The Phoenix Mars Lander detected a larger number of short (∼20 s) pressure drops that probably indicate the passage of convective vortices or dust devils. Near-continuous pressure measurements have allowed for monitoring the frequency of these events, and data from other instruments and orbiting spacecraft give information on how these pressure events relate to the seasons and weather phenomena at the Phoenix landing site. Here 502 vortices were identified with a pressure drop larger than 0.3 Pa occurring in the 151 sol mission (Ls 76 to 148). The diurnal distributions show a peak in convective vortices around noon, agreeing with current theory and previous observations. The few events detected at night might have been mechanically forced by turbulent eddies caused by the nearby Heimdal crater. A general increase with major peaks in the convective vortex activity occurs during the mission, around Ls = 111. This correlates with changes in midsol surface heat flux, increasing wind speeds at the landing site, and increases in vortex density. Comparisons with orbiter imaging show that in contrast to the lower latitudes on Mars, the dust devil activity at the Phoenix landing site is influenced more by active weather events passing by the area than by local forcing.
 The Phoenix Mars Lander [Smith et al., 2008, 2009] landed successfully in the Martian arctic at 68.2°N, 234.3°E on 25 May 2008 and operated actively through 151 sols [Smith et al., 2009]. This corresponds to Ls = 77–148 with Ls = 0 as northern spring equinox. The primary goals for Phoenix were to characterize the local geomorphology, the physical properties of the soil layers and investigate the near-surface ice at the landing site as well as the climate and habitability of the Martian arctic [Smith et al., 2008].
 The spacecraft monitored the daily weather during the mission with the meteorological station (MET). This suite of instruments included a lidar (light detection and ranging) [Whiteway et al., 2008, 2009] and three temperature sensors [Taylor et al., 2008; Davy et al., 2010] built by the Canadian Space Agency (CSA). The thermocouples were situated at different heights on a 1 m vertical meteorological mast (0.25, 0.5, and 1 m above the lander deck). A pressure sensor was received from the Finnish Meteorological Institute and a Danish built so-called Telltale wind indicator [Gunnlaugsson et al., 2008; Holstein-Rathlou et al., 2010] was mounted on top of the meteorological mast. Furthermore, the multispectral Surface Stereo Imager (SSI) [Lemmon et al., 2008] as well as the Robotic Arm Camera (RAC) [Keller et al., 2009] had the ability to detect local weather phenomena such as dust devils.
 Dust devils are convective vortices generated in the lower part of the boundary layer when the atmosphere exhibits a superadiabatic lapse rate due to surface heating, thereby generating convective plumes of rising air parcels that interacts with the ambient vorticity. The rising hot plumes induce a radial inflow of air near the surface due to mass conservation and because of conservation of angular momentum, the inflowing air is accelerated and creates a converging, circulatory flow around the low-pressure core of the vortex [Ryan and Carroll, 1970; Sinclair, 1973; Renno et al., 1998, 2004].
 Some of these vortices obtain horizontal wind speeds large enough for dust particles to be lifted off the surface and into the vortex and thus become dust devils. Dust devils are a common feature in dry regions on Earth, such as hot desert regions [Sinclair, 1969] and in the subarctic [Grant, 1969].
 The occurrence of dust devils and their tracks on Mars was first suggested in the work of Ryan  and has since been observed on several previous landed missions and orbiters. Dust devils were first noted in Viking orbiter images [Thomas and Gierasch, 1985]. The features were roughly 100–1000 m across and extended a few kilometers above the surface.
 Dust devils were not seen in the Viking Lander (VL) 1 and 2 images, but convective vertical vortex candidates were identified in the VL1 and VL2 meteorology measurements by Ryan and Lucich . Forty candidates were observed at VL1 and 78 at the VL2 site with vortex diameters estimated to range between 5 and 950 m, temperature changes of 1–2 K and maximum horizontal wind speeds of 2–42 m s−1. Later, the VL2 data was reviewed and additional 38 possible vortices were identified [Ringrose et al., 2003].
 Evidence for the existence of dust devils in the northern polar region was first examined by Drake et al.  using the MGS Mars Orbiter Camera Narrow Angle camera (MOC-NA) images [Cantor et al., 2002]. This study was performed in preparation for the Phoenix landing in the 65–72°N latitude band. Four longitude regions were considered within this annulus, named region A (90–110 W), region B (220–240 W), region C (275–295 W), and region D (110–130 W); region D was ultimately where Phoenix landed [Smith et al., 2009]. Prior to the Mars Reconnaissance Orbiter (MRO) mission, no active dust devils had been observed at the Phoenix landing latitude [Cantor et al., 2006; Drake et al., 2006]. The MOC-NA images indicated that there were in fact surface features that looked like dust devil tracks, due to their dark albedo and curvilinear morphology, indicating that dust devils can and have formed at this high latitude. Region D had the smallest number of dust devils or wind streaks seen as a percent of the total seen in all regions (6.38%) and the fewest number of images containing these features. Due to the limited coverage of this region by the MOC-NA camera, no overlapping images existed to allow insight into the rate of formation or destruction of these features. As such, it was not clear if dust devils form in the present era or if the tracks seen were a remnant of past eras. It was not until 20 April 2008, just 35 days before Phoenix landed, that the Context Camera (CTX) on MRO imaged active afternoon dust devils within Phoenix's landing ellipse (M. C. Malin et al., MRO CTX spots dust devils at Phoenix landing site, available at http://www.msss.com/msss_images/2008/05/06/, 2008).
 Convective vortices and hence dust devils have a characteristic pressure and temperature signature due to the low-pressure core and the inflow of warm surface air toward the center of the vortex [Sinclair, 1969, 1973; Renno et al., 1998]. From a ground based pressure sensor, e.g., on the Phoenix Mars Lander, this signature will be a distinct pressure dip of the order of ∼20 s when the dust devil passes by, depending on the ambient wind field and the size and location of the vortex. The temperature will sometimes correspondingly increase during the passage of the dust devil. These signatures are seen in both Viking and Pathfinder data [Ryan and Lucich, 1983; Schofield et al., 1997].
 In this work, the near-continuous measurements of pressure and temperature by the MET instrumentation on the Phoenix Mars Lander [Taylor et al., 2009] are investigated to identify the passage of vertically oriented vortex structures such as convective vortices and dust devils at the Phoenix landing site. Similar work was done on Mars Pathfinder pressure data in the work of Murphy and Nelli .
 First, we introduce a vortex identification scheme able to detect the pressure drops. The found events are then fitted to a model to determine the heights and durations of the pressure drops of the events and corrections are applied to the data set. Statistics on the events are made and the results are discussed.
 The pressure and temperature sensors are not the only instruments able to detect and observe dust devils and local weather phenomena at the Phoenix landing site, and supporting data sets from large eddy simulations, the SSI, the Telltale as well as MRO Mars Color Imager (MARCI) are introduced and compared to the results from the pressure data.
1.2. Phoenix Mars Lander Meteorological Observations
 During the Phoenix mission air temperatures were measured at the three levels on the 1 m mast and pressure was measured with a Vaisala Barocap®/Thermocap® system on the deck of the lander, being about 1 m above ground level [Taylor et al., 2008, 2009]. Measurements were made at 0.5 Hz with a sensor noise level of 0.1 Pa and ran almost continuously through the landed mission from sols 0–151 apart from short daily breaks due to data transfers [Taylor et al., 2009]. The atmospheric pressure data shows regular diurnal cycles and during this period of time, the pressure can be seen falling steadily from about 860 Pa to near 720 Pa, due to the deposition of CO2 ice in the south polar region [Taylor et al., 2009]. Furthermore, features associated with meteorological events can be seen, e.g., a low-pressure system around sol 95 (Ls = 120).
Figure 1 shows the top mast temperature data from sols 0–151 (the top sensor has least distortion from the lander deck) [Davy et al., 2010]. The temperature data show regular diurnal cycles as well as a decrease in the mean temperature occurring around sol 80 (Ls = 113) due to the changing seasons.
 During the Phoenix mission the pressure and temperature sensors frequently detected features indicative of dust devils passing over or close to the lander. Just as expected, short duration pressure drops of order 1–3 Pa (and often less) were observed frequently, and accompanied by occasional increases in temperature. Furthermore, dust devils were observed several times near the Phoenix landing site by the Surface Stereo Imager (SSI) and were indicated by the Telltale anemometer as shown later in this work.
2.1. Vortex Identification
 As in the work of Murphy and Nelli , we define a convective vortex identification (or pressure event) as a pronounced, isolated pressure drop. Pressure drops can also be caused by other weather phenomena such as thermal plumes and wind gusts. Due to this, we require sufficient amplitude (>0.3 Pa) and duration (10–20 s) in order to ensure that the feature was not caused by, e.g., small-scale turbulence.
 The pressure sensor was mounted horizontally on the upper payload electronics box (UPEB) with the pressure tube pointing eastward [Taylor et al., 2009]. During the Phoenix mission, winds coming from the NE–SE were 4–5 m s−1 in average [Holstein-Rathlou et al., 2010] and dynamic pressure errors from this are estimated to be <0.2 Pa.
 Later in the mission, stronger winds occurred at midsol (∼10 m s−1) preferably from W [Holstein-Rathlou et al., 2010]. This places the sensor opening at the lee side of the UPEB and makes it a subject to dynamic pressure errors possibly stronger than 0.2 Pa due to vortex shedding from the UPEB. However, vortex shedding from the UPEB is estimated to be in the range of 0.5–8 Hz for winds between 2 and 10 m s−1 [Martinuzzi and Havel, 2004], i.e., on much higher frequencies than the events we are looking for. Furthermore, due to the response time of the pressure sensor (discussed in section 2.2), frequencies higher than 0.5 Hz are damped and thus do not have any effect on the detected data.
 To find the significant pressure events in the raw Phoenix pressure data, we use a running average algorithm with three intervals of each 20 s each: a, b and c. The algorithm calculates the average pressure drop ΔPav = ((a + c)/2 − b) between intervals a, c, and b. If ΔPav is larger than a preset cutoff value, we record the event. The resolution of the Phoenix pressure sensor allows us to distinguish significant events with ΔP > 0.2 Pa since this is well above the noise level. This corresponds roughly to ΔPav > 0.1 Pa and this value is chosen as a cutoff value for the algorithm. This simple method proved sufficient to distinguish all potential pressure events from the background pressure.
Figure 2 shows four of the vortex passages with largest ΔP throughout the 152 sol mission and the shown events are good representatives of the data set. Most events have associated temperature perturbations; however it is noted that there are temperature events of comparable size not associated with the pressure events, and in some cases (e.g., sol 95, Ls = 120) clear temperature anomalies are absent.
 There are several possible causes for this: turbulent temperature fluctuations masking the temperature perturbations of the vortices, passage of dust devils not directly over the lander, distortion of temperature increases due to vertical displacement of warmer air from near the surface, the measurements being close to the ground relative to the expected height scale of the vortices and vertical flow distortions due to lander structures (e.g., solar panels). Some of the large ΔP pressure events (e.g., sol 118, Ls = 132) appear as “double” events, indicating two vortices passing by simultaneously. These are analyzed individually.
 For every pressure event found, we analyze the surrounding pressure and temperature values. Nonsignificant and false events, e.g., from data transfer gaps, are removed by hand.
 In order to easily detect anomalous events such as double events (see Figure 2, sol 118 event) and correct for the time response of the pressure sensor (as described in section 2.2) we need a continuous functional description of the events. The maximum pressure drops ΔP and half maximum durations Γ of all recorded events are therefore determined by fitting a modeled pressure profile to each event. Also the effect of instrument noise can be canceled in this way.
 It can be shown from an approximation to the Rankine vortex model that a Lorentzian function can be used for fitting the detected pressure drops (H. Kahanpää, personal communication, 2009). The Lorentzian function was also found to fit the pressure drops better than the Rankine and Q vortex models. Due to this the pressure event can now be described as
where Praw(t) is the pressure as a function of time t, t0 is the time of the center of the peak, ΔPraw is the pressure drop in the raw data, k is the slope of the background pressure, Γraw is the full width at half maximum (FWHM) duration of the pressure event and P∞ is the background pressure at the time of the pressure event. We use the subscript “raw” to indicate analysis performed on the raw data and without subscript after pressure sensor response time corrections (see below).
 It has to be emphasized that the features do not need to pass directly over the Phoenix Lander to produce a pressure drop with the pressure sensor. The observed pressure drop ΔP and event duration Γ depend on the real pressure drop of the vortex center ΔPcenter, the full width at half maximum width of the vortex W, the closest distance between the lander and the vortex center r0 and the velocity of the vortex compared to the lander U. The relation between these quantities and the observed characteristics of the vortex can be derived from the approximation that the pressure profile of a vortex follows a Lorentzian function:
where r is the distance between the sensor and the center of the vortex. The distance r can be expressed as
assuming that the vortex moves along a straight line with speed U and passes by the sensor at distance r0 at time point t0. Substituting this into the equation above gives
From this we get relations between the observed and actual characteristics of the vortex:
The maximum pressure drop ΔP falls as the closest distance r0 increases, and as the maximum falls, the half maximum duration increases. The pressure minima that are recorded by the lander are therefore not as low as the central pressure minima and the apparent durations are longer than actual durations of the vortices. One problem with the assumed pressure distribution, coupled to our definition of Γ, is that Γ > W/U for r0 > 0. Caution is therefore needed in interpreting U multiplied by “event durations” in Table 1 as effective radii of these vortices.
Table 1. Pressure Events with ΔP > 0.5 Pa During the 151 Sol Mission
Event Duration Γ (s)
2.2. Pressure Corrections
 As described by Taylor et al.  the raw readings of the pressure sensor need some adjustments due to the temperature dependence of the pressure sensor signal, changes in the sensor calibrations and response time of the pressure sensor. The pressure reading depends on the temperature of the Barocap® pressure sensor head. This was monitored by an adjacent Thermocap® temperature sensor mounted on the same printed circuit board (PCB). Because of different thermal contacts with the PCB there was a time lag in the temperature of the Barocap® relative to the Thermocap® and hence the raw pressure readings differ from the actual, processed readings. The procedure used to cancel this effect, called PCOR1, is presented by Taylor et al. . In the analysis presented here the temperature dependence related adjustments by Taylor et al.  have not been applied. Without these adjustments, the raw pressure signal contains artificial slopes of ∼3 × 10−3 Pa s−1 extending for minutes, while even the smallest vortex events generate at least ∼7 × 10−3 Pa s−1 slopes on much shorter time scales and thus the effect of the temperature dependence related issues are not important in this analysis. The vortex identification scheme has been tested using a PCOR1 adjusted data set. The results show no significant difference in the detection of short time scale pressure events compared to tests with raw data. Some artificial events caused by the PCOR1 adjustment were seen and due to this the raw data is preferred for the detection of short time scale pressure events.
 The response time of the pressure sensor was measured in the sensor level calibration tests. The result was a response time of 3.3 ± 0.2 s in dry air. From this result it can be calculated that in the gas composition of the Martian atmosphere the response time is 2.8 ± 0.2 s [Kahanpää, 2009]. In this study we use the value 2.8 s for the response time. The full width at half maximum durations of the shortest detected pressure events are of the same order as the response time. Therefore the instrument response time must have a significant impact on the detected characteristics of the pressure events, especially those with short duration. The actual pressure drop will be larger than detected and the duration of the pressure event will be shorter than detected.
 We considered two different approaches for correcting the effect of response time. One option would have been to invert the effect of response time in the raw data but that would have increased the noise level notably. Instead the time response error is corrected by making use of the approximation that an observed Lorentzian pressure drop will correct into a Lorentzian profile with different maximum pressure drop (ΔP) and duration (Γ). Using the data from [Kahanpää, 2009], the correction factors derived are shown in Figure 3.
Figure 3 also shows an example of how these corrections fit a pressure event on sol 95 (Ls = 120) (also shown in Figure 2). A Lorentzian profile (not shown) was fitted to the raw data. The parameter values of this profile were ΔPraw = 2.96 Pa and Γraw = 10.88 s. Using the correction factors illustrated in Figure 3, the corrected pressure profile is found with ΔP = 3.56 Pa and Γ = 8.56 s. Using the response time of the sensor (2.8 s), an estimate for the detected profile has been calculated from the corrected profile. The calculated profile is almost indistinguishable from the original, fitted profile, and the difference is 0.07 Pa in average. By applying the above mentioned corrections to the events found by the vortex identification scheme, we now have the corrected values ΔP and Γ for all pressure events recorded during the mission.
 The majority of the events have a ΔP close to the cutoff value. To ensure that we only have significant events with a ΔP well above the noise, we apply a cutoff value of 0.3 Pa on the corrected events and our data set now only contain events with ΔP > 0.3 Pa.
 For this data set, the corrected pressure drops are 13% larger in average than the uncorrected pressure drops. The maximum correction is 138% on an event with a Γ duration of 8 s and the minimum correction is 0.5% for an event with a Γ duration of 40 s.
 The pressure events are distributed with an exponentially decaying trend toward events with larger pressure drops. This can be described as N(ΔP) = N0 * exp(−ΔP/ΔP0) where N is number of pressure events, ΔP0 is ∼0.1 Pa and N0 is ∼331 for 0.3 Pa < ΔP < 0.5 Pa. For ΔP > 0.5 Pa, ΔP0 is ∼0.24 Pa and N0 is ∼55.
3. Results and Discussion
3.1. Pressure Events Statistics
 We find 502 vortex identifications with ΔP larger than 0.3 Pa in the raw Phoenix pressure data from sols 0–151. 197 of the vortices have a ΔP larger than 0.5 Pa while 44 of the vortices have a ΔP larger than 1 Pa. The events on Figure 2 are good representatives of the major pressure events occurring during sols 0–151 and show the distinct pressure signature as discussed above. Smaller pressure events with ΔP < 1 Pa generally show the same signatures as those shown in Figure 2 but we see an increase in less smooth events as ΔP approaches the minimum value of ΔP = 0.3 Pa.
 The largest recorded pressure drop was 3.6 Pa. This vortex event was captured on sol 95 (Ls = 120). Table 1 lists vortices with a pressure drop larger than 0.5 Pa with sol, Ls, local mean solar time, pressure drop ΔP and duration Γ. During the mission, we recorded two double events (sols 85 and 118). Both events fit to a combination of two Lorentzian functions (reflected in Table 1) indicating two vortices passing by simultaneously. Both double events have pressure drops too large to be caused by increases in wind speed.
Figure 4 shows observed events per sol as a function of time since landing for all completed sols, i.e., sols 0–151 (Ls = 77–148). ΔP > 0.3 Pa, ΔP > 0.5 Pa and ΔP > 1.0 Pa distributions are shown. In the first 75 sols of the mission (Ls = 77–111) the dust devil statistics are rather homogeneous (an average of 1.7 observed events per sol for ΔP > 0.3 Pa and 0.6 events per sol for ΔP > 0.5 Pa). Around sol 76 (Ls = 111), the average number of events per sol increases significantly (4.9 events per sol for ΔP > 0.3 Pa and 2.0 events per sol for ΔP > 0.5 Pa) and short-term increases in the number of events are observed around sols 80 (Ls = 113), 95 (Ls = 120), 130 (Ls = 138), 140 (Ls = 143) and sol 150 (Ls = 148). A general increasing trend in all distributions can be seen; especially from sol 75 to sol 150 (Ls = 111–148), indicating that the dust devil season at the Phoenix landing site continues after Ls = 148. In addition to this, we also see an increase in the amount of events with large ΔP during the mission. Most of the events with ΔP > 1.5 Pa are occurring after sol 75 (Ls = 111).
 Because the number of observed events per sol increased significantly at Ls = 111, it is reasonable to take a closer look at the statistics in two periods: prior to Ls = 111 (from sols 0 to 75) and after Ls = 111 (from sol 76 to the end of the mission, sol 151). In order to find the time of day where events are most probable, one needs to normalize with the availability of the pressure and temperature data as shown in Figure 5. There are few sols where data are completely missing, and almost every sol there is roughly a 30 min gap due to data transfer. This transfer was usually performed during midsol. Normalizing observed events per hour per sol with the data availability, the results shown in Figure 6 are obtained.
 From sol 0–75 (Ls = 77–111), at most ∼0.2 events with ΔP > 0.5 Pa are observed per hour, while this number increases to ∼0.4 from sol 76–151 (Ls = 111–148). The same pattern can be seen for the ΔP > 0.3 distribution with ∼0.4 events per sol hour from sol 0–75, increasing to ∼0.8 from sol 76–151. The increase is fractionally the same for both distributions and this could indicate that the pressure event size distribution is fixed when the vortex activity changes.
 All diurnal distributions are somewhat bell shaped within the uncertainties; in general most events occur in the noon hours, the part of the sol with peak insolation, and very few events occur during the early morning or late afternoon hours. Both sol 0–75 (Ls = 77–111) distributions have a peak around 1200 LMST while both sol 76–151 (Ls = 111–148) distributions have peaks roughly around 1000 and 1400 LMST with an apparent local dip occurring roughly between 1100 and 1300 LMST. This dip is even larger without the normalization with data availability and since it occurs at the same time as the dip on Figure 5 and the error bars are quite large, we consider this to be an artifact and hence these distributions do peak during the noon hours also.
 Between 2100 and 0100 LMST a few events occur (29 events with ΔP > 0.3). These nighttime events only occur in the first 94 sols (Ls = 77–120) of the mission and other than a handful of ΔP > 0.3 events through the first 50 sols (Ls = 77–99), the majority of the nighttime events occur during sols 77–78 (Ls = 112) and 90–94 (Ls = 118–120), also indicating a connection with local weather phenomena. These events could be regarded as an evidence for the filter cutoff value being set too low. However, there are indications that before sol 110 (Ls = 128), the air that passed over the lander close to midnight did pass over Heimdal crater ∼2 h before [Holstein-Rathlou et al., 2010]. Heimdal crater is the major topographic feature in the vicinity of the lander [Heet et al., 2009] and a possible source of turbulent air. Hence, these mostly small events may be due to turbulence induced by air passing over Heimdal crater, and thus do not share origin with the midsol convective vortices.
 The event width distribution versus ΔP can be seen on Figure 7. The full width at half maximum duration Γ represents roughly half of the total event duration. The longest pressure events have a Γ duration of around 70 s, but most events have a Γ duration between 5 and 15 s. Pressure events with larger ΔP seem to have smaller Γ durations. The explanation for this is probably that the vortices that caused the large ΔP events passed close to the sensor. As explained above, ΔP falls and Γ rises as the closest distance between the vortex and the sensor increases.
3.2. Comparison With Previous Observations
 Compared to the corresponding numbers for Mars Pathfinder data, Phoenix saw fewer events per sol hour, especially in the first half of the mission. In the second half of the mission, the numbers are more similar before and after noon. Pathfinder has a narrower peak with more events during the noon hours, as seen on Figure 6. Compared to previous Mars Pathfinder results [Murphy and Nelli, 2002; Ferri et al., 2003] and Viking results [Ringrose et al., 2003] we see the same patterns in magnitude, duration and temperature correlation with all our major pressure events having the same signature as the Pathfinder events. In addition, we see similar patterns in the diurnal distribution of the convective vortices with most events occurring around noon, a distribution that is also seen in MER observations, [e.g., Greeley et al., 2006] and in terrestrial observations [e.g., Sinclair, 1969].
 We would expect Mars Pathfinder to see more dust devils than Phoenix because of the differences between the local environments of the two landing sites. Mars Pathfinder landed close to equator and the stronger diurnal temperature cycle and different local wind field change the occurrence of the convective plumes and the vorticity, respectively. Both spacecraft landed during northern summer and operated into the late summer/early autumn but Mars Pathfinder operated from Ls = 142–183 whereas Phoenix operated from Ls = 77–148.
 Furthermore, differences in the surface heterogeneities between the two landing sites also have an effect on the amount and size of the convective vortices since heterogeneous surfaces have the potential of producing more intense convective circulation and a larger number of convective vortices than homogeneous surfaces due to temperature gradients [Renno et al., 2004]. Compared to the Pathfinder landing site, the Phoenix landing site is very homogeneous with almost no topographic features on the scale of large boulders or hills [Ward et al., 1999; Heet et al., 2009].
Fisher et al.  note a correlation between peak surface temperatures and dust devils in MOC images from the Amazonis region (36°N). MER Spirit also saw a correlation between a change in daytime surface temperature in the Gusev crater (15S) and dust devil activity [Greeley et al., 2006]. Furthermore, this pattern was seen to some extent by Cantor et al.  in the northern hemisphere in a multiyear survey for dust devils with MOC.
 In our case the high midsol, near-surface air temperatures around sol 40–75 (Ls = 95–111) as seen on Figure 1 do not correlate with high surface heat fluxes [see Davy, 2009, Table 4.1] or with a peak in dust devil activity. During the Phoenix mission the numbers of vortex pressure events detected per day increase after sol 75 (Ls = 111) at which time maximum air temperatures are starting to decline while midsol surface heat fluxes are starting to increase, and the diurnal temperature range is increasing slightly.
Renno et al.  argue that dust devil activity is associated with convective plumes; the depth of the boundary layer and the sensible surface heat flux. In contrast to Renno et al.'s  theory the boundary layer depth, as detected by the lidar, is starting to decline at this time [Whiteway et al., 2009]. To complicate matters we should note that dust levels declined (which would affect heating due to absorption of solar radiation in the boundary layer) and nocturnal clouds started to be detected at the top of the boundary layer at this stage of the mission [Whiteway et al., 2009]. However, an inverse relationship between boundary layer depth and number of vortices is also found in our large eddy simulation studies, conducted without dust absorption.
 Thus, even though we do see some correlation between convective vortex activity and midsol surface heat flux, our results indicate that local forcing is not the primary driver for dust devil and vortex activity at the Phoenix site. In section 3.3 we will present orbiter images indicating a relationship between convective vortex activity and baroclinic weather events at the Phoenix site.
3.3. Correlation With Local Weather Events Observed by MARCI
 The Mars Color Imager (MARCI), onboard the Mars Reconnaissance Orbiter (MRO) spacecraft, is a low-resolution, dual optical, push frame camera system with seven filter band passes spanning from the ultraviolet (260 nm) to the near infrared (750 nm) [Malin et al., 2008]. From MRO's elliptical orbit (255 km × 320 km), with periapsis occurring on the dayside of the orbit, MARCI has an intrinsic nadir imaging resolution of ∼900 m pixel−1 on the dayside. MRO's orbit is also sun-synchronous, providing MARCI dayside equator crossing (south-to-north) nadir imaging at ∼1500 local mean solar time (LMST) on each orbit. The 180° field of view of the MARCI allows for limb-to-limb imaging on each of the 12.8 orbital passes it makes every day, from which a daily global map of the planet is generated in each of the seven filter band passes. The near polar orbit of MRO provides for substantial overlap in coverage between adjacent orbital swathes in the polar regions in each hemisphere during their respective late spring and summer seasons. As a result, areas of interest in the polar regions were imaged from 2 to 24 times per sol depending on their latitude. The Phoenix Lander region was imaged between 2 to 4 times per sol, with half the imaging occurring on the 1500 LMST ascending side of the orbit and the other half occurring on the 0300 LMST descending side of the orbit. This semidiurnal imaging capability along with the MARCI band 1 (420 nm) and band 3 (600 nm) filters, which were designed to help distinguish between condensate and dust clouds [Malin et al., 2001], allows for the monitoring and tracking of traveling weather systems (i.e., condensate clouds and dust storms).
 During the course of the Phoenix Mission, we were able to observe with MARCI the passage of condensate (water ice) clouds and in a couple of cases (diffuse) dust clouds, either directly over the Phoenix Lander site or in very close proximity. The timing of these clouds coincided within a sol of low-pressure events and the peaks in dust devil activity detected by Phoenix (Figure 4) and thereby indicating a strong correlation between regional weather phenomena and dust devil activity.
 In the following, we provide a description of the weather phenomena observed in the MARCI imaging, mainly during the periods when Phoenix detected increases in dust devil activity. A summary is given in Table 2.
Table 2. A Summary of MARCI Observed Weather Events
Type of Event
Comments (Phoenix = PHX)
30 May 2008
Diffuse haze had passed over PHX previous sol
18 June 2008
Diffuse cloud observed a few hundred kilometers to west of PHX.
12–13 August 2008
Passed over PHX from the north.
29 August 2008 to 1 September 2008
Several fronts developed to the south and east, generating AM and PM cloud cover over PHX
4–5 October 2008
Water ice AM and PM clouds associated with a front over and around PHX
5–6 October 2008
Water ice AM clouds associated with a front over PHX
7–8 October 2008
Water ice clouds and diffuse dust haze associated with a front passed over PHX
16 October 2008
Water ice clouds associated with a front over PHX
22 October 2008
Water ice clouds associated with a front over PHX
24 October 2008
Water ice clouds associated with a front over PHX
26 October 2008
Convective water ice cloud streamers over PHX, with dust storm developing to the north.
 The images used are color composites generated using 420 nm, 550 nm, and 600 nm filter band images. They have been polar stereographically projected at resolution of 2 km pixel−1 and the Phoenix landing site is marked with a circle with the north polar layered deposits in the lower right corner.
 The weather activity in the first 75 sols of the mission was relatively calm. According to MARCI imaging two diffuse dust cloud events (sols 3–5 and sols 25–35) occurred during that part of the mission, but as seen on Figure 4 neither of these two events resulted in significant changes in vortex activity at the Phoenix landing site. The two events are described below:
 On sols 3–5 (Ls = 78–79) and 25–35 (Ls = 88–92), diffuse dust clouds passed over the landing site. Both instances started at the polar region correlated with surface ice sublimation. Figure 8 shows a time series around sol 21 (Ls = 86), 16 June 2008, where diffuse dust clouds were observed a few hundred kilometers to the west.
 On sol 19 (Ls = 85), minor dust activity is seen at Olympia Undae. At sol 21 (Ls = 86), the dust activity has moved further south, and on sol 22 (Ls = 87) it is clearly seen that much of the surface frost in Olympia Undae has disappeared. The dust activity reached the Phoenix landing site at sol ∼25 (Ls = 88) and was clearly observed as an increase in optical depth [Lemmon et al., 2010] which lasted ∼10 sols and as a slight increase in pressure levels from otherwise generally decreasing pressures [Taylor et al., 2009].
 The lack of an increased number of vortices when diffuse dust clouds passed over the landing site agree with the observed anticorrelation between dust storms and dust devil activity [Cantor et al., 2006] and happens because the dust absorbs the solar radiation and stabilizes the atmosphere. On sols 77–78 and 93–96 MARCI imaging showed condensate clouds over the landing site that correlate with the two first major peaks in the convective vortex activity:
 On sols 77–78 (Ls = 112), 12–13 August 2008, very diffuse early morning condensate clouds associated with a weak cold front coming down from the north passed over the landing site. These morning clouds gave way each afternoon to clear skies. This correlates with the first of the peaks on Figure 4, with a high number of vortex events on sols 77–78 (Ls = 112).
 On sols 93–96 (Ls = 119–121), 29 August to 1 September 2008, MARCI morning imaging showed early morning water ice cloud formations associated with the development of several low-pressure weak baroclinic systems to the south and east of Phoenix. Some of the water ice clouds persisted through the afternoon on sol 94 (Ls = 120) while the early morning clouds were observed moving off to the east at an average speed of 1–2 m s−1, leaving the landing site well off to the west by sol 98 (Ls = 122). Figure 9 shows a MARCI image from sol 95 (Ls = 120). This weather event correlates well with the days where the highest daily number of vortices was observed. Pressure data shows lower pressures during the passing of this event, correlated with the increase in vortex activity, as seen in Figure 10.
 Several times in the remainder of the mission, such condensate clouds were observed over the landing site. Events in Figure 4 also show a clear correlation between the weather phenomena described below and increased vortex activity with peaks around sol 128 (Ls = 137), sol 140 (Ls = 143) and sol 150 (Ls = 148):
 Diffuse water ice clouds were observed over and around the landing site from sols 128–129 (Ls = 137), 4–5 October 2008, followed by another early morning weak front observed over the landing site on sols 129–130 (Ls = 137–138). This was observed in the afternoon hours as very diffuse water ice clouds.
 Following this, condensate/dust water ice clouds and diffuse dust haze associated with a front passed over Phoenix on sols 131–132 (Ls = 138–139).
 By the end of the first week of October, dust storm activity began to pick up along the edge of the northern perennial cap. These storms were observed well to the east or west of the lander. Additional, condensate water ice clouds were observed to pass directly over or in close proximity to the Phoenix Lander between associated eastward moving fronts on sols 140 (Ls = 143) and 146 (Ls = 146), 16 and 22 October 2008. The early morning imaging on sol 148 (Ls = 147), 24 October 2008, showed water ice clouds associated with a front directly over the Phoenix Lander. Afternoon imaging showed convective clouds to the east and south of the lander associated with the leading edge of the front. By sol 149 (Ls = 147), early morning skies were slightly clearer as the leading edge of the weather front was to the south and east of Phoenix. Afternoon water ice condensate clouds had become optically thicker south of the lander.
 By the next sol, sol 150 (Ls = 148) (26 October 2008), afternoon imaging showed that trailing side of the storm system was over the lander as indicated by the convective cloud streamers extending over the lander site. A local dust storm could be seen coming off the southern boundary of the ice cover outliers to the north. It was this storm system that developed into a regional dust storm that put an end to Phoenix's extended mission. In all instances, with the exception of the mission terminating regional storm that began on sol 150 (Ls = 148), afternoon peak dust and water ice optical depths associated with these frontal systems (which were all derived away from the Phoenix Lander) were <2.1.
 The vortex and dust devil activity seems to be stimulated along or within weak fronts as these are areas of deep convection and vorticity. An example of this is sol 95 (Ls = 120) where we see condensate clouds above the landing site and a substantial increase in dust devil activity (see Figure 4). This is in accordance with MOC observations by Cantor et al.  on dust devil activity stimulated along the periphery of a few storms in northern Amazonis. Thus, we can conclude that compared to the lower latitudes on Mars [Fisher et al., 2005; Greeley et al., 2006] the dust devil activity in the Martian arctic is far more influenced by more active weather systems such as condensates and weak fronts than by changes in temperature.
3.4. Large Eddy Simulations
 In parallel to the Phoenix observations, large eddy simulations (LES) of highly convective boundary layers have been performed to investigate the natural formation of convective vortices in the environment of the lander. For this the NCAR LES [Sullivan et al., 1994] has been adapted for Mars and developed for time series detection of vortices and their physical characteristics. Results show a satisfactory agreement with the observations. Sample result of a simulated vertical vortex, shown in Figure 11, represents a pressure drop of about 0.68 Pa and a temperature increment of about 6 K (Figure 11, top). The diameter (see Figure 11, bottom) and height of the vortex are, respectively, about 120 m and 1100 m. Characteristics of the pressure signature of this vortex are consistent with the Phoenix observations presented in Table 1 if we assume a translation velocity of order 4 m s−1, agreeing with typical winds observed by the Telltale [Holstein-Rathlou et al., 2010].
3.5. Observed Dust Devils and Dust Devil Tracks
 Even with dust devils imaged in the area from orbit before landing (M. C. Malin et al., available at http://www.msss.com/msss_images/2008/05/06/, 2008), the earliest dust devil spotted by the Surface Stereo Imager (SSI) on Phoenix was on sol 104 in a set of panorama images. After this, several dust devil search campaigns were made and dust devils were imaged frequently. Over the course of the Phoenix mission, 1506 images of the horizon were taken, and 562 of those were in the southwest direction where all dust devils were imaged. Only sol 104 (Ls = 125) revealed incidental dust devils while taking a slice of the multispectral panorama. After sol 104, image sets were dedicated to search for dust devils, and at least one dust devil was spotted in 7 of the 12 sets. Image sets consisted of 13 to 50 images of the horizon, taken mostly between 1100 and 1600 local true solar time (LTST). The difference between LMST and LTST was roughly 13 min at sol 0 and roughly 36 min at sol 151.
 To review the images, the contrast was enhanced which allowed very faint wind events, up to a ∼3% difference of the background albedo, to be more easily viewed. The majority of the dust devils were rather faint and did not appear until after this processing.
 During the Phoenix mission a total of 76 dust devil events were found in the SSI images from sol 104 through sol 138 (Ls = 125–142). In the sets of sequential images, dust devils were seen to move between capturing each image. An example of this is seen on Figure 12 that shows a dust devil captured on sol 109 (Ls = 127).
 Based on the observable physical characteristics and the distance traveled between images, there were 37 unique dust devils imaged during the mission. Another 11 events are more likely to be strong gusts of wind that picked up loose regolith rather than actual dust devils. This interpretation was based on the events lacking the revealing characteristics of dust devils, such as vertical lift and ability to stay cohesive for a considerable length of time. As the mission progressed, more wind events were observed which is most likely caused by the changing seasons with weather events passing by the landing site.
 Due to the lack of images, interpretation should be made with some caution. However, Figure 13 shows the success rate of the dust devil observations that started on sol 104 (Ls = 125). Some time variation can be seen from sol 104 to sol 138 (Ls = 125 to 142) with most dust devil captures compared to the amount of images occurring sols 127 to sol 138 (Ls = 136 to 142) which seems to agree with a high dust devil activity at that time.
Figure 13 also shows the same distribution as a function of LTST. Most images were taken during the 1300–1500 h time frame and most dust devils were observed during midsol, peaking in the 1300 LTST hour with 32% of the images including dust devils, confirming the distribution seen on Figure 6. No dust devils were imaged in the 1200 and 1600 LTST hours, but only few images were taken during this period of time.
 As mentioned before, dust devil tracks as well as wind streak features were used by Drake et al.  to measure the predominant wind direction in the different regions at the Phoenix landing latitude. Phoenix ultimately landed in region D, but wind directions for region D were not given in the work of Drake et al.  as the area was not considered a potential landing location for Phoenix at that time. However, wind directions for region D are provided by Holstein-Rathlou et al.  and it can be seen that the predominant wind direction for region D (110–130 W) is W/NW–E/SE. There is a directional ambiguity in the results since the dust devil tracks and wind streaks do not indicate if the winds are blowing, e.g., N-to-S or S-to-N.
 Due to the lack of clear landmarks in the SSI images, it is difficult to calculate the exact distance and traveling direction of the dust devils and interpretation should be made with caution. However, all images with dust devils were taken in the SW direction and based on the movement of dust devils in the image frames, it seems likely that most of them are moving with a component in the W/E direction. This also agrees with observations from the Telltale that saw midsol predominant winds from west in the last part of the mission, after sol 90 (Ls = 118) [Holstein-Rathlou et al., 2010].
3.6. Comparison to Telltale Wind Speed Data
 The number of vortices observed by the pressure events depends on the wind speed to a certain level. With constant area density of vortices, the number of detected pressure anomalies will be proportional to the wind speed until the wind speed is large enough to destroy the dust devil column.
 As previously mentioned, wind speeds at the landing site were measured with the Telltale wind indicator [Gunnlaugsson et al., 2008; Holstein-Rathlou et al., 2010]. During daytime (0600 to 1800 LMST), average wind speeds show correlation with the passing of condensate clouds, and increase toward the end of the mission. This agrees very well with our observations of an increase in large ΔP pressure events with time and confirms the correlation between vortex size and background vorticity or wind shear [Renno et al., 2004].
 During daytime, measured wind speeds show significant dependence on convective turbulence, and change on a few minute time scale. Reasonable average wind speeds Uav are obtained for series containing more than 10 Telltale images. Using daytime data (0600 to 1800 LMST), days with more than one measurement of wind speeds suggest that the daily average can be determined with standard deviation of 0.8 m s−1. It has to be noted that sol 95 (Ls = 120) wind data were insufficient for this analysis.
 Normalizing the number of observed vortices to the daily average wind speed should give a number proportional to the areal density of the convective vortices and using only days where both wind data and vortices observations exist give the results shown on Figure 14. If we assume a width of 100 m then 1 event per sol/Uav is equivalent to roughly 1 event per sol per 10 km2.
 It can be seen that the areal density of the convective vortices shows a generally constant level with periods of increased activity between sols 77 and 93 (Ls = 112 and 119), and two on perhaps two days thereafter, 128 (Ls = 137) and 150 (Ls = 148). All of these sols, condensate clouds were seen passing the landing site in MARCI images, as were shown in section 3.5.
 There were 28 convective vortex or dust devil events observed during image sequences with the Telltale, and these have been used to determine the direction of which dust devils were seen traveling at the landing site [Holstein-Rathlou et al., 2010]. As the Telltale data is acquired in average every 50 s, only a few pressure events can be correlated with wind perturbations. Table 3 shows pressure events where a Telltale image has been taken within 10 s from the minimum pressure dip.
Pressure events with Telltale images taken within 10 s. Listed are the corrected pressure drop and duration, time between pressure minimum and Telltale image (corrected for the time response of the pressure sensor), and associated change in wind vector magnitude.
 On average, the changes in the wind vector magnitude between Telltale images are less than 1 m s−1 during daytime conditions. From Table 3, much higher magnitude changes (Δv) are observed during pressure events. Without having information on how far from the lander the vortices passed, detailed analysis is hampered, and it can only be stated here that the events are consistent with vortices in cyclostrophic balance with scale lengths from 20 to 200 m in diameter.
3.7. Dust Devils and Dust Lifting
 Dust devils are thought to play an important role in the entrainment of dust in the Martian atmosphere. However, it is not fully understood how the fine dust particles (few μm in size) are lifted from the surface and into the atmosphere. For the Martian atmosphere, the threshold friction velocity needed to lift the fine dust is generally estimated to be ∼30 m s−1 [Greeley and Iversen, 1985; Greeley et al., 2003], and low density dust aggregates could be lifted at wind speeds of 10–15 m s−1 [Merrison et al., 2007]. The tangential wind speed of many dust devils approximates the Rankine vortex model [Sinclair, 1973] and as in the work of Renno et al. [1998, 2000], assuming that dust devils to the first order are in cyclostrophic balance, the central pressure drop ΔP can be expressed as
where R is the atmospheric gas constant, T is the mean temperature, Pav is the mean pressure, and v is the maximum tangential wind. Using typical values of T = 240 K, P = 780 Pa, R = 187 J kg−1 K−1 with a wind speed of v = 25 m s−1, this corresponds to a pressure drop of roughly 11 Pa. We did not see any dust devils of this magnitude and few dust devils have been observed or modeled to have such tangential velocities [Ryan and Lucich, 1983; Renno et al., 2000; Toigo et al., 2003]. From this approximation, the dust devil with largest ΔP captured by the Phoenix pressure sensor (with a ΔP of 3.6 Pa) had tangential wind speeds of ∼14 m s−1. The pressure events captured by the Telltale shown in Table 3 indicate wind speeds roughly in agreement with cyclostrophic balance.
 Thus, Phoenix observations seems to be in agreement with the tangential wind speeds of dust devils observed visually at the Mars Pathfinder landing site (0.5–4.6 m s−1) [Metzger et al., 1999] and theoretically calculated from the Pathfinder meteorological data (9.0–17.7 m s−1) [Renno et al., 2000]. For comparison, general wind speed data from Viking [Hess et al., 1977], Pathfinder [Schofield et al., 1997] and also Phoenix [Holstein-Rathlou et al., 2010] all show average wind speeds of around 5–10 m s−1, thus yielding a maximum wind perturbation of less than 30 m s−1. These are all wind speeds that are well below the general Martian dust-lifting threshold [Greeley and Iversen, 1985; Greeley et al., 2003] which indicates that the dust devil induced dust lifting is not only caused by the force from the tangential winds.
 To explain this, it has been suggested that the dust devils lift material by a combination of impact saltation and suction from the low-pressure core [Greeley et al., 2003, 2006]. Another plausible explanation is suggested by Wurm et al. , who describe a method to significantly lower the threshold for dust entrainment by wind by a combination of a solid state greenhouse effect within the upper layer of dust and a thermophoretic effect. These effects in combination might help provide the initial lift necessary for the convective vortex to lift dust and hence become a dust devil even though the wind speeds are smaller than, e.g., 25 m s−1.
 The Phoenix mission has collected a unique set of in situ meteorological data from the Martian arctic [Taylor et al., 2009] and with this data set, the existence of convective vortices and dust devils in the north polar regions of Mars has been confirmed. During the Phoenix mission, 502 vortex identifications with ΔP > 0.3 Pa were found, with most events occurring in the noon hours. The largest recorded pressure drop caused by a vortex event was 3.6 Pa and was captured on sol 95 (Ls = 120). Most events have a Γ duration around 15 s and a correlation between high wind speeds and large pressure events is seen.
 The diurnal distribution of the convective vortices is bell shaped with most events occurring in the noon hours, consistent with the late afternoon collapse of boundary layer turbulence [Tamppari et al., 2008; Tyler et al., 2008]. A general increase with major peaks in the convective vortex activity is seen from sols 75–151 (Ls = 111–148), indicating that the dust devil season continues after Ls = 148 at the Phoenix landing site. During sols 75–151, we also see an increase in the number of pressure events with large ΔP. This correlates with changes in midsol surface heat flux [Davy et al., 2010] and increasing wind speeds as observed by the Telltale [Holstein-Rathlou et al., 2010]. An inverse relationship between boundary layer depth and number of vortices is also seen. Comparisons with MARCI imaging show that the convective vortex and dust devil activity seems to be stimulated along weak fronts and is controlled mainly by active weather events passing by; the same periods where a correlation between vortex activity and Telltale wind data shows increased vortex density. This is different from the lower latitudes on Mars [Fisher et al., 2005; Cantor et al., 2006; Greeley et al., 2006] where the dust devil activity is mainly controlled by local forcing.
 We thank the Phoenix engineering and science teams for all their work leading to a successful mission. The financial support from the Danish Natural Science Research Council for the Danish participation in the Phoenix mission is greatly appreciated. Canadian university participation was supported by Canadian Space Agency grants and contracts. We thank an anonymous reviewer for constructive criticisms and suggestions improving the manuscript substantially. Thanks to Carlos Lange for his help with the vortex shedding and to Søren Larsen for suggestions to the manuscript. L. Tamppari's contribution to the research described in this paper was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with NASA. The Phoenix mission was led by Peter Smith of the University of Arizona, on behalf of NASA and was managed by NASA's Jet Propulsion Laboratory, California Institute of Technology. The spacecraft was developed by Lockheed Martin Space Systems.