Inclusion of mountain wave-induced cooling for the formation of PSCs over the Antarctic Peninsula in a chemistry–climate model

An important source of polar stratospheric clouds (PSCs), which play a crucial role in controlling polar stratospheric ozone depletion, is from the temperature ﬂuctuations induced by mountain waves. However, this formation mechanism is usually missing in chemistry–climate models because these temperature ﬂuctuations are neither resolved 5 nor parameterised. Here, we investigate the representation of stratospheric mountain wave-induced temperature ﬂuctuations by the UK Met O ﬃ ce Uniﬁed Model (UM) at high and low spatial resolution against Atmospheric Infrared Sounder satellite observations for three case studies over the Antarctic Peninsula. At a high horizontal resolution (4 km) the mesoscale conﬁguration of the UM correctly simulates the magni- 10 tude, timing, and location of the measured temperature ﬂuctuations. By comparison, at a low horizontal resolution (2.5 ◦ × 3.75 ◦ ) the climate conﬁguration fails to resolve such disturbances. However, it is demonstrated that the temperature ﬂuctuations computed by a mountain wave parameterisation scheme inserted into the climate conﬁguration (which computes the temperature ﬂuctuations due to unresolved mountain waves) are 15 in excellent agreement with the mesoscale conﬁguration responses. The parameterisation was subsequently used to compute the local mountain wave-induced cooling phases in the chemistry–climate conﬁguration of the UM. This increased stratospheric cooling was passed to the PSC scheme of the chemistry–climate model, and caused a 30–50 % increase in PSC surface area density over study demonstrates that: (i) UM high-resolution (4 km) mesoscale model simulations are able to accurately simulate the large mountain wave-induced temperature 25 ﬂuctuations in the lower stratosphere associated with strong westerly or north-westerly ﬂow over the Antarctic Peninsula, and (ii) UM low-resolution (2.5 ◦ × 3.75 ◦ ) climate model

nor parameterised. Here, we investigate the representation of stratospheric mountain wave-induced temperature fluctuations by the UK Met Office Unified Model (UM) at high and low spatial resolution against Atmospheric Infrared Sounder satellite observations for three case studies over the Antarctic Peninsula. At a high horizontal resolution (4 km) the mesoscale configuration of the UM correctly simulates the magnitude, timing, and location of the measured temperature fluctuations. By comparison, at a low horizontal resolution (2.5 • × 3.75 • ) the climate configuration fails to resolve such disturbances. However, it is demonstrated that the temperature fluctuations computed by a mountain wave parameterisation scheme inserted into the climate configuration (which computes the temperature fluctuations due to unresolved mountain waves) are 15 in excellent agreement with the mesoscale configuration responses. The parameterisation was subsequently used to compute the local mountain wave-induced cooling phases in the chemistry-climate configuration of the UM. This increased stratospheric cooling was passed to the PSC scheme of the chemistry-climate model, and caused a 30-50 % increase in PSC surface area density over the Antarctic Peninsula com-

Introduction
Gravity waves generated by stratified flow passing over orography (mountain waves) that propagate into the stratosphere play a crucial role in the formation of polar stratospheric clouds (PSCs). Adiabatic temperature changes resulting from mountain wave-important source (Dörnbrack et al., 2001;Alexander et al., 2013). Moreover, mountain waves are a significant source of PSCs on the synoptic-scale in both the Arctic and Antarctic due to their advection far downstream of the wave event that formed them (Carslaw et al., 1999;Höpfner et al., 2006;Eckermann et al., 2009;Alexander et al., 2011). 25 The role of PSC particles in polar ozone chemistry is generally well understood. The conversion of passive species of chlorine and bromine to active forms that destroy stratospheric ozone in sunlight takes place on the surface of PSCs (Solomon, 1999). Furthermore, the removal of nitric acid from the atmosphere by sedimentation of PSC Introduction particles leads to a slower conversion of active chlorine back into its passive species (Jensen et al., 2002). The Antarctic ozone hole has profound impacts on the Southern Hemisphere circulation and surface climate during summer (e.g. Orr et al., 2008Orr et al., , 2012Thompson et al., 2011). With the continued implementation of the Montreal Protocol, recovery of the Antarctic ozone hole is generally anticipated by the end of the 5 century. However, model predictions using coupled chemistry-climate simulations give a large range of estimates of the rate and timing of this recovery (Eyring et al., 2013). The fact that the results are model dependent suggests that some mechanisms are not yet fully understood. Similarly, simulations of the ozone hole covering the past few decades obtain a wide range of results, further questioning the value of these predic- 10 tions (Austin et al., 2010). Accurate predictions of the timing are critical as this recovery will reshape Southern Hemisphere climate by no-longer counteracting the effects of increasing greenhouse gases (Polvani et al., 2011). Therefore, to produce accurate simulations of stratospheric ozone depletion, coupled chemistry-climate models must be able to represent PSC formation mechanisms 15 and their attendant ozone-loss chemistry due to localised dynamics such as mountain waves (Cariolle et al., 1989;Carslaw et al., 1998b;Austin et al., 2010). However, current chemistry-climate models have a spatial resolution of some hundreds of kilometres (e.g. Morgenstern et al., 2010) at the equator and are therefore only able to explicitly resolve waves with long horizontal wavelengths (Reinecke and Durran, 2009), i.e. 20 the temperature fluctuations associated with small-scale mountain waves are missing, leading to insufficient PSC formation in the models. Consistent with this is the systematic over-prediction of high-latitude spring time ozone increases in both hemispheres by models (Carslaw et al., 1998b;Eyring et al., 2006).
Mountain wave-induced stratospheric temperature fluctuations can be detected by 25 their associated fluctuations in temperature-sensitive satellite radiance measurements from infrared scanning instruments such as the Atmospheric Infrared Sounder (AIRS) (e.g. Alexander and Barnet, 2007;Hoffmann et al., 2013). As a nadir viewing instrument, AIRS radiance measurements have a high horizontal resolution (14 km at nadir), Introduction enabling waves with short horizontal scales which are unresolved by chemistry-climate models to be visible. On the other hand, AIRS radiance measurements have a limited vertical resolution, meaning waves with short (typically ≤ 10 km) vertical scales are poorly resolved. Comparison between AIRS radiance measurements and model simulated radiance measurements (calculated using the simulated temperature field of the 5 model as input for a radiative transfer model) provides an effective and direct means of validation of the model representation of gravity wave events (Grimsdell et al., 2010).
To improve the simulation of mountain wave-induced PSCs in a chemistry-climate model, the temperature fluctuations due to unresolved (sub-grid scale) mountain waves can be parameterised (e.g. Carslaw et al., 1999;Dean et al., 2007;Wells et al., 2011). 10 In the parameterisation scheme of Dean et al. (2007)  is used to improve the simulation of mountain wave-induced PSCs in the UK Chemistry and Aerosol (UKCA) module, which is the chemistry-climate configuration of the UM (Sect. 5). Our approach firstly requires demonstration of the ability of the parameterisation to reasonably simulate stratospheric temperature fluctuations. This is achieved by using case studies of AIRS measurements to validate high-resolution simulations (us-Introduction  al. (2007) scheme is inserted into the climate configuration of the UM and its temperature fluctuations are assessed by comparing with output from the high-resolution simulations (Sect. 4). We will demonstrate below that the high-resolution simulations are in excellent agreement with the AIRS observations, and can therefore be used as a "truth" with which to investigate the performance of the parameterisation scheme. The 5 paper finishes with a summary and discussion (Sect. 6).
2 Models, mountain wave parameterisation, data, and methodology

Models
The UM is a numerical modelling system based on

Description of the mountain wave parameterisation
By assuming that waves are forced by steady flow over a two-dimensional ridge and 15 that vertical variations of the background atmospheric state are slowly varying (compared to the wave phase), the scheme described by Dean et al. (2007) derives generalised expressions for the maximum and minimum vertical streamline displacement (resulting in cooling and warming, respectively) associated with gravity waves induced by sub-grid scale orography (SSO). These expressions are used to compute the maxi-20 mum negative ∆T − SSO and positive ∆T + SSO temperature fluctuations associated with the displacement, which are derived using the local potential temperature gradient (Wells et al., 2011). The overall temperature fluctuation induced is subsequently calculated as ∆T SSO = ∆T + SSO + ∆T − SSO . Waves are launched at every model grid box over land and at every model time step. 25 The expressions for the maximum and minimum streamline displacement depend on both the wave phase and peak vertical streamline displacement amplitude (hereafter referred to as wave amplitude), which are determined as follows. gation is based on linear theory for hydrostatic waves forced by steady, stably stratified flow over a two-dimensional ridge, assuming that vertical variations of the background atmospheric state are slowly varying. McFarlane (1987) showed that under these circumstances and in the absence of dissipation mechanisms that the vertical evolution of the wave amplitude is determined by the decrease in density of the atmosphere 5 with height and by changes in the horizontal wind speed U (resolved in the direction of the wave vector) and the Brunt-Väisälä frequency N. Dissipation mechanisms such as wave-breaking and critical level absorption are introduced by preventing the amplitude from exceeding the local "saturation amplitude" for which the wave field becomes unstable (= U/NF sat , where F sat is the critical Froude number for saturation). The verti-10 cal evolution of the wave phase is determined by changes in U and N, i.e. the Scorer parameter l (≈ N/U).
To complete the determination of the wave phase and amplitude, their initial values at the top of the blocked layer must be decided. The initial wave phase is set equal to zero. The initial wave amplitude is set equal to the "effective" mountain height h eff (= h − h b , 15 where h is the height of the sub-grid scale mountain and h b is the height of the blocked layer that occurs at low Froude number), i.e. the maximum vertical displacement of streamlines able to pass over the top of the mountain. This is strongly dependent on the direction of the low-level wind relative to the principle axis of the SSO (which preferentially aligns as ridges), and ensures that the surface amplitude is large (small) when 20 the wind is perpendicular (parallel) to a ridge. Here, h = n σ σ, where σ is the standard deviation of the SSO height from the grid-box mean and n σ is a constant (such that n σ σ approximates the physical envelope of the peaks), and where F c is the critical Froude number at which flow blocking is deemed to first occur, and the subscript "0" refers to the surface layer, represented by averaging U and N be-Introduction  Dean et al. (2007) which represents the directional dependence by defining the standard deviation σ of the SSO height in the surface wind direction. The parameterisation scheme is implemented in the climate and chemistry-climate configurations of the UM. The SSO parameters used by the scheme are based on Lott and Miller (1997). In the scheme the parameters n σ , F sat and F c are treated as tuneable. 5 Following an initial sensitivity study to optimise the performance of the scheme (not shown), their values were set to n σ = 3, F sat = 2 and F c = 4.

Data
AIRS (Aumann et al., 2003) is aboard the National Aeronautics and Space Administration's Aqua satellite, which was launched in May 2002. AIRS measures the thermal 10 emissions of atmospheric constituents in the nadir and sub-limb observation geometry. An across-track scan consists of 90 individual footprints and covers a distance of 1765 km on the ground. The along-track distance between two scans is 18 km. The AIRS aperture is 1.1 • , corresponding to a spatial resolution of 13.5 km at nadir and 41 km × 21.4 km at the scan extremes. The AIRS radiance measurements cover 15 wavelength ranges from 3.74 to 15.4 µm with a total of 2378 radiance channels. The absolute error of the radiometric calibration is less than 0.2 %. The noise equivalent delta temperature is about 0.39 K at 250 K scene temperature for the spectral channel (666.5 cm −1 ) considered here. The analyses presented in this paper are based on consolidated version 5 data products made freely available by NASA. The equatorial  Infrared radiance measurements in the 4.3 and 15 µm CO 2 bands are of particular interest for the study of stratospheric gravity waves. These bands get optically thick in the stratosphere and as the CO 2 concentration does not vary substantially in the 18285 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | lower and middle atmosphere, the radiance measurements in these channels are most sensitive to atmospheric temperature. Hoffmann and Alexander (2009) show the temperature kernel functions for the individual AIRS channels covering the 4.3 and 15 µm CO 2 bands. In this study we selected the 666.5 cm −1 radiance channel of AIRS, which is within the 15 µm CO 2 band. The temperature weighting function of this channel is 5 given in Fig. 1, which shows that the brightness temperatures BT are most sensitive to atmospheric temperature at an altitude of 22 km, with full width at half maximum of 9 km. The altitude range covered by the 666.5 cm −1 channel is of particular interest for the formation of PSCs. As the kernel function drops to less than 1 % of maximum sensitivity below 14 km, there is little interference from tropospheric emissions from clouds 10 or water vapour.

Methodology
Three instances of stratospheric mountain waves observed over the Antarctic Peninsula by AIRS, characterised by large amplitude and long vertical wavelength, occurred on 7 August 2011 at 03:40 UTC (case study 1, hereafter CS1), 2 August 2010 at 15 18:59 UTC (case study 2, hereafter CS2), and 14 July 2010 at 20:00 UTC (case study 3, hereafter CS3). These events were simulated by running the mesoscale model for a 48 h period driven by output from global model forecasts (following a 3 h spin-up) initialised on 5 August 2011 at 12:00 UTC for CS1, 1 August 2010 at 00:00 UTC for CS2, and 13 July 2010 at 00:00 UTC for CS3. The mesoscale model output times (integra-  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | cident with strong westerly or north-westerly winds incident to the Peninsula. These winds showed the requisite strengthening with height required for the mountain waves to have long vertical wavelengths which were visible to AIRS (not shown).
To verify the mesoscale model forecasts, the Juelich Rapid Spectral Simulation Code (JURASSIC) radiative transfer model (Hoffmann and Alexander, 2009) was used to compute model simulated AIRS radiances at 666.5 cm −1 . For comparison, both the real and mesoscale model simulated AIRS radiances are subsequently converted into their corresponding BT values. Brightness temperature perturbations ∆BT were computed by removing a background brightness temperature, which was determined by fitting a 4th-order polynomial (e.g. Wu, 2004;Alexander and Barnet, 2007;Hoffmann 10 and Alexander, 2010). This fit removes slowly varying atmospheric signals, e.g. from planetary waves and general scan-angle dependence of radiances due to the sub-limb geometry. For the AIRS measured radiances the fit was carried out for each scan in the across-track direction; for the mesoscale model simulated radiances it was carried out for each latitudinal band of the model grid. In both cases it was found that the 15 fits are well constrained by the data and the process did not introduce any artificial wave-like structures that could obfuscate the results. In order to avoid the suppression of waves with fronts parallel to the fit direction the AIRS measured (simulated) background estimates were smoothed by a 300 km running mean in the along-track (longitudinal) direction. Finally, mesoscale model estimates of ∆BT are re-gridded to 20 the AIRS measurement grid.
The mesoscale model forecasts were repeated using the climate model (i.e. the climate model is initialised using the same Met Office operational analysis and integrated forward in time for 48 h). Comparison of the mesoscale model and climate model simulations of the near-surface winds at the time of the mountain wave events (Fig. 2) 25 shows relatively small differences in the large-scale flow impacting the Peninsula, i.e. the large-scale atmospheric conditions responsible for the initial forcing of the mountain waves are broadly similar in both models. As the mountain wave parameterisation scheme is implemented in the climate model, the temperature fluctuations predicted by ACPD 14,2014 Inclusion of mountain wave-induced cooling the scheme ∆T SSO , as well as the temperature fluctuations explicitly resolved by the climate model ∆T CLIM (computed by removing the background temperature, determined by fitting a 4th-order polynomial), can be assessed by comparing with those from the mesoscale model. Using the mesoscale model simulations enables investigation of the vertical profile of the parameterised output, in particular the vertical evolution of the 5 wave phase, which is not possible at good vertical resolution using AIRS data alone. In the climate model implementation, ∆T SSO is passed solely to the model output to enable its evaluation and is not used by the dynamical core or any other parameterisation scheme.
Finally, to assess the impact of the scheme on PSC microphysics and chemistry, per-10 turbation and control experiments using the chemistry-climate model were conducted.
In the perturbation experiments the mountain wave parameterisation is switched on.
Following Carslaw et al. (1999), only the cooling phase ∆T − SSO of the parameterised temperature fluctuations are coupled to the PSC scheme, meaning that the warm phase is neglected. Carslaw et al. (1999) argue that this approach is physically jus-15 tified as the warming phase of the wave-induced temperature fluctuations is typically of short enough duration that the complete evaporation of the PSC particles is unlikely to occur before temperatures fall again. In addition, evaporation will not occur if the synoptic-scale temperatures are sufficiently low that the warming phase still results in the temperature being below the PSC threshold value. The PSC scheme computes a 20 "total" temperature, used only by itself, by combining the temperature explicitly resolved by the chemistry-climate model T CHEM-CLIM with ∆T − SSO . In the control experiment the mountain wave scheme is switched off. Both the perturbation and control experiments were run for 30 years (following a 30 year spin-up period) for a perpetual year-2000, using prescribed sea-surface temperatures and sea ice concentrations.

Mesoscale model verification
Figure 3 compares maps of measured and mesoscale model simulated estimates of ∆BT for each of the three case studies. In the left panels the measured field ∆BT AIRS shows warm and cold temperature disturbances of amplitude 2-3 K clearly aligned with the western side of the Peninsula mountain ridge, i.e. typical of phase fronts associated 5 with a mountain wave caused by low-level westerly flow passing over the Peninsula and propagating upward in the atmosphere. In the right panels the amplitude and structure of the corresponding mesoscale model field ∆BT MES agrees well with the measurements. Figure 4 compares ∆BT AIRS and ∆BT MES in more detail by examining their variation 10 along the west-east orientated lines displayed in Fig. 3. The mountain wave appears prominently in both fields, with the mesoscale model producing a similar looking temperature disturbance to that measured. There are slight differences in terms of the wave amplitude, e.g. the mesoscale model amplitude in CS3 is slightly larger than that measured.

15
Note that in addition to a coherent mountain wave structure, Figs. 3 and 4 also show highly localised temperature fluctuations. For AIRS these fluctuations are partly due to increasing instrumental noise with low scene temperatures. The nominal noise of 0.39 K at 250 K scene temperature scales to 0.67-0.78 K at 190-200 K, which is more representative for the situations observed here. ing temperature fluctuations, the fairest approach is to compare profiles of ∆T SSO and ∆T CLIM for a particular N48 grid box with the mean and spread (± two standard deviations) of ∆T MES for all the mesoscale model points within the same N48 grid box. This is shown in Fig. 5  The lack of phase tilt is due to the parameterised wave field being represented by a hydrostatic gravity wave launched from an isolated bell-shaped ridge for each gridbox, which is then only propagated vertically through the column of air above. This simplification is also prohibitive in modelling the full downstream response. At N48 resolution the Antarctic Peninsula is multiple grid boxes wide as its resolved orography 5 field is hugely smoothed/flattened (see Fig. 6) and is thus represented in the parameterisation as a series of very similar sub-grid ridges, while in the mesoscale model the Peninsula is resolved as a dominant wide single ridge. Therefore the parameterisation produces a simplified broad response, which has smaller amplitude compared to the mesoscale model, across the Peninsula, whereby any change in phase can only result

Impact of the mountain wave parameterisation on PSC formation
Having shown that the parameterised mountain wave-induced temperature fluctuations are broadly consistent with the mesoscale model results, we can progress to assessing the impact of including the wave-induced cooling phase ∆T − SSO in the chemistryclimate model, again concentrating on the Antarctic Peninsula. In the first instance, 25 we will examine the impact on the temperatures seen by the PSC scheme. Figure 9 shows for July at a height of 21 km the 30 year average difference in the frequency f ACPD 14,2014 Inclusion of mountain wave-induced cooling of the temperature falling below the 195 K and 188 K thresholds for PSC formation of type Ia and II, respectively. The differences are between the frequency based on the explicitly resolved temperature T CHEM-CLIM plus ∆T − SSO from the perturbation run, and the frequency based solely on the explicitly resolved temperature of the perturbation run, i.e. f T CHEM-CLIM +∆T − SSO − f T CHEM-CLIM . The differences in frequency are always positive, 5 which is consistent with only mountain wave cooling being used. The results show that over much of the Peninsula, the impact of the mountain wave cooling is to increase the frequency that the 195 K threshold is exceeded, peaking over its northern tip with a frequency difference of 4 percentage points. By comparison, the impact on the 188 K temperature threshold is even more dramatic, resulting in differences which are both larger and extending much further south, peaking over Alexander Island to the south-west of the Peninsula with a frequency difference of over 6 percentage points. The fact that the differences in 195 K threshold frequency are located predominately over the middle and northern sections of the Peninsula is consistent with the climatological 195 K isotherm of the perturbation run being situated at approximately −75 • 15 latitude (not shown). Hence, any increase in the frequency of temperatures falling below 195 K as a result of the parameterisation can only occur northward of this, i.e. where the large-scale temperature is not already less than 195 K. Similarly, the differences in 188 K threshold frequency which encompass the entire length of the Peninsula are consistent with the model 188 K isotherm being situated southward of the 195 K 20 isotherm (not shown). Figure 10 compares the 30 year temperature distribution based on T CHEM-CLIM + ∆T − SSO of the perturbation run against that of T CHEM-CLIM for the perturbation run for the same N48 grid box used for Figs. 5, 7 and 8, again for July and at 21 km. As expected, inclusion of the parameterised mountain wave cooling shifts the temperature distribution to lower temperatures. In particular, it causes a longer left tail 25 of the temperature distribution which extends down to 177 K (or 5 K colder than the temperature distribution based solely on T CHEM-CLIM ).
The effect of the parameterisation on PSCs is investigated by evaluating the 30 year average difference in PSC surface area density between the perturbation and control ACPD 14,2014 Inclusion of mountain wave-induced cooling simulations (perturbation minus control). PSC surface area density controls the amount of reactive chlorine species produced, which cause ozone destruction. Figure 11 shows the difference in PSC surface area density at a height of 21 km for July. The perturbation run results in increases in surface area density for all PSCs (i.e. combined type I and II) of 6-10 µm 2 cm −3 over the Antarctic Peninsula and > 10 µm 2 cm −3 over the 5 Bellingshausen Sea. Relative to the control run, these are equivalent to increases of more than 50 % over the northern tip of the Antarctic Peninsula, and at least 30 % over the Bellingshausen Sea. The Weddell Sea region shows a non-significant decrease in PSC surface area density. This is not unexpected, even though the first order expectation is for a large-area increase in PSCs. The chemistry-climate model is interactive: changing PSCs change chlorine activation, which impacts ozone loss. Changing ozone alters the heating rates that impact temperatures and circulation. What is diagnosed in Fig. 11 (and related figures) is the difference between two climate equilibrium states for identical boundary conditions (compare e.g. Braesicke et al., 2013). Consequently, what is shown in the figures is locally strongly influenced by the additional parameter-15 isation (adding localised cooling and thus producing more PSCs), but in regions away from the direct impact the response can be determined by feedback mechanisms. Figure 11 additionally separates these differences into their individual contributions from type I and type II PSCs. It is type I (type II) PSCs which are largely responsible for the overall PSC increase over the Antarctic Peninsula (Bellingshausen Sea). Note that 20 significant differences in PSC surface area density were also evident in June, but not in August and September (not shown).

Summary and discussion
This study demonstrates that: (i) UM high-resolution (4 km) mesoscale model simulations are able to accurately simulate the large mountain wave-induced temperature 25 fluctuations in the lower stratosphere associated with strong westerly or north-westerly flow over the Antarctic Peninsula, and (ii) UM low-resolution ( simulations are completely unable to resolve such temperature fluctuations. These fluctuations act as a significant source of localised PSC formation as they enable stratospheric temperatures which otherwise would remain above the temperature threshold for PSC formation, to fall below it. With low spatial resolution a model is unable to resolve such temperature fluctuations, and as a consequence underestimate mountain 5 wave-induced PSCs and the attendant PSC-induced ozone depletion. This in turn is also a good rational for using the negative temperature anomalies only for the call to the chemistry scheme. On a sub-grid scale, going below a threshold temperature will produce additional PSCs, whereas staying above will not trigger PSCs. So for a gridbox averaged PSC coverage only additional incidents below the threshold temperature increase the coverage. Certainly for such an assumption to be true the sub-grid wave train should be slowly evolving horizontally compared to the model time step, thus producing an additional PSC occurrence frequency on the fringes of the synoptic-scale threshold temperature regions, as has been illustrated with Fig. 9 and discussed above. To investigate the impact of temperature fluctuations due to unresolved (sub-grid of independent sub-grid scale ridges which each launch a mountain wave vertically through the column of air above. Moreover, the parameterisation also does not represent trapped mountain lee waves, which can result in localised cooling many hundreds of kilometres downstream (e.g. Dörnbrack et al., 1999). The study followed this by assessing the impact on PSCs using the UKCA chemistry-5 climate model. It was found that adding the wave-induced cooling phase to the resolved temperature had a substantial impact on the frequency and magnitude of low temperatures which satisfy PSC thresholds, resulting in a regional 30-50 % increase in PSC surface area density during July at a height of 21 km over the Antarctic Peninsula and the Bellingshausen Sea. It should be stressed that we were unable to compare these results with observations as: (i) detailed measurements of Antarctic PSCs over a decadal time scale are not available at present (Austin et al., 2010), and (ii) global atmospheric reanalyses do not resolve small-scale temperature fluctuations. Our decision to neglect the wave-induced warming phase might imply that the diagnosed increase should perhaps be considered as an upper bound. However, as mentioned 15 above, the formation of PSCs downstream due to trapped lee waves is currently not represented in the model. Carslaw et al. (1999) remedied this by applying a horizontal "influence function" to simulate cooling downstream of orography, and a similar approach will be considered in future implementations of the scheme in UKCA chemistryclimate model. The simulation of PSC differences both upstream and downstream of 20 the Antarctic Peninsula, and hence removed from the actual region where the parameterisation impacts temperatures directly, is suggestive of a new climate-equilibrium state being established in the model that allow non-local effects to occur. Investigation of this will be the subject of future study. Moreover, the parameterisation might offer a method for improving lower stratospheric model temperatures that more often satisfy 25 conditions for PSC formation, the failure of which was suggested by Austin et al. (2010) to be one of the main reasons for the poor simulation of ozone depletion.
It is worth noting that other biases can affect the ability of chemistry-climate models to realistically simulate PSCs. For example, the failure of many models to represent the Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | effects of non-orographic gravity wave drag can result in unrealistically cold temperatures in the Southern Hemisphere winter stratosphere (Orr et al., 2010), i.e. resulting in synoptic-scale temperatures which fall below the PSC temperature threshold when in reality they should be above it, which as a consequence cause the formation of too many PSCs and associated increased ozone losses (Austin et al., 2003). Moreover, the 5 standard quasi-equilibrium PSC scheme used by the UKCA module does not advect PSC particles (Feng et al., 2011), i.e. the impact on PSC formation of the mountain wave parameterisation scheme is localised to wherever the temperature fluctuations are applied. This means that the occurrence of circumpolar belts of PSCs which have been attributed to mountain wave-induced PSCs over regions such as the Antarctic Peninsula would not be represented. However, future work will investigate replacing the quasi-equilibrium PSC scheme with the full microphysical scheme DLAPSE (Denitrification by Lagrangian Particle Sedimentation), which uses a Lagrangian trajectory scheme and as such is able to transport PSC particles away from the region of formation (Feng et al., 2011). However, the current study illustrates that a more compre- 15 hensive treatment of sub-grid scale mountain waves in a global climate model leads to realistic localised temperature change diagnostics. Subsequently, we have been able to assess and characterise the localised impact of the modelled temperature fluctuations in a comprehensive chemistry-climate model. Further work will investigate the non-localised effects in more detail. 20 Acknowledgements. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Chemistry-climate model simulations of spring Antarctic ozone, J. Geophys. Res., 115, D00M11, doi:10.1029/2009JD013577, 2010 Circulation anomalies in the Southern Hemisphere and ozone changes, Atmos. Chem. Phys., 13, 10677-10688, doi:10.5194/acp-13-10677-2013 Campbell, J. R. and Sassen, K.: Polar stratospheric clouds at the South Pole from 5 years of continuous lidar data: macrophysical, optical, and thermodynamic properties, J. Geophys. Res., 113, D20204, doi:10.1029/2007JD009680, 2008 Figure 1. The temperature weighting function (brightness temperature (K) / temperature (K)) 2 for the 666.5 cm -1 AIRS channel. This function was calculated for a polar winter reference 3 atmosphere, a 1 km altitude grid, and the nadir observation geometry.  CS1 (a, b), CS2 (c, d), and CS3 (e, f) mountain wave events. See Table 1 for dates. The black arrows are wind vectors (for the mesoscale model only 1 in every 40 grid points is shown). The colour shading indicates the wind magnitude. Also shown is the coastline of the Antarctic Peninsula.  CS1 (a, b), CS2 (c, d), and CS3 (e, f) mountain wave events. See Table 1 for dates. The horizontal black lines indicate the latitude band selected for a more detailed comparison, shown in Fig. 4. Also shown is the coastline of the Antarctic Peninsula.  Table 1 for dates. The climate model profile is from the same N48 grid box (i.e. over the high-elevation ridge of the Antarctic Peninsula). The mesoscale model profiles are for all the mesoscale model points within the N48 grid box. ACPD 14,2014 Inclusion of mountain wave-induced cooling  Figure 11. Impact of the mountain wave parameterisation during July at 21 km on PSC 2 surface area density (μm 2 cm -3 ) over the Antarctic Peninsula in the chemistry-climate model. 3 The shading indicates the 30-year average difference in surface area density between the 4 perturbation run and the control run (perturbation run minus the control run) for PSC types I 5 and II (A), type I (B), and type II (C). The contours indicate the 30-year average PSC surface 6 area density from the control run. Hatching denotes significance at the 95% confidence level 7 using a two-tailed Student's T-Test. Also shown is the coastline of the Antarctic Peninsula. 8 Figure 11. Impact of the mountain wave parameterisation during July at 21 km on PSC surface area density (µm 2 cm −3 ) over the Antarctic Peninsula in the chemistry-climate model. The shading indicates the 30 year average difference in surface area density between the perturbation run and the control run (perturbation run minus the control run) for PSC types I and II (A), type I (B), and type II (C). The contours indicate the 30 year average PSC surface area density from the control run. Hatching denotes significance at the 95 % confidence level using a two-tailed Student's t test. Also shown is the coastline of the Antarctic Peninsula.