The meteorology and chemistry of high nitrogen oxide concentrations in the stable boundary layer at the South Pole

Four summer seasons of nitrogen oxide (NO) concentrations were obtained at the South Pole (SP) during the Sulfur Chemistry in the Antarctic Troposphere (ISCAT) program (1998 and 2000) and the Antarctic Tropospheric Chemistry Investigation (ANTCI) in (2003, 2005, 2006–2007). Together, analyses of the data collected from these studies provide insight into the largeto small-scale meteorology that sets the stage for extremes in NO and the significant variability that occurs day to day, within seasons, and year to year. In addition, these observations reveal the interplay between physical and chemical processes at work in the stable boundary layer of the high Antarctic plateau. We found a systematic evolution of the large-scale wind system over the ice sheet from winter to summer that controls the surface boundary layer and its effect on NO: initially in early spring (Days 280–310) the transport of warm air and clouds over West Antarctica dominates the environment over the SP; in late spring (Days 310–340), the winds at 300 hPa exhibit a bimodal behavior alternating between northwest and southeast quadrants, which is of significance to NO; in early summer (Days 340–375), the flow aloft is dominated by winds from the Weddell Sea; and finally, during late spring, winds aloft from the southeast are strongly associated with clear skies, shallow stable boundary layers, and light surface winds from the east – it is under these conditions that the highest NO occurs. Examination of the winds at 300 hPa from 1961 to 2013 shows that this seasonal pattern has not changed significantly, although the last twenty years have seen an increasing trend in easterly surface winds at the SP. What has also changed is the persistence of the ozone hole, often into early summer. With lower total ozone column density and higher sun elevation, the highest actinic flux responsible for the photolysis of snow nitrate now occurs in late spring under the shallow boundary layer conditions optimum for high accumulation of NO. This may occur via the non-linear HOX–NOx chemistry proposed after the first ISCAT field programs and NOx recycling to the surface where quantum yields may be large under the low-snow-accumulation regime of the Antarctic plateau. During the 2003 field program a sodar made direct measurements of the stable boundary layer depth (BLD), a key factor in explaining the chemistry of the high NO concentrations. Because direct measurements were not available in the other years, we developed an estimator for BLD using direct observations obtained in 2003 and step-wise linear regression with meteorological data from a 22 m tower (that was tested against independent data obtained in 1993). These data were then used with assumptions about the column abundance of NO to estimate surface fluxes of NOx . These results agreed in magnitude with results at Concordia Station and confirmed significant daily, intraseasonal and interannual variability in NO and its flux from the snow surface. Finally, we found that synoptic to mesoscale eddies governed the boundary layer circulation and accumulation pathways for NO at the SP rather than katabatic forcing. It was the small-scale features of the circulation including the transition from cloudy to clear conditions that set the stage for short-term extremes Published by Copernicus Publications on behalf of the European Geosciences Union. 3756 W. Neff et al.: The meteorology and chemistry of high nitrogen oxide concentrations in NO, whereas larger-scale features were associated with more moderate concentrations.


Introduction
The Investigation of Sulfur Chemistry in the Antarctic Troposphere (ISCAT) field programs in the austral summers of 1998 40 and again in 2000 (Davis et al. 2004a; Davis et al. 2001) discovered unexpectedly high atmospheric nitrogen oxide (NO) concentrations at the South Pole (SP). These early investigations suggested that the high NO levels were associated with continuous sunlight, shallow stable boundary layers, downslope flow from the east Antarctic plateau, and associated nonlinear NO X -HO X chemistry that resulted in long NO X lifetimes, (Davis et al. 2004a; Davis et al. 2008)). It was also argued that the high levels of NO at the SP, compared to other polar sites, were partly due to the long fetches for air parcels in the 45 katabatic flow from the Antarctic high plateau and the lack of a diurnal cycle, thus allowing continuous photolysis of snow nitrate. A major field program followed in 2003 to re-examine this phenomenon, the Antarctic Tropospheric Chemistry Investigation (ANTCI). This new study involved ground-based measurements at SP as well as aircraft probing over more extensive areas of the Antarctic Plateau in 2003 ) and then again in 2005 (Slusher et al. 2010). These field studies reinforced the earlier ISCAT results and introduced evidence suggesting that fast recycling mechanisms for 50 redeposited nitrate could be important given that photolysis of buried nitrate appeared inadequate to explain the sustained levels of atmospheric NO X . In combination with meteorological factors, this recycling enabled NO X levels which routinely reached and/or exceeded several hundred pptv. Because of significant daily, seasonal, and inter-annual variability in NO observed in the previous field studies this paper will examine in greater detail the influence of large-to small-scale meteorological circulations on the boundary layer and associated chemical factors that set the stage for high NO episodes. A 55 key feature of the boundary layer at the South Pole is that it is statically stable throughout most of the year and often very shallow which leads to the confinement of surface chemical emissions near the surface except during higher-wind events and cloudy periods when a deeper internally well-mixed inversion layer is often present. Important factors also influencing the near-surface chemical environment include the winter-to-summer seasonal cycle in wind and temperature as well as interannual variability and synoptic variability in cloudiness (important to the surface energy budget and the boundary layer 60 structure), low snow accumulation, and total column ozone and the radiative changes associated with the breakup of the stratospheric polar vortex and ozone hole in the spring. Past work (Neff, 1999) found evidence for the effect of stratospheric ozone depletion on the tropospheric circulation in the Austral spring over the interior of Antarctica. A key question then was the potentially combined effects of changes in the radiative environment (via UV photolysis) and concomitant changes in the near-surface meteorology affecting NO. The complexity of the potential processes affecting NO are well captured graphically 65 in Davis et al. (2008, Figure 2) where they identify atmosphere-surface exchange processes, plateau drainage, loss mechanisms associated with continental outflow, lower latitude transport into the continent, and boundary layer-free troposphere exchange as key meteorological processes. However, earlier work did not examine year-to-year meteorological variability and effects on NO. In addition, the step-wise linear regression that we show later demonstrates that the dominant parameters for NO vary from year to year reflecting the variability in weather regimes affecting NO at the SP. Thus, the new 70 finding we will show is that not only many variables are important but that their relative importance varies from year-to-year To address some of these processes, we analyze previous data sets complemented by a new data set collected in 2006-07 that extends over much of the seasonal cycle. Past work emphasized the relationship between high NO and shallow boundary layers Neff et al. 2008). Here, the primary focus is on providing a broader perspective of the origin and evolution of high NO episodes, their relationship to stable boundary layer meteorology, the unique large-scale meteorology 75 of the high Antarctic plateau, the influence of the changes in the stratosphere in recent decades, and the effect of variability within seasons as well as from year-to-year. se, effort uses analyses of previous data sets complemented by a new unreported data set collected in 2006-07 that extends over much of the seasonal cycle. The foci here are the origin of high NO episodes and their relationship to stable boundary layer meteorology, meteorological variability within seasons as well as from yearto-year 80 2. Boundary layer meteorology, NO, and the seasonal cycle, and boundary layer meteorology In this section, we examine the meteorological factors controlling boundary layer depth (BLD) and NO variability, over hourly, intraseasonal and interannual time scales, using the summer field program data from 1998, 2000, 2003, and 2006-07. Because NO was the one species measured consistently in all field studies cited, it will be used in most of the follow-on discussions (NO and NO 2 were both measured in [2006][2007]. This analysis benefits from a repeatable weather cycle at SP 85 and the lack of a diurnal solar insolation cycle. The weather cycle typically includes a period of advection of warm air, higher wind speeds, and clouds with a deep boundary layer followed by clearing skies, strong radiative surface cooling and formation of shallow statically stable boundary layers (Neff, 1999). These changes occur over normal synoptic time scales of a few days and over planetary wave time scales of 10-60 days (Yasunari and Kodama 1993). During periods of warm air advection and cloudy conditions, the surface inversion can be fairly deep and well-mixed as the warm air adjusts to a colder 90 ice surface. In these cases the boundary layer is often several hundred meters deep (Neff, 1980). This is in contrast to clear during the "evening (low sun elevation)" period (Frey et al. 2015a) (Bonner 2015;Bonner et al. 2010). A detailed study of the surface energy budget and boundary layer evolution at Concordia Station (King et al., 2006) shows a convective boundary layer and positive net radiation lasting eight hours (averaged over December and January). This convective period developed when the solar elevation angle was greater than 25 o , in contrast to the lack of thermal convection at the SP where 100 the maximum solar elevation angle is always less than this. Frey et al. (2015) show a specific example for 9 January at Concordia when a stable boundary layer forms with large increases in NO shortly after 1800 LT when the solar elevation angle decreases to less than 20 o . Similarly King et al. (2006) show a sodar record from 28 January with a stable boundary layer until 0800 and after 1730 LT when solar elevation angle less than ~22 o . For these reasons air masses arriving over long distances at the SP are likely to have encountered some convective mixing enroute and may need to be considered in future 105 model simulations of surface chemistry.
The synoptic weather variability at SP occurs in the context of a strong seasonal cycle from early spring to summer as the sun rises and the omnipresent surface inversion weakens to less than 10% of its winter value (Neff 1999) implying only weak katabatic influence during the Austral summer (discussed in more detail below). Also of Of particular interest in our analyses are the major shifts in large-scale weather patterns that occur from early-October to 110 mid-or late-December, coincident with the weakening of the circumpolar trough, warming of the stratosphere and breakup of the polar vortex and the stratospheric ozone hole. The dynamical consequences of this these transitions also affect boundary layer behavior that is dependent on cloud fraction, surface radiative balance, inversion strength, as well as upper-level winds and associated surface pressure gradients. In our discussions we will use the grid direction convention at SP where grid north is aligned with the Greenwich Meridian ( Fig. 1).

Overview of the four field seasons
The first three field seasons have been described in some detail (Davis et al 2001(Davis et al , 2004 whereas a fourth more extensive observation of the seasonal cycle in NO, obtained in 2006NO, obtained in -2007 has not yet been published. The four seasons of NO data are shown in Fig. X1 and their statistics in Table X1. In 1998, the ozone hole broke up very late as seen in Fig. 1a but the observational program was also delayed until early December. In 2000 (Fig. 1b), total column ozone increased well 120 before the beginning of the field season and NO remained relatively low. In both 2003 (Fig. 1c) and 2006, the ozone hole broke up about the same time but higher column ozone was more persistent in 2006 until late December. In Fig. 1d

The seasonal cycle, upper-level winds and their influence on surface flows
As the stratosphere over Antarctica begins warming around the time of the spring equinox, prevailing winds at the height of 135 the summer tropopause over the South Pole (~300 hPa) evolve in three stages as shown in Fig. 1 2 (based on rawinsonde data for the period 1990 to 2013). In Fig. S1 we provide more detailed wind-rose plots for each subseason as well as cloud fraction as a function of 300-hPa wind direction for extremes in daily cloud fraction of 8-10 tenths and 0-2 tenths.
In Early Spring (JD280-310), 300-hPa winds are largely from the direction of west Antarctica and associated with high cloud fraction and advection of warmer air at the South Pole (Neff 1999). This period coincides with the maximum in the 140 semi-annual oscillation when the circumpolar trough is most intense (Meehl 1991 Wedd photolysis and recycling (e.g. Frey et al. 2015).
The increase in synoptic weather activity at this time suggests increased likelihoodthe possibility of transport of NO precursors from the ocean, surrounding sea ice and more northerly latitudes to the interior of the continent. The increase in 150 synoptic weather activity at this time suggests the possibility of transport of NO precursors from the ocean, surrounding sea ice and more northerly latitudes to the interior of the continent. During this period it has been noted that atmospheric nitrate typically increases in mid-October whereas surface nitrate increases later in mid-November although this may reflect higher fluxes of reactive nitrogen early in the season compared subsequent nitrate deposition rates (Erbland et al. 2013). Erbland et al (2015) also suggest, as 155 an estimate, that half of the annual average nitrate comes from stratospheric sources and half from longrange transport (although there are no measurements of this). Meanwhile, it has been suggested that increased surface nitrate at Concordia Station in 2006 has statospheric origins (Traversi et al. 2014). In fact, back trajectories for this period show them passing just to the east of SP in early November 2006 with higher nitrate observed on 13 November at Concordia Station whereas during our last observational field 160 program we observed higher NO on 18 November at SP. For austral winters Neff (1999) calculated isobaric geostrophic temperature advection at the SP from averaged rawinsonde profiles and found advective warming associated with NW to SW winds aloft and cooling associated with SE winds aloft.
As the season evolves during the Late Spring (JD310-340), the winds become semi-bimodal as shown in Fig. 1 Neff (1999), this Late Spring period also has the lowest average cloudiness (see also Fig. S8 for more recent data). Furthermore, for winds greater than 10 ms -1 at 300 hPa, between 1990 and 2010, in the 45 o intervals centered on 157.5 o SSE and 337.5 o NW, the average surface inversion strength (from rawinsonde soundings) and surface wind speed are 170 6.3 o C and 3.5 ms -1 and 2.5 o C and 5.4 ms -1 , respectively. Thus, in addition to reduced cloudiness during this period, surface wind speeds are less and surface-based inversion strengths are larger for 300-hPa winds from the SE. As seen in Fig. S1, clear-sky conditions byare most likely with winds aloft from ESE to SE.
In Early Summer, the occurrence of SSE winds decrease while the frequency of winds from the NNW increases. (For the Early Summer period, the results are similar at 2.2 o C and 3.1 ms -1 and 0.3 o C and 5.9 ms -1 , respectively). With the higher 175 frequency of winds from the northwest in early summer with increased cloudiness, low static stability and higher surface wind speeds, high NO should be less likely on the average, after mid-December. Conversely, periods of low-cloud fraction are consistent with increased radiative cooling of the surface and the stronger surface temperature inversions that result are consistent with higher NO.
The difference in surface wind speed and inversion strength for the NW and SE modes can be explained in terms of the 300 180 hPa geopotential height (GPH) patterns corresponding to the NW-SE two modes of upper level winds (Fig. S1S2). For NW winds, low GPH lies between the Ross and Weddell Seas over West Antarctica with higher GPH over the continental interior. For SE winds, lower GPH lies in the eastern hemisphere extending over most of the high Plateau. Although somewhat different than the winter pattern in Neff (1999) where for SE winds aloft a strong ridge of high pressure extended from the Ross Sea over portions of east Antarctica, the large scale pressure gradient force still reveals either an on-shore or 185 off-shore orientation in the two cases. In the case with low pressure centered mostly over high terrain, downslope flows would be less likelylikely be weaker whereas the inverse would be true for off-shore pressure gradient forces.
The frequency of occurrence of SE winds at 300 hPa can, however, varyvaries from year to year and from decade to decade as reported in (Neff 1999). Figure 2 3 shows the year-to-year variability in the occurrence of SE winds for the more recent  (Charlton et al. 2005). follows the almostFigure 23 shows the semi-decadal variability discovered earlier documented previously for the period 1961-1998 (Neff, 1999). For each of the four observational periods (approximately mid-November through December for consistency) the frequencies of occurrence of winds at 300 hPa from the SE were as follows: 27% in 1998, 2% in 2000, 24% in 2003, and 42% in 2006. For each of these periods when NO was recorded at SP, 195 the number of hours of NO>250 pptv were as follows: 42% in 1998, 4% in 2000, 56% in 2003, and 29% in 2006 Winds aloft from the SE quadrant often coincide with a transition to easterly winds at the surface. Figure 3 shows that these easterly surface winds were most prevalent in 2003 and 2006, intermediate in 1998, and minimal in 2000. Also, averaged over the four field seasons, highest NO occurred when surface winds were from the east to southeast in concert with winds aloft from the SE. Figure Figure S3a reveals strong interannual variability in the occurrence of surface winds with an easterly component as well as an upward trend exceeding 10% beginning in the late 1990s. In this case, we are only looking at the period of Late Spring whereas in Fig. 2, the data 220 extended to the end of December. Figure S3b shows similar interannual variability in SE winds at 300-hPa as well as multidecadal variability but no systematic trend. In Fig. S3c, we show that extremes in the frequency of occurrence of winds from the SE at 300 hPa correspond to extremes in the occurrence of surface winds from the SE with r 2 =0.4. Comparing Fig. S3 with Fig.2 which spanned the period mid-November to the end of December, shows that easterly surface winds occurred primarily after mid-December in 1998: However, in that year, the ozone hole breakup was delayed until the end of December 225 and hence optimum boundary layer conditions favorable for high NO occurred with still-high actinic flux as seen in Fig. 1 where the 1998 field study started later in the seasonal cycle.
The surface inversion strength also varies strongly with the season as shown in Fig. S4a where we have plotted the average inversion strength from 1961 to 1998 (Neff 1999) as well as daily data for 2006 showing typical short-term variability. (The inversion strength was determined from the daily or twice-daily rawinsonde temperature profile, measuring the difference 230 from the 2-m tower temperature to that at the first relative maximum in the rawinsonde profile). Because of the common assumption that the winds over Antarctic are largely katabatic in nature, motivated mainly by wintertime observations in coastal areas, we also show in Fig. S4a the geostrophic wind equivalent of inversion strengths of 4 o C (1 ms -1 ) and 18 o C (5 ms -1 ) assuming a terrain slope of 0.001, characteristic of the interior of the ice sheet and derived from (Ball 1960) by Neff (1980. From the figure, geostrophic winds aloft of only 1 ms -1 will dominate over katabatic forcing from mid-December to 235 mid-January. In Fig. S4b we show the distribution of wind speeds at 300 hPa for December, averaged from 1998 to 2014. This figure shows that almost all synoptic scale winds in December at 300 hPa imply surface synoptic pressure gradients substantially greater than the katabatic ones associated with summer inversion strengths (Neff, 1980;Neff, 1999, Sec. 4.2;Parish and Cassano, 2003). The conclusion from the previous figures is that synoptic and mesoscale pressure gradients at the surface dominate over those driving katabatic flows over the interior of Antarctica, especially in the period November-240 January and that 300-hPa winds from the SE provide a significant control over the boundary layer depth important to high NO. Past work has suggested that katabatic trajectories may have been responsible for long fetches over which NO could accumulate (Davis et al. 2004a). However, from these arguments it appears more likely that meso-to synoptic-scale circulations define accumulation pathways over the interior of Antarctica in the late spring and summer and that these pathways vary significantly on intraseasonal to interannual time scales. Furthermore, as can be seen comparing surface wind 245 directions for maximum observed NO in Fig. 3 with the large-scale topography in Fig. 1b, the surface flows align largely along constant topographic contours between South Pole and Concordia Station (along 123 o E) rather than downslope. Of note, high NO, up to 400 pptv, along 110 o E was found for several hundred km from the SP on 4 December 2003 in aircraft measurements ). CHowever, this was an unusual period starting in late November with a deep boundary layer, high winds as well as high NO, in particular, on 30 November . At the same time high filterable 250 nitrate occurred with high methanesulfonate (MSA) suggesting marine origins (Arimoto et al., 2007). Also, Arimoto et al.
found trace elements normally associated with continental origins and thus could not rule out external influences during this period (although for most of the experimental period the preponderance of evidence pointed to recycling of NO from the surface ).

2.3
The Changes associated with the spring breakup of the polar stratospheric vortex

255
The warming of the stratosphere between October and December is important for changes in the large scale circulation as well as to boundary layer behavior at the South Pole (Neff 1999). It has been noted by others that the final warming of the southern stratosphere has a significant influence on the tropospheric circumpolar circulation (Black and McDaniel 2007) and over the interior of the continent (Neff 1999). Furthermore, with the final warming, downward coupling (at Wave-One scales) from the stratosphere to the troposphere ceases (Harnik et al. 2011)   consequence of delays in recent decades, the actinic flux has increased in the spring as a result of a persisting ozone depleted column coupled with a higher sun elevation angle. Figure 1 suggested that during the late spring/early summer period of higher sun elevation angle and thus higher potential actinic flux, the late-spring period had an almost 50% higher occurrence of SE winds compared to early summer. A question remains, however, is whether the distribution of winds aloft in Fig This implies even less favorable conditions for high NO during the summer period. This change in the circulation may be due associated with to the stronger tropopause strength in the 1960s and 1970s (average difference of 6.7 o C between 300 hPa and 200 hPa) versus recent decades (1.6 o C). that allows winds in the lowermost stratosphere to be less isolated from those in the 300 upper troposphere.

Seasonal variation in cloudiness
Shallow boundary layers occur most often with clear skies. With cloudy conditions the boundary layer is usually deep and well mixed ). It has been noted that cloudiness has generally increased over Antarctic in recent decades (Neff 1999;Schnell et al. 1991). Figure S8 shows an updated version of Plate 5 in Neff (1999)  However, the average cloud fraction can vary within a season and from year to year. For example, the average cloudiness for days 310-340 (late spring) was 6.5, 5.1, 3.9, 5. 1 for 1998, 2000, 2003, and 2006 respectively. In 1998, the late spring period 315 was particularly cloudy compared to average conditions. However, after a late start to observations (see Fig. 1), higher NO occurred intermittently later in December during days with clear skies, light winds and higher actinic flux (due to the delay in the breakup of the ozone hole that year. The intermittency in 1998 is also reflected in the very limited values of NO in excess of 500 pptv (only 4% compared to 25% in 2003, Table 1) The relative minimum in cloudiness marks the transition in the dynamical regime favoring transport of clouds from West Antarctica to one favoring transport from the area of the Weddell 320 Sea. Following this minimum, cloudiness again increases in concert with decreasing inversion strength. This period also occurs at the time of the sea ice minimum implying a shorter moisture transport path from the open ocean to the interior.
Because the total column ozone increases at the same time, the lower actinic flux coupled with higher cloud fraction and its effect on the boundary layer will not be as favorable for high NO concentrations. 4c5c. Figure 4c 5c shows the direct solar irradiance for JD 335-340 (December 1-5) where decreases indicate the presence of clouds which affect the surface radiation balance. The daily average cloud fraction (green bars -from the SP climatological record) shows close inverse agreement with the average direct solar irradiance. Starting on JD 337, skies clear, leading to the 335 rapid formation of a surface temperature inversion below 10m as seen on the 22-m tower. Several hours after the cooling starts, the winds shift from N to SE (winds not shown), the inversion strengthens and hourly NO increases to nearly 800 pptv (1200 pptv in 10-min averages) despite a modest decrease in actinic flux. What is interesting in this case is the fact that even though the photolysis rate is decreasing, NO is increasing locally. What is unknown is whether high NO was generated earlier away from the SP and then transported into the site. This will be discussed more fully in Section 3.2. These results 340 still suggest the dominant role of a strong surface inversion, shallow boundary layer and potential non-linear NO X chemistry as suggested by Davis et al. (2004).

2.5
The role of local topography

350
Although discussions of the slope flows at the SP generally assume an idealized sloped plane of great fetch, the local topography is much more complex as seen in Fig. 56. In fact, the terrain rises 250 m in the first 150 km to the east of the SP (slope~0.002) just adjacent to an extensive plateau (slope~0.0003) that begins 150 km east of the SP and extends 200 km to the foot of the relative steep central Antarctic dome that rises to 4 km ASL to the northeast. In addition, a small "air drainage" basin about 100 km in extent lies to the southeast. The average surface wind direction from the northeast follows 355 terrain isopleths for several hundred km rather than flowing down a slope as is normally envisioned. However, while the average surface wind direction at SP is from the northeast along the terrain, the wind direction actually fluctuates between north (with warmer temperatures) and east (with colder temperatures). In the winter, the difference in the centers of the wind direction distributions for warming-and cooling-events is 90 o (Neff 1999). The distribution for our summer experimental periods is narrower with maxima at 20 o and 70 o and a relative minimum at 40 o . Figure S9 shows

365
We have shown the dependence of NO on 10-m wind speed and direction (Figs. 3 4 and S9). However, in both 2003 and 2006, prolonged and high NO occurred in the last half of November (Fig. S101). This raises the question of whether antecedent conditions set the stage for high NO. In particular, past studies (Mahesh et al. 2003) have noted the systematic change in wind speed between November and December and also concluded that blowing snow was prevalent at the South Pole when the wind exceeded about 7.5 ms -1 which was more likely in November. Figure S11 S10 shows the 1-min wind 370 speed distributions between November and December for the four ANTCI seasons revealing the much higher prevalence of winds in excess of 7.5 ms -1 in November. Table S1 shows a more detailed breakdown of the number of hours of winds exceeding 7.5 ms -1 for October, November, and December for each of 1998, 2000, 2003, and 2006. In addition, hours of relatively clear skies (direct irradiance >80% of maximum) are indicated on the temperature plot with hash marks plotted at -20 C. and daily cloud fraction less than 6/10 ths at -15 C on the temperature time series panel. In 395 general, high direct irradiance associated with a lack of clouds, implies a net loss of radiation from the surface because of the cold dry atmosphere that prevails over the SP, high albedo, and the lower sun angle compared to other polar sites.
In each of these cases in 2006, higher NO follows the clock-wise rotation of the wind past 60 o together with slight-tomoderate drops in temperature (see Fig. 4). The duration in each case ranges from less than a day to multiday periods.
Episodes "A to C" follow high wind episodes by 2 to 3 days. Episode A in Fig. 6 7 reveals a rapid increase in NO as the 400 wind shifts to the east and weakens, followed by a rapid decrease as the wind increases to 10 ms -1 and shifts to northerly.
Episode "B" follows several periods of winds reaching 10 ms -1 until JD320 (16 November) followed by lighter winds from easterly directions and persistently high NO for almost ten days. During the period of high NO in "B", winds at 300 hPa were consistently from the SE, an artifact of a persistent, elongated trough of low pressure over the higher terrain grid NE of the South Pole (see also Fig. S1S2). Episode "C" was explored in the example shown in Fig. 4 5 and will be discussed more 405 fully below. Episodes "D-G' represent short excursions of NO and all occur with lower wind speeds and occasional easterly directions. Episode "H" is included because of its contrast with early periods and its occurrence as one of the few "events" available during the summer period: "H" begins with the temperature exceeding -20 C and then dropping to between -30 C and -25 C with the arrival of a cold front and clear skies. The highest temperatures, above -20 o C, occurred during totally overcast skies, a weak inversion, and NO concentrations less than 100 pptv. Later, there were periodic wind shifts to the 410 east, clear skies, light winds and colder temperatures, and higher but modest levels of . In this case, NO levels remained only moderate despite weak easterly winds, increased inversion strength, a shallow boundary layer, and a longer potential fetch.

NO levels before and after the final warming of the stratosphere
Of particular interest in 2006 is the fact that Episode 'C" occurred just prior to the final warming of stratosphere (Fig. 45). As It should be noted that we did not consider 1998 or 2000: in 1998, the breakup of the vortex was much delayed whereas in 2000 it was very early (Harnik et al. 2011;Neff 1999). Figure

Meteorological and other potential local and external influences on NO concentrations
450  revealed that there were well defined meteorological regimes that favored high NO concentrations that involved wind directions aloft from the southeast, surface winds from the east, and a surface energy balance that was strongly influenced by clear sky conditions. Past work has indicated that the SP is not isolated from the influence from more northerly latitudes. For example, high methanesulfonate was observed from 28 November to 2 December --an indication of marine sources (Arimoto et al. 2008). This period had the highest NO values recorded in the entire series of field programs (Davis et 455 al. 2008) similar to those found at Concordia in a similar time frame in 2011 (Frey et al. 2015). Furthermore, as suggested in (Arimoto et al. 2008) air arriving at the South Pole may have had its origins in continental regions (due to much higher concentrations of the elements Pb, Sb, and Zn than in previous field programs) as well as in the ocean off Wilkes Land on the far side of the continent.
Other antecedents may include transport of NO X in the free troposphere that then mixes to the surface supplying nitrate or 460 alternatively, that air previously in contact with the surface was enriched from snowpack emissions upwind of the station winter/early spring insofar as October-early November is a period of vigorous transport from latitudes north of Antarctica following the maximum in the semi-annual oscillation and as evidenced by high cloud fraction (Neff 1999).

480
Episode "C" is quite interesting insofar as it is preceded by a period of high winds and a rapid surface wind shift from north to southeast following the winds at 300 hPa. Figure 7 8 shows this case in more detail using wind data from the South Pole and AWS stations Henry (100 km along 0 o E) and Nico (100 km along 90 o E) as seen in Fig. 5. These data show that as the wind speed aloft drops, the surface wind speed also drops to less than 0.5 ms -1 albeit sequentially first at Henry, then Nico, and finally at the South Pole. The increase in NO follows a surface wind direction shift and increase in speed from the SE 485 that appears first at Henry and then four hours later at the South Pole. These shifts are quite abrupt whereas that at Nico is more gradual. The increase in NO at the South Pole does not coincide with the drop in wind speed to 0.5 ms -1 (which would have implied local accumulation or local sources) but rather follows during an increase in wind speed with the wind direction shift to the SE (from the basin shown in Fig. 56.) This suggests a hypothesis that this basin to the SE is an accumulation area for high NO that is exhausted after an hour or two (see Fig.78: with wind speeds of 2-4 ms -1 , transport distances of 7 to 14 km 490 per hour might be inferred).
Tracing the progression of the event through the AWS and the South Pole, suggests that the event is triggered by a mesoscale disturbance propagating from the NE that leads to boundary layer changes. However, referring to Fig. 7d8d, it appears that clearing skies quickly result in a low-level radiation inversion and associated shallow BL in which NO can accumulate. The 22-m tower data show that most of the radiative cooling is felt below 10 m. Of interest in this case is the greater contribution 495 of mesoscale processes relative to the potential influence of katabatic forcing. Figure 8 9 shows geopotential and wind field at 650 hPa from ERA-I (~300-400 m above the surface at the SP) during this event at 12Z, JD 337 (3 December) between 90 o S and 75 o S. In this case a small mesolow formed to the west of the SP and migrated rotated cyclonically to just north of the SP over 24 hours. It then intensified further north bringing air over longer distances from the E to NE back to the station but with a reduction of NO to about 200 pptv over the following day. The rawinsonde at 1048Z on Day 337 recorded winds from 500 the NE at 8 ms -1 below 100 m. By 2120Z the rawinsonde showed winds increasing in the lowest 100 m approaching 10 ms -1 .
It is instructive to compare these winds to those estimated from katabatic arguments. An estimate of the katabatic acceleration can be obtained from: acceleration = 9.8 ms -2 * ∆T/T * sin(α). With ∆T=2C, T=250C, and terrain slope, sin(α)=0.002 (from Fig. 56), the acceleration is ~0.5 ms -1 per hour. Four hours at this acceleration rate would lead to a wind speed of 2 ms -1 and a transport distance of several tens of km. In this case katabatic effects can account for only a small 505 portion of the local circulation and Fig. 8 9 shows the dominance of meso-to-synoptic scale eddies in producing transport across the ice sheet. In particular, the small meso-low moved clockwise around SP until at 18Z it was in position to accelerate easterly flow at the surface. With NO decreasing to 200 pptv following the initial peak of 800 pptv in the prolonged NE flow, one might reasonably assume that the initial extreme in this case resulted from short-term boundary layer effects but that prolonged levels of 200 pptv are fetch related and reflect accumulation and/or dilution upstream or transport 510 from longer distances.
This case also highlights a unique feature of the South Pole, namely the low sun angle through the summer that allows a net loss of radiation from the surface in non-cloudy conditions (Carroll 1982). This is unlike the situation at sites at 75 o S and 80 o S where a convective boundary forms during the "daytime" such as Concordia Station (King et al. 2006) or Dome Argus (Bonner 2015). The above example of the role of the rapid onset of surface radiative cooling as clouds clear is analogous to 515 observations from Concordia Station (Frey et al. 2015a), which is subject to a strong diurnal cycle and show a similar rapid collapse in BLD in the late afternoon as the surface cools through long-wave radiation. Even though the actinic flux decreases dramatically relative to its value at local noon at Concordia, the BLD collapses even more rapidly as the production of NO into the BL continues, leading to increases in NO to 0.3 to 1.0 ppbv (Frey et al. 2015). It was also concluded by Frey et al. (2015) that the rapid increase in NO with the collapse of the boundary layer in the early evening suggests that nonlinear 520 HO X -NO X chemistry need not be invoked in these cases. The case in Fig. 7 9 suggests that smaller enhancements of NO (say, less than 200 pptv) occur with longer fetches when NO can accumulate but large excursions such as those in Fig. 7 8 respond to rapid changes in the surface energy budget that lead to very shallow inversion layers. The subsequent evolution of NO would then depend on chemical processes as well as changes in boundary layer mixing, such as those due to increased wind speeds. Figure 8 also suggests limitations in correlating surface NO with wind shifts at 300 hPa: In this case the peak in 525 NO occurs before the shift in winds aloft; by the next rawinsonde 12 hours later NO has dropped substantially. Given the apparent sensitivity to surface cooling under clear sky conditions, Fig. 9 10 shows the relationship between the magnitude of direct solar irradiance, NO, and surface wind direction. In Fig. 9a10a, NO>400 pptv only occurs when skies are essentially cloud-free whereas in Fig. 9b10b, wind directions > 45 degrees are associated with clear skies. Conversely, more persistent cloudy conditions and surface winds from the west to NNW are associated with lower NO. The few outliers for wind 530 azimuths >45 o indicate the presence of occasional scattered clouds (in hourly data).

BLD calculated for 1998, 2000, 2003, and 2006 data
Given the success of the multiple linear regression technique described in Appendix A which used 22-m meteorological tower data with 2003 sodar data to develop regression equations which were then tested against independent data from 1993, we computed hourly BLD for each of the four years during which NO measurements were recorded. The main time period 535 addressed by these estimates was from mid-November (or somewhat later if data collection started later) to the end of December (for consistency). The next step was to test these data for the inverse relationship between NO and BLD as was found for 2003 (Davis et al. 2004a;Neff et al. 2008). If one bins the data to average out short-term fluctuations (Davis et al. 2004a;Neff et al. 2008) one finds a close inverse relation as shown in Fig. 1011. Here we averaged BLD for 100 pptv bins of NO. Earlier results (Davis et al. 2004a;Oncley et al. 2004) applied empirical surface-layer similarity relations to 540 observations of turbulence scales in the surface-flux layer to determine BLDs in the 2000 data set. In their results they found BLD between ~80m and ~500m (Davis et al. 2004a) when they binned NO and BLD in sequential 30-point averages. In Fig.   1011, for 2000, our binned BLD data fell between 28 m and 100 m, a factor of ~3 less in total range. A key argument for high NO at the SP has been the presence of shallow boundary layers which Fig. 10  One potential explanation for this difference is the high snow accumulation rate in the summer at Summit Station of 5.0 to 7.5 cm per month in the summer, maximizing in July (Castellani et al. 2015).
Results from 2003 stand out in Fig. 10. Recapping the findings of Neff et al. (2008), an unusual period of high NO and/or deeper BLD occurred from 24 November to 4 December. This period was unusual in that strong winds from the east 550 extended from the boundary layer to several kilometers aloft. As described in ) the boundary layer was well mixed up to a capping inversion at about 100 m after an initial shallow phase. This was also a period of high methanesulfonate (MSA) and filterable nitrate that were likely of marine origin either from the Weddell Sea or from across the high eastern ice sheet from Wilkes Land and/or continental regions (Arimoto et al. 2008). During this period there were three bursts each lasting 30 to 50 hours of easterly wind and decreasing levels of NO with each burst (765 pptv, 512 pptv, 555 290 pptv, respectively). The first burst had NO>900 pptv coincident with high nitrate (0.5 μg m-3) and high MSA (0.046 μg m -3 ). The highest MSA occurred toward the end of the first easterly burst (Arimoto et al. 2008).

Estimating seasonal NO x emission fluxes for the SP region
The development of a robust method for estimating BLD opens the door to estimating seasonal NO x emission fluxes. This is based on the assumption that the surface flux is in balance with the photochemical loss of NO x in the overlying BL column.

560
These flux estimates require several factors to be considered. First, the total BL NO x must be extrapolated from the nearsurface measurement of NO. While only NO was measured in previous years, the 2006 measurements included both NO and NO 2 . Based on those measurements, NO/NO x was found to be 0.65 +/-0.08. This is consistent with NO x estimates derived from model calculations in previous years (Chen et al. 2001;Chen et al. 2004). Those calculations also indicate that the NO x lifetime exhibits nonlinear behavior increasing from ~7.5 hours when NO x is less than 200 pptv to more than 20 hours when 565 NO x exceeds 600 pptv (Davis et al. 2004a). Along with these effects, tethered balloon measurements at SP during ANTCI 2003 show that the exponential roll off in NO concentrations from the surface to the top of the boundary layer can often be as much as a factor of five for very shallow boundary layers Neff et al. 2008). Using these vertical profiles, the BL column NO x abundance can be derived from surface NO observations. Given that the roll off in NO may not always be as severe as a factor of five, these are considered to be conservative estimates of the total BL NO x .

570
Seasonal NO x emission fluxes are generated by taking daily average values of NO measurements and BLD estimates (based on Appendix A). Then based on the assumptions listed above, a total BL NO x and associated lifetime are estimated. The flux is then equivalent to the total BL NO x divided by the lifetime. It is important to do this calculation daily since changes in BLD can cause total BL NO x and lifetime to vary in different ways. Results for the four seasons as summarized in Table 1 II are 12.5, 6.3, 13.7, and 9.1×10 8 molec cm -2 s -1 for 1998, 2000, 2003, and 2006, respectively (Frey et al. 2015b;Frey et al. 2013). These larger values also exhibit a larger range of temporal variability due to the diurnal cycle at Dome C.
The range of seasonal flux values and daily variability is smaller than might be expected based on Fig. 1011, but it is 585 important to note that under shallow BL conditions, nonlinear growth in NO x abundance can occur as radical concentrations become suppressed and NO x lifetime increases (Davis et al. 2004b). Nevertheless, the differences between years are significant and point to other potential factors affecting the seasonal NO x emission flux. One factor that has already been discussed at some length is the interannual difference in the breakup of the polar stratospheric vortex and associated ozone hole. With regard to nitrate photolysis, this is particularly important to determining the available UV actinic flux in the early 590 Austral spring over the Antarctic plateau. As shown in Fig. S101, this breakup happened much earlier in 2000, the year with the smallest seasonal flux. A second related factor has to do with the potential for recycling of NO x between the firn and atmosphere. As described in the following paragraph, secondary NO x emissions due to recycling are likely to grow in importance throughout the season and are likely influenced by primary emissions in the early spring.
The NO x emissions flux represents contributions from both primary and secondary release of nitrogen from the snow.

595
Primary release represents the photolysis and ventilation of NO x from nitrate initially deposited through snowfall. This nitrate can be much harder to release due to cage effects (Noyes 1956) associated with incorporation into the bulk ice phase. By contrast, secondary release of redeposited nitrate should recycle through the atmosphere much faster due to several factors.
First, the redeposited nitrate is at the snow surface where higher actinic fluxes are available. Second, being adsorbed to the surface of ice crystals, redeposited nitrate should be more easily photolyzed. In a laboratory experiment by (Marcotte et al. 600 2015), photolysis of nitrate on the ice surface was enhanced by a factor of three over nitrate in the bulk ice phase. Finally, redeposition of reactive nitrogen is in the form of both nitric and pernitric acid (Slusher et al. 2002). Pernitric acid would be even more readily photolysed than nitric acid and its relative importance would also increase at lower temperatures. While representing these effects with any realism is beyond the scope of this analysis, more detailed modeling and representation of the factors influence NO x emissions are being investigated by other groups such as in (Bock et al. 2016;Frey et al. 2015b).

605
Future laboratory and field studies are needed to further illustrate the complex mechanisms controlling snow NO x emissions and the chemistry above snow surface.

610
Earlier work Neff et al. 2008), primarily based on 2003 data which included direct sodar measurements of boundary layer depth presented a straightforward conceptual model that linked high NO to shallow boundary layers, light winds, and stronger surface inversions. In our examination of four seasons of observations, explanation of the initiation and evolution of high NO episodes as well as intra-seasonal to interannual variability proved more challenging. Using four spring-to-summer seasons of observations, We we have described the influence of the synoptic-to-mesoscale weather 615 patterns and their seasonal cycle on stable boundary layer characteristics at the South Pole in the spring-summer period that set the stage for high NO episodes on the Antarctic plateau. These included 1) The relative unimportance of katabatic forcing compared to the accelerations due to synoptic and mesoscale scale pressure gradients from November through January. In fact, visualizations of near-surface airflow using ERA-I (Dee et al. 2011) revealed complex mesoscale circulations that belied any simple explanation of accumulation pathways for NO.

620
2) The effect of clearing skies locally that lead to rapid radiative losses and the rapid formation of very shallow inversion/boundary layers and high NO. Given observations only at the SP, the geographical extent of such radiatively driven boundary layers is unknown but worthy of further field observations. Unfortunately aircraft measurements of NO in 2003  were not permitted between -40 o W and 120 o E (the clean air sector). However, the one flight along 120 o E showed NO in excess of 400 pptv between 100 and 400 km from the SP as NO concentrations were dropping at the 625 SP.
3) The three-phase transition in spring for 300-hPa winds over the SP. These three phases corresponded to 1a) a late winter regime of transport of moisture over west Antarctica to the interior when the circumpolar trough is at its maximum and opens the possibility for the transport of NO precursors from northerly latitudes, 2b) a semi-bimodal regime with 300 hPa winds alternating between northwest and southeast quadrants during November and early December as part of the seasonal cycle, 630 followed by 3c) a summer regime favoring 300-hPa winds from the Weddell Sea and warmer cloudy conditions. During the second phase, 300-hPa winds from the southeast favored clear skies, light surface winds and shallow inversions conducive to high NO concentrations at the same time the total column ozone was still low allowing higher actinic fluxes. 300-hPa winds from the northwest favored warm-air advection and cloudy conditions resulting in deep boundary layers and low NO concentrations. Because these boundary layer characteristics follow the same seasonal cycle as temperature, it is difficult to 635 separate out the temperature dependence of nitrate photolysis rates from December into January.
Because of the importance of SE winds aloft to high NO episodes Wwe examined the frequency of their occurrence of SE winds at 300-hPa and as well as winds at the surface from the east to south during late spring (Days 310-340) from 1961 to 2013 and found strong interannual variability. At 300 hPa the interannual variance was significant but modulated on decadal time scales with a maximum in the period 1995-2010. At the surface, the interannual variance doubled with an upward trend 640 of 30% in the mean in the last 30 years. We also found that extremes in the frequency of occurrence of SE winds at 300 hPa correlated well with extremes in easterly surface winds at the surface (r 2 =0.4). This suggests that continuous multi-year monitoring may be advisable to avoid sampling bias in the interpretation of the chemical environment over the high Antarctic plateau. We also examined wind distributions aloft between the periods 1961-1980 and 1990-2010 and found no systematic change except that related to the weakening of the tropopause and increased coupling of winds between 300-and 200-hPa.

645
We concluded that the delay in the breakup of the ozone hole in recent decades has allowed higher actinic fluxes to coincide with the optimum time in spring when boundary layer conditions are conducive to high NO concentrations.
We also examined NO data collected in 2006 confirming the systematic relationship between NO concentrations and wind direction shifts to easterly with lower temperatures, lower wind speeds, and shallow boundary layers. In comparing 2006-07 with other years, we found similar frequencies of easterly surface winds in 2003, fewer in 1998, and the lowest in 2000. We 650 carried out a case study from 2006 that occurred just prior to the final stratospheric warming. It was at this time that a minimum in cloudiness occurred, a minimum that is persistent in the long-term climatology over 50 years although subject to some synoptic "noise." Over a few hours in this case study, winds aloft rotated 180 o from northerly to southerly (in a grid sense), a mesoscale cool front moved from the northeast (observed in two AWS stations 100 km from the Pole and captured by ERA-I reanalysis) and skies cleared. With a clearing sky, a surface inversion formed below 10 m and NO climbed in 10-655 min averages of up to 1200 pptv with light surface winds from the southeast. NO>400pptv lasted only eight hours in this case with sustained moderate NO (150-400 pptv) lasting for another 48 hours in a sustained northeasterly flow (see Fig. 8).
This case also reveals the difficulty in correlating peaks in NO directly with upper-level winds insofar as the peak in NO occurs in the gap between rawinsonde observations. Perhaps more important are these transitions in upper level winds that signify a direct connection to large scale dynamical changes in the atmosphere over the plateau. In addition,

660
We argued that the source area for extreme NO concentrations was a shallow basin that extended about 100 km from the South Pole to the southeast. In this case we hypothesize that the short-term peak in NO occurred, not from transport in a long-range katabatic circulation but rather originated from local, subtle terrain accumulation areas in the vicinity of the South Pole. This was coupled with rapid radiative losses from the snow surface that created a strong surface inversion below 10 m.
In some respects, this situation is similar to the peaks in NO that occur at more northerly sites such as Concordia Station 665 where an evening peak in NO occurs with the collapse of the daytime convective boundary layer into a shallow stable boundary layer. In such cases long fetches are not necessary to achieve extremes in the concentrations of NO. However, in 2006, only 15% of the hourly NO data exceeded 400 pptv whereas 53% lay between 100 and 200 pptv. It is in this latter range that accumulation in long fetches associated with stable boundary layers and meso-and synoptic-scale circulations may be most relevant. However, without observations further away for the SP, we can't rule the possibility that high NO is first 670 generated in a larger area of light winds and shallow boundary layers upstream of the SP and then with increases in wind speed and boundary layer depth undergoes dilution. In addition, further away from the SP where convective mixing is possible during part of the diurnal cycle, further dilution is also possible. These results suggest further studies of chemical and boundary layer processes along pathways more complicated than predicted by katabatic flow arguments, particularly with the subtle topography to the east and southeast of the SP.

675
We developed a method to estimate boundary layer depth using 20-m meteorological tower data together with direct depth measurements in 2003 and tested against an independent data set collected in 1993.
Step-wise linear regression using wind speed and direction, temperature and the 2-to-20 m tower temperature difference produced regression equations that accounted for about 70% of the variance with direct observations. These equations were applied to all four seasons of NO data and confirmed the inverse depth relationships found in earlier studies. However, we also found significant year to year 680 differences in magnitude of these inverse relationships that suggested interannual variability in the surface fluxes of NO. To calculate NO x fluxes, the BLD data were used to obtain an estimate of the BL column NO x abundance (based on 2006 data where NO/NO x ~0.7). The abundance was derived from surface NO observations using a range of values of fall-off with height obtained from tethersonde profiles in 2003 and calculated using the increased lifetime of NO X for very shallow boundary layers. For each season, we found the average NO (scaled at 33.3 m from a least-squares fitting of NO versus 685 BLD -1 for each of four seasons) which followed a nearly linear relationship with the seasonally averaged fluxes of NO X confirming the consistency of our analysis. The source of the year-to-year variability in fluxes was not fully resolved but the variability was consistent with similar results at Concordia Station.

695
In the main sections of the paper we discussed the large-and meso-scale influences on local micrometeorology and chemistry that can set the stage for enhanced levels of NO. Past work has implicated shallow boundary layers as a key ingredient in high NO episodes (Davis et al. 2004a). These shallow boundary layers were associated with lower wind speed, changes in wind direction, colder temperatures and greater static stability as indicated by the temperature difference between 2 and 22 m on a nearby tower (Fig. 4). Although a shallow BLD was proposed as a major factor underlying high-NO episodes, the only 700 direct BLD measurements were those carried out in 2003 using a sodar in combination with surface turbulence measurements and balloon profiling . However, such supporting measurements were not available for other field seasons.
With this background, we examined regression analysis of NO and BLD against the various meteorological variables available to us for all four field seasons including the more recent data from 2006. This establishes some of the basic relationships between NO and BLD for the routine meteorological variables. However, because potential collinearity 705 between a number of the variables, we also pursued step-wise linear regression to find a subset of the variables that best account for NO and BLD for each observational period. This approach eliminates redundant variables. This approach then identified a simple subset of variables to develop prediction equations for BLD through multiple linear regression (MLR).
Establishing a relationship between NO and BLD, it then becomes possible to explore the unique chemical mechanisms which might be responsible for the high concentrations of NO at SP as compared with other sites and with the here-to-for 710 under prediction by chemical models Frey et al. 2015b).
The most routine hourly variables available at SP include wind speed (WS) and direction (WD), temperature (T), and temperature difference across a 20-m tower (ΔT). We also considered hourly direct solar irradiance (DR), net long-wave irradiance NIR, and solar zenith angle (SZA). (The variables DR, NIR, and SZA were available from www.esrl.noaa.gov/gmd.) Cloud fraction (CF) was available as a daily average from station climatological records whereas 715 bulk stability (ΔT B ) was obtained from the twice daily rawinsonde in summer. Sodar data gathered in 1993 (Neff 1994) provided an additional test data set with which to evaluate the effectiveness of the 2003 MLR results. These data were stored as facsimile records rather than digitally [unlike the 2003 data where automated processing was used ].
Also the sodar in 1993 was operating in a Doppler wind measuring mode so the pulse resolution was coarser (~30 m) so echo layers and hence BLD appeared to extend higher by one half the pulse length or 15 m. To correct for this effect 15 m was 720 subtracted from the BLD estimates.
Linear regression results are shown in Table A1 for NO in each of the four field seasons along with results for measured BLD given in Table A2 Fig. S4a. However, based on Fig. S4a, katabatic forcing will be small compared to that due to large scale weather systems.
During the later period, JD340-361, WS, WD, ΔT, and cloud fraction CF play bigger roles.  Fig. 9a, the simple linear regression between NO and DR in Table I does not capture the threshold effect evident in the figure. (Cases with high DR in Fig. 9a and low NO usually occur under high wind conditions with deep boundary 735 layers.) The dependence of BLD on tower-measured ΔT is greater than on the bulk inversion temperature difference ΔT B which is typically measured over several hundred meters: This is consistent with the fact that the BLD is usually much shallower than the background temperature inversion and more consistent with the ΔT measured on the 22-m tower. When ΔT is dominant over ΔT B it indicates that the shallow temperature inversion is likely due to surface radiative cooling (with clearing skies) rather than synoptic weather changes. Comparison of these correlations with those from the more extended 740 data in 1993 sheds some insight into the appropriate variables to use in our MLR analyses (Table III).
From obtained over average depths of 542m, 579m, and 622m for each month respectively. This demonstrates the strong seasonal cycle in inversion strength and the changing seasonal impact on katabatic forcing of surface winds (see Fig. S2a). In December the near-surface (2-22m) inversion is much stronger than the bulk value from the rawinsonde suggesting that any slope effects will reflect nearby terrain variations such as those shown in Fig. 5. It should be noted that the effect of WD on 750 BLD is greatest in the transition month of November. With a weaker inversion strength in the summer, those winds (December) respond mostly to meso-and synoptic scale weather influences such as cloudy versus clear skies and/or high versus low wind speeds (Neff 1999) rather than katabatic forcing.

A.1 Linear regression
Because of the potential covariance of meteorological variables (e.g., often when the wind direction shifts to the SE, 755 wind speed, ΔT and temperature all decrease: see Fig. 6), we used stepwise linear regression (using SPSS) of NO and BLD with principal meteorological data as summarized in Table A4. The value of stepwise regression is that it orders the results by magnitude of the importance of variables to the regression. In Table A4, step-wise regression reveals year-to-year variability in important variables depending on the year-to-year variability of dominant meteorological regimes. It should also be noted that the step-wise regression process can produce a different sequence of regression coefficients than those 760 obtained from individual regressions as can be seen in comparing Table A1 and  fits with r 2 =0.96. For 1993, the data appear to have two modes perhaps because of a weather regime shift not captured by the MLR method using just a few meteorological variables (e.g., there was a major change in bulk inversion strength ΔT B around mid-December). In this case the two equations still agree quite well with r 2 =0.93.

A.2 Results
We have extended the past analysis of boundary-layer depth ( Stepwise linear regression showed that the principal variables affecting both NO and BLD were wind speed and direction, temperature, and low-level static stability as measured on a 22 m tower. We then used these variables to develop a multiple linear regression equation for BLD using sodar data from 2003 and simple meteorological tower measurements. When we tested the regression equation derived from 2003 data against independent sodar data collected in 1993, we accounted for 60-800 70 percent of the variance in estimating BLD. We then applied these equations to each year of NO data to examine the NO response to BLD. When the data were binned in 100 pptv bins (Fig. 10), we found a linear fit between NO and 1/BLD with regression fits accounting for 88% to 97% of the variance confirming past results. When we looked at the binned data for Tables  915   Table X  6.8 x 10 9 4.7 x 10 9 6.1 x 10 3 2.9 x 10 13 3.2 x 10 4 9.06 x 10 8 2003 11/22-12/27 10 x 10 9 6.5 x 10 9 8.1 x10 3 5.2 x10 13 3.8 x 10 4 1.37 x 10 9 2000 11/15-12/31 3.2 x 10 9 2.1 x 10 9 8.3 x 10 3 1.7 x 10 13 2.7 x 10 4 6.30 x 10 8 1998 12/1-12/31 7.8 x 10 9 5.4 x 10 9 7.5 x 10 3 4 x 10 13 3.2 x 10 4 1.25  removed.    primarily from the direction of West Antarctica and the Amundsen Sea (~270 o ). In late spring, 300-hPa winds become bimodal in direction, with higher probabilities from ~157.5 o and ~337.5 o and fewer from the west. Past work has indicated that winds from 157.5 o bring cooler temperatures and lighter wind speeds to the South Pole: this direction corresponds to an inland directed synoptic scale pressure gradient. Conversely, 300-hPa winds centered on 337.5 o are typically associated with higher surface 960 winds, increased cloudiness, and warmer temperatures: this direction reflects an off-shore directed synoptic pressure gradient. In early summer, winds at 300 hPa increasingly favor a north-northwest direction from the Weddell Sea.  . Data gaps at the end of December are due to either missing NO and/or meteorological data. Shaded areas and labels A-H are cases where NO >300 pptv. Case "C" is that highlighted in Fig. 4 which occurs at the time of stratospheric warming and total column ozone 1010 increase. The wind azimuth of 60 o is also noted (corresponding to Fig. 3 for directions associated with higher NO). Periods of clear skies are indicated on an hourly basis by hash marks at -20 C on the temperature plot: Note the increase in temperature around JD 320 associated with cloudy skies, increase in wind speed, and shift in wind direction.  Figure 6). In (b) we indicate wind speed and direction at 300 hPa (red diamonds) at the actual launch times of the rawinsondes (2-3 hours prior to reporting times of 00Z and 12Z). Of note is the decrease in surface wind speeds from over 8 ms -1 (sufficient for blowing snow) that follow the same trend as at 300 hPa. We have also indicated the BLD 1020 calculated later in Section 4.3 that shows the rapid decrease in BLD. and anticyclonic winds over the high plateau. Lows and highs are indicated and the location of a small cyclonic circulation at 6-hr intervals on JD337 that brought easterly winds to SP. By 12Z on JD 338, the low to the NNE had intensified and produced a consistent NE winds to SP