New Estimation of the NOx Snow‐Source on the Antarctic Plateau

To fully decipher the role of nitrate photolysis on the atmospheric oxidative capacity in snow‐covered regions, NOx flux must be determined with more precision than existing estimates. Here, we introduce a method based on dynamic flux chamber measurements for evaluating the NOx production by photolysis of snowpack nitrate in Antarctica. Flux chamber experiments were conducted for the first time in Antarctica, at the French‐Italian station Concordia, Dome C (75°06'S, 123°20’E, 3233 m a.s.l) during the 2019–2020 summer campaign. Measurements were gathered with several snow samples of different ages ranging from newly formed drifted snow to 6‐year‐old firn. Contrary to existing literature expectations, the daily average photolysis rate coefficient, JNO3¯ , did not significantly vary between differently aged snow samples, suggesting that the photolabile nitrate in snow behaves as a single‐family source with common photochemical properties, where a JNO3¯ = (2.37 ± 0.35) × 10−8 s−1 (1 σ ) has been calculated from December 10th 2019 to January 7th 2020. At Dome C summer daily average NOx flux, FNOx , based on measured NOx production rates was estimated to be (4.3 ± 1.2) × 108 molecules cm−2 s−1, which is 1.5–7 times less than the net NOx flux observed previously above snow at Dome C using the gradient flux method. Using these results, we extrapolated an annual continental snow sourced NOx budget of 0.017 ± 0.003 Tg · N y−1, ∼ 2 times the nitrogen budget, (N‐budget), of the stratospheric denitrification previously estimated for Antarctica. These quantifications of nitrate photolysis using flux chamber experiments provide a road‐map toward a new parameterization of the σNO3−(λ,T)ϕ(T,pH) product that can improve future global and regional models of atmospheric chemistry.

developed a nitrate air-snow exchange model and tested it against the summer observations at Dome C, which demonstrated that co-condensation, that is, the simultaneous condensation of water vapor and trace gases at the air-ice interface, was the most important process to explain nitrate incorporation in snow. Bock et al.'S (2016) model works well at cold sites on the Antarctic Plateau, where air temperatures are well below freezing and in fact below the eutectic point year-round, and where no snow melt occurs. However, the model does not reproduce the summer observations at the coast, where the temperature, relative humidity and concentration of aerosol in air and snow are much higher than that on the Plateau, and where snow surface melt is possible, strengthening previous theory of several photolytic domains aforementioned. Nonetheless, Chan et al. (2018) developed a new model and concluded that winter air-snow interactions of nitrate between the air and skin layer snow can be described as a combination of non-equilibrium surface adsorption and co-condensation on ice, coupled with solid-state diffusion inside the grain, similar to Bock et al. (2016). In addition, Chan et al. (2018) were able for the first time to reproduce the summer observations on the Antarctic Plateau and at the Coast, concluding that it is the equilibrium solvation into liquid micro-pockets, based on Henry's solubility law, that dominates the exchange of nitrate between air and snow at warmer sites, that is, where the temperatures are above the eutectic temperature.
To date, no consensus can be found in the literature about the different forms of nitrate that would allow us to reduce the modeled NO x E F uncertainties. Models usually use the quantum yield of 0.003 molecules 1 photon E found by Chu and Anastasio (2003), therefore often under-predict the observations, for example, at night time at Dome C . A better parameterization of nitrate photolysis quantum yield as a function of snow micro-physical properties and nitrate form (ion or 3 HNO E molecule) is needed to improve the models.
A major difficulty encountered for estimating NO x E F is that it depends upon two components: transport (both inward and outward to the snowpack) and photochemical production/loss. Previous field studies did not separate   Ca. 8.7 cm per Year (Picard et al., 2019) NO E sensitivity of the instrument (Section 2.3).
The snow samples contained in the scufa box was homogenized again and weighed once transferred in the chamber to obtain the density by gravimetry. 8.5-12.0 kg of snow were used to fill up the chamber. The densities calculated for each experiment ranging from 0.32 to 0.45 g 3 cm E agreed well with previous observations (Gallet et al., 2011).
A mechanical scale (TERAILLON Nautic) installed on a stable surface and leveled was used for the weighing of the snow before and after each experiment. A mean average loss of 1.1% E was measured representing about  E 100 g of snow sample. The losses observed are not significant in our case for the need of weight corrections.
A total of eight snow samples ( E 25 3 cm E each) were taken randomly in the chamber before the FC was sealed, for each experiment. The chamber was then buried in the snowpack, taking care to disturb the surrounding snow as little as possible, and connected to the zero-air flow. The production from the snow was then monitored continuously after the setup of the FC, for 2-4 days.
To investigate the possibility of nitrate stratification and denitrification during the experiments, samples ( E 25 3 cm E each) at 2 cm snow depth resolution were collected from the box after each experiment (4 samples for each 2 cm layer were taken resulting in 40 samples).

Data Processing
The NO x E measurements from the IBBCEAS were corrected for the chamber blank (0.068  E 0.012 nmol 1 mol E ) before being further processed for validation. A 2h-running mean was calculated and the standard deviation ( mean E ) was determined within the same window. The data falling beyond 2 ×  mean E were discarded, which resulted in less than 6% E rejection. Clouds passing by would impact NO x E and would not be detected as outliers. However, the radiative conditions were the same from 1 day to another during the FC experiments. The measurements were then averaged every 20 min, corresponding to the best performance of the instruments to achieve the ultimate NO x E detection limit of 30 pmol 1 mol E (Barbero et al., 2020).
As mentioned in Section 2.3, the 3 O E inside the chamber was monitored using a UV Photometric 3 O E analyzer (Thermo Scientific, Model 49i) connected by tubing to an automatic snow tower platform which housed the switching manifold and the analytical equipment. To eliminate the response time after the switching manifold, only the last three minutes of measurements, when concentrations reach steady-state, were used and averaged, giving one measurement of ozone concentration every 2 h. The data were then interpolated linearly every 20 min to match the resolution of the NO x E measurements. (details of the calculations can be found in Appendix D). Due to this insignificance, we concluded that the potential impact from ozone on our results was negligible, therefore, we decided that there was no need to discuss the ozone data. Figure 4 shows the nitrate concentration in ng 1 g E of snow measured in each snow sample before and after the experiments. Drifted snow was found to be up to 30 times more enriched in nitrate than the mean of all the rest of the snow samples, likely due to being blown and stirred by the strong wind (dark blue colors, Figure 4). The pristine snow from the 25 km site shows a rapid decrease in nitrate concentrations with depth (red colors, Figure 4), with less than 50 ng 1 g E of nitrate remaining in the 30-40 cm layer, while local snow shows fairly uniform concentrations over the topmost 50 cm (green colors, Figure 4).
The bottom layers at the two sampling sites do not match in depths due to a sampling error. The exponential decrease of nitrate concentration from the surface to depth is a common feature on the Antarctic plateau, determined by photolysis denitrification occurring in the photic zone. And indeed, the concentrations in the 25 km South snow samples are in agreement with previous observations taken at Dome C shortly after Concordia station construction (Erbland et al., 2013;France et al., 2011).
A part from the top layer sample, the local snow samples appeared enriched in nitrate compare to the pristine snow samples, phenomena probably due to site pollution as demonstrated by Helmig et al. (2020). Additionally, the local sampling area is known to have been very locally contaminated during the previous summer campaign of 2018-2019 by technical activity as shown on Figure F1 given in Appendix F. Therefore, the results given by the local 2-7 cm layer sample, corresponding to a snow aged from the previous summer campaign (Table 1), will be considered with caution.As explained in Section 2.5, four profiles were taken in the flux chamber after each experiment with a 2-cm resolution to observe a potential stratification that would have occurred during the experiments. Samples were analyzed twice with the IC system and the results are shown in Figure 5. The variability observed with depth for the four different profiles of each snow sample is lower than the standard deviation of the measurements. It is reasonable to state that no apparent stratification occurred during the experiments, thus sensible to conclude that our nitrate reservoir is constant.A strong diurnal variability in the NO x E production from the snow is observed for each experiment (Figure 6), with a minimum around midnight and a maximum around local noon, following the daily UV radiation cycle. An apparent proportionality is observed between the nitrate concentrations contained in the snow samples and the amplitude of NO x E mixing ratios produced as a snow with 10 times the concentration of nitrate seems to produce 10 times more NO x E ( Figure 6). Second, an exponential decrease in NO x E mixing ratio during the experiments is also observed, potentially due the reservoir denitrification. Assuming one photolyzed molecule of  3 NO E produces one molecule of NO x E (reactions R1 to R5) the denitrification occurring during the experiments represents on average 0.12  E 0.08% E of the initial amount of nitrate, therefore negligible with respect to the initial nitrate reservoir (more details are provided in Appendix E).
Two main patterns of temporal variations are observed for all experiments in Figure 6: an oscillation driven by the diurnal cycle of UV radiation, called steady-state regime in the following sections, to which is superimposed a slower exponentially decreasing trend with a maximum of NO x E mixing ratio on the first day, called the transitory regime. Below, both regimes are discussed separately. For the stationary regime, a comparison between the different experiments is made with the aim of better characterizing the nitrate photolysis. In a of snow] at the two sampling sites before and after the FC experiments. In blue colors, the drifted snow, in red colors, the pristine snow, that is, 25 km South, and in green colors, the local snow. The error bars correspond to one standard deviation,  E 1 E , over samples measurements: before experiment = average of 8 samples analyzed twice, and after experiment = average of 40 samples analyzed twice, both after data processing explained in Section 3. second part, the exponential decrease is studied within the same experiment in an attempt to explain the transitory regime.

Steady-State Regime Study
Flux chambers are a useful tool for directly measuring production rates of a defined sample. Combining their results with other observations could lead to the estimation of a flux area.
] on the chamber depth for each experiment. Samples were collected with a 2-cm resolution and analyzed twice using the IC system. Blue, red, green and orange colors correspond to four different profiles collected at the end of each experiment, and black colors to the average profile. Error bars correspond to the standard deviation, 1 E , observed for each sample repetition analysis.  (Jones et al., 2000), are reported in Table 2.The NO x E production of the FC experiments is proportional to the nitrate concentration initially present in the snow samples as shown in Table 2, where linear regression for the pristine snow shows an 2 R E = 0.691; and, excluding the local 2-7 cm layer, the linear regression for the local snow shows an 2 R E = 0.999 (0.659 without the drift sample), note that the linear regressions were calculated with their intercepts forced at zero. This is consistent with previous finding (e.g., Grannas et al., 2007). The local snow sample at 2-7 cm depth exhibits a lower NO x E production, and we strongly suspect that this sample contained absorbing contaminants emitted by the station activities (more details can be found in Appendix F). The NO x E production of snow with similar nitrate concentrations, calculated over one day, is very similar for different snow sources (Figure 7). It can be observed that for two different types of snow of the Antarctic Plateau, the NO x E production is very similar (green triangles and red squares, Figure 7). However, such similarity in production does not exist when it is compared with the Neumayer experiment (Jones et al., 2000), which was located on the western coast of Antarctica but at similar latitude (70 E S) as Concordia station (75 E S) (solid blue circles, Figure 7). Additionally, the condi- Figure 6. NO   340 nm) for the measurement period (10/12/2019 to 07/01/2020), which is three times lower than the value usually used in the models. This contradicts previous studies of Davis et al. (2008) and Meusinger et al. (2014) which proposed two nitrate domains: an easy photolabile nitrate fraction (i.e., adsorbed on the surface of ice) and a more difficult to successfully photolyze fraction (i.e., incorporated within the ice crystal lattice). However, our observations are consistent with Bock et al. (2016) and Chan et al. (2018) which proposed a single mechanism responsible for the incorporation of nitrate in the snow at cold sites such as Dome C.

Transitory Regime Study
The transitory regime represents the decreasing exponential trend observed over the few days of the FC experiments, with a maximum of NO x E gas phase mixing ratios the first day observed in Figure 6. NO x E productions decrease from one day to the next with no special pattern observed between the experiments: for the drifted snow The total NO x E production (  NO x tot E P ) observed in Figure 6 can be expressed as the sum of the steady state regime production (  NO x steady E P ) and the transitory regime production (  NO x trans E P ), Equation 6. Although the hypothesis of two types of nitrate was disregarded in Section 4.2 because it is not compatible with our findings, where and older snow presents similar photolysis rate coefficient as a younger snow, we still want to question if this transient regime could be produced by a minor family of nitrate that could be depleted more rapidly than the bulk nitrate. Therefore, the hypothesis of two nitrate populations is explored with It is possible that a thin surface layer is denitrified much faster due to a phenomenon of amplification of the photolysis in this layer, (Traversi et al., 2017), but that was not captured with a coarse sampling resolution of 2 cm. Simpson et al. (2002) modeled the ratio of the in-snow actinic flux to the incident down-welling actinic flux as a function of extinction optical depth (proportional to depth for homogeneous snow) within the snowpack. Snow increases the actinic flux within the topmost layers of the snowpack, due to the high albedo in the UV region.
The actinic flux at the snow surface is equal to the atmospheric actinic flux above the snow, whereas in the first few mm, depending on SZA, the actinic flux is either increased or decreased. For SZA =  0 E , the actinic flux is enhanced by the conversion of direct light to diffuse light (Chan et al., 2015). Madronich (1987) argued that the maximum enhancement factor is four times higher than the incident actinic flux. Using the TARTES optical radiative transfer model, relative actinic fluxes for each snow samples at 305 nm were simulated (Figure 9), using specific surface area (SSA) as a function of snow depth following Gallet et al. (2011). No significant differences between the up-welling and the down-welling irradiance are observed (solid orange curve and dotted dashed green curve, Figure 9), making the travel of the light nearly isotropic within the snowpack (i.e., the same properties in every direction), even near the surface.
The actinic flux is thus relatively smooth and ranges between 1.2 and 2.8 times the flux received by the snowpack at 305 nm for the 0-20 cm layer. This variation range is far too small to explain a localized layer with highly photolabile nitrate that would be depleted over the three days of the experiment given the initial abundance of nitrate in all the samples.

NO E
) would allow to characterize the sample and constrain porosity. However, here we report for the first time the F NO x estimate using our FC approach at Dome C Table 3, and the results from our approach and the flux gradient method  fall within the same order of magnitude, confirming the usefulness of the FC approach. A combined experiment of in-snow dynamic FC, as described in this work, with surface-snow dynamic FC, as described in Figure 1    L the length of the reference day (more details on the calculation are found in Appendix H). In addition, the variability of the SZA, associated to the maximum amplitude of the NO x E production, should be considered in the extrapolation and it is represented here by a cosine ratio of the solar noon sun's position for the day of interest over the one of reference. This calculation is deduced by Equation 11 hereafter (Finlayson-Pitts & Pitts, 2000;Forsythe et al., 1995). ) (Ehhalt et al., 2018). Nevertheless, this comparison on a continental scale with a nitrogen source of the Antarctic atmosphere, such as the stratosphere denitrification, is more useful to understand its significance.
Indeed, the evidence for denitrification in the Antarctic stratosphere is well established since the 90s (Deshler et al., 1991;Fahey et al., 1990;Mulvaney & Wolff, 1993;Salawitch et al., 1989;Santee et al., 1995;Van Allen et al., 1995). Savarino et al. (2007), estimated a N-budget of (0.035 coming from the stratosphere denitrification, only 2 times less the N-budget of the snowpack denitrification estimated here, making the snowpack source a rather substantial source. However, McCabe et al. (2007), by studying the oxygen nitrate isotopic composition, suggested that nitrate was the result of a mixture of 25% E stratospheric and 75% E tropospheric origin, making the total nitrate source around 0.032 Tg· . Therefore, the recycled nitrogen from the snowpack nitrate photolysis would represent approximately half of the nitrogen atmospheric input. Furthermore, Erbland et al. (2015) simulated that  E 20% E of the total nitrate mass fluxes, that is, stratospheric and tropospheric inputs as well as photolytic recycling, was exported.
Two scenarios are conceivable: (i) all the NO x E produced by the snowpack denitrification are re-oxidized, making the balance null and closing the nitrate budget in Antarctica; (ii) photolysis destruction of snow nitrate is potentially an important denitrification factor of the Antarctic cryosphere. That said, note that this extrapolation to the continent is a first order estimate as it does not consider the differences on the E e -folding depth and the quantum yield variations with temperature and pH, especially at the coast; neither the latitude's differences modifying the ratios used in Equation 11. Indeed, Noro and Takenaka (2020) recently showed that at a coastal site called H128 34'E) that is located approximately 100 km away from the Japanese Syowa Station in East Antarctica, 50% E of the nitrate on surface snow is lost by photolysis. Additionally, they suggest that a photic zone of 45 cm m E ) were typical of the summer climatology observed at Dome C. Two episodes of strong northeast wind (up to 10 m 1 s E ) occurred     Appendix H: Length of Days at Dome C Figure H1. Daylength calculation was made using a Center for Biosystems Modeling (CBM) model (Forsythe et al., 1995). The model estimates daylength with error less than one minute within 40 degrees of the equator and less than seven minutes within 60°, described by Equations H1, H2 and H3:     0.2166108 2arctan(0.9671396 tan(0.00860( 186))) J (H1)    arcsin(0.39795cos( )) (H2) sun with respect to the horizon, chosen for this calculation is: sunrise/sunset is when the top of the sun is apparently even with horizon, giving E p = 0.8333. Figure H1 represents the length of day for one year at Dome C.