Global investigation of the Mg atom and ion layers using SCIAMACHY/Envisat observations between 70 and 150 km altitude and WACCM-Mg model results

. Mg and Mg + concentration ﬁelds in the upper mesosphere/lower thermosphere (UMLT) region are retrieved from SCIAMACHY/Envisat limb measurements of Mg and Mg + dayglow emissions using a 2-D tomographic retrieval approach. The time series of monthly mean Mg and Mg + number density and vertical column density in different latitudinal regions are presented. Data from the limb mesosphere-thermosphere mode of SCIAMACHY/Envisat are used, which covers the 50 5 to 150 km altitude region with a vertical sampling of ≈ 3.3 km and latitudes up to 82 ◦ . The high latitudes are not observed in the winter months, because there is no dayglow emission during polar night. The measurements were performed every 14 days from mid- 2008 until April 2012 . Mg mum, which is more clearly pronounced than for SCIAMACHY, and globally higher peak densities than the SCIAMACHY results. Although the peak densities are not in agreement, the vertical column densities agree well, because SCIAMACHY and WACCM proﬁles have different widths. The agreement between SCIAMACHY and WACCM results is much better for Mg + with both showing the same seasonality and similar peak densities. However, there are also minor differences, e.g. WACCM showing a nearly constant altitude of the Mg + layer’s peak density for all latitudes and 25 seasons.

of Mg + this fraction is estimated to be negligibly small (Vondrak et al., 2008, Fig. 14.). Mg + is formed by charge exchange of Mg with the main ion constituents at this altitude, i.e. NO + and O + 2 . The dominant loss process for Mg is the reaction with O 3 , which leads to stable oxides, hydroxides and carbonates as reservoir species. Mg + is lost by reactions with O 3 but also by reactions with N 2 and other trace gases (CO 2 , H 2 O) (see, e.g. Fig. 9 in Plane and Whalley (2012) for a schematic diagram of the most important reactions). 60 The metal molecules condense and form meteoric smoke particles (e.g. Hunten et al., 1980;Kalashnikova et al., 2000;Saunders and Plane, 2006). These meteoric smoke particles are thought to act as nucleation nuclei. Heterogeneous nucleation of clouds is thought to play an important role in the formation of noctilucent clouds (NLC) (see, e.g. Rapp and Thomas, 2006) in the polar summer mesopause region and also in the formation of polar stratospheric clouds (PSC) in the polar winter stratosphere (see, e.g. Voigt et al., 2005;Curtius et al., 2005). PSC have an important role in the ozone chemistry and the depletion of stratospheric ozone. In a dynamical equilibrium of meteoroid input and the above mentioned loss reactions, the neutral Mg layer is formed between 85 and 90 km altitude and the Mg + layer is formed slightly above the Mg layer (see, e.g. Plane and Helmer, 1995;McNeil et al., 1998 andWhalley, 2012). 70 The MLT is not readily probed by in-situ instrumentation, as balloons fly too low, and satellites have to fly higher because of the atmospheric drag that reduces the satellite's operation time. As a result, only rockets carrying in-situ instrumentation are able to access this region (see Sect. 6 for references). These are, however, expensive and rare. Metal atoms and ions have very strong absorption coefficients and oscillator strengths (see, e.g. Kramida et al., 2012). As a result, although 75 their densities even in the peak region are only a few thousand particles per cubic centimeter at most (for Na, Fe, Mg, less for others), they are very strong emitters of resonance fluorescence.
Mg and Mg + spectral lines are observed at wavelengths below 290 nm. This wavelength region is strongly affected by ozone absorption in the stratosphere, so that observations from ground are not possible. For this reason the knowledge of the Mg and Mg + content in the MLT is poor. 80 In this study the dayglow emissions of Mg and Mg + , detected by the satellite experiment SCIA-MACHY (Scanning Imaging Absorption Spectrometer for Atmospheric CHartographY) on Envisat (Environmental Satellite) are used. The Mg and Mg + densities are retrieved by mathematical inversion of these measurements using a radiative transfer model. Mg and Mg + profiles had already been retrieved from SCIAMACHY by Scharringhausen et al. (2008a,b). As a result of these investigations 85 a novel optimised measurement mode for the purpose of better covering the peak region of Mg and Mg + has been established, and the observations are used for this study.
The resulting new data products are compared in this study with other Mg and Mg + measurements, as well as with WACCM-Mg simulations. WACCM-Mg combines different models: a Meteor Input Function model is coupled to a meteor differential ablation model, which provides the 90 metal input in the MLT. The metal release into the upper atmosphere is then used as one input to simulations using the Whole Atmosphere Community Climate Model (WACCM) (Marsh et al., 2013b) including a new chemistry model of Mg and Mg + (Plane and Whalley, 2012). WACCM is a general circulation model that incorporates interactive chemistry for both neutral and ion species.
In Sects. 2 and 3 the SCIAMACHY/Envisat satellite observations and the Mg and Mg + density re-95 trieval are briefly described. In Sects. 4 and 5 the results obtained for Mg and Mg + are presented and discussed. In Sect. 6 these findings are compared to other measurements. In Sect. 7 the WACCM-Mg model simulations of Mg and Mg + are introduced and the model simulations are compared with the SCIAMACHY dataset. In Sect. 8 a possible connection between Mg and Mg + results and NLC is discussed. Finally, Sect. 9 summarizes the results obtained in this study. 100 2 Instrument and algorithm SCIAMACHY, a satellite-borne grating spectrometer on Envisat, was launched by ESA (European Space Agency) on 28 February 2002 (see, e.g. Burrows et al., 1995;Bovensmann et al., 1999).
Envisat flies in a sun-synchronous orbit with a 10:00 a.m. descending node. The duration of one orbit is roughly 100 min (15 orbits per day). Note that the local time of SCIAMACHY limb obser-105 vations changes with latitude. It is ≈ 11 a.m. at 60 • N and 9 a.m. at 60 • S and, as shown in Fig. 1, varies strongly in the vicinity of the poles. The highest latitude covered by SCIAMACHY limb observations of scattered light is ≈ 82 • .
Data from SCIAMACHY's limb MLT-mode with tangent altitudes between ≈ 50 and ≈ 150 km covered in 30 consecutive ≈ 3.3 km steps are used in this study. Dependent on latitude, consecu-110 tive limb measurements have a latitudinal separation of up to 8 • , because nadir measurements are performed between two MLT-measurements. However, there is also a small latitude offset between MLT measurements in two consecutive orbits, to improve the meridional sampling for zonal averages.
MLT-measurements were performed from mid 2008 until the loss of contact to Envisat in April 115 2012. They were performed roughly every two weeks for 15 consecutive orbits, which corresponds to one day of consecutive measurements. This means, that measurements on 84 single days were performed, which is roughly a quarter year of single day data. This is much less than 4 years of single day data, and there are no continuous daily observations in the MLT-mode. Apart from a few days, where some orbits are missing in the data, all longitudes are equally covered (sampling every 120 24 • of longitude).
The two-dimensional retrieval algorithm uses all measurements of one orbit and retrieves the metal densities from each emission line of the species on a latitude and altitude grid. For averaged data, a reference orbit for geolocations is used, which is formed from the geolocations of the data that is averaged. Measurements with similar latitudes and local times as the measurements of the reference 125 orbit are averaged. For Mg the spectral line at 285.2 nm is used. For Mg + the spectral lines at 279.6 nm and 280.4 nm are used. A comprehensive discussion of the algorithm was provided by Langowski et al. (2013). Langowski et al. (2013) also present an extensive discussion of systematic and statistical errors and, e.g. also explain why only data for altitudes above 70 km altitude is used in this study.

130
An improvement compared to the analysis described by Langowski et al. (2013) is achieved by correcting the measured spectra for the polarization response of the instrument. This is briefly discussed in Sect. 3. As the main focus of this study is on the results of the retrieval algorithm, this discussion is kept short and more details can be found in Langowski (2014).
The first measurements at the start of the sunlit part of each orbit (high/mid northern latitudes) are 135 contaminated by solar stray-light, which can be seen in the dark signal measurement at 350 km tangent altitude. We therefore exclude these measurements and also adjacent measurements, for which the contamination can not be clearly excluded. When using the remaining measurements, some of the features similar to the ones resulting from this error source are still observed. However, these features are not dependent on the orbit phase like the ones resulting from the stray-light contamina- The actual latitude grid for calculating the densities has a larger extent than the shown results. This is because the outermost latitude intervals have been cut off to reduce edge effects, which may, however, affect the neighboring latitudes. In Sect. 8, where near polar effects are discussed less data 145 is cut off.

Polarization correction
SCIAMACHY is sensitive to the polarization of the incoming radiation. This was known before launch and wavelength dependent correction parameters were measured. The majority of backscattered radiation from the Earth's atmosphere is produced by Rayleigh-scattering, which polarizes 150 the incoming unpolarized solar radiation. The resonance fluorescence process of metals results in nonzero polarization of radiation even if the incoming beam is unpolarized, depending on the change of angular momentum between the electronic states that are involved in the fluorescence process.
The polarization of the resonance fluorescence process from metals is well described by Hamilton (1947) and can also be found in Chandrasekhar (1960) with a slightly different nomenclature in the 155 used formulas.
The incoming solar radiation is assumed to be unpolarized, which means that the parallel and perpendicular intensity components are equal in any observation plane. Following (Chandrasekhar, 1960, Eq. (259), page 51) the incoming radiation components parallel and perpendicular to the scattering plane are transformed as follows: The parallel component is dependent on the scattering angle θ, which leads to nonzero polarization, if E 1 = 0. The parameters (E 1 ,E 2 ) are (1,0) for Mg at 285.2 nm, (0.5,0.5) for Mg + at 279.6 nm and (0,1) for Mg + at 280.4 nm. The coordinate system independent degree of polarization P is , and the polarization angle χ is perpendicular to the scattering plane.

165
The response of SCIAMACHY depends on the polarization state of the radiation, so that an accurate absolute calibration requires the effect to be corrected for. Until recently, this was not possible because the available set of calibration data for the polarization sensitivity which is provided with the Level 1 data suffered from severe errors, in particular for the limb measurement geometry and studies for an improvement of this effect in the calibration are still being done (see, e.g. Krijger et al. 170 (2014) and Liebing et al. (2013)).
From in-flight data it is possible to derive polarization sensitivities in the wavelength range below 300 nm by comparing measured reflectances with the ones calculated by a vector radiative transfer model (SCIATRAN 3.1, see, e.g. Rozanov et. al. (2014)) and then fitting the differences to the expected polarization. The results of this fit have been applied here together with Eq. (1) and a 175 transformation to the SCIAMACHY polarization reference frame.
To derive the polarization sensitivities an ozone profile has to be assumed. The error due to the unknown actual ozone distribution is estimated to be quite large. This is why a consistency check to this approach has been performed here. Assuming the unpolarized Mg + line at 280.4 nm (D 1 line) to yield exact results, a correction factor for the line at 279.6 nm (D 2 line) can then be determined 180 requiring that the densities retrieved from both Mg + lines match. Assuming that the polarization sensitivity does not change between 280 and 285 nm wavelength, a correction factor for Mg can be calculated knowing that the degree of polarization P D2 of the D 2 line is just half of the degree of polarization P Mg of the Mg line.
In cases where the Mg + signal is not dominated by noise, the results from both methods agree 185 with each other well for Mg. Similarly the Mg + densities from both lines retrieved using the updated polarization correction for SCIAMACHY are in good agreement. We therefore conclude that the polarization correction for the 280-285 nm region improved the calibration of the data considerably.

Seasonal variations of Mg
The seasonal variations of Mg and Mg + densities on a latitude and altitude grid are investigated.

190
Prior to averaging, measurements at night or those having noise peaks in the investigated spectral region, caused by highly energetic particles hitting the detector (e.g. in the Southern Atlantic Anomaly region), are removed from the data set by filtering. The limb spectra of up to 15 daily single orbits are zonally averaged before running the retrieval algorithm, to improve the signal to noise ratio.
To further reduce the noise in the results, data for individual months are averaged for the entire 195 measurement period.
The retrieval algorithm produces densities on an altitude and latitude grid with 40 equidistant day-side latitude intervals between 82 • S and 82 • N and 80 altitude intervals with 1 km thickness between 70 and 150 km. Additionally, there are latitude intervals for the part of the orbit, where the satellite moves northwards (see Fig. 1), to overcome ambiguities of identical latitudes but different 200 local time at high latitudes.
The monthly averaged results for Mg are shown in Fig. 2. Mg shows a peak at around 90 km with densities up to 1500 cm −3 . There is a strong month-to-month variability in the latitudinal distribution. The annual mean results for Mg are shown in Fig. 3. In the annual mean, the densities at low latitudes are higher than at the high latitudes in the peak region. At the highest southern latitude 205 covered there are increased densities above 100 km, which, however, are most likely explained by the fact that at this latitude the number of measurements is smaller and the statistical error thus larger. Figure 4 shows the seasonal variation of the vertical profile for low, mid and high latitudes, and  The seasonal variation of the Mg peak altitude is estimated to be less than 5 km.
The VCD of Mg varies between 0.5 × 10 9 cm −2 and 3.5 × 10 9 cm −2 and is around 1.75 − 2.5 × 10 9 cm −2 for most latitudes and times. The VCD between 40 • N and 40 • S is higher than for higher latitudes, which are only covered in the hemispheric summer and show a summer minimum, which however is only clearly observed in the Northern Hemisphere. The seasonal variation is small com-215 pared to the mean of the vertical column densities. This relatively small variation is required to retrieve the correct mean value, because of non-linearities in the retrieval, as comprehensively discussed and explained by Langowski et al. (2013).

Seasonal variations of Mg +
For Mg + the same approach for data averaging is used as described for Mg in Sect The densities on an altitude and latitude grid for the monthly averaged results are shown in Fig. 6.

225
The Mg + densities peak at altitudes in the range of 95-105 km with peak values of 500-4000 cm −3 .
The peak densities show a seasonal variation with a summer maximum between 25-45 • in both hemispheres. In summer the highest peak altitude is roughly at 45 • (N and S). It occurs around 105 km, which is up to 10 km higher than the lowest peak altitude at the equator and higher than that at the high latitude region. There is a minimum at the equator and also a second maximum in peak 230 altitude in the mid latitudes in the winter hemisphere, that, however, is more variable than the one in the summer hemisphere.
The latitudinal variation of Mg + is also seen in the average over all available measurements in 2008-2012, which is shown in Fig. 7. The latitudinal dependence of the peak altitude seems to be symmetric to the equator. However, there is an asymmetry in the peak density, with larger densities 235 for the Northern Hemisphere (NH) than for the Southern Hemisphere (SH). However, this effect is compensated by a broader profile in the SH, so that the differences in the vertical column densities are smaller.
A reason for this may be Lorentz force induced transport processes together with the very different mapping of geomagnetic coordinates to geographic coordinates (the magnetic southpole at around 240 65 • S is further away from the geographic pole than the magnetic northpole at around 82 • N).
At the upper vertical profile edge up to 110 km the density is higher at mid latitudes than at the equator, which can be explained by the density maximum at mid latitudes being also at higher altitudes, than at the equator. However, above 110 km altitude this behaviour is reversed and the equatorial densities are larger than the densities at mid latitudes. This may be explained by enhanced 245 vertical upward transport in the equatorial region, which will be discussed in Sect. 7.2. Figure 8 shows the seasonal variation of the vertical Mg + profile for low, mid and high latitudes, and Fig. 9 shows the corresponding VCD. In each of the latitude regions, a maximum in peak altitude is found during summer in the corresponding hemisphere. The VCD varies between 1 × 10 9 cm −2 and 6 × 10 9 cm −2 . A seasonal cycle with a summer maximum is observed and the strongest varia-  were found. According to Grebowsky and Aikin (2002) approximately 50 flights of rocket-borne mass spectrometers had been made until 2002, probing the region between 80 and 130 km. Results of these flights for Mg + can, e.g. be found in Istomin (1963), Narcisi andBailey (1965), Narcisi (1971), Aikin and Goldberg (1973), Philbrick et al. (1973), Zbinden et al. (1975), Herrmann et al. (1978), Kopp and Herrmann (1984), Kopp et al. (1985b,a), Kopp (1997) and Roddy et al. (2004).

265
Date, local time, latitude and reference publications of these and other rocket flights can also be found in table 1 of Grebowsky et al. (1998). Figure 11 shows reproduced plots of vertical number density profiles of Mg + . The original sources of the data for these plots are listed in table 1.
A disadvantage for the comparison with the new dataset presented in this study is that most of these measurements were performed during special events having E s layers (sporadic electron lay-270 ers), aurora, meteor showers, stratospheric warmings or NLC present. Sporadic Mg + layers often occured between 105-110 km and/or at around 120 km. The Mg + peak altitude in most of these measurements can be found between 90 and 95 km altitude and the full width at half maximum (FWHM) of the layers is on the order of 5-10 km, but sometimes the FWHM is only 1 km.
When comparing in-situ mass spectrometer measurements with satellite remote sensing results,

275
it has to be noted that the in-situ measurements are localized and limited to the direct vicinity of the rocket. In contrast, remote sensing techniques typically cover a large volume with horizontal distances along and perpendicular to the viewing direction of several hundreds of kilometers. This results in smoother layers with a larger FWHM for the remote sensing method. Still, the width of the Mg + layers as well as the peak densities presented in this study are in good agreement with 280 the in-situ rocket measurements. However, the strong latitudinal dependence of the peak altitude observed in Fig. 8 as retrieved from SCIAMACHY measurements is not found in the in-situ rocket data.
In addition to the in-situ measurements with rockets, there are also airglow measurements available from rockets, the space shuttle and satellites. However, to retrieve density information from this 285 method radiative transfer models as well as inversion techniques and computational power is needed.
As a result, slant column information rather than profiles has been retrieved and made available in the first remote sensing studies (see, e.g. Boksenberg andGérard, 1973 or Gérard andMonfils, 1974). In Anderson and Barth (1971) (Summer, ≈ 40 • N) the region up to 106 km altitude is scanned during a sporadic E s layer event, and the peak altitude of the Mg + was not observed during this flight, i.e. it 290 was higher than 106 km. No Mg signal above the instrumental noise was observed in these spectra.
The region above the peak altitude from 150 km up to the F-layer and above was investigated by Gérard and Monfils (1978), Fesen and Hays (1982a), Mende et al. (1985), Gardner et al. (1995) and Gardner et al. (1999) and typically shows less than 100 cm −3 Mg + ions at 150 km altitude. This is in good agreement with the profiles described in Sect. 5. In some cases higher densities are observed 295 in the profiles retrieved from the SCIAMACHY limb observations. These are explained by retrieval artifacts on the edge of the retrieval grid and vertical constraints tuned for the main peak. These effects result in small oscillations, compared to the main peak, in regions with lower densities. Since these artifacts appear at the upper edge of the profile it also implies that there is a significant density above the highest tangent altitude for Mg + .

300
In Minschwaner et al. (2007) a combined NO and Mg + retrieval for satellite limb measurements is shown, which is in good agreement with the results in Sect. 5 in terms of the Mg + concentrations at peak altitude and at the upper edge of the profile. However, the Mg + peak altitude is at 90 km.
Taking into account the coarser sampling (every 7 km), higher statistical errors and a tangent height offset (±4 km) in Minschwaner et al. (2007), the agreement is reasonable. Accurate tangent height A time series of Mg + vertical columns covering several years and retrieved from SBUV nadir measurements was presented by Joiner and Aikin (1996). These measurements were performed approximately 1 day per month, with a spectral resolution of 1.13 nm and a spectral sampling ev-320 ery 0.2 nm (compare to SCIAMACHY with ≈ 0.22 nm resolution and sampling every ≈ 0.11 nm).
The results in Joiner and Aikin (1996) are in very good agreement with the results obtained in this study, especially when comparing Figs. 10 and 11 in Joiner and Aikin (1996) with Fig. 9 in this study. Figs. 12 and 13 show redrawn VCD timeseries from different sources for a quick and easy comparison of the results of Mg + and Mg (Please see the original sources for more details).

325
Mg was also investigated in Joiner and Aikin (1996). However, the average VCD for these profiles, where the signal was significant, is 4 × 10 10 cm −2 , which is a factor of 10 more than the VCD in Satellite measurements with long time series and daily coverage are available from GOME and GOME-2 in nadir mode and from SCIAMACHY in nadir and nominal limb mode. All 3 instruments 330 have a similar spectral resolution. VCD of Mg and Mg + were retrieved from the GOME dataset by Correira et al. (2008Correira et al. ( , 2010 and Correira (2009). Figures 1 and 2 in Correira et al. (2008) show Mg and Mg + VCD for 1996 and 1997 as well as the Mg + to Mg ratio for latitude intervals from 0-10 • and 30-40 • for both hemispheres. For the low latitudes, where there is less seasonality the VCD for Mg + are about 6-7×10 9 cm −2 , which is higher than in Fig. 9. The Mg densities are about 335 3×10 9 cm −2 , which is similar to that in Fig. 5. In the equatorial region the Mg columns agree better, and a higher VCD in the nadir results for Mg + can be explained by the thermospheric part of Mg + which is not part of the VCD in Fig. 9. At mid latitudes Mg + shows a strong seasonal cycle with a summer maximum in Correira et al. (2008). This seasonality is quite symmetric for both hemispheres with higher VCD in the SH. Furthermore, the summer maximum at mid latitudes 340 does not exceed the VCD at low latitudes. For Mg a summer maximum is observed, which is more pronounced in the SH. The Mg + to Mg VCD ratios in Correira et al. (2008)  More differences between our retrievals and the results by Scharringhausen (2007) can be found in the Mg data. This is because the data product retrieved using the Mg line is more affected by 360 the radiative transfer improvements made in this study than the product retrieved using the Mg + lines. For example, there was no correction of the Ring effect, the filling in of Fraunhofer lines by inelastic Raman scattering in the Earth's atmosphere (see, e.g. Grainger and Ring (1962)), in Scharringhausen (2007), which led to high densities below 80 km and a density maximum at the lower edge of the retrieval boundaries at 70 km. This inelastic scattering contribution additionally 365 adds a seasonal variation to the dataset.
Furthermore, the Mg line at 285.2 nm is much more affected by self absorption of the emission, which was not considered in Scharringhausen (2007), so the Mg VCD were smaller than reported here. However, most of the important findings from the further analysis of the Scharringhausen data set are only weakly affected by these differences and are still valid.

370
In summary, it can be concluded that the Mg/Mg + results presented here are often in good agreement with previous satellite and rocket instruments. The most striking difference to previous measurements is the strong latitudinal dependence of the peak altitude of Mg + with differences of up to 10 km for different latitudes.

WACCM-Mg model
The first global 3D model of magnesium, which is similar to other metals (e.g., Na in Marsh et al. WACCM is a high-top chemical-dynamical model, which simulates the altitudes from the Earth's surface up to 140 km Marsh et al., 2007Marsh et al., , 2013b. Here we use the specified dynamics SD-WACCM, which is nudged by the GEOS5 meteorological dataset (including tempera-385 ture, specific humidity, horizontal winds) below 60 km. This is the same version used by Feng et al. Magnesium constitutes 9.6% of chondritic meteorites by mass, so the equivalent interplanetary dust particle mass is 0.338 t d −1 . This is an extremely low number, which indicates that Mg ablates relatively inefficiently from meteoroids.
The peak ablation height is around 95 km. The fluxes are obtained by scaling the Mg MIF to 400 match the observed Mg + column density (not to the Mg column density, as Mg + is better known from earlier satellite studies). This requires dividing the Mg ablation flux by a factor of 15, relative to that of Na (Marsh et al., 2013a). Since the relative chondritic abundance of Mg is 16 times that of Na, this means that the ablation fluxes of Mg and Na are similar in the model. Similarly, when modelling the Fe layer using WACCM, Feng et al. (2013) had to reduce the Fe MIF relative to Na 405 by a factor of 4. The most likely explanation is that more extreme differential ablation is occurring.
As shown in Vondrak et al. (2008) -see Fig. 12 -the more refractory elements such as Mg and Fe ablate less efficiently than Na from slow meteoroids with entry speeds below 20 km s −1 .
The MIF used in the present study was calculated from an astronomical dust model where the average speed is around 30 km s −1 (see Feng et al. (2013) and Marsh et al. (2013a) for further 410 details). Hence, one implication of these relatively small injection rates of Mg and Fe is that a significant fraction of meteoroids enter the atmosphere from near-prograde orbits, with speeds close to the minimum of 12 km s −1 .
Tables 2-4 list the Mg chemical reactions and their rate coefficients added to the standard chemistry scheme in this study. A schematic diagram of the Mg + /Mg chemistry appears as Fig. 9 in Plane and Whalley (2012   The seasonal cycle of the Mg + layer is more easily explained than that of the Mg layer, and is mainly driven by the availability of the charged main atmospheric constituents NO + and O + 2 for charge exchange reactions. The Mg + maximum thus can be found in summer, particular at high latitudes. A summarizing schematic diagram of the neutral and ionized magnesium chemistry is shown in Fig. 9 of Plane and Whalley (2012).

475
In contrast to the SCIAMACHY measurements, which are made at one particular local time, daily averaged output is used for WACCM in Figs. 14-16. We also co-located the WACCM dataset to SCIAMACHY local time and latitudinal coverage, but found only small differences to the non co-located daily average mean of WACCM (not shown). There is a diurnal variation of the vertical column density in the WACCM data, but nearly no diurnal variation for the the vertical profile shape.
480 Figures 17 and 18 show the direct comparison of SCIAMACHY and WACCM annual means for Mg and Mg + . We used the co-located data for both plots, since these are the only plots with significant differences between not co-located daily means and co-located data, for latitudes higher than 40 • . Figures 19 and 20 show the comparison for the equatorial vertical profile only.
The peak density of Mg for WACCM in the global annual mean profile is higher than for SCIA-485 MACHY. However, the peak width of the SCIAMACHY profile is larger than for WACCM. Therefore, the VCD of SCIAMACHY and WACCM are similar, which shows, that determining the Mg input rate so that the Mg + VCD are in agreement also leads to an agreement in the neutral Mg VCD. The altitude of the maximum number density of Mg at the equator is at 90 km for both SCIA-MACHY and WACCM. In the high latitude region, which is only covered by SCIAMACHY in 490 summer, however, the maximum number density altitude of WACCM is about 2 − 4 km lower than for SCIAMACHY. Considering the 3.3 km vertical sampling in the SCIAMACHY measurements the ≈ 3.5 km vertical resolution of WACCM in the MLT and the altitude grid step size of 2 km for WACCM and 1 km for SCIAMACHY, this difference is small, but nevertheless significant.
A possible retrieval issue, which might explain the wider SCIAMACHY peak is a strong Ring 495 effect influence (Grainger and Ring, 1962) for the Mg spectral line. At the lower and upper peak edge, the SCIAMACHY number densities are not zero. This may be related to the requirement in the retrievals that the densities must be positive, which is needed for stability in the retrieval. However, the densities are smaller than the single day error there (error estimation for the Mg and Mg + dataset has been done by Langowski et al. (2013), e.g. Figs. 21-23). When averaging the data, it must be 500 kept in mind that the error for the close to zero density region does not become smaller, as this is rather a systematic error induced by initial statistical errors, than simply a statistical error.
The agreement of WACCM and SCIAMACHY is better for Mg + . The WACCM Mg + peak and column densities are of similar magnitude as the SCIAMACHY results and also show the same seasonal variability with a summer maximum. There are also small differences: the global density 505 maximum, which is found in the summer hemisphere and which moves over time to the other hemisphere, is not reduced in the WACCM data when passing the equatorial region during equinox and also has a stronger extension to the higher latitudes in summer. In addition, a second maximum appears in the winter hemisphere at high latitudes in the WACCM dataset. SCIAMACHY makes no measurements during polar night and so this second maximum is not observed. The WACCM results 510 do not show the strong seasonality and latitude dependence of the peak altitude, that can be found in the SCIAMACHY results. The peak altitude of 95 km in WACCM is in slightly better agreement with the rocket data than with SCIAMACHY. The differences of the vertical profile of WACCM and SCIAMACHY are small especially at the equator (See, Fig. 20).
The lower peak edge of Mg + is in better agreement than for Mg. The impact of the filling-in 515 of the Fraunhofer lines in the atmosphere by rotational Raman scattering by air molecules is much weaker for Mg + than for Mg. The agreement of the peak edge becomes worse with larger distance from the peak altitude. Nevertheless, the FWHM of the peak is in good agreement.
A part of the differences between the WACCM results and the SCIAMACHY measurements may be attributed to transport processes of charged particles in electromagnetic fields in the lower ther-520 mosphere. The equatorial electron distribution with transport from the equator to higher latitudes has been observed in ionospheric soundings. It is known as the equatorial anomaly or Appleton anomaly (Kendall and Windle, 1965). Typically there are fewer electrons at the equator than at 20 • on either side of the equator due to this transport.
Similar transport processes have also been observed for the ions and in situ satellite measurements 525 of the metals above 120 km, e.g. have been carried out using the Atmospheric Explorer satellite.
A discussion of this data set can be found in Grebowsky and Aikin (2002).
The vertical upward transport of Mg + near the equator has been discussed, e.g. in Hanson and Sterling (1972). The sun is passing the equatorial region from east to west. The ionisation in the thermosphere is strong for low solar zenith angles (SZAs) which leads to denser plasmas there.

530
The heavier ions and the lighter electrons gyrate in different directions in the Earth's magnetic field pointing from north to south at the equator. This leads to a charge separation, which produces an eastwards pointed electric field and a westward current of the electrons. The motion of a single ion particle with charge q can be separated into the gyration in the magnetic field B, and the motion of the guiding center. For a force F , which is perpendicular to the magnetic field, the guiding 535 center moves perpendicular to B and F with the drift velocity The magnetic field along the meridians and the electric field along the equator lead to an E × B-Drift perpendicular to both in the radial direction, which is different in sign for electrons and ions and results in a strong upward polarization field. This polarization field accelerates the ions to higher altitudes and even 540 above the F-layer. However, this effect can only explain an upward transport of Mg + within ±3 • N geomagnetic latitude. Fesen et al. (1983) showed that additionally neutral meridional winds have to be taken into account in order to explain vertical transport also at higher latitudes up to 30 • and this was experimentally shown, e.g. in Fesen and Hays (1982a,b) and Gérard and Monfils (1978).
The annual means of SCIAMACHY Mg + densities (see Fig. 7) show higher densities above 545 110 km at the northernmost and southernmost latitudes due to less coverage and therefore higher statistical errors. However, at high northern latitudes and at the ascending node side, which is cut off in the shown results, high densities are retrieved when also using the stray-light contaminated measurements. This is shown in Fig. 21. The results as well as the input raw data show low densities below 90 km and at different altitudes in between the high signal region, which makes a differentia-550 tion between a stray-light effect and a true metal emission complicated. Should this feature be real it could be explained by the cleft ion fountain found by Lockwood et al. (1985), which describes the transport of charged particles along the magnetic field lines, which strongly converge at the pole.
This transport may lift charged particles up to several Earth radii until they become neutralized and sink down if they are heavy enough and are not quickly ionized again.  Fig. 22 changes during the summer and shows a weak uplift and thinning of the layer (especially in June and July), which 560 however is hard to qualify as significant, as Mg shows strong variations and in addition the latitudinal edge region of the retrieval bears potential for systematic errors, since, e.g. the latitudinal smoothing constraint used does not work well at the edge. An uplift of the layer, may also be attributed to transport due to the upper mesospheric meridional circulation (Note that in Figs. 22 and 24 less data is cut out at the high latitudes, so that more months are covered, with the tradeoff of more edge 565 artifacts, especially in the south).
The VCD, however, is significantly reduced at polar latitudes during the summer and the reduction is stronger in the NH than in the SH. In contrast to the WACCM-Mg model results Mg + does not show the highest densities in summer at high latitudes, but at mid latitudes. NLC, which are not considered in WACCM-Mg, occur at polar latitudes during a 3 month period in the summer 570 hemisphere.
NLC can be detected in many different ways. For example they show an increased amplitude of backscattered radiation at the altitude of their appearance. This is used here to quantisize the amount of NLC. A two step algorithm is used (for details see Langowski (2014)). First it is determined whether a NLC is detected at all. Afterwards the amount of NLC is determined only for 575 the cases, where a NLC is identified, as otherwise this quantification produces wrong results. The quantification is illustrated in Fig. 23. At a certain wavelength window (here we use 458 to 552 nm), where the signal is dominated mainly by Rayleigh-scattering the average signal is formed for each tangent altitude of the MLT-measurements. NLC are typically observed as peaks at around 84 km.
The background spectra, which results only from Rayleigh-scattering without NLC is fitted and sub-580 tracted from the limb radiance. The remaining signal is summed up and the resulting quantity is called NLC radiance here.
VCD of Mg and Mg + and the NLC radiance are shown in Fig. 24. The NLC radiance is well anticorrelated with both the Mg and Mg + densities at high latitudes. However, this does not unambiguously imply that the lower VCD of Mg and Mg + are dominated by Mg being taken up by NLC 585 particles, as other effects may play a more dominant role. For example, the seasonal variation with a summer minimum for Mg is even more pronounced in the WACCM-Mg (also for Na and Fe) model, which does not consider NLC, than in the SCIAMACHY results and in the SCIAMACHY results the Mg density is also reduced at regions, e.g. 50 • N in April, where no NLC occur. The Mg + layer is formed at the top of the NLC particle region and only weakly overlaps with the region, where 590 NLC are formed. Furthermore Lorentz force induced upward transport may reduce the Mg + density at the poles.
In summary it is interesting to note that the VCD of both Mg and Mg + are lower in summer during NLC episodes but further investigations are needed to understand the origin of the lower Mg and Mg + densities to decide whether loss to NLC or other processes are dominant.  Mg + shows a clear seasonal cycle with a summer maximum in the peak density, most pronounced 605 at mid latitudes. Mg + shows a seasonal variation of the peak altitude with higher altitudes in the summer, as well as a latitudinal dependence of the peak altitudes with up to 5-10 km higher altitudes at mid latitudes compared to the equatorial peak altitude of 95 km.
The SCIAMACHY results are reasonably consistent with previously published rocket profiles and significantly extend the amount of data of Mg and Mg + . The peak density values and the observed 610 seasonality for Mg + are in good agreement. However, the rocket measurements do not show the latitudinal dependence of the peak altitude, but typically show peak altitudes at slightly below 95 km.
The SCIAMACHY results were compared with WACCM-Mg model results. For Mg WACCM shows a clearer seasonal cycle with a winter maximum at the poles than SCIAMACHY. The WACCM Mg peak densities are roughly a factor 2-3 bigger than the SCIAMACHY peak densi-615 ties. The peak altitude and peak shape are in good agreement with SCIAMACHY, which shows a slightly wider profile. The combination of higher peak densities for WACCM, but wider peak profile for SCIAMACHY leads to very similar values for the VCD.
The agreement between WACCM and SCIAMACHY is generally better for Mg + than for Mg.
Both SCIAMACHY and WACCM Mg + results show a clear seasonal cycle with a summer maxi-620 mum and the peak densities are in good agreement. However, similar to the limited number of the rocket measurements, WACCM does not show a latitudinal dependence of the peak altitude. The peak shape of Mg + agrees well for WACCM and SCIAMACHY in the high density region, where the SCIAMACHY error is low.
During the arctic summer Mg and Mg + show a density minimum at high latitudes. This is in 625 coincidence with the occurence of NLC, but our results do not allow the conclusion that the Mg and Mg + reduction at high latitudes during summer is related to the occurrence of NLC. Further investigations involving model simulations are required to address this question.
Zbinden, P. A., Hidalgo, M. A., Eberhardt, P., and Geiss, J.: Mass spectrometer measurements of the positive ion composition in the D-and E-Regions of the ionosphere, Planet. Space Sci., 23, 1621Sci., 23, -1642Sci., 23, , 1975.   (2012) R27 Mg + + e − → Mg + hν k27 = 1.0 × 10 −12 Plane and Helmer (1995) R28 Mg + O2 + → Mg + + O2 k28 = 1.2 × 10 −9 Rutherford et al. (1971) R29 Mg + NO + → Mg + + NO k29 = 8.2 × 10 −10 Rutherford et al. (1971) R30 MgN2 + + O2 → MgO2 + + N2  to-month variations. The maximum peak altitude is 5-10 km higher than the minimum peak altitude. For mid latitudes, the peak altitude in the winter hemisphere is still higher than the peak altitude at the equator (see also  and shows a summer maximum, which is in good agreement with Correira et al. (2008). Day of year  (Fig. 1, cyan). 0.5 × 10 9 cm −2 and 8 × 10 9 cm −2 and show a clear seasonal cycle with a winter maximum, which is most pronounced at the poles. Mg + VCD vary from 1×10 9 cm −2 to 5 × 10 9 cm −2 , with most VCD between 2 × 10 9 cm −2 and 3 × 10 9 cm −2 . A clear seasonal cycle with a summer maximum for Mg + is observed. 140 km is 2.3 × 10 9 cm −2 for SCIAMACHY and 2.5 × 10 9 cm −2 for WACCM. Note that the statistical error for SCIAMACHY estimated by Langowski et al. (2013) (Fig. 21) is ≈ ±200 cm −3 for the low density region above 100 km and below 80 km and ≈ ±400 cm −3 at 90 km. stray-light contaminated measurements at the start of the sunlit part of each orbit are also used here extending the coverage to higher northern latitudes and even the ascending node side. High Mg + densities are observed at the north pole. See Fig. 1 and Sect. 4 2nd paragraph for an explanation of the x-axis labels. Note that this is an older plot, which shows data with a wrong polarization correction, which is, however, unimportant for the effect shown here.   observed in neighbouring regions without NLC and can also be explained by other processes. (Note that the NLC retrieval includes more latitudes from the ascending node side).