered and ice-free surfaces is observed.The mapping is based on the assumption that the reflectance in the 2.64 µ msaturated absorption band of the surface CO ice is mainly due to the lightscattered by aerosols. The atmospheric CO (respectively H O) gas absorptionat 2.7 (respectively 2.6 µm) has negligible impact. In this case the reflectancefactor varies monotonically as a function of the optical depth for a given set ofphotometric angles. Therefore, the optical depth can be unambiguously determ-ined by comparing the observed reflectance factor at 2.64 µm with a referencelook-up table. A method for selecting pixels free of dust contamination has beenderived from the relationship between the observed reflectance factor at 1.08 µmand the optical depth modelled from the reflectance at 2.6 µm. Correlation oflow frequency spatial variations of optical depth with altitude can be modelledwith a well-mixed dust atmospheric component with a scale height of 11 km.
As a conclusion the two methods are complementary since our approachspecifically treats ice free areas (mineral surfaces), areas dominated by H O iceor areas where remains CO ice provided that the latter shows specific spectralproperties. Conversely the method of Vincendon et al. (2008) is restricted toareas entirely covered by CO deposits with the supplementary condition thatthe latter are not contaminated by dust and water. The common strength ofboth methods relies in the ability to provide estimation of the AOD for eachpixel of a single image, i.e. at a fixed geometry. As regards the uncertainties,8imitations on the knowledge of the optical properties of aerosols will inducea possible, systematic, and uniform bias in the maps. For the first method,empirical tests have shown that this bias is likely small since the mean effectiveoptical path length is moderately dependent on the single scattering albedo andthe phase function of the aerosols in the range of models proposed by authors inthe last few decades (Korablev et al., 2005). For the second method, the overallabsolute level of the optical depth could be biased by a factor up to 1.36. Forboth methods the assumption of a lambertian surface has little effect on theresults. Otherwise the relative uncertainty linked with stochastic errors in themeasurements or in the modelling is of the order of 10%.
2. Analysis of a time series of OMEGA observations
The imaging spectrometer OMEGA on board Mars Express has acquired acomprehensive set of observations in the near-infrared (0.93-5.1 microns) in thesouthern high latitudes of Mars from mid-winter solstice (Ls=110°, December2004) to the end of the recession (Ls=320°, November 2005) of martian yearMY=27. These observations provide a global coverage of the region with a timeresolution ranging from 3 days to one month and a spatial resolution rangingfrom 700 m to 10 km.pixel -1 . We refer to Langevin et al. (2007) for a completedescription. We have systematically processed a subset of 284 observations fromL S =220° to 280° by using the two complementary methods of Section 1. As aresult, we obtain a series of corresponding τ k aer maps in the image space (opticaldepth τ aer of the atmospheric dust at a reference wavelength of one micron) thatwere normalised according to a reference altitude of 0 km and a scale height of H scale = 11 km in order to correct for changes due solely to varying atmosphericheight because of topography: τ k det = τ k aer exp ( h/H scale ) The timescale between two maps that partially overlap is frequently between0.5° and 1° of Ls. 9 .2. Specific definitions related to the recession of the south seasonal deposits
In the following a special attention is paid to the possible relationshipsbetween atmospheric dust activity and the seasonal ice regression in spring.The latter phenomenon has been monitored from orbit for decades. Differentconcepts, now being recognised widely, were introduced to ease its description.Since we use these concepts in our own study it is useful to give at this pointsome definitions. The first are relative to the passage of the sublimation frontthat can be detected in one given location in the visible, the infrared, or thenear-infrared ranges (see Schmidt et al. (2009) for a summary). The time evol-ution of three physical quantities - respectively the albedo, the temperature,and the strength of a CO ice diagnostic absorption band, all related to thesurface - is parametrised by an analytical model. The “crocus date” coincideswith the inflection of the parametric curve, supposed to coincide with CO dis-appearance. The “crocus line” is the set of locations, where the crocus datesare equal to a given solar longitude Ls. Some studies also suggested that theedge of the seasonal cap can be more precisely described as a transition zonewhere patches of CO ice and dust coexist geographically over a certain spatialextent. In Schmidt et al. (2009) the transition zone is characterised using CO detection by the OMEGA instrument in the near infrared. The “outer (respect-ively inner) crocus line” is defined as the set of locations that contain for a givendate a CO ice coverage ≈
1% (respectively ≈ frost deposits that are regressingtoward the high latitudes during spring and beginning of summer. The nexttwo definitions pertain to the interior of the south seasonal cap in spring. UsingTES data, Kieffer et al. (2000) define the “cryptic region” where CO ice (atlow temperature) has a low albedo and also recesses faster. The “cryptic sector”occupies longitudes between 60°E and 230°E whereas the “anti-cryptic sector”is the complementary (longitudes between 140°W and 60°E). The “anti-crypticsector” notably contains the permanent cap.10 .3. Integrating the AOD maps into a common geographical grid These maps were independently integrated onto a common geographical gridgenerated from the Hierarchical Equal Area isoLatitude Pixelization (HEALPix, http://healpix.jpl.nasa.gov , (Górski et al., 2005)) of Mars southern hemi-sphere at different spatial resolutions. Such an integration makes it easy tocreate a mosaic at a given date or to build a time evolution curve at a givenlocation. The resolution of the grid is expressed by the parameter N side whichdefines the number of divisions along the side of a base-resolution bin that isneeded to reach a desired high-resolution partition. The total number of binsequal to N pix = 12 × N side .Two built-in properties of HEALPix - equal areasof discrete elements of partition, and Iso-Latitude distribution of discrete areaelements on the sphere - make it easy to map any point of coordinate ( lat, long ) into the corresponding bin. The latter can also be tagged and addressed bya single integer. If one considers in addition a partition of time according todiscrete solar longitudes - those of the observations - any space-time data canbe conveniently stored into a two dimensional array (the Primary IntegratedData Array, PIDA). Its X dimension corresponds to the bin number and its Ydimension corresponds to the time index. Consequently the τ k det maps - eachcorresponding to a given image and thus date - are integrated, one at a time, bymapping all the pixels where the AOD evaluation has succeeded to the appro-priate line of the array. In case several pixels fall into the same bin, their valuesare averaged. After completion of the operation, the whole collection of mapshas been integrated into a common geographical grid providing a mean periodof AOD sampling for each bin that depends basically on the latitude. As illus-trated by Figure 1 in the case N side =64 (1.0°.pixel -1 ) the mean period rangesfrom less than two sols for latitudes higher than 80°S to approximately six solsat latitudes in the interval 65-75°S. An additional dimension can be optionallyadded to the PIDA by considering the local time of acquisition for each pixel.Then a division of the martian sol into three equal Local Time (LT) intervals isadopted: 0-6, 6-12, and 12-18. The interval 18-24 is abandoned since it is onlymoderately populated by the OMEGA spatio-temporal points of acquisition.11 .4. Describing time evolutions Figure 2 shows as an example the time evolution of the AOD for a bin chosenin the anticryptic longitude sector (see definition in Section 2.2). It was plottedby extracting a column of the PIDA. A noticeable day to day variability (onedata point to the next) can be immediately noted. This can be expected for avery dynamic environment such as the polar atmosphere in spring and beginningof summer. Variability could also be accounted for by random errors that affectthe AOD retrieval but, with a theoretical root mean square of the order of ≈ . , they can only explain a part of it. A gap between L S =228° and L S =235°and irregularities in the sampling are also evident in the plot. Indeed coverageof the area by the OMEGA sensor had necessary some limitations due to orbitalcharacteristics, planning constraints, and episodic OMEGA malfunctions. Tomitigate the difficulties induced by the previous factors on the analysis, weconsider that any of these temporal signals consists of two contributions: a meantrend of τ k det versus time, i.e. the baseline, and a highly variable component, i.e.the variability around the baseline. The former can be calculated by regressionprovided enough data points are available. The latter is just the differencebetween the original signal and the mean trend. Having a model for the seasonaltrend in the form of a regression function allows to fill the gaps in the PIDAthat is then restricted to the baseline component (Modified Integrated DataArray, MIDA). Nevertheless this is done at the expense of the spatial coverageas explained below. We now define the baseline of any time evolution curve asthe curvilinear object that minimises its root mean square distance with thedata points (continuous line in Figure 2). The typical local curvature of thebaseline is controlled by a characteristic time scale that is fixed at ∆ L s =5°.Such parameter acts as a threshold that separates what we consider to be aseasonal trend from what is the day-to-day variability.In order to calculate the baseline, we use Support Vector Machine (SVMs)Vapnik (1998) a popular machine learning method for classification, regression,and other learning tasks. Traditional polynomial fit is not suited to model themean trend with its typical ups and downs which will require using high orders.12n addition SVM-based regression allows us to control the characteristic timescale of the modelling. Consider a set of data points, { ( x , y ) , . . . , ( x l , y l ) } where x i ∈ R n is a feature vector, y i ∈ R is the target output, and l is thenumber of points available. Under two given parameters C > and (cid:15) > andthe choice of a kernel function K , the standard form of the Support VectorRegression ( (cid:15) -SVR) is: min α,α ∗ ( α − α ∗ ) T Q ( α − α ∗ ) + (cid:15) l (cid:80) i =1 ( α i + α ∗ i ) + l (cid:80) i =1 y i ( α i − α ∗ i )subject to e T ( α − α ∗ ) = 0 , ≤ α i , α ∗ i ≤ C, i = 1 , · · · , l, where Q ij = K ( x i , x j ) , and e = [1 , . . . , T is the vector of all ones. Thesolutions of the previous optimisation problem is expressed in the form of l support vectors α i − α ∗ i and a constant b such that the R -valued approximateregression function is: f ( x ) = l (cid:88) i =1 ( − α i + α ∗ i ) K ( x i , x ) + b The self-adaptation of (cid:15) -SVR to any kind of curvilinear object is a decisiveadvantage in our case. For implementation of the regression we use the libraryof support vector machine routines LIBSVM (Chang and Lin, 2011), one ofthe most widely acclaimed SVM package. We choose a radial basis function: K ( x i , x j ) = exp (cid:16) − γ (cid:107) x i − x j (cid:107) (cid:17) where the parameter γ controls the width ofthe function and is directly related to the characteristic time scale γ = 2∆ L s .Regarding our regression, it should be noted that the feature vectors reduce toscalars (solar longitudes) that must be linearly mapped into [0 , following a re-quirement of the (cid:15) -SVR algorithm. The boundaries of the previous interval cor-respond respectively to the minimum and maximum solar longitude consideredin our study. The parameter (cid:15) appears in the cost function and accommodatesthe dispersion of the data. In our case it is fixed at 0.05. Finally parameter C is a regularisation term that is usually set to 1.0 by default. The LIBSVM13outine for (cid:15) -SVR directly outputs α − α ∗ and b allowing us to build the modelof the mean trend of τ k det versus time for any bin of the HEALPix grid for whichenough points are available and sufficiently distributed in the period of interest.The bins for which these criteria are satisfied fall predominantly poleward of the70 th parallel, the precise number depending only slightly on the chosen spatialresolution of the grid. We find that N side =64 is the best compromise betweenthe spatial resolution and the number of curves to model. Facilities associated to HEALPix provide a means to represent each line ofthe PIDA or MIDA on a geographical map according to different projections. Inparticular the orthographic projection is the most suited among those availablefor the representation of the southern polar region of Mars. Prior to the mappingof the MIDA, we average along the Y dimension of the array all the valid valuesfalling in a given solar longitude interval (of width 10° for the first two mosaicsand then 2° for the following): L S =220-230°, 230-240°, 240-242°, and so on. Inaddition, subsequently to the mapping, we superpose on the map the positionof the Seasonal South Polar Cap (SSPC) crocus lines as determined by Schmidtet al. (2009) at the beginning of each time interval of interest. As a result weobtain a series of 22 mosaics (at a spatial resolution of N side =64, i.e. 1.0°.pixel -1 )that compiles the modelled version of the observations. Discussion of the resultswill be principally based on these mosaics. Map projections of the individuallines of the PIDA lead to one AOD map at a spatial resolution of N side =1024,i.e. 1/16°.pixel -1 , for each OMEGA observation. A selection of these maps(Figures 12 to 16) is marginally examined to get some hints about the lowestlatitudes and around the cap edge. τ k det Even though the modelling of the seasonal trend is performed in a pixel-wisemanner, the spatial coherency of the time series of mosaics is excellent as can beseen in Figures 3 and 4. Then we may expect that in the MIDA, baselines canbe gathered, based on similar shapes, into a limited number of classes. To test14his hypothesis the whole collection of baselines is processed by kmeans classi-fication provided by the R statistical package ( ).The main difficulty to overcome is the prior evaluation of the class number N class . For that purpose, several runs of the kmeans routine are conducted in-dependently with an increasing value for this unknown input parameter. Thenstatistical tests Sugar and James (2003) are performed through the genera-tion of a likelihood function depending on N class and peaking at the most likelyvalue ˆ N class for the previous parameter. We found ˆ N class =4 though with a poorseparability of the classes (inter-class variance only accounts for 32 % of the totalvariance of the data). This is to be expected when studying atmospheric condi-tions that transition progressively at the global scale of our investigation fromone regime to the next. Nonetheless Figures 5 and 6 respectively demonstratethat a good spatial coherency and seasonal baseline separability is achieved forthe four classes which boundaries must be considered as indicative only. In Section 2.4 we define the variable component as the difference betweenthe original signal and the mean trend respectively extracted from the PIDAand MIDA for the same (x,y) coordinates. We simply define an estimator of theday-to-day variability magnitude by calculating for each valid HEALPix bin theroot mean square (RMS) of these differences over non-overlapping contiguousintervals of solar longitudes 10° in duration. By mapping the estimator in thegeographical space according to the orthographic projection, six mosaics areobtained that indicate the most and least active locations for each time period(Figure 7). The relevance of the mosaics has been checked by comparing them toconjugated maps indicating the integrated number of original image pixels thatfalls into each bin of the PIDA for L S =220-230°, 230-240°, 240-250°, 250-260°,260-270°, and 270-280°. No correlation is found for any of the time periods. Inaddition to producing mosaics, a spatial averaging of the estimator is performedseparately over the four spatial regions conjugated to the classes distinguishedin the previous Section resulting in distinct temporal trends (Figure 8 upper15raph). As regards to spatial heterogeneity, we perform its assessment over thesame four spatial regions. For that purpose we first calculate the root meansquare of the variable component over a given region at each date, - time index- of the PIDA. We then obtain four temporal curves at full time resolution thatare smoothed by a sliding average operation (width of the window: 10° L S ).The result is then sampled at the solar longitudes 225°, 235°, 245°, 255°, 265°,and 275° (Figure 8 lower graph).
3. Trends of atmospheric dust opacity for the high southern latitudes
Figure 5 displays the contours of the four principal spatio-temporal unitsthat have emerged from classifying the collection of seasonal trends followed bythe dust optical depth. The first unit roughly spans the area between meridians90°E and 210°E and parallels 70°S and 85°S. It corresponds basically to theportion of the southern polar layered deposits (SPLD) known as the “crypticregion” (see definition in Section 2.2). It is characterised by a steady rise of τ k det from ≈ ≈ S =270°. The rate of increase is moderate for (cid:46) L s (cid:46) ° , before beingaccentuated. Once passed the maximum, one can observe a noticeable drop inthe curve. Unit 2 is composed basically of two non contiguous area, the mostextended being between meridians 210°E and 300°E and parallels 70° and 80°:Parva Planum, Argentera Planum. The eastern most part of the area is actuallyslightly deported towards the lower latitudes. The least extended area of unit 2is centred around (78°S,100°E; western part of Promethei Planum). As regardsto its temporal behaviour, this unit can be distinguished by a steady rise of τ k det that starts at higher values than unit 1 ( ≈ ≈ S =265°). Then τ k det declines to some degree. Unit 3 is quite contiguous even though it displays acomplex shape. The major part of it lies principally between longitudes 270°Eand 30°E and at latitudes as high as 87°S and as low as 70°S including DorsaArgentera. It also comprises the area occupied by the so-called Mountain of16itchell located near 70°S, 40°E (James et al., 2000) and an “arm” of SPLDaround the south permanent cap at longitudes between 120°E and 270°E. Units2 and 3 basically coincide with the anti-cryptic region. The atmospheric opacityof unit 3 is already quite high at the beginning of our period of interest around ≈ S =240°, and increases afterwards up to amaximum of ≈ S =255°. Then it decreases down to a minimum of ≈ S =272°. Our curve indicates the beginning of a reversal afterwards, butit is not possible to say if it is significant. Finally unit 4 surrounds the quasitotality of unit 3 as a “stripe” passing through (clockwise) the permanent cap andthe nearby SPLD, the Argentera Planum, the Sisyphi Planum, and the DorsaBrevia. A striking anti-correlation between the temporal behaviours of unit 3and 4 can be noticed. Until L S =240°, the dust opacity remains basically at thesame level ( ≈ ≈ S =250° ) when the baseline of unit3 reaches its maximum ( ≈ ≈ S =275°for unit 4, ≈ The spatio temporal units with their distinct temporal signatures are thesummarised expression of the dust loading of the atmosphere depicted withmore details by the time series of mosaics built from the MIDA (explained inSection 2.5 and displayed in Figures 3 and 4). In addition one can appreciate theregression of the seasonal deposits by looking at the simultaneous displacementof the inner and outer crocus lines.
Cryptic region.
When the latter region, which is nearly in a one to one cor-respondence to unit 1, is entirely covered by CO frost ( (cid:46) L s (cid:46) ° ),the overlying atmosphere AOD is the lowest of the entire seasonal cap. As theinner crocus line progresses through the region toward the high latitudes - fasterthan in any other longitude sectors - patches of more dusty atmosphere appearabove the area that are just defrosted. By L S =254°, only the central part of thecryptic region displays AOD values below ≈ S =262°the emergence of a regional AOD maximum centred at (80°S,165°E) that growsin value and spatial extent and reaches its height at L S =272-274°, and startsto disappear afterwards but not completely. Thus this regional enhancementof dust opacity is present for at least ≈
20 days. In the most western part ofthe cryptic region around the location (78°S,100°E; western part of PrometheiPlanum), there is an area of the same sort but that has been active even earlier(L S =230°). As result it is classified as belonging to unit 2 rather than to unit1. Anti-cryptic region.
Unit 2 which falls principally in the latter region sees inmany places a rise of atmospheric opacity since the beginning, i.e. even be-fore the passage of the crocus lines. Defrosting has only a modest influenceon the rise and several regional maxima’s develop. The most important one isalong longitude 270°E and migrates by ≈
5° towards the higher latitudes betweenL S =252° and L S =280°. Unit 3 is the portion of the seasonal cap that also showsrelatively high and growing values of the overlying atmosphere AOD at least for (cid:46) L s (cid:46) ° . However contrary to unit 2, this trend then starts to reversewith the crossing of the inner crocus line. Consequently the subsequent preval-ence of ice-free terrains seems correlated with a substantial apparent clearingof the atmosphere. At the end of the period of interest we observe a regionalminimum of dust opacity around (75°S,330°E; Dorsa Argentera). We distinguishtwo sub-units for unit 4 (see Figure 5). At the lowest latitudes that we coverin the “anti-cryptic” sector (unit 4a) an early increase of AOD (modest in amp-litude) followed by a decline is also observed linked to the passage of the innercrocus line. After L S =250-270° with the passage of the outer crocus line, theAOD starts to recover reaching ≈ ≈ S =262-264°. Such a propagating front ex-plains the major characteristics of the temporal behaviour of units 3 and 4 andtheir anti-correlation. In Section 2.7 an estimator of the magnitude of the day-to-day variability isput forward and mapped at a resolution of 1.0°.pixel -1 for six non-overlappingcontiguous intervals of solar longitudes 10° in duration (220-230°, 230-240°, 240-250°, 250-260°, 260-270°, and 270-280°; Figure 7). The area where the resultingmosaics display meaningful values is more restricted in general than the area forwhich we have models of the mean trend of τ k det versus time. Indeed calculationof the estimator requires even more OMEGA observations per time intervalsthan for the model. Nevertheless this is not a limiting factor for studying theday-to-day variability poleward of 70°S latitude. For (cid:46) L s (cid:46) ° thevariability is low ( ≈ S ≈ (cid:46) L s (cid:46) ° ) during whichhigh variability is widespread among all sectors. Finally ( (cid:46) L s (cid:46) ° ) thedust activity decreases in intensity especially inside the remaining SSPC andfor latitudes equatorward of ≈ (cid:46) L s (cid:46) ° each spatio-temporal unit has a verydistinct signature. The atmospheric dust cover above the “cryptic” region (unit1) displays the highest temporal variability in conjunction of being the secondmost spatially segregated. Nevertheless a look at Figure 6 reminds us that theabsolute level of AOD is in general relatively lower than anywhere else. At thecontrary above the “anti-cryptic” sector including the “Mountains of Mitchell”(unit 3) the dust cover is much more spatially uniform also implying a lowday-to-day variability. According to the latter indicator alone, units 2 and4 are intermediate. After L S ≈ In this section our goal is to characterise local time dependencies of at-mospheric optical depth during spring and beginning of summer for the highsouthern latitudes. In Section 2.3 we mention that an additional dimension canbe optionally added to the PIDA by considering the local time of acquisitionfor each OMEGA pixel. A division of the martian sol into three equal LocalTime (LT) intervals is adopted: 0-6, 6-12, and 12-18. Identically to what isdescribed in Section 2.4 we model the mean trend of τ k det versus L S althougheach LT interval is now treated independently. As a result any systematic shiftaffecting the AOD correlated to the diurnal cycle should be dissociated fromthe variability associated to random τ k det estimation errors and day-to-day met-eorology for example. The drawback of adding the LT dimension is that thenumber of samples per bin of the HEALPix grid is now much reduced. The binsfor which enough points are available and sufficiently distributed in the periodof interest fall predominantly poleward of the 80 th parallel, the area that wasscrutinised the most frequently by OMEGA. Finally the modelling leads to thegeneration of three MIDA, one per LT interval. Prior to further processing,we average along the Y dimension of each array all the valid values falling inthe reference solar longitude intervals: L S =220-230°, 230-240°, 240-242°, etc.At this point quantifying the spread and the ordering of the triplet of seasonal20olour regime Local time pattern of dust activityYellow B − < B − < B − Magenta B − < B − < B − Red B − < B − < B − Cyan B − < B − < B − Table 1: Correspondence between the main colours appearing in Figures 9 and 10 and thelocal time pattern of dust activity. τ k det values ( B − , B − , B − ) - one per baseline - is performed for each eli-gible bin and each L S range. In addition a better visualisation of the results isachieved by means of a special colour coding. We assign to each bin a distinctiveprimary hue and a luminosity depending on the ordering of the triplet and onits variance var LT as follows. Red, magenta, blue, cyan, green, and yellow cor-respond respectively to the six possible ordering, i.e. permutations, among thethree baselines. The luminosity is a continuous real number in the [0,1] intervalthat is set to min( √ var LT / . , . The denominator of the previous fractionrepresents three times the standard deviation of √ var LT when doing the stat-istics on the whole dataset. By mapping our colour coded joint estimators ofordering and variance in the geographical space according to the orthographicprojection we obtain a time series of mosaics (Figures 9 and 10). Bright andsubdued colours indicate significant, respectively undetectable, LT variability.Table 1 summarises the local time pattern of dust activity for each color regimeeffectively realised in the series of mosaics.At the beginning of the covered period L S =220-230° a clear dichotomy sep-arates the “cryptic sector” of longitudes with high diurnal variance and the“anti-cryptic” sector with much less pronounced values of var LT . Neverthelessin both cases we observe two regimes: B − < B − < B − (magenta) inthe western part (0-150°E) and B − < B − < B − (yellow) in the east-ern part (150-300°E). The remaining area shows no sign of diurnal variabilityuntil L S ≈ (cid:46) L s (cid:46) ◦ the portion of unit 1 (seeSection 2.6) covered by the map stays predominantly in the yellow regime eventhough the middle part shows very low values of var LT . Meanwhile closer to21he pole over a quite extended area we rather have B − < B − < B − (“red” regime). In the interval (cid:46) L s (cid:46) ◦ a transition can be observed.Then for (cid:46) L s (cid:46) ◦ the predominance of the red regime is obvious every-where except in the sector 240-270°E where the ordering remains in the regime B − < B − < B − (cyan) until L s (cid:46) ◦ . In any case the variance ismoderate to low. After L S ≈ S ≈ Further insights into the dust activity going on around the cap edge andat lower latitudes emerges from the examination of individual projected AODmaps (Section 2.5). Information provided by each map is enhanced by providingtwo companion maps associated to the same observation. In the first, an RGBcomposition of the top-of-atmosphere (TOA) martian reflectivity measured byOMEGA at three wavelengths (0.7070, 0.5508, and 0.4760 µ m) is shown. It isstretched so as to make visible the dust clouds as yellowish hues against mineralsurfaces. Nevertheless it should be noted that such a stretch completely satur-ates the image over the the seasonal deposits. In the second map, a colour scaleis used to indicate the local time of pixel acquisition. Figures 12 to 16 show a se-lection of nineteen observations out of the initial collection of 284 observationsthat illustrate different types of representative situations. In Figure 12 eachobservation corresponds to a line in a Table, the RGB composition, the AODmap, and the LT map being respectively arranged in the first, second and thirdcolumn. In the other four Figures, each observation corresponds to the RGBcomposition and the AOD map alone, both superimposed on a context mosaicdepicting the level of seasonal AOD. The pair is identified as well as situatedin time (solar longitude Ls) and space (arrows). The following description isbased on the entire collection of individual observations but illustrated with the22election.In the interval L S ≈ τ k det ≈ L S ≈ L S ≈ L S ≈ L S ≈ τ k det ≈ τ k det up to ≈ L S ≈ L S ≈ L S ≈ τ k det occur more fre-quently during the second part of the “night” and in the morning (ORB2002_1,ORB2010_1, ORB2012_1, Fig.15, and ORB2079_2, Fig.16). Then surface fea-tures in the visible are subdued behind a diffuse veil of scattering aerosols withrelatively elevated TOA reflectivities. Note that sometimes the development ofthe dust cloud is readily appreciable both in the AOD map and the RGB com-23osition as in the pair of observations ORB1930_2, ORB1941_2, Fig.14 bothacquired during the afternoon. For regions 2 and 4a the area close to defrost-ing area undergo locally intense dust activity ( τ k det ≈ ≈
10° of solar longitudeshow low to moderate AOD corresponding to quite clear conditions in the visibleimages (ORB2047_3, Fig. 16).
In this last sub-section, we confront briefly the evolution and variability ofthe AOD given by on the one hand our study (OMEGA images in MY 27) and,on the other hand, by the study of Toigo et al. (2002); Imamura and Ito (2011)(TES and MOC images in MY 24, 25 and 26 ).Basically Toigo et al. (2002) present two (TES) maps of dust opacity at 9µm. In addition they also discuss a discontinuous series of MOC daily globalmosaics in color. The first TES map depicts the situation at Ls ≈ S ≈ (cid:46) L s (cid:46) °) cover large strips of Mars, joining across the capareas of opposite longitudes at moderate latitudes (Section 3.5). The regionsexternal to the cap (latitudes equatorward of ≈ S ≈ ≈ S ≈ (cid:46) L s (cid:46) , Imamura andIto (2011) found that dust clouds emerge repeatedly (every 10-20 sols) froman area delimited by latitudes 70–80°S and longitudes 240–300°E. Then theymove westward and reach the region in the longitude sector 60–120°E to finallydisappear. The overall longitude range of the disturbance, 60–300°E, coincideswith elevated terrains in the south polar region, and with the increase of dustoptical depth observed at Ls=270-280° that encircles the south pole. The quasi-periodic behaviour (period of 10–20 sols) of the dust optical depth disturbancethat is recognised in the latitude band 70-80°S by Imamura and Ito (2011) intheir Hovmöller diagrams is absent from our curves of seasonal AOD even forthe main source regions (Unit 2 until L S ≈ . Discussion Figure 11 proposes a qualitative summary of the principal results obtained inSection 3. The four main spatio-temporal units are followed in parallel accordingto the solar longitude. In addition we indicate which time interval is concernedby the passage of the crocus lines. A series of indicators - the seasonal trend of τ k det (Section 2.4), the day-to-day variability var D D and spatial heterogeneity(Section 2.7), as well as the ordering and spread var LT of the triplet of baselinevalues ( B − , B − , B − ) (Section 3.4) are reported. In the present sectionthese indicators are now conjointly discussed. The goal here is to propose scen-arios, i.e. succession in time of mechanisms or conditions that plausibly controlthe atmospheric dust activity for different area of the high southern latitudes inspring. First we introduce several mesoscale and microscale circulation processesbased on earlier works. Then the expression, expected conditions of occurrence,and dust lifting capabilities of such processes are phenomenologically linked withthe evolution and variability of the AOD observed with OMEGA in conjunctionwith the state of the regressing seasonal cap. This is the method by which weput forward hypotheses regarding which circulation process(es) likely controlthe atmospheric dust activity for a given unit and for a given time range. Theseproposed scenarios, which are illustrated by a series of figures (13 to 16), couldbe tested by future high-resolution atmospheric mesoscale modelling. Katabatic Winds (KW) are drainage atmospheric flows that form in the low-est atmospheric levels ( ≈ one hundred meters) when cooled dense air isaccelerated down sloping terrains by gravity, overcoming the opposingalong-slope pressure gradient. Martian conditions conducive to this be-haviour are the extremely low temperatures of the seasonal deposits inthermodynamical equilibrium with the overlying CO gas and nighttimeradiative cooling of surface and atmosphere. In both cases they are max-imum at night hours when the temperature inversion above surfaces is thehighest (Spiga, 2011; Kauhanen et al., 2008).26 hermal Circulations are triggered by lateral temperature heterogeneitiesdue to soil thermophysical contrasts but most importantly due to icy vs.bare surface contrasts. They can cause high surface friction and verticalmixing thus injecting dust in the atmosphere (Spiga and Lewis, 2010; Siiliet al., 1999). Atmospheric flow modellings (Holton, 2004) demonstratethat a rough proportionality exists between the typical size of the temper-ature heterogeneities and the height reached by the convection cells. Notethat the latter own a dynamical surface branch oriented from the lowestto the highest temperatures and a return flow, i.e. blowing in oppositedirection, at higher altitude. In particular Toigo et al. (2002) have studiedthe development of the so-called cap winds (CW), analogues to terrestrialsea breeze circulation, generated by the strong thermal contrasts that ex-ist along the retreating edge of the southern spring polar cap during themiddle to late spring. Their three-dimensional numerical modelling atmesoscales demonstrate that surface wind stresses are sufficient at somelocation during specific local time intervals to initiate movement of sand-sized particles and hence dust lifting over quite large area. Thus theyrecognise cap edge winds, a mesoscale structure ( ≈ − km), as animportant factor for the development of dust storms near the cap edge.However the influence of a having a gradual transition between the innerand outer crocus lines instead of a neat cap edge and the availability ofdust and/or sand particles are not addressed in their paper.Note that Siili et al. (1999) performed a set of numerical experiments with thetwo-dimensional Mars Mesoscale Circulation Model that shed some new lightabout the previous two processes. In particular they demonstrate that strongnear surface winds potentially capable of lifting dust from the surface can beachieved (i) in daytime along a slope which lower section is frozen and uppersection is ice-free: “anabatic winds” (ii) in nighttime over a fully ice-coveredslope or defrosted in its lower section: “katabatic” winds. Inclusion of atmo-spheric dust ( τ = 0.3) reduces the daytime ice-edge forcing - the upslope flow27s attenuated - while the nocturnal downslope flow is enhanced. Consequentlythermal circulations, in the form of anabatic winds, can be enhanced by localrelief in the defrosting area. Interestingly, spatial segregation of frozen and icefree terrains in this transition zone can be controlled by topography throughslope orientation and profile for example. Daytime Convective Turbulence (DCT) is particularly well developed inthe Martian boundary layer during the afternoon under the influence ofthe heated surface. Even in situations of weak large-scale and mesoscalewinds, this thermally-driven convection might itself cause significant ver-tical mixing and wind gustiness. Two worth-mentioning manifestations ofconvective turbulence are “convective gusts” found near walls of convectivecells and convective vortices linked with “dust devils”.
Synoptic Scale Circulation at large-scale (>100s kilometers) is character-ised by inter-hemispheric meridional Hadley Cells and mid-latitude baro-clinic waves, the latter features occasionally accompanied by dust frontsextending over thousands of kilometers.
For unit 1, that well corresponds to the “cryptic” region, we distinguish foursuccessive phases of activity.In the very early phase of defrosting ( (cid:46) L s (cid:46) ° ), the seasonal AODis the lowest of the entire seasonal cap, and we have B − < B − < B − (i.e. a “yellow regime”) with dispersed values of var LT , meaning that, stat-istically, there is more dust in the atmosphere during the early hours thanduring the morning and the afternoon with local variations of the spread. For (cid:46) L s (cid:46) ° the mantle of seasonal CO deposits is still quite continuousand exhibits a very compact texture in the form of a translucent slab (Kiefferet al., 2000). Nonetheless it is then superficially contaminated by a large amountof dust making it barely detectable by its spectral signature in the shortwaveinfrared (Langevin et al., 2006). We put forward the hypothesis that power-ful katabatic winds (KW) hurtling down existing topographical slopes up to28 couple hundreds meters above the surface likely cause high surface stresses,lifting the dust that contaminates the ice. At the same time these winds likelyevacuate the dust downstream in the form of dust clouds lying low in the at-mosphere leading to an under-estimated optical depth by our method (Doutéet al., 2013). By the same token we could also explain the fast superficial clean-ing undergone by the CO icy deposits at the ground as noted by Langevinet al. (2006) between (cid:46) L s (cid:46) ° . As for the requirement of significantslopes, we find in the area the mouth of two chasma and the south polar layereddeposits display their most pronounced scarps down to the surroundings plains.The katabatic mechanism would be especially efficient for (cid:46) L s (cid:46) ° .In the next phase, starting at L S ≈ (cid:46) L s (cid:46) ◦ a transition between the “yellow” and “red” regimes oc-curs. That means that, thereafter, the AOD is statistically higher during theafternoon than during the morning and more so than during the early hourswith low to moderate spread (see also section 3.4 and Table 1 for further de-scriptions). Despite the growing activity of the second phase, the increase ofseasonal AOD is late compared to unit 2. This is somewhat surprising since unit1 is the sector for which the inner crocus line is progressing the fastest. Ourinterpretation relies on the fact that, with the wide separation between the in-ner and outer crocus lines, spatially segregated defrosting patterns cover a veryextended area for the considered time period (cid:46) L s (cid:46) ° (Schmidt et al.,2009). As a consequence during daytime, and at the ≈ kilometre scales (smallscales), local temperature contrasts are huge promoting thermal circulations butlimited in their vertical extent. They become stronger with LT injecting moredust during the afternoon. Besides during nighttime, because of reduced lateralthermal contrasts and temperature inversions above high-standing icy surfaces,the katabatic wind regime could recover for a few hours sweeping the dust down-stream. Conjunction of the two phenomena - katabatic winds (KW) much moreefficiently than small scale thermal circulation (STC) - restrict atmospheric up-ward mixing and dust accumulation above the region to some degree.29y L S ≈ S =272-274°. That corresponds to the maximum levelof mean seasonal AOD for the unit which is then also the most spatially segreg-ated in terms of τ k det . Unit 1 evolves from a red regime to a regime for which thenight hours display significantly higher AOD than during daytime even thoughopacities stay persistently at high level in agreement with the individual obser-vations (Section 3.5). These elements forge the following assumption for thethird phase. The conjunction of two mechanisms for atmospheric dust enhance-ment in the region could be envisioned. The advection and upward diffusion(AD2) of dust from the retreating distant edge of the cap is the first potentialmechanism. As explained before preferentially katabatic winds but also possiblycap edge winds (CW) both able to carry dust blow outward from area at higherlatitudes that are still in the second phase. Thermal gusts and dust devils as-sociated to the daytime convective turbulence (DCT) (Spiga, 2011; Spiga andLewis, 2010) is the second potential mechanism.The fourth and last phase corresponds to the decline of seasonal AOD afterL S ≈ For unit 2 we distinguish three successive phases of activity.The first one stretches from the beginning of the covered period to L S ≈ deposits.From a starting value of ≈ (cid:62) (cid:46) L s (cid:46) ◦ the climb of seasonal AOD levelscontinues at the same overall rate while the crocus lines sweep the unit. Thedefrosting area that presents a geographical mixture of ice-free and still frozenterrains implying high thermal contrasts is more limited in extent than for unit1, of the order of 2 to 5° in latitude. Day-to-day variability var D D is alsoprogressively on the rise with greatest values in the defrosting area. Meanwhileseveral regional maxima develop after the passage of the crocus line, and thepeak of atmospheric opacity due to aerosol is reached at L S ≈ var D D declines soon followed by the seasonal level of AOD. Such a behaviour,equivalent to phase 4 of unit 1, has likely the same origin: less efficient dustlifting by gusts and dust devils (DCT). This unit is characterised by a first phase of gradually enhanced dust opacityrather similar to the one experienced by unit 2 until L S ≈ S ≈ S ≈ S (cid:38) (cid:46) L s (cid:46) ◦ in the area of the “Mountain of Mitchell”(then in their last phase of defrosting) and around Dorsa Argentera around L S ≈ We think that this unit can be divided into two sub-units each one beingcharacterised by specific mechanisms despite their very comparable but likelycoincidental behaviour. Sub-unit 4a mainly corresponds to the lowest latitudesthat we cover in the “anti-cryptic” sector whereas sub-unit 4b lies basically inthe region of the south permanent polar cap (SPPC).Sub-unit 4a very much behaves like unit 3 but in advance of seasonal phase.In the period (cid:46) L s (cid:46) ◦ advection of dust at high altitudes from thelower latitudes is the preferred mechanism for explaining relatively high AODlevels. Around L S ≈ S ≈ ≈ S ≈ S ≈
5. Summary
The OMEGA instrument has acquired a comprehensive set of observationsin the near-infrared (0.93-5.1 microns) at the high southern latitudes of Marsfrom mid-winter solstice (L S =110°, December 2004, MY 27) to the end of therecession (L S =320°, November 2005). We systematically process a subset ofthese observations from L S =220° to 280° by performing the inversion of radi-ative transfer schemes described in Douté et al. (2013) and Vincendon et al.(2008) on every top of the atmosphere reflectance spectra forming the OMEGAimages. As a result, we obtained a series of maps depicting the distribution ofaerosol optical depth (AOD) noted τ k aer . The maps were normalised in orderto correct for changes due solely to varying atmospheric height because of to-pography. They were independently integrated into a common grid generatedfrom the Hierarchical Equal Area isoLatitude Pixelization of Mars southernhemisphere. Such an integration, that can be performed at different spatialresolutions, allows to build the time evolution of the AOD for each spatial binof the HEALPix partition. We then separate two contributions: a mean trend34f τ k det versus time, i.e. the baseline, and a highly variable component, i.e. theday-to-day variability around the baseline with an adapted method. We also im-plement a data processing to isolate any dependencies to the local time (diurnalcycle) of a given bin. The HEALPix system provides means to map the valuesof atmospheric opacity or of any derived quantity for a certain date or averagedover a time interval according to different geographical systems. Thus we gen-erate time series of orthographic mosaics depicting spatio-temporal distributionof the seasonal mean values, the variance and the local time dependence of theAOD. Besides, although the modelling of the seasonal trend is performed in apixel-wise manner, the spatial coherency of the mosaics is excellent. Then wemay expect that seasonal baselines can be gathered, based on similar shapes,into a limited number of classes. This was confirmed by performing a kmeansclassification that gives four main types of seasonal trends followed by the AOD τ k det in four units segmenting the southern polar region. Following this com-plete analysis of the data , each spatio-temporal unit was studied carefully forsearching trends of atmospheric dust opacity.As a result a synthetic view (Figure 11) of dust activity in the south polaratmosphere in mid spring to early summer has been established. From thiscompilation of observations we propose hypothesis regarding the origin of aerosolactivity of the southern polar region. Different mechanisms are invoked at avariety of spatial scales in conjunction with the regression stage reached bythe seasonal deposits. In particular we suspect that two mechanisms mightplay a major role for lifting and transporting efficiently mineral particles andcreate dust events or storms: (i) nighttime katabatic winds (KW) at locationswhere a favourable combination of frozen terrains and topography exists (e.g.unit1, ° (cid:46) L s (cid:46) °) (ii) daytime mesoscale thermal circulation at theedge of the cap, i.e. cap winds (CW), when the defrosting area (transitionzone) is sufficiently narrow (e.g. unit2, ° (cid:46) L s (cid:46) °). Indeed this kind ofbreezes could be inhibited, should the width of the transition zone be broad,i.e. in the absence of a sharp thermal boundary of regional proportions (e.g.unit4a, L s (cid:38) °). Thermal circulation at smaller scales (STC) due to the35igh thermal contrasts associated to segregated terrains in the transition zoneexist and could also pick up some dust but likely leads to limited vertical andhorizontal transport (e.g. unit3, ° (cid:46) L s (cid:46) °). Far from the seasonal cap,gusts and vortexes associated with Daytime Convective Turbulence (DCT) overice-free terrains may not be an efficient mechanism to inject large amount ofdust in the atmosphere explaining the clear decline of the AOD after L S ≈ S ≈ S ≈ S ≈ S ≈ Acknowledgements
We thank the OMEGA team at Institut d’Astrophysique Spatiale for sup-port with sequencing and data downlink activities. This work is supported bya contract with CNES through its Groupe Système Solaire Program. We are36rateful to X. Ceamanos and A. Spiga for their reading of the manuscript and forfruitful discussions. We would like to warmly thank both anonymous review-ers for their tremendous work in reviewing the article and their constructivecomments.
References
Basu, S., Richardson, M.I., Wilson, R.J., nov 2004. Simulation of the Martian dustcycle with the GFDL Mars GCM. Journal of Geophysical Research (Planets)109 (E18), 11006–+.Chang, C.-C., Lin, C.-J., 2011. LIBSVM: A library for support vector machines. ACMTransactions on Intelligent Systems and Technology 2, 27:1–27:27, software availableat .Douté, S., Ceamanos, X., Appéré, T., 2013. Retrieving atmospheric dust opacity onmars by imaging spectroscopy at large angles. Planetary and Space Science 85 (0),38 – 52.URL http://opac.inria.fr/record=b1122647
Horne, D., Smith, M. D., Mar. 2009. Mars Global Surveyor Thermal Emission Spec-trometer (TES) observations of variations in atmospheric dust optical depth overcold surfaces. Icarus200, 118–128.Imamura, T., Ito, Y., 2011. Quasi-periodic dust events in the summertime south polarregion of mars. Icarus 211 (1), 498 – 503.URL ames, B., P., Cantor, A., B., Malin, andEdgett, M. C., andCarr, K., H., M., Daniel-son, E., G., Ingersoll, P.andDavies, A., E., M., Hartmann, K., W., McEwen,S.andSoderblom, A., A., L., Thomas, C., P., Veverka, J., apr 2000. The 1997 SpringRegression of the Martian South Polar Cap:Mars Orbiter Camera Observations.Icarus 144, 410–418.James, P. B., Cantor, B. A., Davis, S., 2001. Mars orbiter camera observations of themartian south polar cap in 1999 - 2000. J. Geophys. Res. 106 (E10), 23635 – 23652.Kahre, M. A., Murphy, J. R., Haberle, R. M., Jun. 2006. Modeling the Martian dustcycle and surface dust reservoirs with the NASA Ames general circulation model.Journal of Geophysical Research (Planets) 111, 6008.Kauhanen, J., Siili, T., Järvenoja, S., Savijärvi, H., Sep. 2008. The Mars limited areamodel and simulations of atmospheric circulations for the Phoenix landing area andseason of operation. Journal of Geophysical Research (Planets) 113, 0.Kieffer, H. H., Titus, T. N., Mullins, K. F., Christensen, P. R., 2000. Mars south polarspring and summer behavior observed by TES: Seasonal cap evolution controlledby frost grain size. J. Geophys. Res. 105 (E4), 9653 – 9700.Korablev, O., Moroz, V., Petrova, E., Rodin, A., 2005. Optical properties of dust andthe opacity of the martian atmosphere. Advances in Space Research 35 (1), 21–30.URL Langevin, Y., Bibring, J.-P., Montmessin, F., Forget, F., Vincendon, M., Douté, S.,Poulet, F., Gondet, B., Jul. 2007. Observations of the south seasonal cap of marsduring recession in 2004-2006 by the omega visible/near-infrared imaging spectro-meter on board mars express. J. Geophys. Res. 112, –.Langevin, Y., Doute, S., Vincendon, M., Poulet, F., Bibring, J.-P., Gondet, B.,Schmitt, B., Forget, F., Aug. 2006. No signature of clear co2 ice from the /‘cryptic/’regions in mars’ south seasonal polar cap. Nature 442 (7104), 790–792.URL http://dx.doi.org/10.1038/nature05012
Schmidt, F., Douté, S., Schmitt, B., Vincendon, M., Bibring, J.-P., Langevin, Y., TheOMEGA Team, Apr. 2009. Albedo control of seasonal South Polar cap recession onMars. Icarus 200, 374–394.Siili, T., Haberle, R. M., Murphy, J. R., Savijarvi, H., 1999. Modelling of the combinedlate-winter ice cap edge and slope winds in mars hellas and argyre regions. Planetaryand Space Science 47 (8-9), 951 – 970.URL
Spiga, A., Aug. 2011. Elements of comparison between Martian and terrestrial meso-scale meteorological phenomena: Katabatic winds and boundary layer convection.Planet. Space Sci.59, 915–922.Spiga, A., Lewis, S. R., 2010. Martian mesoscale and microscale wind variability ofrelevance for dust lifting. International Journal of Mars Science and Exploration 5,146–158. ugar, C., James, G., 2003. Finding the number of clusters in a dataset. Journal ofthe American Statistical Association 98 (463), 750–763.URL Toigo, A. D., Richardson, M. I., Wilson, R. J., Wang, H., Ingersoll, A. P., Jul. 2002.A first look at dust lifting and dust storms near the south pole of Mars with amesoscale model. Journal of Geophysical Research (Planets) 107, 4–1.Vapnik, V., 1998. Statistical Learning Theory. Wiley, New York.Vincendon, M., Langevin, Y., Poulet, F., Bibring, J.-P., Gondet, B., Jouglet, D.,OMEGA Team, Aug. 2008. Dust aerosols above the south polar cap of Mars as seenby OMEGA. Icarus 196, 488–505. igure 1: Mean period (in sols) of aerosol optical depth sampling for a given bin by thetime series of OMEGA observations integrated into a common geographical HEALPix grid ofresolution 1.0°.pixel -1 . τ k0 ae r Solar longitude ( ° ) observation (78S, 312E)model of seasonal evolution Figure 2: Time evolution of the AOD for a given bin of the PIDA (crosses) and modelling ofits baseline by (cid:15) -SVR (plain line).
Figure 3: Time series of orthographic mosaics depicting from L S =220° to L S =260° the spatialdistribution of seasonal mean values for the aerosol optical depth at 1 µm. The inner (respect-ively outer) crocus line of the South Seasonal Polar Cap is coloured in red (respectively inblue). Figure 4: Same as in Figure 3 but from L S =260° to L S =280°. egion 1 Region 2Region 3 Region 4a and 4b Figure 5: Approximate spatial boundaries of the “cryptic” and “anti-cryptic” regions as wellas of the four main classes of seasonal trends followed by the AOD τ k det in the southern polarregion. The boundaries appear respectively as plain black, plain white and striped contours.For unit4 the width of the stripes is different for sub-units 4a (large) and 4b (fine). SeeSections 3.1 and 3.2 for a detailed description. .20.30.40.50.60.70.80.91220 230 240 250 260 270 280 τ k0d e t ( s ea s on a l ) Solar longitude ( ° ) region 1region 2region 3region 4 Figure 6: Characteristic seasonal baseline of the four main classes of seasonal trends followedby the AOD τ k det in the southern polar region. See Section 3 for a detailed description. Figure 7: Time series of orthographic mosaics depicting from L s =220° to L s =280° the spatialdistribution of day-to-day variance values for the aerosol optical depth at 1 µm. The inner(respectively outer) crocus line of the SSPC is coloured in red (respectively in blue). .060.080.10.120.140.160.180.2220 230 240 250 260 270 280 τ k0d e t ( M ea n V a r i a b ilit y ) Solar longitude ( ° ) region 1region 2region 3region 40.060.080.10.120.140.160.180.2220 230 240 250 260 270 280 τ k0d e t ( S p a ti a l i nho m og e n e it y ) Solar longitude ( ° ) region 1region 2region 3region 4 Figure 8: For each of the four spatio-temporal units, the average seasonal evolution of theday-to-day variability (up) and the seasonal evolution of spatial heterogeneity (down).
Figure 9: Time series of orthographic mosaics depicting from L S =220° to L S =260° the spatialdistribution of local time dependency for the aerosol optical depth at 1 µm. We assign to eachbin a distinctive primary hue and a luminosity depending on the ordering of the triplet ofLT τ k det values and on its variance var LT as follows : 1: red B − < B − < B − , 2:magenta B − < B − < B − , 3: blue B − < B − < B − , 4: cyan B −
AD1 AD1 AD1 AD1
DCT DCT DCT CW
AD2
KW CW
AD2 AD1 CW Ls (°) KW STC STC STC local maximum of AOD local minimum of AOD
DCT
Steady rise of AOD Steady rise of AOD especially aUer the passage of the crocus line High and growing AOD clearing of the atmos. clearing of the atmos. Progressive recovery Moderate rise AOD Decrease Steady rise of AOD
Figure 11: Synthetic view of the atmospheric dust activity in the high southern latitudes ofMars in mid-spring to summer for MY27. AOD Aerosol Optical Depth. AD1 and AD2: ad-vection of dust respectively by high altitude return flows and cap winds (CW). KW: katabaticwinds. DCT: daytime convective turbulence in the boundary layer. STC small scale thermalcirculation in the transition zone. Y and R respectively yellow and red regimes. D2D: day today. The markers of local maxima/minima of the AOD are related to those of Figure 6 .
RB1781_1 L S =224.45 a b c ORB1849_1 L S =236.41 d e f VIS TOA reflectivity AOD LT
Figure 12: A selection of two global OMEGA observations and associated products. The leftcolumn displays an RGB composition of the TOA martian reflectivity in the visible whichis stretched so as to reveal dust in the atmosphere as yellowish hues. The central columndisplays the Aerosol Optical Depth map at 1 micron. The right column displays a map whichcolour scale is used to indicate the local time of pixel acquisition. See the text for details.
RB1905_2 Ls=246.351° ORB1889_2 Ls=243.51° ORB1925_1 Ls=249.9° ORB1892_2 Ls=244.042°
KW CW+AD2 AD1 DCT STC τ aero
Figure 13: Composition aimed at illustrating and supporting the discussion. The global mapof the seasonal level of AOD for Ls=244-246° serves as the background and also gives theposition of the two crocus lines. Markers are superposed that indicate the possible occurrenceof the different dynamical mechanisms mentioned in the text for the period Ls=240-250°. Inthe legend, AD1 and AD2 stand for advection of dust respectively by high altitude returnflows and cap winds (CW). KW are katabatic winds, DCT daytime convective turbulence inthe boundary layer, and STC small scale thermal circulation in the transition zone. Finally aselection of pairs of products derived from individual OMEGA observations are presented andput in their geographical context. The RGB composition is the top-of-atmosphere martianreflectivity at three visible channels which is stretched so as to reveal dust in the atmosphereas yellowish hues. The map is the corresponding Aerosol Optical Depth at 1 micron. See thetext for details. aero
ORB1941_2 Ls=252.749° ORB1930_2 Ls=250.794° ORB1944_2 Ls=253.283° ORB1958_3 Ls=255.774° ORB1943_2 ls=253.105°
Figure 14: Same as Fig. 13 but for the period Ls=250-260°
RB2010_1 Ls=264.985° ORB2006_3 ls=264.287° ORB2002_1 Ls=263.569° ORB2012_1 ls=265.339° ORB2007_1 Ls=264.455° τ aero Figure 15: Same as Fig. 13 but for the period Ls=260-270°.
RB2079_2 Ls=277.123° ORB2047_3 Ls=271.519° ORB2055_1 Ls=272.917° ORB2073_2 Ls=276.074° ORB2063_3 Ls =274.328 τ aeroaero