Composition changes after the " Halloween " solar proton event : the High-Energy Particle Precipitation in the Atmosphere ( HEPPA ) model versus MIPAS data intercomparison study

We have compared composition changes of NO, NO2, H202,03, N 2 0, HNO3 , N2 05 , HN04, CIO, HOCI, and CIONO 2 as observed by the Michelson Interferometer 30 for Passive Atmospheric Sounding (MIPAS) on Envisat in 5 the aftermath of the "Halloween" solar proton event (SPE) in October/November 2003 at 25-0.01 hPa in the Northern hemisphere (40-90°N) and simulations performed by the following atmospheric models: the Bremen 2D model (B2dM) 3e and Bremen 3D Chemical Transport Model (B3dCTM), the ,o Central Aerological Observatory (CAO) model, FinROSE, the Hamburg Model of the Neutral and Ionized Atmosphere (HAMMONIA), the Karlsruhe Simulation Model of the Middle Atmosphere (KASIMA), the ECHAM5/MESSY 40 Atmospheric Chemistry (EMAC) model, the modeling tool 15 for S01ar Climate Ozone Links studies (SOCOL and SOCOLi), and the Whole Atmosphere Community Climate Model (WACCM4). The large number of participating models allowed for an evaluation of the overall ability of atmo4e spheric models to reproduce observed atmospheric perturba20 tions generated by SPEs, particularly with respect to NOS, and ozone changes. We have further assessed the meteorological conditions and their implications on the chemical response to the SPE in both the models and observations by comparing temperature and tracer (CH4 and CO) fields. 25 Simulated SPE-induced ozone losses agree on average 5 within 5% with the observations. Simulated NO v enhancements around 1 hPa, however, are typically 30% higher than Correspondence to: B. Funke (bemd a iaa.es) indicated by the observations which can be partly attributed to an overestimation of simulated electron-induced ionization. The analysis of the observed and modeled NO y partitioning in the aftermath of the SPE has demonstrated the need to implement additional ion chemistry (HNO 3 formation via ion-ion recombination and water cluster ions) into the chemical schemes. An overestimation of observed H2O2 enhancements by all models hints at an underestimation of the OH/HO 2 ratio in the upper polar stratosphere during the SPE. The analysis of chlorine species perturbations has shown that the encountered differences between models and observations, particularly the underestimation of observed CIONO2 enhancements, are related to a smaller availability of CIO in the polar night region already before the SPE. In general, the intercomparison has demonstrated that differences in the meteorology and/or initial state of the atmosphere in the simulations causes a relevant variability of the model results, even on a short timescale of only a few days.


Introduction
Energetic particle precipitation has important implications on atmospheric chemistry.In particular, protons and associated electrons, generated during solar eruptions, cause sporadically in-situ production of NO X and HOx radicals involved in catalytic ozone destruction.These solar proton events (SPEs) thus represent an important Sun-Earth connection which contributes to the natural ozone variability.The quasi-instantaneous increase of odd nitrogen and hydrogen Funke et al.: HEPPA intereomparison study due to SPEs induces perturbations of the chemical composition of the middle atmosphere on a short-time scale.In this sense, SPE-induced perturbations of the atmospheric composition represent an ideal natural laboratory for studying 60 stratospheric and mesospheric chemistry (see also Jackman "s and McPeters, 1987).In recent years, there have been two large SPEs (October/November 2003 and January 2005) (Jackman et al., 2008) which have been intensively observed by several in-65 struments on different satellite platforms, including, for ex-120 ample, NOAA 16 SBUV12 and HALOE data (Jackman et al., 2005a,b;Randall et al., 2005); MIDAS, GOMOS and SCIA-MACHY on Envisat (L6pez-Puertas et al., 2005a,b;von Clarmann et al., 2005;Orsolim et al., 2005;Seppala et al., zo 2004;Rohen et al., 2005); and MLS on AURA (Verronen125 et al., 2006).In particular, during late October and early November 2003, three active solar regions produced solar flares and solar energetic particles of extremely large intensity, the fourth largest event observed in the past forty years 75 (Jackman et al., 2005b(Jackman et al., , 2008)).During and after this event,,ao often called "Halloween" storm, the MIPAS instrument observed global changes (e.g. in both the Northern and Southern polar regions, during day and nighttime) in the stratospheric and lower mesospheric composition.This includes ao enormous enhancements in NO., e.g., in NO and NO2 , and 135 large depletions in 0 3 (L6pez- Puertas et al., 2005a) as well as significant changes in other NOy species, such as HNO3, N2 0 5 , CIONO 2 (L6pez- Puertas et al., 2005b), and N20 (Funke et al., 2008).In addition, there also have been ob-e5 served changes in CIO and HOCI as evidence of perturba-140 tions by solar protons on the HO, and chlorine species abundances (von Clarmann et al., 2005).Several model studies, aiming at reproducing observed short-and medium-term composition changes after this pargo ticular event (Jackman et al., 2008;Verronen et al., 2008;,45 Funke et al., 2008;Baumgaermer et al., 2010;Egorova et al., 2010) and evaluating SPE-induced long-term effects (Jackman et al., 2009) have been carried out in the past.The High-Energy Particle Precipitation in the Atmosphere 95 (HEPPA) model vs. data intercomparison initiative has ,5o brought together scientists involved in atmospheric modeling using state-of-the art general circulation models (GCMs) and chemistry-transport models (CTMs) on the one hand and scientists involved in the analysis and generation of observa-100 tional data on the other hand.The objective of this commu-i55 nity effort is (i) to assess the ability of state-of-the-art atmospheric models to reproduce composition changes induced by particle precipitation, (ii) to identify and -if possibleremedy deficiencies in chemical schemes, and (iii) to serve 105 as a platform for discussion between modelers and data pro-m ducers.This is achieved by a quantitative comparison of observed and modeled composition changes after particle precipitation events, as well as by inter-comparing the simulations performed by the different models. 110 In this study we report results from the intercompari-;55 son of MIPASIEnvisat data obtained during 26 October -30 November 2003, before and after the Halloween SPE, at altitudes between 25-75 km (25-0.01hPa) with simulations performed using the following GCMs and CTMs: the Bremen 2d Model (B2dM) (Sinnhuber et al., 2003b;Winkler et al., 2009), the Bremen 3d Chemical Transport Model (B2dM and B3dCTM) (Sinnhuber et al., 2003a), the Central Aerological Observatory (CAO) model (Krivolutsky and Vyushkova, 2002), FinROSE (Damski et al., 2007b), the Hamburg Model of the Neutral and Ionized Atmosphere (HAMMONIA) (Schmidt et al., 2006), the Karlsruhe Simulation Model of the Middle Atmosphere (KASIMA) (Kouker et al., 1999), the ECHAM5IMESSy Atmospheric Chemistry (EMAC) model (Mckel et al., 2006), the modeling tool for SOlar Climate Ozone Links studies (SOCOL and SO-COLi) (Egorova et al., 2005;Schraner et al., 2008;Egorova et al., 2010), and the Whole Atmosphere Community Climate Model (WACCM4) (Garcia et al., 2007).Among the species affected by SPEs we focus here on NO, NO2, H2 02 , 03 , N2 0, HNO 3 , N 2 05 , HN04 , CIO, HOCI, and C10NO 2 .For these species a significant perturbation well above the detection limit has been observed by MIPAS.We have further assessed the meteorological background conditions in both the models and the real atmosphere as observed by MIPAS by comparing temperature and tracer fields (CH4 and CO).Although SPE-induced composition changes during the Halloween event have been reported in both hemispheres, we restrict our analysis to the Northern hemisphere (NH) in the latitude range 40-90°N where most pronounced effects have been observed and composition changes can be well distinguished from the background variability.Apart from the initial particle forcing leading to atmospheric ionization, SPE-induced composition changes are controlled by several other factors such as the neutral and ion chemistry responsible for the repartitioning of primarily generated species, the background composition interfering with the chemical repartitioning, and the meteorological/dynamical conditions.The large number of controlling factors and their interaction introduce a significant spread in the model results and make their analysis difficult.In order to reduce the model variability and to make differences between the simulations more traceable, we have simplified the intercomparison setup such that a common particle-induced ionization source has been used in all models.These ionization rates, accounting for protons (154 eV-500 MeV) and electrons (154 eV-5 MeV) have been provided by the AIMOS model (Wissing and Kallenrode, 2009).Different model responses to the particle forcing are hence reduced to differences of the intrinsic model properties, e.g. chemical and dynamical schemes.A major aim of this paper is the assessment of these differences and their implications on the models' ability to correctly describe particle precipitation effects which represent an important source of natural, solar-induced climate variability on short and mid-term scales.Additionally, conclusions on the quality of the description of the ex- ternal forcing provided by the ionization model can be drawn from the overall agreement of the short-time response of pri-Zoe marily generated constituents (i.e.NOj The paper is organized as follows: In Section 2 we give an overview on MIPAS observations and data products used in this study, followed by Section 3 describing the ionization model AIMOS and Section 4 describing the participat-210 ing global circulation and chemistry transport models.The intercomparison method is described in Section 5, followed by the discussion of the results (Section 6).

MIPAS observations 215
The Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) is a mid-infrared Fourier transform limb emission spectrometer designed and operated for measurement of atmospheric trace species from space (Fischer et al., 2008).It is part of the instrumentation of the European Envi-220 ronmental Satellite (ENVISAT) which was launched into its sun-synchronous polar orbit of 98.55°N inclination at about 800 km altitude on I March 2002.MIPAS passes the equator in a southerly direction at 10.00 a.m.local time 14.3 times a day, observing the atmosphere during day and night with 225 global coverage from pole to pole.The instrument's field of view is 30 km in horizontal and approximately 3 km in vertical direction.MIPAS operated during October/November 2003 at frill spectral resolution of 0.035 em' (unapodized) in terms of full width at half maximum.During this pe-2e riod, MIPAS recorded a rear-viewing limb sequence of 17 spectra each 90 s, corresponding to an along track sampling of approximately 500 km and providing about 1000 vertical profiles per day in its standard observation mode.Tangent heights covered the altitude range from 68 down to 6 km 2ss with tangent altitudes at 68, 60, 52, 47, and then at 3 km steps from 42 to 6 km.Trace gas profiles have been retrieved from calibrated ge-2ce olocated limb emission spectra with the scientific MIPAS level 2 processor developed and operated by the Institute 017240 Meteorology and Climate Research (IMK) in Karlsruhe together with the Instituto de Astrofisica de Andalucia (IAA) in Granada.The general retrieval strategy, which is a constrained multi-parameter non-linear least squares fitting of measured and modeled spectra, is described in detail in von Clarmann et al. (2003).Its extension to retrievals under consideration of non-LTE (i.e., CO, NO, and NOz) is described in Funke et al. (2001).Non-LTE vibrational populations of these species are modeled with the Generic R.Adiative traNsfer AnD non-LTE population Algorithm (GRANADA) (Funke et al., 2007) within each iteration of the retrieval.
In contrast to previous work describing MIPAS observations of composition changes during the Halloween SPE (L6pez-Puertas et al., 2005a,b;von Clarmann et al., 2005), we base our analysis here on reprocessed IMK/IAA MIPAS data which have substantially improved with respect to previous data versions.These improvements include updates in the L I B processing (version 4.61162 instead of 4.59) performed by ESA as well as changes in the L2 processing performed at IMK/IAA.The new data set also offers full temporal coverage over the period of interest (26 October-30 November 2003).In the following, we summarize the improvements of the retrieval setups for each species/parameter and characterize the data used in our analysis in terms of estimated single measurement precision and vertical resolution obtained from the full width at half maximum of the rows of the averaging kernel (AK) matrix (Rodgers, 2000).The AK diagonal elements are also discussed as a measure of the sensitivity of the retrieval at a given profile grid point to the "true" profile.Values close to zero (typically c0.03) indicate that there is no significant sensitivity to the retrieval parameter at the corresponding altitude and hence are excluded from our analysis.This value may appear unreasonably small but since IMK/IAA retrievals are not constrained by optimal estimation (Rodgers, 2000) but by a first order smoothing constraint using a Tikhonov (1963) formalism, low values do not hint at a large a priori content of the retrieval but only at extensive smearing of information over altitude.A detailed discussion of systematic retrieval errors can be found in previous works describing the individual constituent retrievals which are referenced in the following.1).The single measurement pre-300 eision ranges from 0.5 K to 1.5-2.5 K above the stratopause.Vertical resolution is 3-4 kin below 1 hPa and 5-7 km above.Meaningful data are obtained in the whole vertical range of interest (25-70 km).
The single measurement precision ranges from 10-20ppbv3m in the upper stratosphere to 50-70 K above and below (see Figure 1) Vertical resolution is 3-6 km below 0.03 hPa and slightly higher above.Meaningful data is obtained in the whole vertical range of interest (25-70 km).31S 2.3 CO CO data versions used here are V30-CO-9, V30-CO-10, described in detail in Funke et al. (2009), as well as the,,, most recent version V30-CO-11 (see Table 1).Improvements implemented in the latter version include an extended set of spectral fitting windows resulting in a better precision and vertical resolution in the lower and middle stratosphere.
In the period of interest, the temporal evolution of MIPAS CO abundances at 60-90°N indicate polar winter descent of... mesospheric air masses of about 10 km around 1 hPa (see Figure 1).The single measurement precision ranges from 20-30% above 1 hPa to 70-80% in the lower stratosphere.Vertical resolution is 6-12 km below 0.1 hPa.Meaningful data are obtained in the whole vertical range of interest (25-28c70 km).330

NO
We use version V30-NO-14 (see Table 1), available for the whole time period.This version has substantially improved335 with respect to the retrieval setup described in Funke et al. (2005) and the data discussed in Lopez-Puertas et al. (2005a) by i) the use of log(vmr) instead of vmr (volume mixing ratio) in the retrieval vector, ii) a revised correction scheme for line of sight variations of the NO, partitioning close to the terminator, and iii) joint-fitted vmr horizontal gradients340 at constant longitudes and latitudes.NO increases of several 100 ppbv have been observed at 60-90°N during the intense proton forcing during 29 October-4 November in the upper stratosphere around 0.2 hPa (see Figure 1).Above, NO increases were mainly produced by polar winter descent of Lip- per atmospheric air masses, resulting in vmrs up to 1 ppmv345 below 70 km.The single measurement precision is of the order of 10%.Vertical resolution ranges from 4 to 8 km below 70 km.Meaningful data are obtained in the whole vertical range of interest (25-70 km).

NO2
NO 2 data versions used here are V30-NO2-11, V30 -NO2-13, and V30-NO2-14 (see Table 1).Including the same modifications as described above for NO, these versions have substantially improved with respect to the retrieval setup described in Funke et al. (2005) and the data discussed in Lopez-Puertas et al. (2005a).While differences between the latter two versions do not affect noticeably the data characteristics, a modified regularization scheme and tenninator treatment implemented after version V30-NO2_11 gave rise to non-negligible differences in the newer versions with respect to the previous setup.These differences are visible in the vertical resolution in the mesosphere and middle stratosphere (see Figure 1, third column) and go along with generally smaller vmrs around the terminator at 70°N around 0.1 hPa.Similar to NO, increases of 50-80 ppbv were observed during the proton forcing in the upper stratosphere, descending by approximately 10 km until the end of November.Polar winter descent of NO, led to mesospheric NO 2 increases of more then 100 ppbv, particularly in the second half of November.The single measurement precision is of the order of 5-10%.Vertical resolution ranges from 4 to 8 km below 70km.Meaningful data are obtained in the whole vertical range of interest (25-70 km).

N20
We use version V30_N2O_I2 (see Table 1), available for the whole time period.This version, which has already been used for the previous analysis of N 2 0 abundance changes during the Halloween SPE (Funke et al., 2008) and differs from other versions by a relaxed regularization above approximately 40 km which allows for vertically resolving the upper stratospheric and mesospheric enhancements.At 60-90°N, these enhancements of around 5-7 ppbv appeared around 30 October and descended during November to the middle stratosphere (see Figure 2).The single measurement precision ranges from 0.5 ppbv in the upper stratosphere to 2 ppbv in the mesosphere.Vertical resolution is 4-6 km.
Meaningful data are obtained in the whole vertical range of interest (25-70km).

HNO3
We use version V30-HNO3-9 (see Table 1), available for the whole time period, and which is based on the retrieval setup described in Wang et al. (2007).HNO 3 increases of around 3 ppbv tip to altitudes of 0.1 hPa during the proton forcing and a further buildup at slightly lower altitudes at the end of November are visible in Figure 2, consistent with previous findings (Lopez-Puertas et al., 2005b).The single measurement precision ranges from 0.1 ppbv in the middle stratosphere to 0.35 ppbv around the stratopause.Vertical resolu- Fig. 2. Same as Figure 1, but for N 2 0, HNO3, N205, and HN04.
tion is 3-4 km below 12 hPa and 7-10 km above.Meaningful below approximately 0.3 hPa (52 km).data are obtained below 0.1 hPa (60 km).365 2.9 HN04 2.8 N205 N2 0 5 data versions used here are V30_N2059 and V30_N205_10 (see Table 1), all based on the retrieval setup described in Mengistu Tsidu et al. (2004).Differences between both versions are of minor nature and do not affect376 noticeably the data characteristics.N2 0 5 increases related to the proton event are mainly visible in Figure 2 in the second half of November around 2-0.5 hPa, consistent with previous findings (Lopez-Puertas et al., 2005b).The single measurement precision ranges from 0.05 ppbv to 0.15 ppbv 375 in the middle stratosphere.Vertical resolution is 5-7 km below 2 hPa and 5-7 km above.Meaningful data are obtained We use version V30-HN04-12 (see Table 1) which differs from the original retrieval setup described in Stiller et al. (2007) by the application of a weaker regularization in the middle stratosphere, where most pronounced SPE effects are expected.Unfortunately, this version is sensitive to systematic oscillations in the radiance baseline related to an imperfect gain calibration of the instrument (see also Stiller et al., 2008).In consequence, retrieved HN0 4 profiles are systematically biased during each gain calibration period (typically a few days) with a randomly changing magnitude from one calibration period to another.The variable bias is noticeable in the temporal evolution of the observed Fig. 3. Same as Figure 2, but for 03, H 2 0 2 , CIO, HOCI, and CIONO2.HN0 4 distributions at 60-90°N (see Figure 2) as sharp increases/decreases in the upper stratosphere, coincident with 380 the onsets of new gain calibration periods (i.e., 28 October, 10 November, and 24 November).Therefore, we restrict our analysis of SPE-related HN0 4 increases in Section 6 to data 39c observed during one particular gain calibration period, 28 October-5 November, covering the onset of the proton forcaas ing which led to short-term HN0 4 increases of the order of 0.15 ppbv (hardly visible in Figure 2).competing in weeks with seasonal mesospheric 0 3 buildup (see Figure 3).Also, the previously reported NO.
-induced losses at lower altitudes are seen on a mid-term scale.The single measurement precision ranges from 0.1 ppmv around the stratopause to 0.25 ppbv above and below.Vertical reso-440 1ution is 3-4 km below 1 hPa and 5-7 km above.Meaningful data are obtained in the whole vertical range of interest (25-70 km). 445

H2O2
We use version V301-1202_4, available for the whole time period (see Table 1) which is based on the retrieval setup described in Versick (2010).H2 02 increases up to 0.15 ppbv 451 have been observed at 60-90`N during the intense proton forcing on 29 October-4 November in the upper stratosphere around 0.2 hPa (see Figure 3).The single measurement precision in the middle stratosphere ranges from 0.1 to 0.2 ppbv, being thus of the order of the observed enhancements.In consequence, averaging is required for the analysis.Vertical resolutions larger than 10 km indicate that no relevant information on the vertical distribution of the middle/upper stratospheric enhancements can be extracted from the measurements.Meaningful data are obtained below approximately 0.2 hPa (55 km).460 2.12 CIO CIO data versions used here are V30-CLO_10 and,,, V30-CLO_11 (see Table 1), all based on the retrieval setup described in Glatthor et al. (2004).Differences between both versions are of minor nature and do not affect noticeably the data characteristics.As in the case of HN0 4 , CIO data is affected by systematic oscillations in the radiance baseline related to an imperfect gain calibration of the instrument, however, to a lesser degree than in the case of HN0 4 .The sin-4,0 gle measurement precision ranges from 0.2 ppbv in the lower stratosphere to 0.7 ppbv around 2 hPa, being thus higher than 100% at the C10 peak height (see Figure 3).In consequence, averaging is required for its analysis.1), all based on the retrieval setup described in von Clarmann et al. (2006).Differences between both versions are of minor nature and do not affect noticeably the data characteristics.HOC1 increases of around 0.3 ppbv show up in Figure 3 immediately after the main proton forcing at the beginning of November, consistent with previous findings (von Clarmann et al., 2005).The single measurement precision ranges from 0.05 to 0.1 ppbv around 2 hPa.Vertical resolution 8-12 km below 2 hPa and coarser then 15 km above.Meaningful data is obtained below approximately 0.5 hPa (40 km).

CIONO2
CIONO 2 data versions used here are V30_CLONO2-1 I and V30-CLONO2_12 (see Table 1), all based on the retrieval setup described in H6pfner et al. (2007).Differences between both versions are of minor nature and do not affect noticeably the data characteristics.CIONO 2 increases of around 0.5 ppbv are visible in Figure 3 after the main proton forcing above 5 hPa and last until the end of November.This is consistent with previous findings based on data version V 1_CLONO2_1 (L6pez-Puertas et al., 2005b) in qualitative terms, however, the peak height of the increases is slightly higher (N5 km) in the newer data versions included here.This difference is mainly related to a change of the heightdependent regularization strength in order to allow for more sensitivity at lower and higher altitudes.The single measurement precision ranges from 0.06 to 0.12 ppbv, increasing with altitude.Vertical resolution is 5-8 km below 2 hPa and 12-14 km above.Meaningful data is obtained below approximately 0.5 hPa (40 km).

Ionization rates
The model intercomparison is based on ionization rates calculated with Atmospheric Ionization Module OSnabruek (AIMOS).The reason is to avoid different model results due to different ionization rates as to better understand the differences in the dynamical and chemistry schemes of the models   under assessment.AIMOS calculates ionization rates due to precipitating solar and magnetospheric particles.The altitude range of calculated ionization rates is defined by the energy range of the particles considered, which is specific to the satellite instruments used.The data used here and their altitude coverage are listed in Table 2. Given by the altitude range of this study, the focus lies on solar particles.As particle precipitation strongly depends on the geomagnetic field,,,, the model accounts for different spatial precipitation zones.
A detailed description on AIMOS can be found in Wissing and Kallenrode (2009).AIMOS is composed of two parts.One describes the spatial particle flux on top of the atmosphere while the seconds,, calculates the resulting ionization rate.Both parts will be discussed in the following.

Spatial particle flux
The particle flux on top of the atmosphere is measured by the TED and MEPED instruments on POES 15(16 as well as the 52o SEM instrument on GOES 10.As all particle measurements are in-situ, the main challenge is to derive a global coverage at any time.Inside an empirically determined polar cap where particle precipitation is homogeneous, the high energetic particle flux from GOES and the mean flux values from 525 polar cap crossings of the POES satellites are used.Outside Soo the polar cap, particle precipitation depends on geomagnetic latitude, geomagnetic activity and local time.Therefore, mean precipitation maps for the POES TED and MEPED channels, based on a 4 year data set, have been produced, 53o sorted by the geomagnetic Kp -index and local time.These mean precipitation maps represent the spatial distribution, including, e.g., the movement of the auroral oval.According to the recent Kp -level, the mean precipitation maps are selected and scaled to recent POES particle flux.535 In summary, the first part of the model describes the incoming particle flux at every grid point.The spatial resolution is 96 zonal cells, divided into 48 meridional sections.Regions of similar particle flux are combined as, e.g., the polar cap.Given by the scaling of the mean precipitation maps, the temporal resolution is limited by the POES orbit and has been set to 2 h.

3,2 Modeling ionization rates
The second part of AIMOS is the atmospheric particle detector model, which simulates particle interactions based on the GEANT4-Simulation Toolkit (Agostinelli et al., 2003).GEANT4 provides Monte-Carlo based algorithms to model energy depositionlionization of protons and electrons.The atmospheric detector model is divided into 67 logarithmically equidistant pressure levels, ranging from sea-level to 1.7 x 10 -5 Pa.Since the atmospheric parameters (density, altitude, composition and temperature) depend on latitude, season and solar activity, model versions for 80°N, 60°N, 60°S and 80°S, 3 different F1Q7 flux values and 4 different months are used.These parameters are adopted from the HAMMONIA (Schmidt et al., 2006) and MSIS (Picone et al., 2002) models.The ionization rates for mono-energetic and isotropic particle ensembles are determined.As a final step, the mono-energetic ionization rates are combined with multiple power-law fits of the particle flux at various regions.
Figure 4 shows the temporal evolution of the resulting ion pair production rates averaged over 40-90'N during the pe-Funke et al.: HEPPA intercomparison study riod of interest.The latitudinal distribution of the ionization rates of protons and electrons, respectively, is shown in Fig. 5 for the 28 October 2003.
The ionization rates should provide a similar forcing forego 5L5 all models, therefore the original data set has been adopted to every model grid.The data set and the adoption routine for a used specific grid is available at http:llaimos.physik.uos.de.545 The Bremen two-dimensional model is based on the twodimensional transport, chemistry and radiation model for -eoo merly described in Sinnhuber et al. (2009) and Chipperfield and Feng (2003).It uses the dynamical core of the so-called two-and a half-dimensional" model THIN AIR (Kinners-55o ley, 1998), which calculates temperature, pressure, and horizontal transport on isentropic surfaces, interactive with the model chemistry.The model covers the altitude range fromso5 the surface to 100 km in 29 isentropic surfaces, providing a vertical spacing of about 3.5 km.The horizontal resolution 555 is about 9.5 degree.Stratospheric dynamics are forced by the amplitudes of waves 1 to 3 of the Montgomery potential from meteorological analyses with a repeating annual cycle for the sao period of May 1980 to April 1981.There is no quasi biennial oscillation (QBO) in the model, i.e., the modeled tropical 500 stratospheric wind is always in a weak easterly state.In this sense, the Bremen 2d model is a two-dimensional chemistryclimate model which is forced to repeat a very similar scenarioby the repeating annual cycle of the Montgomery potential.505 The chemistry is based on the SLIMCAT chemistry (Chipperfield and Jones, 1999), but adapted for the use in the mesosphere in several ways: (1) above 50km, no family ap-,,o proach is used; (2) H 2 O and CO2 are treated as short-lived species explicitly, and 112 is varied as well, to provide a re-57o alistic description of mesospheric HOx and CO. (3) NOx and HOx production by atmospheric ionization is parameterized based on Porter et al. (1976) and Solomon et al. (1981), sze i.e., 1.25 NOx are produced, of which 45% are produced as N, and 55% as NO, and up to 2 HOx are produced per ion 575 pair depending on pressure and ionization rate, equally distributed to H and OH.Ionization due to Galactic Cosmic Rays in the stratosphere has been included based on Heaps,,, (1978); the additional ionization due to solar and magnetospheric particles is considered by introducing atmospheric 580 ionization rates of protons and electrons provided by the AIMOS model (see section 3).
All reaction and photolysis rates are taken from Sander et al. (2006).The Bremen 2d model has been used in the past to investigate the impact of large solar proton events 035 585 on the composition of the middle atmosphere (Sinnhuber et al., 2003b;Rohen et al., 2005;Winkler et al., 2008).For the HEPPA intercomparison, the two-dimensional model has been combined with a one-dimensional model sharing the same description of chemistry-in the following way: 25 model runs with the two-dimensional model are carried out at different longitudes, to take into account the tilt of the geomagnetic poles.For every MIPAS measurement used in the intercomparison, a one-dimensional modal run is started initialized with output of the 2-dimensional model rums interpolated to the geo-location of the measurement, at local noon of the day before the measurement took place.Onedimensional model runs are then carried out until the time of the measurement, providing model output at the exact time and geo-location of the measurement.

Bremen 3d Chemistry and Transport model (B3dCTM)
The Bremen three-dimensional Chemistry and Transport Model is a combination of the Bremen transport model (Sinnhuber et al., 2003a) and the chemistry code of the Bremen 2D model (Sinnhuber et al., 2003b;Winkler et al., 2008), which is based on the SLIMCAT model (Chipperfield and Jones, 1999).
The model has 28 isentropic levels ranging from 330 to 3402K (approx.10-60km) and has a horizontal resolution of 3.75° x 2.5°.Output is provided hourly.The vertical transport across the isentropes is calculated through diabatic heating and cooling rates.These rates are calculatedusing the radiation scheme MIDRAD (Shine, 1987).The horizontal transport is driven by external wind-fields.Advection is calculated by using the second order moments scheme by Prather (1986).Meteorological data, such as horizontal wind speeds and temperatures, are taken from ECMWF ERA Interim (Simmons et al., 2006).
The model calculates the behavior of 58 chemical species, using a family approach for short-lived species (HO., NOx, Ox, CLOx, BrOx and CHO.).It includes about 180 gas phase, photochemical, and heterogeneous reactions and uses the recent set of recommendations for kinetic and photochemical data established by the Jet Propulsion Laboratory (Sander et al., 2006).
To account for ion chemistry reactions within the neutral code, the production of NOx and HOx is parameterized as suggested by Porter et al. (1976) and Solomon and Crutzen (1981).Hence 1.25 N atoms and about 2 HOx are produced per ion pair.Atmospheric ionization due to solar and magnetospheric particles is considered by introducing atmospheric ionization rates of protons and electrons provided by the AIMOS model (see section 3).

CAO
The Central Aerological Observatory (CAO) model consists of a CTM and a 3D dynamical core with a horizontal resolution of 10° x 10° and vertical resolution of 2 km.The  Sander et al. (2003).The annual and daily variations of the solar zenith angle at a given point and its dependence on the height above the Earths surface were taken into account.For zenith angle higher than 75°, Chapmans functions have been used in accordance with Swinder and Gardner (1967).Photolysis rates have been recalculated685 every hour during the integration of the model.A family approach (Turco and Whitten, 1974) has been used for solving the chemical equations, including O X (03+0+0(tD)), NO, (N+ NO+NO2+NO3+2N205+HNO3+HN04+C10NO2), Cl, (C1+C10+OC10+C100+HOCI+HCI), and HO, 690 (H+OH+ H0 2 +2H2 O2).Other long-lived species (N20, CC14 , CFC13 , CF2 C12 , CH303, CH4, H2 O, H2 and CO2) were included also in simulations.The CAO model applies additionally to electron and proton-induced ionization also ionization rates caused by alpha-particles provided by the-AIMOS model (Wissing and Kallenrode, 2009).
The vertical profiles of molecular oxygen and air density were fixed during photochemical calculations.Heterogeneous removal for H 2 O 2 , HNO3 , HCl, and HN0 4 was... included in the troposphere.We assumed also fixed mixing ratios for long-lived and chemical families components 665 at lower and upper boundaries during calculations in order to formulate the boundary conditions.Corresponding mixing ratio values were taken from Park et al. (1999).An accu-706 rate, non-diffuse method for three-dimensional advection of trace species suggested by Prather (1986) was used to solve 670 the continuity equation for each transported species (families and long-lived species).The chemical constituents were initialized with profiles obtained from a one-dimensional model (Krivolutsky et al., 2001).Wind components used for transport by advection were obtained from the simulation with 710 675 the 3D dynamical model for each day of the year.Daily averaged global zonal, meridional, vertical wind components, and temperature have then be used in the CTM rums.More details concerning the chemical 3D model can be found in Krivolutsky and Vyushkova (2002).This model was used to study the response in composition and dynamics after the July 2000 SPE (Krivolutsky et al., 2006).The CAO simulation included in this study covers the period 26 October-4 November.

FinROSE
FinROSE is a global 3-D chemistry transport model (Damski et al., 2007a).The model dynamics are from external sources except the vertical wind, which is calculated inside the model using the continuity equation.In this study FinROSE has 35 vertical levels (0-65 km), a horizontal resolution of 10° x 5°a nd uses ECMWF Interim analyses (Simmons et al., 2006) for dynamics.Output is provided every 3 hours.The model produces distributions of 40 species and includes about 120 homogeneous reactions and 30 photodissociation processes.Chemical kinetic data, reaction rate coefficients and absorption cross-sections are taken from look-up-tables based on the Jet Propulsion Laboratory compilation by Sander et al. (2006).Photodissoeiation frequencies are calculated using a radiative transfer model (Kylling et al., 1997).The model also includes formation and sedimentation of polar stratospheric clouds (PSCs) and reactions on PSCs.Tropospheric abundances are given as boundary conditions and long-lived trace gases are relaxed towards long time trends.The FinROSE model applies additionally to electron and proton-induced ionization also ionization rates caused by alpha-particles provided by the AIMOS model (Wissing and Kallenrode, 2009).

4,5 HAMMONIA
The Hamburg Model of the Neutral and Ionized Atmosphere (HAMMONIA) is a 3-dimensional GCM and chemistry model covering an altitude range from the surface up to 1.7 x 10 -7 hPa.A detailed description of the model is given by Schmidt et al. (2006).Simulations of particle pre-cipitation effects use a modified version of HAMMONIA.It treats 54 photochemical, 139 bi-and termolecular, 5 ionelectron recombination, and 12 ion-neutral reactions involving 50 neutral and 6 charged (O + , 0^ , N+ , Nz , NO + , e-) components.Neutral bi-and termolecular reactions and the corresponding rate coefficients are taken from Sander et al. (2006).Photochemistry involves 7 ionizing and dissoeiat -770 ing reactions through solar irradiance of wavelengths shorter than Lyman-alpha using a parametrization of Solomon and Qian (2005) and observed solar spectral irradiance.Additionally, 6 ionizing, dissociating, and exciting reactions represent the direct influence of precipitating primary and sec-775 ondary particles on thermospheric chemistry.Corresponding reaction rates are calculated using the particle induced ionization rates and branching ratios given by Roble and Ridley (1987) and Rusch et al. ( 1981).Below 10 -3 hPa, particle im- pact on chemistry is represented by the production of N( 2 D), 780 73o N(4 S) and HO.. Here, HAMMONIA uses parametrizations of Jackman et al. (2005a) based on formulations of Porter et al. (1976) and Solomon et al. (1981).The simulations use 67 pressure levels and a horizontal resolution of T31 (96 longitudes x 48 latitudes).Output is provided every 2 hours.785 Up to 179hPa, the model is relaxed to ECMWF analyzed temperature, divergence, vorticity, and surface pressure.

KASIMA
The KASIMA model is a 3D mechanistic model of the mid-700 dle atmosphere including full middle atmosphere chemistry (Kouker et al., 1999).The model can be coupled to specific meteorological situations by using analyzed lower boundary conditions and nudging terms for vorticity, divergence and temperature.Here we use the version as described by Reddmann et al. (2010).It has a horizontal resolution of ab0ut795 5.6 0 x 5.6° with 63 pressure levels between 7 and 120 kin and a vertical resolution in the lower stratosphere of 750m, gradually increasing to 3.8 km at the upper boundary.The frequency of output is every 6 hours.The model is nudged to ECMWF analyses below I hPa.A numerical time step of 8o0 12 min was used in the experiments.The chemistry is calculated up to 90 km, above which only transport is applied.The chemistry uses JPL 2002 data (Sander et al., 2003) and is calculated up to 90 km, above which only transport is applied.
The chemical fields of long-lived tracers have been initial-805 ized from a multi-annual run starting in the year 1960.For the HEPPA experiments, the transport scheme has been revised to allow transport of the members of chemical families NOx and HO, individually in the mesosphere.In addition, the ozone hearing rate is calculated interactively.The rate 8,o constants of the gas phase and heterogeneous reactions are taken from Sander et al. (2003).For the production of HO, the parameterization of Solomon et al. (1981) is used, for the production of NO,;, 0.7 NO molecules are produced per ion pair and 0.55 N atoms in ground state, including reactions8,5 of N+02 , NTNO, N -I-NO2 .The HNO3 production from proton hydrates (de Zafra and Smyshlyaev, 2001) has been modied to be dependent on actual ionization rates.

EMAC
The ECHAM5/MESSy Atmospheric Chemistry (EMAC) model is a numerical chemistry and climate simulation system that includes sub-models describing tropospheric and middle atmosphere processes and their interaction with oceans, land and human influences (Mckel et al., 2006) (2008).For the HEPPA comparison, two versions of SOCOL have been used.One is with parameterized production of odd nitrogen, where for the NO, sources, the fact that 1.25 NO molecules were produced were taken into account (Porter et al., 1976), for the HOX sources, the table given by Solomon et al. (1981) has been used.The other version (SOCOLi ) includes the chemistry of ionized species.SOOCLi is described in Egorova et al. (2010).As sources for ionization the model uses galactic cosmic rays (Heaps, 1978), energetic electron precipitation, solar proton events and observed solar irradiance.SOCOLi takes into account 580 reactions involving 43 neutral of the oxygen, hydrogen, nitrogen, carbon, chlorine and bromine groups, electrons, 31 positive and 17 negative charge species including clusters of 02 + , H+ and NO-.The rate constants of the gas phase and heterogeneous reactions are taken from 85o Sander et al. (2000).SOCOL and SOCOU models apply additionally to electron and proton-induced ionization also ionization rates caused by alpha-particles provided by AIMOS model (Wissing and Kallenrode, 2009).This choice is based on the assumption that AIMOS model describe all physical 855 processes relevant to particle precipitation during the event.
Output is provided at the local time and location of the Ml-PAS overpass.Above that level the forcing is reduced linearly, so that the model is free-running between 50 km and the model top at approximately l 35 km (4.5x 10 -6 hPa).Heating rates and photolysis are calculated using observed daily solar spec-845 tral irradiance and geomagnetic activity effects in the auroral region are parameterized in terms of the Kp index (Marsh et al., 2007).A description of simulations of the effects Sao of solar proton events using an earlier free-running version of WACCM and comparison with measurements is given in Jackman et al. (2008Jackman et al. ( , 2009)).The standard WACCM chemistry is described and evaluated extensively in WMO (2010).Reaction rates are from Sander et al. (2006).For these simulations we have modified the N+NO2 reaction to include two additional pathways as described in Funke et al. (2008).
It should be noted that both WACCM and HAMMONIA use the same chemical solver based on the MOZART3 chemistry Kinnison et al. (2007), include the same set of ionized species, and use the parameterized EUV ionization rates from Solomon and Qian (2005).For these simulations the latter parameterization has been extend to include the photoionization of CO2 in the EUV.Proton and electron ionization rates, used in the nominal simulation, are taken from AIMOS, however above 5 x 10 -4 hPa (-100 km) ionization from electrons is instead calculated by the WACCM parameterized aurora.An additional simulation using proton ionization, only, has also been performed (in the following denoted as WACCMp).The HO, production per ion pair is included in WACCM using a lookup table from Jackman et al.
(2005b, Table 1), which is based on the work of Solomon et al. (1981).It is assumed that 1.25 N atoms are produced per ion pair and divide the N atom production between ground state, N(4 S), at 0.55 per ion pair and excited state, N( 2 D), at 0.7 per ion pair (Jackman et al., 2005b;Porter et al., 1976).In this study, WACCM constituent and temperature profiles were saved at the model grid point and time-step (model time-step is 30 minutes) closest to each of the MIPAS observation locations.

Intercomparison method
In order to reduce errors related to the different sampling of the MIPAS observations and gridded model data (i.e., B3dCTM, CAO, FinROSE, HAMMONIA, KASIMA, and EMAC) , we have linearly interpolated the model results to  the MIPAS measurement locations and times, as well as to the corresponding pressure levels of the vertical retrieval grid 915 885 of the species under consideration.This approach has the further advantage that diurnal variations of particular species are implicitly taken into account.Comparison of MIPAS measurements and model results requires the transformation of modeled profiles to MIPAS altitude resolution.Based on the B9a formalism by Rodgers (2000), we calculate the model profiles adjusted to MIPAS resolution x ad,, as 920

X ad, =AX mod+(I-A ) X a > (1)
where A is the MIPAS averaging kernel matrix, x,,od is the original model profile, I is unity, and xa is the a priori in-995 formation used in the MIPAS retrievals.Assuming that the 925 altitude resolution of the models is much finer than that of the MIPAS retrievals, the comparison of x adj and MIPAS measurements is not affected by any smoothing error.This procedure has been applied to each model result re-900 sampled at the corresponding measurement location.As93o an example, Figure 6 compares HOCI zonal mean distributions at 40-90°N, averaged over the period 29 October to 4 November 2003, as observed by MIPAS and as modeled by WACCM with and without application of averaging ker-9r,5 nels.In the latter case, the vertical distribution is broader935 and slightly shifted towards lower altitudes, similar to the retrieved MIPAS profiles.Also, the absolute vmr peak values are smaller than without application of the averaging kernel.
It should be noted that the apparent better agreement of the 910 maximum vmr values between MIPAS and the uneonvolveda4a WACCM simulations are related to the fact that background HOCI vmrs are underestimated in the model.The relative vmr increase related to the SPE is in better agreement when comparing observed and convolved model data (see also Section 6).
6 Results and discussion

Meteorological background conditions
Meteorological background conditions, particularly the thermal structure and the prevailing dynamics, can have an important impact on the magnitude and spatial distribution of SPE-induced composition changes.Temperature differences between models and observed data have, on the one hand, a significant impact on SPE-related chemistry due to involvement of highly temperahrre-dependent reactions (i.e., N+02 or NO2+03 ).On the other hand, meridional transport and mixing, depending largely on the development stage of the early winter polar vortex, control the redistribution of air masses between polar night and illuminated regions and hence, the efficiency of photochemical losses.Also, the strength of polar winter descent plays an important role in the vertical redistribution of some species on the time scale of the intercomparison period.
Figure 7 shows the MIPAS temperature zonal mean distribution at 40-90°N averaged over the period of the main proton forcing, 29 October to 4 November 2003, and the corresponding differences between the models and the observations.It is evident that models which are driven or strongly forced by assimilated meteorological data tip to the upper stratosphere (i.e., B3dCTM, FinROSE, KASIMA, and WACCM) reproduce reasonably well the observed temperatures below approximately 1 hPa.On the other hand, free-running models (B2dM, CAO and SOCOU) and those which are nudged to meteorological in the troposphere,   only (EMAC and HAMMONIA) tend to overestimate the observations inside the polar vortex by more than 15 K around approximately 1 hPa or slightly below.Slightly too high stratopause temperatures are found in EMAC, CAO, B3dCTM, and FinROSE simulations.In the polar meso-,,, sphere, temperatures are underestimated by HAMMONIA (up to 15 K) and SOCOU (more than 25K).The temporal evolution of observed polar temperatures (70-90°N) and the corresponding differences between model and observations are shown in Figure 8.No significant trend in both, observa-,65 tions and model data, can be observed during the period of interested, while short-term temperature fluctuations of more than 10 K compared to the observations, most likely related to differences in the planetary wave activity, show up particularly in the case of the free-running or weakly nudged models (B2dM, CAO, HAMMONIA, and SOCOU).
Differences in the magnitude of meridional redistribution between models and observations have been assessed by comparing CH 4 zonal mean distributions provided by all models except CAO.Since the global stratospheric CH4 abundances differ noticeably among the models, we used the relative meridional CH4 anomaly as indicator for meridional redistribution rather than absolute vmrs.Figure 9 shows the observed and modeled meridional CH 4 anomalies at 40-   90°N averaged over the whole period.A pronounced gradient in the observed anomalies around 60°N indicates the 970 early winter vortex boundary.In general, the vortex bound-9B9 ary position is well reproduced by all models, although there are significant differences in the overall CH 4 change from mid-latitudes to the pole between the models.Strongest latitudinal gradients (i.e., weakest redistribution) were found 975 in the KASIMA simulations, while smallest gradients (i.e., 985 strongest redistribution) are visible in HAMMONIA.The reason for the underestimation of meridional redistribution in KASIMA is not fully understood, particularly because other ECMWF-driven models have simulated considerably stronger mixing.The too strong mixing in HAMMONIA is most probably related to wave-1 activity, being present in the whole time period.The vertical distribution of the observed CH4 meridional anomaly shows a broadening in the stratopause region (1-0.1 hPa), indicating a weakened transport barrier at the vortex top.This behavior is reproduced by the models in general although there are differences with respect to the altitude and magnitude of the broadening region.
In B2dM, it is shifted slightly upwards while the opposite is observed in EMAC, SOCOL, and SOCOU simulations.In these latter models, meridional redistribution seems also to be slightly overestimated around the stratopause.It should be noted that our analysis of CH 4 meridional anomalies does not allow to distinguish between meridional redistribution by eddy diffusion and large-scale transport by planetary waves, the latter being of higher importance for the redistribution of air masses between polar night and illuminated regions.
The variability of the polar vortex strength has been assessed by comparing the temporal evolution of the relative change of CH4 abundances with respect to 26 October averaged over 70-90°N (see Fig. 10).The observed evolution indicates a vortex intensification and descent in the lower and middle stratosphere while a CH 4 increase above 0.3 hPa, particularly during the proton forcing at the beginning of November, hints at an increase of meridional mixing in the mesosphere.This general behavior is qualitatively reproduced by the models, although important differences with respect to the vertical structure and magnitude exist.These differences have to be taken into account when analyzing the temporal evolution of SPE-induced composition changes Carbon monoxide is an ideal tracer for upper stratospheric and mesospheric dynamics.Particularly, it allows to identify air masses which have descended from the upper mesosphere and contain enhanced NO , , related to energetic electron pre-so45 cipitation (EEP).Since polar winter descent of NO x generated by EEP prior to the SPE event, is not resolved by all models and since we focus here on SPE-related effects, observed NO, enhancements due to descending upper mesospheric air masses perturb our analysis and should hence boo5o excluded.MIPAS CO observations provide an excellent criterion for identification of EEP-related enhancements (see Section 6.2).
CO distributions also allows for the characterization of descent and vortex perturbations by large-scale wave activityo55 and isentropic mixing across the vortex boundary in the upper stratosphere and mesosphere.Figure I 1 shows the observed and modeled temporal evolution of CO abundances average over 70-90°N.In general, the continuous decrease in altitude of CO vmr isolines in the upper stratosphere,oso related to polar winter descent, agrees well in models and data.Around 1 hPa, polar air masses descended approximately 5 km in both models and observations during the time period under investigation.
A higher variability is found in the mesosphere.Ob-1035 served CO abundances decreases around the beginning ofio65 November, at the same time when CH4 increased significantly (see Fig. 10).A pronounced CO increase occurred around 20 November, hinting at enhanced descent and vortex intensification.Modeled CO distributions show a different temporal evolution in the mesosphere, although some similarities can be found.For instance, EMAC, KASIMA and07o WACCM reproduce the CO increase in late November, how- ever, with a smaller magnitude and slightly shifted in time.A CO decrease around 0.1 hPa at the beginning of November, as observed by MIPAS, is also visible in SOCOL, SOCOU, and -to a lesser extent -in WACCM simulations.
In contrast to the observations, these modeled decreases occur nearly instantaneously on 31 October, suggesting that the simulated CO changes might be related to the proton event rather than dynamical modulations.Indeed, CO is removed by the reaction with OH, which is strongly enhanced during the SPE at nighttime.The isolation of a possible SPEinduced chemical CO loss from dynamical effects is difficult in both, observations and simulations.Nevertheless, we have analyzed the observed CO abundances at fixed CH4 levels in the vertical range of 0.2-0.05hPa in order to exclude CO variations related to isentropic mixing or meridional redistribution.CO abundances observed simultaneously with CH4 vmrs of less than 40 ppbv decreased by approximately 1 ppmv from 29 October to 1 November, thus suggesting a chemical removal of the order of 10% which could be related to enhanced OH.The CO decreases found in the WACCM simulations have a similar magnitude, while SOCOL and SOCOU simulations show a CO decrease around 30%.

Enhancements of NO Y and N20
The most important impact of proton precipitation on the middle atmosphere is the immediate formation of NO, (_ NO + NOz) via dissociation of molecular nitrogen by ionization and subsequent recombination with oxygen.Due to its relatively long chemical lifetime in the stratosphere, SPEinduced NO, enhancements have a strong potential to de-Funke et al.: HEPPA intercomparison study plete ozone on a mid-to long-term scale via catalytic cycles.A fraction of excess NOx produced by proton forcing is subsequently buffered into NO, reservoir species (i.e., N205, HNO3 , and CIONO 2) by a series of chemical processes (see next section) at different time scales.In general, the NO, deactivation is very slow in the upper stratosphere.At lower altitudes, however, observed HNO 3 and CIONO2 increases immediately after the onset of the proton forcing during the Halloween event indicate a much faster conversion.In order to assess the agreement of observed and modeled SPErelated odd nitrogen enhancements, we have thus compared, at first instance, total NO,, (= NO + NO 2 + HNO3 + 2N205 + CIONO 2 + HN04) rather than NO,.Since meridional redistribution is an issue (see discussion in the previous subsection), we have analyzed area-weighted averages of NOy enhancements with respect to 26 October within 40-90°N, i.e., in an area that entirely covers the source region.
As a first step, we analyze the instantaneous NOy enhancements during the main proton forcing around 29 October-t November.Figure 12 shows the observed and modeled NOy enhancements during this period, ranging from a few ppbv in the middle stratosphere to several 100 ppbv in the mesosphere.The agreement between the observations and the multi-model average, the latter providing a measure of the overall ability of current atmospheric models to reproduce SPE-related NOy increases, is reasonable, exhibiting differences below 30% in the whole altitude range.There is, however, a systematic overestimation(underestimation) of the models around I hPa (above 0.3 hPa).

1130
The NOy underestimation of the models above 0.3 hPa could be related to an overestimation of NO photolysis, the principal NOy loss mechanism in the sunlit mesosphere.It has been pointed out by Minschwaner and Siskind (1993) that absorption of solar irradiance by thermospheric NO, be-0135 ing significantly enhanced during SPEs, has an important impact on the photolysis rates of nitric oxide in the middle atmosphere.
The systematic behavior of the NO y overestimation around 1 hPa suggests that these differences are related -ab140 least partly -to the simulated ionization rate profile.In this pressure range, there are uncertainties in the modeling of electron precipitation at 300 keV to 5 MeV.Electrons contribute there to the total ionization within 40-90°N by approximately 15%.As the highest electron channel on POES,14e does not provide data up to 5 MeV, the energy spectra was extended according to Klassen et al. (2005).However, the overestimated NO, at I hPa questions the extended spectra.In addition, the energy range of the highest electron channel rnepM is not known for sure (private communica -1150 tion, Janet Green, NOAA) and it might be smaller than the published 300 keV-2.5 MeV (Evans and Greer, 2000).A smaller energy range would give an increase to NO y production at 0.1 hPa.A possible overestimation of electron 1125 ionization is also supported by the better agreement of thel55 WACCM simulation without electrons (WACCMp) with the NO, production efficiency (ratio of the net NO, increase and the integrated initial N production) during the period of the main proton forcing (28 October-1 November) from box model calculations for night-and daytime conditions (solid and dashed black lines, respectively), assuming initial atmospheric conditions as observed by MIPAS in the polar cusp region.The following variations for dark conditions are also shown: a 20 K temperature increase (red) and decrease (blue), a factor of 2 increase of 0 3 (solid green), and initial NO x abundances set to zero (dashed green).observations.Above 2 hPa, this simulation shows approximately 20% less enhanced NO y than the nominal simulation, including protons and electrons.Additional ionization by alpha particles, included in CAO, FinROSE, SOCOL, and SOCOU contributes only by approximately 5% to the total ionization within 40-90°N, hence increasing the SPE-related NOy enhancements only marginally.
On the other hand, there is an important spread of up to 100% among the modeled NO, enhancements.This is surprising, given that all models use the same AIMOS ionization rates and, thus, rather similar modeled NO y enhancements are expected at least during the first days of the main proton forcing.NO y enhancements are most strongly overestimated (up to 100%) by SOCOLi, SOCOL, and CAO in the stratosphere around 1 hPa.In the mesosphere, smallest NO, increases are obtained by B2dM and EMAC (up to 80% less than observed), while SOCOL and SOCOLi simulations overestimate the MIPAS enhancements by 50%.
In order to investigate possible reasons for the spread among the model results, a more detailed look into the NO, production mechanism is required.Generally, it is assumed that each ion pair produces 1.25 atomic nitrogen atoms, distributed between the electronic ground state N( 4 S) and the excited N( 2 D) state with a branching ratio of 0.45 and 0.55, respectively (Jackman et al., 2005b).The value of 1.25 atomic nitrogen atoms per ion pair has been adapted by all models involved in this study, except for EMAC and SO-COLL In the latter model, N production is implicitly modeled by means of the involved ion chemistry scheme.In EMAC, an altitude-dependent N production has been as-Fig.14.Spatial distributions of observed and modeled NO y at I hPa averaged of the period of the main proton forcing during 30 October-I November.The average precision of MIPAS observations is also shown (upper second panel from the left).
sumed which has been determined empirically by the adjustment of the simulations to observed NO2 and N2 0 abun- dances (Baumgaertner et al., 2010).The resulting N production profile is slightly higher than that used by the othemao models in the upper stratosphere (around 1.5 N per ion pair) and considerably lower in the mesosphere (less than 0.3 N per ion pair) which explains to a major extent the behavior of the EMAC NO Y enhancements compared to other models.
An important source of variability in the NO y productioni85 is related to the reaction paths of the produced atomic nitrogen in its ground and excited states.While the reaction of N( 2 D) with oxygen to form NO is very fast such that practically all N( 2 D) is immediately converted to NO below the thermosphere, the corresponding reaction of the nitrogenl6o ground state

N(4S)+02 --} NO +O (R1)
is slower and highly temperature-dependent.Hence, it com,,95 peter with other reactions, namely: both destroying NO,;.As a consequence, only a fraction of the initially produced NO, remains available after the proton forcing.This fraction depends strongly on temperature due to reaction R1 and to a lesser extent on the repartitioning between NO and NO 2 , driven by illumination and odd oxygen availability.In order to assess the sensitivity of the SPE-related NO, production to these parameters, we have integrated the relevant chemical equations for the period 28 October-1 November with a simple box model including AIMOS ionization rates and assuming initial atmospheric conditions as observed by MIPAS at 70-90°N before the SPE, as well as the N( 4S)IN(2 D) branching ratio recommended by Jackman et al. (2005b) .The modeled NO, enhancements have then been compared to a similar simulation, but setting the rate coefficient for reactions R2 and R3 to zero (i.e., assuming that all initially produced NO survives).
The ratio of both simulations reflects the NO, production efficiency.It is shown in Fig. 13 for nighttime and daytime conditions (solid and dotted black lines, respectively), showing maximum value of 0.55-0.7 around the stratopause and smaller values (0.15-0.4) above and below.Reduced values below the stratopause are related to the background NO;,: if initial NOX abundances are set to zero, the production effi-  1200 ciency increases with pressure to values close to unity in the red and blue lines in Fig. 13).On the other hand, assuming Lower stratosphere (see Fig. 13, dotted green line).A temper,zcs a two times higher ozone abundance results in an increase of ature increase (decrease) of 20 K results in an enhancement the NO, production efficiency by only a few percent.(reduction) of this quantity by approximately 30-50% (see The chemical scheme described above (including a N(4S) and N ( 2 D) branching ratio of 0.45 and 0 .55) has been employed in most of the atmospheric models included in the intercomparison, with some exceptions: B3dCTM and CA0250 use a family approach which implies the immediate conversion of all atomic nitrogen to NO (equivalent to a ratio of 1 in Fig. 13), explaining -at least partly -the relatively high NO,, increases above 2 hPa in these models.Also Fin-ROSE applies a family approach, however, in this model it is1255 implicitly assumed that all N( 4S) produced by ionization destroys NO via reaction R2, resulting in a net NO, production of 0.25 per ion pair (i.e., an altitude-independent production efficiency of 0.2 in Fig. 13).However, although a considerably smaller NOy production is hence expected, FinROSE1260 model results show more excess NO Y than found in the observations.EMAC uses a N( 4 S) and N( 2 D) branching ratio of approximately 0.2 and 0 .8, respectively.Box model calculations using this atomic nitrogen branching yield weakly altitude-dependent NO, production efficiencies of 0.6 -0.8,1265 considerably higher than the nominal efficiency of -0.2 in the mesosphere.Therefore, the smaller atomic nitrogen production in the mesosphere applied in EMAC is partly compensated by the modified N( 4 S) and N( 2 D) branching ratio.
As shown above, temperature differences might explaim270 the differences of the NO Y enhancements simulated by the remaining models.B2dM underestimates the observed temperatures in the mesosphere by about 15 K, consistent with the relatively low NO Y enhancements compared to the other models and observations, there.In contrast, HAMMONIAl275 and SOCOU simulations, exhibiting relatively low mesospheric temperatures, show much larger NO Y enhancements.Stratospheric temperatures are significantly overestimated by B2dM, CAO, HAMMONA, and SOCOLL However, only the latter model shows stratospheric NO Y enhancements weII1260 above the model average.Thus, temperature differences among the models cannot be the only reason for the spread encountered in the modeled NOY enhancements.Therefore, we have looked at the spatial NO Y distribution in order to investigate if the spread in the modeled N Oy1265 could also be related to dynamical effects.Figure 14 shows the observed and modeled NOY distributions in the upper stratosphere ( 1 hPa) averaged over the period 30 October- s November.The spatial extension of the modeled NO Y enhancements exhibits pronounced differences.In some cases, NO Y enhancements are confined to the polar region northward of 70°N (i.e., B2dM, KASIMA, FinROSE) while in other cases they extend even to regions equatorwards of 50°N (i.e., SOCOL and SOCOLi).Taking into account that the spatial extension of the source region is the same in all models, these differences must be related to transport acting on a very short time scale.As discussed above, SOCOU shows higher NO Y averages than other models with similar stratospheric temperatures (i.e., HAMMONIA, B2dM).The spa- tial NO Y distribution of SOCOU at I hPa indicates strong wave activity resulting in a deformation of the pole-centered shape of the NO Y distribution.Thus, it cannot be excluded that tropical NOY' transported into the 40-90°N region, contributed to the large NO Y enhancements identified in this simulation.Further, the fast transport of SPE-generated NO, out of the source region in the SOCOU simulations might result in a higher net NO, production since NOx destruction by reactions with atomic nitrogen (R2 and R3) is then less efficient.
During the following month, the SPE-induced NO Y enhancements were transported downwards with the meridional circulation, forming a NO Y layer around 45 km at the end of November (L6pez-Puertas et al., 2005a).At the same time, NO., generated by continuous EEP in the lower thermosphere, reached the upper stratosphere and began to merge with the upper part of the SPE-induced layer (see also Fig. 1).This behavior is not reproduced by the models since low and mid-energy EEP is not included in the majority of the models.In order to facilitate the comparison of observed and modeled SPE-induced NO Y enhancements in the following month after the proton forcing, we have excluded those parts of all observed and modeled NO Y profiles where MIPAS CO abundances were higher than I ppmv.This value has been chosen such that the major fraction of EEP -induced NO Y enhancements has been filtered out without removing too many MIPAS locations, particularly at higher altitudes.
Figure 15 shows the temporal evolution of the observed Fig. 17.Spatial distributions of observed and modeled NO,, at 2 hPa averaged of the period 20-27 November.and modeled NOY enhancements (related to the SPE, only) with respect to 26 October within 40-90 QN for the following month.While the magnitude of the enhancements is generally larger than in the observations and further shows a significant spread related to the differences in the NOY pro ;325 duction during the proton forcing (see discussion above), the observed evolution of the SPE-induced NO, layer is well reproduced by all models in terms of vertical distribution and relative vmr decrease.A more detailed look into the temporal NO Y evolution of individual models shows that smaller,33o fluctuations can be attributed to dynamical variability.
It is interesting to notice that the WACCM simulation without electron-induced ionization yields a much better agreement with the observations than the nominal simulation throughout the period under investigation (see Figure 16).1335Additional NO Y buildup related to electron-induced ionization is even more pronounced during the second event (4-5 November) below 0.4 hPa compared to the main proton forcing (see right panel of this Figure ).This excess production during the second event dominates the NO Y overesti ,,,o mation of 5-10 ppbv encountered in the nominal WACCM simulation during the following weeks.The meridional distributions of the observed and modeled NOY enhancements exhibit important differences towards the end of November (see Figure 17, showing NO Y1 5 distributions at 2 hPa averaged over 20-27 November).The observed and modeled latitudinal gradients correlate well with the meridional CH4 anomalies (see Fig. 9), highlighting the important role of mixing and large-scale transport.The meridional redistribution of the SPE-induced NO Y en-,35o haneements, particularly the transport out of the polar night region, has important implications on the NOY repartitioning 1320 which is to a major part driven by photochemistry (see next subsection).
An interesting detail of the observed evolution of SPEinduced NO, enhancements (Fig. 15, upper left panel) is the appearance of several "spikes" at mesospheric altitudes, which are temporally correlated with peaks in the ionization related to high energy (>300 keV) electron precipitation, the most pronounced event occurring on 21 November.Ionization by high-energetic electrons is included in the models which, however, do not reproduce such sudden NO Y increases.It is therefore unlikely the observed mesospheric NOY peaks are related to in situ production by EEP associated to the Halloween event.Instead, they could be related to residual contributions of descending NO, from the upper mesosphere which have not completely been filtered out.It should be noted that the observed CO temporal evolution (see Figure 11) indicates particularly strong descent around 20 November.
A fraction of the NO, deactivation by reaction with atomic nitrogen during the proton forcing discussed above occurred via reaction R3, giving rise for the buildup of N 2 0. Upper stratospheric and mesospheric nitrous oxide increases up to 7 ppbv have been observed by MIPAS during the Halloween SPE and have been attributed to this reaction channel (Funke et al., 2008).Reasonable agreement with CMAM model calculations has been obtained by assuming that only half of the products of reaction R3 is N 2 0 and 0, while the other half is N2 and 02 .Figure 18 shows the observed and modeled N20 zonal mean enhancements averaged over the period of the main proton forcing (29 -31 October).Except for FinROSE and B2dM, which do not include the reaction channel R3, N2 0 increases are simulated by all models.The observed enhancements, however, are generally overestimated by a factor 2 to 10, except for EMAC which shows smaller N 2 0 in- creases than observed by MIPAS.In the latter model, thisi375 can be clearly attributed to the modified N( 4 S) and N(2D) branching ratio (see discussion above).Except for WACCM, the remaining models do not include the additional reaction channel of R3, responsible for the formation of N2 and 02, which has been included in the CMAM simulations (Funke et al., 2008).But even when taking into account a reduction by a factor of 2 of the simulated enhancements, these mod,,,, els tend to overestimate the observations and further show a significant spread among the individual results.As in the case of NOy , also the total SPE-induced N2 0 production depends on temperature, NO, partitioning, and dynamical redistribution.However, a dominant relationship of none of these quantities with the differences of the magnitudes of th0385 modeled N2 0 increases can be established.

Repartitioning of nitrogen species 1390
After having assessed the observed and modeled total NO, and N 2 0 enhancements generated by the Halloween event, we analyze in this subsection the repartitioning of initially produced nitric oxide into other NOy species in the aftennath of the SPE.1395

NOx
The conversion of the excess NO generated by the proton forcing into NO 2 acts on a very short timescale (seconds to minutes) and is controlled at dark conditions by the reactions giving rise to a NO2/NO, ratio close to one in the stratosphere, but decreasing in the mesosphere due to the availability of atomic oxygen.This decrease occurs at higher altitudes in the polar night region compared to midlatitudes.Figure 19 shows the observed and modeled nighttime NO 2/ NOX ratios averaged over the initial SPE period.The observed decrease of this ratio above 0.3 hPa at midlatitudes and 0.1 hPa in the polar region is generally well reproduced by the models which resolve the mesosphere, except for B2dM and EMAC, which both overestimate the polar NO 2 fraction at these altitudes.The higher mesospheric NO 2 abundances in these two models might be related to lower atomic oxygen concentrations at high altitudes and/or less efficient mixing between polar night and illuminated regions.At sunlit conditions, photolysis of NO 2 and higher atomic oxygen abun- dances shift the NO2/ NO, ratio to lower values compared to dark conditions.Figure 20 shows the observed and modeled daytime ratios.The observed values are well reproduced by B2dM, SOCOL, and SOCOU, while other models tend to overestimate the polar upper stratospheric and meso1420 spheric NO2 fraction close to the terminator.These differences in the NO, partitioning among the models and observations highlight the difficulties in drawing conclusions on the SPE-induced total NO, enhancements from the comparison if only one of its components is considered.Figure 21 shows the observed and modeled temporal evolutions of the N20 5 enhancements with respect to 26 October averaged over 70-90°N.A stratospheric N 2 05 buildup, being most pronounced in the second half of November, is simulated by all models, qualitatively reproducing the observed behavior.The modeled N2 05 increases are, however, generally overestimated (except for KASIMA) and exhibit a wide spread among the models.Taking into account that the magnitude of the N2 05 increase depends on various factors such as NO,; availability, temperature, ozone abundances, and the efficiency of N2 05 -> HNO 3 conversion (see below), a large spread of the model results is expected.B2dM and EMAC, however, overestimate the observed N 2 05 increases by factors of 4 and 6, respectively.While in the case of B2dM the extraordinarily high N 2 05 amounts can be explained by the very pole-centered distribution of the precursor NOx , implying insignificant photochemical Losses in the source region (see also discussion below), the reason for the unreasonably high N 2 05 abundances of ap to 12 ppbv in the case of EMAC is still under investigation.N 2 05 enhancements simulated by CAO until 4 November are likely to be caused by seasonal variations rather than by the SPE.1465 Two distinct HNO3 enhancements were observed by MI-PAS in the aftermath of the Halloween SPE (Lopez-Puertas et al., 2005b).The first one, reaching vmrs around 2 ppbv, occurred immediately after the SPEs at altiudes above 40 km and has been initially attributed to the gas-phase reaction 1411 NO2 + OH + M -^ HNO 3+ M. Verronen et al. (2008), however, have shown that the instantaneous HNO 3 increase after the proton forcing can only be reproduced by model calculations including ion-ion recombination between NO 3and H+ cluster ions.The second enhancement of 1-5 ppbv started around 10 November and lasted until the end of De 1475 cember.Also in this case, attempts to reproduce the magnitude of the observed increases by model calculations including gas phase chemistry only, have failed (Jackman et al., 2008).1180 Figure 22 shows the observed and modeled temporal evolutions of the HNO 3 enhancements with respect to 26 October averaged over 70-90°N.Consistent with previous findings, the first instantaneous enhancement is considerably un-derestimatedby all models, except FinROSE, which includes the ion chemistry proposed by Verronen et al. (2008).This model, however, overestimates the observed increases by up to a factor of 3. The overestimation below.50 km is surprising, given that 1-D simulations with the Sodankyla Ion and Neutral Chemistry (SIC) model which includes the ion-ion recombination were found to be in good agreement with the same MIPAS dataset at these altitudes.It should be noted, however, that different ionization rates have been used in the SIC calculations of Verronen et al. (2008), and -probably even more important -that FinROSE uses a parameterization of the ion-ion recombination included in the full ion chemistry scheme of the SIC model.
The second enhancement, occurring around 15 November at 1-2 hPa, is only reproduced by KASIMA, however, overestimating the observed increases by a factor of 3. Contrary to other models, KASIMA simulations account for HNO3 formation via water cluster ions (B6hringer et al., 1983) combined with heterogeneous reactions on sulfate aerosols by means of a parameterization provided by de Zafra and Smyshlyaev (2001).At lower altitudes (i.e., below 10hPa), midterm HNO3 increases are visible in the observations, as well as in the B2dM, 133dCTM, EMAC, FinROSE, and ,485 WACCM model results.These increases are not related to the SPE and can be explained by seasonal variations, In order to assess the repartitioning of the main NOy,15,o species towards the end of November in a more quantitatively way, we have analyzed their relative contributions to 1490 the total NO y .This is necessary because of the encoun- tered differences in the total amount of SPE-induced excess NOy among the different models and the observations.Duc,5,5 to the observed conversion of N2 05 into HNO 3 , we have looked, as a first step, at the relative contribution of the sum 1495 of both reservoir species to NO y .Observed and simulated zonal mean (2N 2 05+HNO 3)INO v ratios, averaged over the period 15-30 November, are shown in Fig. 23.The observed5ze ratio of 0.28 at the peak height (w0.2 hPa, see Fig. 15) of the NOy enhancements in late November (indicated by a 1500 black line in Fig. 23) is very well reproduced by all models, except B2dM and EMAC.As discussed above, the disagreement found in these models is produced by too efficient buildup of N 2 05 (see Fig. 21).The differences OF525 the (2N20 5+HNO 3)iNOy ratio in the observations and the 1505 B2dM simulations are, however, much less pronounced than those encountered in the absolute N 2 05 abundances: while B2dM N2 0 5 exceeds the observed amounts by a factor of 4, the modeled 2N 2 05+HNO3 contribution to NO y at its peak height is around 40%, exceeding the observed contribution by only a factor of 0.5.The N 2 0 5 overestimation in this model is hence mainly related to the higher amounts and more pole-centered distribution of the precursor NO 2 , In contrast to B2dM , EMAC simulations obtain more than 90% of the available NO y at its peak altitude in the form of N205.This contribution decreases with altitude, but still exceeds 30% in the mesosphere.Other models show, in some cases, a minor overestimation of the reservoir species fraction which can be partly explained by differences in the modeled temperatures and ozone abundances, controlling the efficiency of reaction R6.
The repartitioning between HNO 3 and N 2 05 has been assessed by comparing the observed and modeled zonal mean HNO31(2N2 O5 +HNO3) ratios averaged over the period 15-30 November (Fig. 24).As expected, the observed ratio is strongly underestimated above approximately 10hPa by all models, except KASIMA.The qualtitative agreement of KASIMA simulations and MIPAS observations is very good, particularly regarding the vertical shape of this ratio.The Fig. 22. Temporal evolution of area-weighted averages of HNO3 changes with respect to 26 October 2003 in WAS observations and model simulations at 70-90°N, as well as differences between modeled and observed averages.Solid contour lines reflect 1 ppbv steps.The significance of observed HNO 3 changes (in units of ( 7) is shown in the upper second panel (from the left).The MIPAS ratio of 0.28, encountered at altitude of the maximum of the SPE-induced NO,, layer, is indicated by a black line.Fig. 24.Zonal mean HNO 3 /(2N205+HNO3) ratios averaged over the period 15-30 November in MIPAS observations and model simula- tions.Regions with observed vmrs of 2Nz0s+HNO 3 smaller than 0.1 ppbv have been omitted.modeled ratio, however, exhibits a positive bias of 0.2 with respect to the observations, most pronounced in the polar re-560 gion.We conclude that the HNO 3 formation via water cluster ions and/or heterogeneous reactions on sulfate aerosols, both included in KASIMA by means of the parameterization of de Zafra and Smyshlyaev (2001), is the responsible mechanism for the observed HNO 3 enhancements in late Novem ;565 her.However, some further work is required to adjust the parameterization quantitatively to the measurements.

Minor NO,, species
Also minor NOy species were found to be enhanced in the of-1570 termath of the Halloween SPE due to the repartitioning of initially produced NO.. L6pez-Puertas et al. (2005b) reported CIONO 2 enhancements up to 0.4 ppbv a few days after the proton forcing from MIPAS observations.These observations are compared to the model simulations in Section 6.6,75 together with observations of other chlorine species.MI-PAS has also observed enhanced HN0 4 during the first days of the Halloween SPE which have not been reported so far.These increases can be attributed to the termolecular reaction Since at the altitude of the HN0 4 enhancements (around 2-3 hPa) SPE-related increases of the precursor NO 2 are relatively small (-2 ppbv) compared to the background NO2 abundance, the observed HN0 4 changes are mainly driveru1565 by enhanced H09 abundances, and hence, represent an indicator of SPE-generated HOx in the middle stratosphere.At dark conditions, HO. is in steady state even during a SPE, and its abundance is hence directly proportional to atmospheric ionization.Stratospheric HN0 4 is destroyed duringsan the day by photolysis and by reaction with OR Nighttime losses are negligible under quiescent conditions, and even during SPEs, OH-related HN0 4 destruction is small compared to its production via reaction R8.Due to problems with the gain calibration, particularly affecting this species (see discussion in Section 2), we restrict our analysis to data from the gain calibration period 28 October -5 November.Figure 25 shows the observed and modeled zonal mean distributions of HN0 4 vmrs during the first four days of the proton forcing (29 October-1 November).Model results for pre-SPE conditions (26 October) are also shown.HN0 4 model output is not available from CAO, EMAC, HAMMONIA and FinROSE.Polar upper stratospheric enhancements of up to 0.18 ppbv are visible in the observations on 29 October, decreasing until I November by about 20%.The HN0 4 enhancements are also simulated by the models in the first days of the SPE, however, generally overestimating the measurements.The overestimation is most pronounced in the B2dM, B3dCTM, and WACCM simulations (a factor 2-3), while SOCOU shows smaller HN0 4 increases.The HN04 peak height is located at somewhat lower altitudes in SOCOU which might be related to the relatively high abundances inside the ambient HN04 layer around 5 hPa.Both, the SPE-related and ambient peaks can not be vertically resolved and merge together after the application of MIPAS averaging kernels.The moderate decrease of HN04 in the following days is qualitatively reproduced by all models except B2dM.In this particular model, the HN04 enhancements are confined to the polar night region, hence experiencing less photochemical losses.
The differences in the magnitude of the HN04 enhancements in the simulations and the observations can partially   . 25. Observed and modeled zonal mean HN0 4 vmrs for pre-SPE conditions (26 October) and during the main proton forcing (29 October-1 November).Solid contour lines reflect 0.1 ppbv steps.Mote that WAS observations from 26 October have been omitted due to gain calibration problems.
be explained by differences in the abundances of the precur-a weaker degree of denoxification compared to the observasor NOz.During the main proton forcing, modeled NO 2 tions already before the SPE event.Aditionally, differences abundances at 70-90°N at the HN0 4 peak height are on av-in the E102 availability might also play an important role erage 50-100% higher than the observed ones (not shown).
in explaining the behavior of modeled HN0 4 At the peak s^6 The NO 2 overestimation in the models is mainly related to6B6 heightof the HN0 4 enhancements and in the absence of sun-  It might also be possible that the HO, partitioning is affected by ion chemistry.Several ion chemistry reactions are665 known which transfer H into OH -, and therefore might act as a sink of H0 2 ; one reaction is known which transfers OH into H0 2 .While it is beyond the scope of this investigation to determine whether these reactions really significantly affect the partitioning between odd hydrogen species, it might676 be worthwhile to investigate this point in the future.

Ozone loss
One of the most important aspects of the model-data inter,675 comparison of SPE-induced composition changes during the Halloween event is the evaluation of the ability of the models to reproduce the observed ozone destruction caused by acceleration of catalytic HO, and NO, cycles.SPE-induced ozone losses have been observed by a variety of space-borne,6 instruments during several of the stronger events of the past two solar cycles (see Jackman et al., 2000, for a review).Two different types of ozone destruction could be distinguished: HO.-related short-time losses, acting principally in the mesosphere during the event itself, and NO.-related mid76e5 term losses in the stratosphere which can last up to several months in the polar winter atmosphere.Such a behavior has also been observed by MIPAS in the aftermath of the Halloween event.L6pez-Puertas et al. (2005a) reported HO.drivenmesospheric ozone losses up to 70% and NO.-driven696 stratospheric losses of around 30%, the latter lasting for more than 2 weeks in the Northern hemisphere.
Figure 26 shows the observed and modeled temporal evolutions of the relative 0 3 changes with respect to 26 October, averaged over 70-90°N.The mesospheric ozone losses,695 above 0.3 hPa, which exhibit two distinct peaks related to the main proton events on 29 October and 4 November, are well reproduced by most of the models.Also the stratospheric 03 losses during the following month, peaking around I hPa, are qualitatively reproduced by the simulations, however, with a more pronounced spread of the model results.This is not surprising since these losses are driven by NOx which exhibits important differences between the models, particularly during the second half of November (see sections 6.2 and 63).Further, NOx-induced ozone loss is driven by reaction R4 which is very sensitive to temperature differences.The midterm evolution in the mesosphere is characterized by ozone buildup which is related to seasonal variations (summer to winter transition), and which is generally more pronounced in the model simulations compared to the observations.
In order to assess observed and modeled short-term ozone depletion in a more quantitative way, we have compared profiles of relative ozone changes at 70-90°N, averaged over the period of the main proton forcing (28 October-4 November), in Figure 27.The agreement between observations and the multi-model average is excellent in the mesosphere, indicating a very good overall ability of the models to reproduce HO.-related ozone losses under SPE conditions.Also, the models themselves agree reasonably well in this altitude range, except for B2dM.In the stratosphere, where NO.relatedlosses are dominant, the agreement between the models is worse, though the model average is very close to the observations within 5%.Ozone depletion around 1 hPa is overestimated by EMAC and B3dCTM.CAO and SOCOL results indicate a somewhat smaller ozone loss throughout the stratosphere.WACCM simulations performed with and without electron-induced ionization (WACCM and WACCMp, respectively in Figure 27) indicate an additional ozone loss induced by electrons in the order of 5% above 2hPa.It is interesting to notice that the agreement of the WACCM simulations with the observation is better when excluding the electron contribution.
Figure 28 shows the corresponding zonal mean distributions.Observed mesospheric losses extend to around 60°N in consonance with the expected cut-off latitude of proton precipitation.This latitudinal distribution is well reproduced by the models.132dM shows a mesospheric ozone buildup poleward of 80°N related to seasonal changes, which overcompensates HO.-related losses at these particular latitudes.This behavior, which can be attributed to deficient meridional mixing in the polar region, give rise for the apparent underestimation of mesospheric ozone losses of B2dM in Figure 27.Oscillations encountered in the CAO ozone changes above I hPa at 40-50°N are related to the background 0 3 and are not caused by the SPE.
The latitudinal extension of observed and modeled stratospheric ozone losses around I hPa correlates well with the area of NO, increases shown in Fig. 14   70°N.It is interesting to notice that B2dM simulations show no NO.-inducedozone loss in the upper stratospheric pola11720 night region, in contrast to the observations and other models.Indeed, the NO, catalytic cycle is expected to be inefficient at dark conditions since NO 2 is not reconverted to NO.Strong mixing is hence required in order to obtain a homogeneous ozone distribution in the polar stratosphere as found in the observations.Ozone increases occur in the SOCOL simulation below 1 hPa which can be related to intrusions of mid-latitude air into the polar region, over-compensating the SPE-induced ozone losses.
Figure 29 shows profiles of stratospheric mid-term ozone changes at 70-90°N, averaged over the period 16-26 November.As expected, modeled ozone depletions have a larger spread than during the main proton forcing, ranging from 10 to 50% at the peak height.The model average, however, is in very good agreement with the observed depletion of 30% at I-2hPa.Only minor differences of 5% are found at its maximum.WACCM simulations with and without electron-induced ionization, however, suggest that these remaining differences could be significantly reduced when excluding the electron contribution to atmospheric ionization.
Figure 30 shows the corresponding zonal mean distributions.Generally, the magnitude of the stratospheric ozone loss at 70-90°N is anti-correlated to its latitudinal extension which, in turn, is linked to the spatial distribution of the SPErelated NO, layer (see Fig. 17).Meridional redistribution is hence a key factor for explaining the differences in the modeled ozone depletions shown in Fig. 29.In particular,"" SOCOL simulations indicate strong meridional distribution around 1 hPa, resulting in polar higher ozone abundances than in the other models, despite of the relatively high NOy availability shown in Fig. 17.There, NO.-driven ozone loss is partly compensated by in-mixing of 0 3 -rich air-masses from lower latitudes.1760 Observed mesospheric ozone changes in late November are characterized by a pronounced increase around the polar night terminator which is related to the buildup of the third ozone maximum (Marsh et al., 2001).This rapid buildup is responsible for the short lifetime of HO.-related ozone depletion at these altitudes.Only in the polar night region, re-9766 duced ozone abundances are found until the end of November.This behavior is well reproduced by EMAC, KASIMA, WACCM and, to a lesser extent, HAMMONIA.B2dM behaves in an opposite way.
In summary, SPE-related short-and midterm ozon0770 changes are well reproduced by the atmospheric models on average, though individual model results can vary significantly due to differences in dynamical and meteorological background conditions.The good agreement between models and observations in the mesosphere can be interpreted775 as a verification of the parameterization of HO, production by atmospheric ionization included in the models.On the 1750 other hand, WACCM simulations with and without electroninduced ionization suggest that the agreement between models and observations could be even improved when excluding 760 the electron contribution to atmospheric ionization.

Enhancements of H2O2
MIPAS observed H2 O2 increases of short duration immediately after the Halloween SPE in polar night stratosphere.H2 O2 is formed by the reaction H02 + H02 --3 H2O2 + 0 2 (R10) and is hence -together with HN0 4 -an indicator for SPEgenerated HO, in the stratosphere.During daytime, it is photolyzed within several hours to a day, or destroyed by the reaction H2 02 + OH -+ H2 O + HO2 .
(RI 1) Chemical nighttime losses are negligible at quiescent conditions.The availability of OH during periods of proton forcing allows for H2O2 destruction also at night.These losses, however, are most important above the stratopause.In the dark stratosphere, reaction RI 1 is expected to deplete H202 by less than 10%.Therefore, observed H2O2 increases are primarily driven by the production mechanism R10.
Model output of H2 O 2 is available from B2dM, B3dCTM, FinROSE, HAMMONIA, KASIMA, and WACCM.Figure 31 shows observed and modeled zonal mean H2O2 changes during the period of the main SPEs (28 October-4 November).The observed increases of up to 0.1 ppbv are considerably overestimated by the simulations by a factor of 4-7.This huge difference between observed and modeled H2 O2 increases can hardly be explained by a possible overestimation of the ionization rates by a factor of 1.2-2, as suggested from the comparison of NOy increases.Although Fig. 31.Zonal mean H2O2 changes with respect to 26 October averaged over the period 28 October-4 November in MIPAS observations and model simulations.Solid contour lines reflect 0.1 ppbv steps.The significance of observed H2 O2 changes (in units of ( 7) is shown in the upper second panel (from the left).
H2O2 production depends quadratically on H0 2, total HOX1B,o scales with the square root of the ionization rate due to reaction R9, being the principal chemical loss mechanism at nighttime.Thus, 4-7 times lower ionization rates would be ,785 required in order to reduce modeled H2 O2 increases to the observed values.As already mentioned in the discussion of HN04 enhancements, the availability of H0 2 during night,e,5 time SPE conditions is largel y controlled by the HO-, partitioning.At the peak height of the H 2 02 increases (0.5-1790 1 hPa), this dependence is even more pronounced than at the pressure levels of the HN0 4 enhancements (2-3 hPa) due to the increasing OH contribution to HO, with altitude.Thus,,820 the disagreement of observed and simulated H2O2 hints at an underestimation of the OHIHO 2 ratio in the upper polar 1795 stratosphere during the proton forcing.Alternatively, H2O2 formation by reaction R10 might be significantly overestimated in the models.1525 Meridional transport to illuminated latitudes, where H2O2 is photochemieally destroyed, could also affect the magni-1800 tude of the SPE-related enhancements.H 2 O 2 distributions simulated by B2dM, which has a very strong mixing barrier, might hence experience less photochemical losses thatr830 in other models.In fact, B2dM enhancements are more confined to the polar night region.Other models, however, show ,805 a similar meridional distribution as observed by MIPAS.It is thus unlikely, that differences in the efficiency of photochemical losses related to transport can explain the pronounced835 differences between observed and modeled H2O2 enhancements.HCI can also be incorporated into negative ions, from which chlorine is released mainly in the form of atomic chlorine or chlorine monoxide.There are also reverse reactions releasing HCI, however, it has been shown in a recent publication (Winkler et al., 2009) that during large solar proton events, chlorine activation dominates, and negative ion reactions can act as a significant sink of HCl, and a source of active chlorine.Atomic chlorine is rapidly converted to CIO by In the polar night stratosphere, where SPE-generated HO, is dominated by H0 2 , CIO is further converted to HOCI: CIO +H02 ^ HOC1+02. (R14)t860 The chemical lifetime of nighttime HOCI is very long below the stratopause.Above and at sunlit conditions, HOCI is removed by the reaction HOCI+OH -C10+H2O The mid-term evolution of polar ambient CIO during the period of the Halloween event is characterized by a continuous decrease related to seasonal variations (see Fig. 3) which makes the analysis of SPE-induced changes on a longer timescale difficult.Therefore, we restrict our analysis to the period of the main proton event on 29-31 October.Figure 32 shows observed and modeled changes of the CIO zonal mean distribution, averaged over these days, with respect to 26-27 October.CIO increases of -0.1 ppbv are found in the MIPAS observations at latitudes around 60°N in qualitative agreement with the previous analysis of von Clarmann et al. (2005).These enhancements are reproduced by none of the models.Evidently, simulated daytime HO, increases are too small compared to the ambient HO,; abundances to alter noticeably the CIO availability.Although the observed enhancements are significant at the 2(T-level with respect to the average measurement precision (see Fig. 32, second panel), this important difference between the observations and the simulations should be carefully interpreted due to a possible systematic bias related to gain calibration errors in the measurements (see Section 2), particularly because the observed CIO change has been calculated from temporal aver- 1R80 ages belonging to different gain calibration periods.In the polar night region, both, observations and models show a CIO decrease.The observed CIO reduction of up to 0.2 ppbv,9o5 is considerably underestimated by the simulations, except for CAO.The latter model overestimates the CIO reduction by ,855 approximately a factor of 10.The unreasonably large CIO depletion in CAO is related to a high CIO availability before the SPE and goes along with a CIONO 2 buildup of a similar magnitude (see below).The reason for the higher" background clo concentrations in this particular model is still 1890 under investigation.In contrast to the observations, the CIO decreases obtained by B2dM, EMAC, and SOCOL are not pole-centered but shifted slightly to lower latitudes.The remaining models (except CAO) produce a very similar C10191s signal. 189fi The differences of observed and modeled CIO changes at latitudes poleward of 70°N are related to the background CIO abundances.Fig 33 shows the zonal mean distribu-,920 tions of CIO vmrs on 26-27 October prior to the onset of the proton forcing.CIO vmrs of more than 0.4 ppbv have 1500 been observed around 2 hPa in the entire NH with a slight decrease poleward of 70°N.Maximum abundances were found at 60-7WN, exactly at the same latitudes where the CIO in^925 creases during the following days occurred.Although we cannot exclude that the observed CIO in this latitude range is affected by gain calibration errors, this coincidence is somehow remarkable.In principle, the enhanced CIO abundances around 60-70GN can be related to differences in the latitudinal distributions of daytime OH and O, the first being responsible for CIO production and the latter for CIO removal.
Modeled CIO abundances do not show this enhancement around 60-70°N.Simulated CIO vmrs are also generally lower by 50% than those observed by MIPAS (except for FmROSE and CAO) and exhibit a pronounced decrease towards the polar night region.In some models (e.g., B2dM and EMAC) CIO has disappeared nearly completely at the pole.It is thus not surprising, that modeled CIO depletions at 70-90°N are less pronounced than in the observations in absolute terms.The much stronger modeled decrease of CIO towards the polar night region during pre-SPE conditions seems to be related to an overestimation of CIO losses.Since the sequestering into the Cl2 02 dimer is inefficient around 2 hPa and simulated HOCI or CIONO 2 distributions before the SPE do not indicate a conversion of CIO to these species, it is most likely that CIO is more efficiently converted to HCl than indicated by the observations.The faster conversion in the models might be related to the reaction path CIO+OH --^ HCI+Oz which has an uncertainty of its rate constant of several 100% (Sander et al., 2006).However, also dynami-,95o cal reasons (i.e., differences in the magnitude of meridional mixing) cannot be excluded.
The temporal evolution of observed and modeled HOCI changes at 70-90°N until mid November is shown in Fig. 34.HOC started to increase rapidly on 29 October, reaching956 values around 0.25 ppbv, and diminished after 1 November within a few days.A smaller second increase occurred on 3 November related to the second, weaker SPE.The simulations show generally smaller enhancements (approximately 30% less on average), except FinROSE.This model overes-"" timates significantly the observed enhancements by nearly a factor of 2. There, HOCI abundances remain enhanced after the SPE for nearly one week and show a second, even more pronounced enhancement around I 1 November.and 32) indicates that ambient CIO is quickly converted to HOCI via reaction R14 during nighttime in the presence of proton forcing.However, HOCI increases are higher than the corresponding CIO losses, resulting in a net increase of active chlorine by approximately 2 ppbv in the observations and most of the models.This can be explained by SPE-related chlorine activation via reaction R12.FinROSE, however, overestimates the chlorine activation by a factor of 3.
The sharp decline of the HOCI enhancements after the proton forcing observed by MIPAS, and also reproduced by most models, must occur in the sunlit atmosphere close to the polar night terminator, since losses via reaction R15 are negligible in the polar night stratosphere after the SPE.This is also the reason for the relatively long lifetime of the HOCI enhancements in B2dM where meridional redistribution is weak.This is not the case in the FinROSE model.There, the long lifetime of the HOCI enhancements related to the SPE, as well as the second buildup around 1I November, seem to be caused by an underestimation of chemical losses of HOCI. 10,0CIONO 2 is removed mainly by photolysis in the sunlit atmosphere and, to a lesser extent, by reaction with atomic oxygen.Due to its pressure dependence, CIONO 2 formation by1995 reaction R16 is more effective at lower altitudes.Enhanced NO2 availability related to the SPE, however, is increas-1975 ing with altitude, leading to a peak height of the observed CIONO 2 enhancements around 3 hPa (-36 km).This is slightly higher than reported by L6pez-Puertas et al. (2005ak000 who based their analysis on an older MIPAS CIONO 2 data version than used here. 1580 The temporal evolution of observed and modeled CIONO 2 changes at 70-90°N until the end of November is.... shown in Fig. 36.The observed enhancements of 0.4 ppbv after the SPE remained in the stratosphere for about two weeks.After a sudden decrease on 13 November, CIONO2 1985 abundances raised again on 19 November, reaching a second, weaker maximum around 22 November.The modeled CIONO 2 increases are generally smaller (except CAO, seezolo discussion above) and show a different temporal evolution.
The CIONO 2 underestimation in the simulations, particu-1990 larly during the first enhancement starting on 1 November, is related to the reduced CIO availability compared to the observations.Figure 37 shows the corresponding zonal mearto15 distribution of the observed and modeled CIONO 2 enhancements averaged over 1-5 November.From the observations, it is clear that CIONO 2 is principally formed in the polar night region where high NO 2 abundances are available and no photochemical losses occur.Most of the model simulations, except CAO, SOCOU and WACCM, show negligible enhancements there.Instead, CIONO 2 formation occurs around 70°N, were daytime losses are still small but CIO is available, however, with a considerably smaller magnitude than observed.SOCOU and WACCM simulations, which have a similar latitudinal distribution of CIONO 2 changes as observed, exhibit higher CIO abundances in the polar night region than other models.
The observed temporal evolution of the CIONO 2 changes in the second half of November is better captured by models based on ECMWF-and MERRA-driven meteorology up to the stratosphere (i.e., B3dCTM, FinROSE, KASIMA, and WACCM), which hints at a strong impact of vortex dynamics on the CIONO 2 abundances.Wave-driven vortex excursions to illuminated latitudes, alternated by reformation of a pole-centered vortex, are mainly responsible for the CIONO 2 variability and particularly for the decrease around 16 November.The descending NO 2 layer, formed during the SPE, acts as a reservoir for continuous CIONO 2 formation in the following weeks after the SPE.Due to the reduced CIO availability in the polar stratosphere towards the end of November, additional CIONO 2 buildup is observed only around 60-70°N, in agreement with most of the model re2o4o suits (not shown).B2dM and HAMMONIA, however, show very small CIONO 2 increases in the second half of November.In the first model, this is related to the confinement of the NO 2 layer to high latitudes, where no CIO is available.In the latter model, strong meridional mixing led to a dilution 2oa5 of the SPE-generated NO 2 layer, such that insufficient NO2 was available for additional CIONO 2 buildup.

Conclusions 2050
We have compared stratospheric and mesospheric composition changes observed by MIPAS in the NH during and after the Halloween proton event with simulations performed with state-of-the-art atmospheric GCM and CTM models.The large number of models participating in the intercompar2o55 ison exercise allowed for an evaluation of the overall ability of atmospheric models to reproduce observed atmospheric perturbations generated by SPEs, particularly with respect to NOy and ozone changes.This model validation represents a mandatory first step towards an accurate implementation of particle precipitation effects in long-term climate simulations.It has also allowed to test and identify deficiencies in the chemical schemes, particularly with respect to nitrogen and chlorine chemistry, being relevant for stratospheric ozone.
Observed SPE-related short-time increases of the minor species HN0 4 and H2 O 2 have been identified for the first time and are qualitatively reproduced by the simulations.The observed enhancements of 0.2 and 0.1 ppbv, respectively, are overestimated by the models on average.Both observations and simulations give further evidence for an SPE-induced CO depletion.A clear isolation of these chemical losses from dynamical variability, however, is difficult.
In general, atmospheric models are able to reproduce most of the observed composition changes.In particular, simulated SPE-induced ozone losses agree within 5% with the observations on average.This excellent agreement is found on a short-term scale (HO x-driven) in the mesosphere, as well as on a mid-term scale (NO x -driven) in the stratosphere.Sim-Fig.37. Zonal mean CIONO 2 changes after the main proton forcing (1-5 November) with respect to 26-27 October in WAS observations and model simulations.Solid contour lines reflect 0.1 ppbv steps up to 0.6 ppbv and 0.2 ppbv steps above.ulated NO, enhancements around 1 hPa are on average 30% higher than indicated by the observations, while an underestimation of modeled NO Y of the same order was found in the mesosphere.The systematic behavior in the stratosphere suggests that these differences are related to the simulatedo85 ionization rate profile shape.WACCM simulations without inclusion of electron-induced ionization yield much better agreement with the observations around 1 hPa than the than the nominal model run including both electrons and protons throughout the period of interest.An upper stratospheric ex2000 Bess NOY production by electron-induced ionization of 5-10 ppbv could be identified from these simulations, particularly after the minor second event around 4-5 November.The excess ozone loss related to electron-induced ionization has been estimated to be around 5%.Again, better agreement with the observations was achieved when excluding the elec 2095 tron contribution to ionization in the WACCM simulations.
Our comparisons hence suggest that the modeled electroninduced ionization during the Halloween event is considerably overestimated.This might be related to the uncertainties in constraining the electron energy spectrum from availabl&"' particle observations.The impact of chemical NO losses due to reaction with atomic nitrogen (R2) on the SPE-induced NO, increases has been studied in detail.An important dependence of the net NOY generation on temperature and background NO,; due to this mechanism has been identified.In the stratosphere, SPE-related NOY increases are reduced (enhanced) by approximately 10% if temperatures were 10 K lower/higher.This behavior might be of relevance for future implications of SPE effects on climate when considering a stratospheric cooling trend related to climate change.The reduced NO,, production efficiency related to reaction R2 also implies limitations for models using family approaches in their chemical schemes, since this mechanism of NO Y destruction is not taken implicitly into account in these models.
The analysis of the observed and modeled NO }, partitioning in the aftermath of the Halloween SPE has clearly demonstrated the need to implement additional ion chemistry into the chemical schemes, Short-term HNO 3 increases can only be reproduced by model calculations including ion-ion recombination between N,03_ and H-cluster ions (Verronen et al., 2008).The partitioning of HNO3 and N 2 05 in the following weeks after the SPE is significantly underestimated by the models that do not include HNO3 formation via water cluster ions (136hringer et al., 1983).However, fur-ther work is required to tune the parameterizations of these 2105 mechanisms such that quantitative agreement with the observations can be achieved.The overestimation of observed H2O2 and HN04 en-?"' haneements by the models hints at an underestimation of the OH/HO 2 ratio in the upper polar stratosphere during the pro-2110 ton forcing.Further work is required to analyze in detail possible reasons for this behavior.The analysis of SPE-induced ,ee changes of the chlorine species CIO, HOCI and CIONO2 has shown that the encountered differences between models and observations, particularly the underestimation of ob-2115 served CIONO2 enhancements, are related to a smaller availability of CIO in the polar region already before the SPE.2170 In general, the intercomparison has demonstrated that differences in the meteorology and/or initial state of the atmosphere in the simulations causes an important variability of 2,20 the model results, even on a short timescale of only a few days.The model responses to the proton perturbation thus1" show a significant spread.On the other hand, this sensitivity of the simulated atmospheric responses to the background conditions, indicated by the spread in the model results, also 2,25 implies that the real atmospheres response to proton events?,sodepends strongly on the actual conditions.Future HEPPA model-data intercomparison activities will focus on the assessment of indirect effects of energetic particle precipitation related to polar winter descent of upper 2130 atmospheric NO,; generated by electron precipitation.ThW85 is motivated, on one band, by the higher potential of indirect effects to influence middle atmospheric composition on longer time scales compared to direct effects (i.e., SPEs) and, on the other hand, by its large variability related to dynam 2,90 2135 ical modulations, making its representation in current atmospheric models challenging.

Fig. 4 ,
Fig.4, Temporal evolution of area-weighted averages (40-90°N) of AIMOS ion pair production rates for protons and electrons during the period of interest.Electron ionization rate below I hPa is induced by bremsstrahlung only.

Funke
Fig. 6.Effect of application of averaging kernels (AKs) to the model data on the example of MIPAS and WACCM4 HOCI zonal mean distribution (40-90°N) averaged over the period 29 October to 4 November 2003.

Fig. 9 .
Fig. 9. Relative meridional CH 4 anomaly in MIPAS observations and model simulations at 40-90°N averaged over the whole time period.

Fig. 10 .
Fig. 10.Temporal evolution of CH 4 changes with respect to 26 October 2003 in MIPAS observations and model simulations averaged over 70-90°N.
Fig. 12. Area-weighted averages (40-90°N) of observed and modeled NO, enhancements during 30 October-I November with respect to 26 October (left) and relative deviations of modeled averages from the MIPAS observations (right).Thick solid and dashed lines represent model multi-model mean average and MIPAS observations, respectively.WACCMp denotes the WACCM simulation including proton ionization, only (excluded from the multi-model mean).
Fig.13.NO, production efficiency (ratio of the net NO, increase and the integrated initial N production) during the period of the main proton forcing (28 October-1 November) from box model calculations for night-and daytime conditions (solid and dashed black lines, respectively), assuming initial atmospheric conditions as observed by MIPAS in the polar cusp region.The following variations for dark conditions are also shown: a 20 K temperature increase (red) and decrease (blue), a factor of 2 increase of 0 3 (solid green), and initial NO x abundances set to zero (dashed green).

Fig. 15 .
Fig. 15.Temporal evolution of area-weighted averages of NOy changes with respect to 26 October 2003 in M1PAS observations and model simulations at 40-90'N, as well as differences between modeled and observed averages.The confidence of observed NOy changes (in units of a) is shown in the upper second panel (from the left).Note that observations exhibiting CO abundances higher than I ppmv have been omitted in the averaging in order to exclude EEP-induced contributions to the NOy enhancements.See text for further details.

e
Fig. 18.Zonal mean N 2 0 changes with respect to 26 October 2003 in MIPAS observations and model simulations averaged over the period 29-31 October.The significance of observed N2 0 changes (in units of a) is shown in the upper second panel (from the left).
limiting reaction of this conversion is R6 which-35 exhibits a strong temperature dependence.NT20s enhancements around 1-2 ppbv, appearing 10-15 days after the Halloween event, have been observed by MIPAS around 70-90°N and have been attributed to the repartitioning of SPEinduced excess NO,(1-6pez-Puertas et al., 2005b).This conversion is further accelerated in the course of November by the growth of the polar night region, reducing the efficiency of N2 0 5 losses by photolysis.

Fig. 20 .
Fig. 20.Zonal mean daytime NO 2/NO, ratios averaged over the period 28 October-15 November in MIPAS observations and model simulations.

Fig. 21 .
Fig. 21.Temporal evolution of area-weighted averages of N 2 0 5 changes with respect to 26 October 2003 in WAS observations and model simulations at 70-90°N, as well as differences between modeled and observed averages.Solid contour lines reflect I ppbv steps.The confidence of observed N2 0 5 changes (in units of T) is shown in the upper second panel (from the left).

Fig. 23 .
Fig. 23.Zonal mean (2N20 5 +HNO3)INOy ratios averaged over the period 15-30 November in MIPAS observations and model simulations.The MIPAS ratio of 0.28, encountered at altitude of the maximum of the SPE-induced NO,, layer, is indicated by a black line.

Fig
Fig.25.Observed and modeled zonal mean HN0 4 vmrs for pre-SPE conditions (26 October) and during the main proton forcing

Fig. 26 .
Fig. 26.Temporal evolution of area-weighted averages of relative 0 3 changes with respect to 26 October 2003 in MIPAS observations and model simulations at 70-90°N, as well as differences between modeled and observed averages.The significance of observed Os changes (in units of (Y) is shown in the upper second panel (from the left).

Fig. 28 .
Fig. 28.Zonal mean relative 0 3 changes with respect to 26 October averaged over the period 28 October-4 November in MIPAS observations and model simulations.Solid contour lines reflect 20% steps.The significance of the observations is also shown (second top panel from the left, in units of (T).
Fig. 29.Area-weighted averages (70-90°N) of observed and modeled relative 0 3 changes with respect to 26 October during 16- 26 November.Thick solid and dashed lines represent model mean average and MIPAS observations, respectively.WACCMp denotes the WACCM simulation including proton ionization, only (excluded715 from the multi-model mean).

Fig. 30 .
Fig. 30.Zonal mean relative Oa changes with respect to 26 October averaged over the period 16-26 November in MIPAS observations and model simulations.Solid contour lines reflect 20% steps.

6. 6
Enhancements and repartitioning of chlorine speciesEnhancements of the chlorine species CIO, HOCI, and CIONO 2 have been detected by MIPAS in the aftennath of the Halloween SPE in the NH polar stratosphere (vonClarmann et al., 2005;L6pez-Puertas et al., 2005b).Short-tern CIO and HOCI increases of the order of 0.2 ppbv occurred immediately after the onset of the proton forcing on 29 October.CIONO 2 increases up to 0.4 ppbv appeared approximately 2 days later, remaining in the stratosphere for several weeeks.SPE-related HOCI increases have also been observed by MLS on Aura during the January 2005 proton event(Damiam et al., 2008).These enhancements were accompanied by a HCI decrease of similar magnitude, thus clearly demonstrating SPE-induced chlorine activation.The conversion of HCI to active species occurred in presence of enhanced OH via the reaction HCI +OH --> CI+11 2 O.(R12)

Fig. 32 .
Fig. 32.Zonal mean CIO changes during the main proton forcing (29-31 October) with respect to 26-27 October in WAS observations and model simulations.Solid contour lines reflect 0.05 ppbv steps.The significance of observed CIO changes (in units of Q) is shown in the upper second panel (from the Left).

Fig. 34 .
Fig. 34.Temporal evolution of area-weighterd averages of relative HOCI changes with respect to 26 October 2003 in MIPAS observations and model simulations at 70-90°N, as well as differences between modeled and observed averages.The significance of observed HOCI changes (in units of Q) is shown in the upper second panel (from the left). 1965

Figure 35
Figure 35 shows observed and modeled changes of the HOCI zonal mean distribution averaged 29-31 October with respect to 26-27 October.The pronounced anti-correlation of HOCI increases and CIO decreases (compare Figures 35 and 32) indicates that ambient CIO is quickly converted to Fig. 35.Zonal mean HOCI changes during the main proton forcing (29-31 October) with respect to 26-27 October in MIPAS observations and model simulations.Solid contour lines reflect 0.1 ppbv steps.

Fig. 36 .
Fig. 36.Temporal evolution of area-weighted averages of relative CIONO 2 changes with respect to 26 October 2003 in MIPAS observations and model simulations at 70-90°N, as well as differences between modeled and observed averages.Contour lines reflect 0.5 ppbv steps.The significance of observed CIONO2 changes (in units of (T) is shown in the upper second panel (from the left).

Table 1 .
Funke et al.:HEPPA intercomparison study Used WAS data versions (indicated by the last digits of the retrieval version) for all species on a daily basis within the period 26 October-30 November 2003.

Table 3 .
Summarized description of the models involved in this study.
CTM calculates the concentrations of 30 minor components, involved in 70 chemical and 35 photochemical reactions, in the range 0-90 km.Output is provided hourly.The reaction rate constants, absorption cross-sections, solar radiation in-680 tensity, and quantum outputs were assigned in the tabulated form according to Funke et al.: HEPPA intercomparison study . In this altitude region, ozone depletion is restricted to latitudes poleward of