We have studied the short-term effect of the October–November 2003 series of solar proton events on the middle atmosphere. Using the proton flux measurements from the GOES–11 satellite as input, we modeled the effect of the precipitating particles between 26 October and 6 November with a one–dimensional ion and neutral chemistry model. Then we compared the results with ground-based radio propagation measurements, as well as with NO2 and ozone profiles made by the GOMOS satellite instrument. The very low frequency signal experiences up to −7 dB absorption during the largest solar proton event, subsequently varying with time of day during the recovery phase. The model and radio propagation observations show very good agreement, suggesting that the model is capturing the impact of solar protons on the ionosphere. The model results show order-of-magnitude changes in odd hydrogen and odd nitrogen concentrations, as well as ozone depletion varying from 20% at 40 km altitude to more than 95% at 78 km. The magnitude and altitude distribution of ozone depletion is found to depend not only on the flux and energy of the protons but also on the diurnal cycle of atomic oxygen and ozone-depleting constituents so that the largest depletions of ozone are seen during sunrise and sunset. The after-event recovery of ozone is altitude-dependent because of the differences in the recovery of odd hydrogen and odd nitrogen and also because of a relatively faster ozone production at higher altitudes. The modeled and measured NO2 profiles agree well at altitudes 35–60 km, particularly during times of large concentrations observed after the solar proton event onset. A comparison of the time series of ozone depletion shows a good agreement between the model and observations.
 Solar proton events (SPE) correspond to solar coronal mass ejections (CME) during which a large amount of protons and heavier ions are emitted [Reames, 1999], sometimes toward the Earth. Solar protons entering the Earth's magnetosphere are guided by the Earth's magnetic field and precipitate into the polar cap areas [Patterson et al., 2001]. Since the protons can have very high energies, up to tens of MeVs, they deposit their energy in the mesosphere and stratosphere. Thus they provide a direct connection between the Sun and the Earth's middle atmosphere. SPEs are relatively sporadic (only a few large solar proton events with proton energies sufficient to penetrate down to the stratopause region reach Earth during a solar maximum) but are extreme examples of solar forcing on the middle atmosphere.
 The Sodankylä Ion Chemistry model (SIC) is a one-dimensional chemical model designed for ionospheric D-region studies. It solves for concentrations of 63 ions and 13 neutral species in the altitude region 20–150 km. The SIC model includes an extensive range of positive and negative ion chemistry so that the effects of ion chemistry on the neutral atmosphere can be modeled without parameterization. During a solar proton or electron precipitation event the model can be used to monitor the day-to-day and also diurnal variation of HOx/NOx increases, as well as ozone depletions which have been reported by Aikin and Smith  to be dependent on the diurnal cycle of HOx constituents. The dependence of HOx induced ozone loss on the solar zenith angle has also been discussed by Solomon et al.  and Jackman and McPeters .
 Very low frequency (VLF) radio signals used in communication systems are known to be severely affected during SPEs [Cummer et al., 1997]. The phase and amplitude of VLF signals are modified during an SPE as a consequence of the enhanced ionization in the atmosphere changing the propagation conditions in the Earth-ionosphere waveguide. We use the SIC model electron density profiles as an input to the Long Wave Propagation Code (LWPC) to model the behavior of VLF signals traveling through the geomagnetic polar cap (see Clilverd et al.  for a description of this technique). In the present case we compare observations of the Iceland (NRK) VLF transmitter received at Ny Ålesund, Svalbard with SIC-based propagation modeling results.
 The GOMOS (Global Ozone Monitoring by Occultation of Stars) instrument on board the European Space Agency's Envisat satellite measures vertical profiles of several middle atmospheric gases, including O3 and NO2. GOMOS makes several hundred occultations per day with good global coverage, including the polar areas. The advantage of stellar occultation technique is that the measurements can be made in both day and night conditions with no strict restriction on the local time of the measurements. Therefore GOMOS is a powerful instrument for monitoring the diurnal variation of ozone in the middle atmosphere.
 In this paper we use the VLF radio signal as well as NO2 and O3 measurements together with the Sodankylä Ion Chemistry model to study the effect of the October–November 2003 solar proton events on the D-region ionosphere and neutral middle atmosphere. For the first time, measurements of SPE-induced ozone loss were obtained from the nightside of the atmosphere [see also Seppälä et al., 2004]. This is important because most of the ozone-destroying HOx and NOx reaction cycles depend on solar light so that significant differences in ozone depletion with respect to diurnal cycle can be expected.
2. October–November Solar Storms
 The storm period in October 2003 began with a X1-class (X1: peak of 0.1–0.8 nm x-ray flux = 1 × 10−4 Wm−2) flare observed on the west side of the solar disc on 26 October. This flare was followed by a halo-type coronal mass ejection (CME) but did not greatly enhance the flux of high energy particles in the near Earth space. On 28 October a second flare was observed in the eastern part of the solar disc. This X17-class flare was also associated with a halo-type CME which this time caused a remarkable enhancement in the particle flux measured in the geostationary orbit (NOAA NGDC–Solar Terrestrial Physics, http://www.ngdc.noaa.gov/stp/stp.html) triggering a severe geomagnetic storm and a solar proton event at the Earth (flux of E ≥ 10 MeV particles >104 particles cm−2 s−1 sr−1). Another CME was again observed a few days later on 2 November, and on 4 November the Sun released the largest flare ever observed which saturated the x-ray detectors on the GOES satellites. This flare was later estimated to be a X28-class flare. However, studying the effects that the flare had on VLF signals propagating under the flare-modified ionosphere, a new measurement has indicated this event was in fact X45 ± 5 (peak of 0.1–0.8 nm x-ray flux = 45 × 10−4 Wm−2) [Thomson et al., 2004].
 After the large flare and the following CME, the solar activity started to decline. However, the high-energy particle flux stayed well above the usual background levels (flux of E ≥ 10 MeV particles < 1 particle cm−2s−1sr−1) until mid-November.
3. Sodankylä Ion Chemistry Model
 The Sodankylä Ion Chemistry Model (SIC) was applied first by Burns et al.  in a study of EISCAT radar data, and thereafter by, e.g., Turunen , Rietveld et al. , Ulich et al. , and Verronen et al. . A detailed description of the original SIC model, in which only ion chemistry was considered, can be found in the work of Turunen et al. . The latest version solves the concentrations of 63 ions, including 27 negative ions as well as 13 neutral species (O(3P), O(1D), O3, N(4S), N(2D), NO, NO2, NO3, HNO3, N2O5, H, OH, and HO2) between the altitudes of 20 and 150 km, with 1-km resolution, taking into account several hundred chemical reactions and external forcing due to solar radiation, electron and proton precipitation, and galactic cosmic rays.
 The background neutral atmosphere is generated using the MSISE-90 model [Hedin, 1991] and tables given by Shimazaki . The former provides altitude profiles of N2, O2, Ar, He, and temperature with 1-km resolution for any given set of time, geographic location, magnetic Ap index, and solar F10.7 flux. The latter provides concentrations of O2(1Δg), N2O, H2, H2O, H2O2, HNO2, HCl, Cl, ClO, CH3, CH4, CH2O, CO, and CO2 for noon and midnight conditions at altitudes 10, 15, 20, 25, 30, 45, 60, 80, and 100 km, which are then converted into altitude profiles of 1-km resolution by interpolation. For the 1-km Shimazaki-based profiles, interpolation with respect to solar flux is used to make the transition from day to night and vice versa. The concentrations of H2O and CO2 are calculated using fixed volume mixing ratio profiles, the default values are 5 ppmv (below 80 km) and 335 ppmv, respectively.
 The numerous sources of reaction rate coefficients for the ionic reactions are listed in the work of Turunen et al.  along with the additions listed in the work of Verronen et al. . The negative ion chemistry scheme and the ion-ion recombination coefficient have been recently checked and revised according to and references therein Kazil et al. . The neutral chemistry includes 59 reactions of the modeled neutral species, for which the rate coefficients have been updated according to Sander et al. . Most of these reactions are listed in the work of Verronen et al. . The additions and changes made for this study are presented in Table 1.
Table 1. Recent Additions and Changes in the Neutral Chemistry Schemea
The coefficient of R4 is that of Sander et al.  reduced by 50%. The coefficient of R5 is the lower limit given by Sander et al. . Notation: M is any atmospheric molecule and T is temperature. Units for reaction R7 are cm6 s−1 and for the other reactions cm3 s−1.
 The model includes a vertical transport code, described by Chabrillat et al. , which takes into account molecular and eddy diffusion. Within the transport code the molecular diffusion coefficients are calculated according to Banks and Kockarts . We use a fixed eddy diffusion coefficient profile, which has a maximum of 1.3 × 106 cm2 s−1 at 102 km.
 The SIC model can be run either in a steady-state or a time-dependent mode. For this study we used the time-dependent mode which exploits the semi-implicit Euler method for stiff sets of equations [Press et al., 1992] to advance the concentrations of the chemical species in time. Vertical transport and chemistry are advanced in 15-min intervals during which background atmosphere and external forcing are kept constant. In the beginning of every interval all modeled neutrals, except the short-lived constituents O(1D) and N(2D), are transported. Next, new values for solar zenith angle, background atmosphere, and ionization/dissociation rates due to solar radiation and particle precipitation are calculated. Finally, the chemistry is advanced.
3.1. Proton Forcing
 We have undertaken ionization rate calculations using GOES-11 satellite proton flux data, available from the NOAA National Geophysical Data Center World Wide Web server at www.ndgc.noaa.gov/stp/stp.html. GOES satellites measure integrated proton fluxes above seven threshold values: 1, 5, 10, 30, 50, 60, and 100 MeV. An integrated proton flux can be described by an exponential rigidity relation [Freier and Webber, 1963], which we used to convert the GOES measurements to differential proton fluxes over the range 600 keV to 2000 MeV. The ionization rate calculation is based on proton energy-range measurements in standard air [Bethe and Ashkin, 1953]. The energy-range relation for protons can be written
where R is the range, E is the proton energy, and a and b are parameters set by measurements. Using the following algorithm, originally presented by Reid , we calculate energy deposition rates on each altitude: The stopping power of a proton with initial energy E and pitch angle θ at altitude h is
is the remaining range at altitude h. The total concentrations n0 and n(h) are taken at ground level and at altitude h, respectively. The last term in equation (3) is the standard air range, i.e., energy, lost by a proton in penetrating the atmosphere down to altitude h. By dividing equation (2) by mean ionization energy Δε, taken to be 36 eV [see, e.g., Rees, 1989, p. 40], and multiplying by the proton flux F(E) and then integrating over energies and angles, we get the total ionization rate for each altitude from
The total ionization rate is divided between N2, O2, and O, according to their relative concentrations and cross sections [Rees, 1982]. Then these three ionization rates are divided between the ionization and dissociative ionization processes using the branching ratios given by Jones  to obtain the production/loss rates for the individual species.
 As a results of proton precipitation, atomic nitrogen is produced from N2 in dissociative ionization by primary protons and in dissociation by secondary electrons. In addition, ion chemistry produces both atomic nitrogen and nitric oxide. For the production of atomic nitrogen it is important to consider the branching between the ground state N(4S) and the excited state N(2D) because the former destroys while the latter produces NO. Table 2 lists the reactions producing N(2D) in the SIC model. HOx production is a result of ionization, water cluster ion formation, and subsequent neutralization. The SIC model considers also an additional route to HOx production via HNO3 formation due to ion-ion recombination and subsequent photodissociation of HNO3 [Solomon et al., 1981; Aikin, 1997].
Table 2. N(2D)-Producing Reactions, Rates/Rate Coefficients, and Branching Ratios Used in the SIC Modela
Rate/Rate Coefficient, cm−3 s−1
Notation: p*p = primary proton, e*s = secondary electron, e = thermal electron, and Q = total ionization rate due to proton precipitation.
 For this study, we selected the location 70°N/0°E and an altitude range of 20–120 km. Before modeling the October–November 2003 solar proton events, we set up the SIC model for quiet time conditions equivalent to late October by repeating a diurnal cycle until convergence, i.e., no significant change between the last and previous cycle. This initialization of the model typically requires 8–12 cycles.
 The level of solar illumination is relatively low during the modeled period. The solar zenith angle at noon is about 84° so that the photodissociation processes are diminished particularly in the stratosphere. This results in a relatively low production of photolysis products, such as O(1D). At the stratopause, the sunrise and sunset occur at approximatively 0700 LT and 1700 LT. The zenith angle is 90° at about 0845 LT and 1515 LT.
 The minor gases of the background atmosphere are an important part of the modeling because they can be used to control the initial level of the modeled constituents. For example, changes in amounts of H2O (source of HOx), N2O (source of NOx), or ClOx could affect the amount of ozone. Test runs for this particular modeling case indicate that the amount of ozone is quite insensitive to changes in N2O because the reaction with O(1D), producing NO, is diminished due to relatively low level of solar radiation. On the other hand, changes in H2O and ClOx have a clear effect on ozone at certain altitudes.
 During the initialization, to improve comparison with GOMOS ozone measurements, we made adjustments to the model background atmosphere and diffusion boundary conditions: (1) Daytime Cl and ClO values given by Shimazaki  were doubled around 30 km where the sensitivity of ozone to changes in ClOx is largest, new values being 6.1 × 104 and 1.0 × 108 cm−3, respectively. (2) Below 83 km the water vapor amount was increased from the default 5 ppmv to 7.5 ppmv between 20 and 70 km. At 71 and 83 km values 7.5 and 3 ppmv were used, respectively, and a linear interpolation in between. (3) At 120 km the atomic oxygen downward flux was set to 1011 cm−2s−1. (4) Two reaction rate coefficients were adjusted; see Table 1. The comparison with GOMOS is presented in section 5.2.
 After initialization, we introduced the proton forcing, i.e., included ionization rates calculated from GOES proton flux data. The fluxes were taken to be isotropic. The location of our modeling (70°N/0°E) corresponds to a high magnetic latitude of 68°N. Therefore as a first approximation, we can neglect the magnetic cutoff effects [see, e.g., Hargreaves, 1992, pp. 351–361]. Figure 1 shows the calculated ionization rates for 26 October to 7 November.
 We started the modeling at 0000 UT on 26 October and modeled the event until 7 November, 0000 UT. We call this the proton run. We also repeated the same modeling run without proton forcing. This we call the control run.
4.1. Subionospheric VLF Propagation
 Very low frequency (VLF) (3–30 kHz), long-wave radio signals are used in communication systems, for example, between ground stations and submarines. The signals used in communication systems are generated by high power transmitters but VLF signals can also be generated by natural processes such as lightning. The VLF signals propagate in the waveguide formed by the Earth's surface and the bottom of the Earth's ionosphere located between 50 and 100 km [Barr et al., 2000], that is, subionospherically. Therefore changes in the ionosphere produce changes in the received amplitude and phase of the VLF signals.
 The signals coming from distant locations can be monitored by VLF receivers set up in different locations around the Earth. In this study, we have used the VLF receiver at Ny Ålesund, Svalbard, Norway (78°54′N, 11°53′E, L = ∼18), to monitor the VLF signal coming from Reykjavik, Iceland (NRK, 37.5 kHz). The great circle path from NRK to Ny Ålesund (see Figure 2) crosses through the magnetic polar cap area, close to the SIC modeling location, allowing the signal to be influenced by ionospheric changes caused by the proton precipitation.
 To study the signal propagation conditions, we have used the Long Wave Propagation Code (LWPC) [Ferguson and Snyder, 1990] provided by the Naval Ocean Systems Center (NOSC) to model the NRK VLF signal. To calculate the signal amplitude and phase at the reception point, LWPC needs electron density profile parameters that define the ionospheric conditions. These parameters are calculated from electron density profiles provided by the SIC model results [Clilverd et al., 2005] made during the proton run. Thus we are able to compare the observed NRK to Ny Ålesund amplitude variations during SPE conditions with the output of the LWPC propagation model. The LWPC calculations are done for the same NRK to Ny Ålesund propagation path and the same transmitter frequency, using the SIC electron density profiles to define the changing ionospheric conditions during the SPE.
4.2. NO2 and O3 Measurements by GOMOS
 GOMOS is a stellar occultation instrument on board the European Space Agency's Envisat satellite [Bertaux et al., 1991, 2004; Kyrölä et al., 2004]. Launched from French Guiana on 1 March 2002, Envisat carries a total of 10 instruments. These include three atmospheric chemistry instruments: GOMOS, MIPAS (Michelson Interferometer for Passive Atmospheric Sounding), and SCIAMACHY (SCanning Imaging Absorption SpectroMeter for Atmospheric CHartographY).
 GOMOS consists of a star tracker, used to guide the pointing system, two spectrometers working in the ultraviolet-visible-near-infrared (UV-vis: 248–690 nm and IR: 750–776 nm, 916–956 nm), and two photometers (blue: 470–520 nm and red: 650–700 nm). GOMOS measures the light of a star as the star descends through the Earth's atmosphere. The incoming light travels through the atmosphere and is absorbed and scattered by various molecules along its path. Knowing the absorption features of the different atmospheric gases their altitude profiles can be calculated using advanced inversion methods [Kyrölä et al., 1993]. The wide spectral range of GOMOS enables inversion for vertical profiles of O3, NO2, NO3, H2O, O2, neutral density, and aerosols. The two photometers allow measurement of temperature profiles with high vertical resolution and can also be used to study turbulence in the atmosphere. The altitude range of the measurements is between 10 and 100 km for ozone and, because of their low abundance in the mesosphere and above, 10–50 km for the other gases. The measurement vertical resolution is better than 1.7 km. Using stars as a light source instead of sunlight, measurements from both dayside and nightside of the Earth can be attained (henceforth referred as bright limb and dark limb measurements, respectively). In addition to stellar occultation, the main measurement technique, GOMOS is also able to measure scattered solar light in the Earth limb.
 The altitude range and error of GOMOS measurements are dependent on the star temperature and magnitude. For example, in the case of ozone the bright and hot stars give best accuracy with typical uncertainties of the order of 5% around both the primary and secondary maximum of ozone. The stratospheric ozone product of GOMOS has been extensively validated by Meijer et al. , who concluded that GOMOS nighttime measurements between 19 and 64 km agree very well with ground-based and balloon-sonde measurements, showing a small negative bias of 2.5–7.5%. On the other hand, GOMOS data quality was found to be strongly dependent on the illumination of the limb through which the star is observed, the bright limb measurements showing a strong negative bias of 18–33% and a limited altitude range.
 In this study we have used GOMOS measurements from the Northern Hemisphere polar area from geographic latitudes 65° to 75° before and during the great storm period of October–November 2003. All NO2 and ozone profiles were produced using GOMOS processing prototype version 6.0a. We selected stars with magnitude better than 2.834 and required the star temperature to be higher than 7000 K and 10,000 K for the dark and bright limb, respectively. We are aware of the negative bias in the daytime ozone measurements. However, we assume that by selecting the measurements in such a way that the difference in solar zenith angle is small during the period of the study, the daytime measurements should have equal relative bias allowing the change of ozone during the proton event to be monitored correctly.
5. Results and Comparison
 The modeled response of the middle atmosphere to the solar proton event is presented in Figure 3. Negative ion concentrations are enhanced by several orders of magnitude during the periods of extreme forcing. This is directly dependent on the ionization rate and thus can be used for monitoring of the magnitude of the forcing below 80 km. The events on 26–27 October, 28–31 October, 2–4 November, and 5–6 November are easily identified.
 HOx (H + OH + HO2) displays a strong diurnal cycle below 80 km. The HOx response to proton forcing is clear at night, with order-of-magnitude changes on 29–31 October and 3 November. During daytime the natural production from water vapor is high and thus SPE effects can only be identified during the most intense ionization periods [see also Solomon et al., 1981]. HOx recovers relatively fast (in a day), which is demonstrated after the end of the largest events, on 1 November as well as on 5 November. Above 80 km the impact of protons is small on HOx levels. It should be noted that around 70 km the background altitude profiles of H2O2 used in the model could result in an overestimation of the HOx concentration and ozone depletion by a few percent.
 NOx (N + NO + NO2) displays a quite different response from that of HOx. Once affected by proton forcing, NOx stays at an elevated level. The recovery is slow because of the long chemical lifetime of NOx especially at high solar zenith angles [see, e.g., Brasseur and Solomon, 1986, pp. 256]. On 26–27 October, NOx is enhanced at 60–80 km, while the largest event on 28–31 October leads to enhancements of several hundred per cent at altitudes above 40 km. In contrast, the effect of the 2–4 November and 5–6 November events is small on the already elevated NOx levels.
 The bottom of Figure 3 shows the relative difference in Ox (O + O3) between the proton and control runs, i.e., 100 × (proton/control − 1). Ox is depleted between 40 and 85 km, with clear differences in the magnitude of depletion with respect to the diurnal cycle and altitude. Substantial Ox loss occurs at sunset of 28 October. During night the Ox depletion is moderate because most of the Ox loss cycles depend on the amount of atomic oxygen. Ox does not recover, however, because it is not significantly produced at night. Even greater loss is seen at sunrise of 29 October, which is followed by a recovery at 55–75 km during the noon and afternoon hours. The maximum depletion is reached just after sunset, with a 95% reduction in the Ox values at 78 km. During daytime on 30 October, Ox partly recovers but is again depleted during sunset. The next day, 31 October, a recovery again occurs after which the Ox values stay on 0–30% lower level until the next event on 2–4 November.
 Ox depletion is driven by the proton forcing. The amount and altitude distribution of the depletion is dependent on the flux and energy of the protons. It is also dependent on the diurnal variation of HOx. The HOx reaction cycles have the greatest impact on Ox during sunrise and sunset when (1) there is significant enhancement of HOx due to proton forcing and (2) atomic oxygen is available. Ox recovery occurs during noon hours, when the photodissociation processes are most effective, because of the high background production of HOx and the production of Ox at that time. The recovery is relatively slow and displays an altitude dependence, i.e., the recovery is generally faster at higher altitudes, because the solar zenith angle is high throughout the day. NOx does not show significant diurnal variation but the reaction cycles destroying Ox require atomic oxygen. Therefore the NOx cycles are most effective during daytime. The nearly constant 20% Ox depletion at 50 km, beginning on 29 October at noon and lasting beyond 6 November, is for the most part due to the increase and persistence of NOx.
5.1. VLF Signal Modeling and Comparison
 The average quiet-day behavior (Quiet Day Curve, QDC) of the NRK signal received at Ny Ålesund was determined from the average of 4 days before and after the period of high magnetic activity. This is shown in Figure 4a as the solid line. The QDC shows an essentially “v”-shaped variation with a minimum amplitude at 1100 UT and a maximum at 1900 UT. The total amplitude change is about 8 dB throughout the day. The equivalent QDC determined from calculating the NRK propagation conditions with the LWPC VLF propagation code (diamonds), using the quiet-time SIC electron density profiles to define the ionospheric profile [e.g., Clilverd et al., 2005], shows low amplitudes during the daytime (0700–1600 UT), and high amplitudes during the night (1700–0600 UT). The SIC based amplitude range is about 7 dB. Clearly, the quiet-time SIC model is providing electron density profiles that can well represent the radio propagation conditions inferred from observing the behavior of the NRK signal. Although not shown here, we also studied the Norway (JXN, 16.4 kHz) signal to Ny Ålesund in a similar way (see Figure 2 for path information). Very good agreement was found between the observations and the modeling in this case.
 The SPE studied here starts at 1215 UT, on 28 October. Figure 4b shows that the NRK signal from the first half of the day (solid line) initially follows closely the QDC determined previously (dashed line). However, just prior to the SPE start the signal is elevated in amplitude relative to the QDC as a result of the precipitation caused by the previous small SPE that occurred on 26 October.
 The SPE proton fluxes peak on 29 October (see Figure 1), before gradually returning to low levels by the end of 31 October. During this period the effect of the SPE on the observed NRK amplitude can be seen as varying signal absorption, with values exceeding QDC levels at times, and returning to close to QDC levels on 31 October. This is shown in detail in Figure 4c by the solid line. The calculated SPE effect on VLF propagation conditions driven by the changing SIC model electron density profiles is also shown in this figure (diamonds). These have been calculated using the LWPC VLF propagation code. The calculations show good agreement with the observations, in that there is initially absorption of the signal to about −4 dB, followed by varying levels of absorption as the signals recover to QDC levels during 31 October. Such good agreement suggests that the SIC model is capturing the impact of the solar protons on the ionization and ion chemistry of the high latitude winter atmosphere, as well as the quiet-time prestorm conditions.
5.2. GOMOS Comparison
Figures 5–6 present comparisons between the SIC model results and GOMOS measured profiles. The comparisons discussed here are made with GOMOS weighted mean profiles, which are represented in the figures by the dashed lines.
 Before the solar proton events the agreement between the nighttime NO2 profiles is reasonably good between 35 and 60 km, the model predicting 50% lower and 200% higher values at 35 and 45 km, respectively. Above 60 km the measurements show order-of-magnitude higher values. It should be noted that the accuracy of the GOMOS NO2 measurements becomes worse with increasing altitude due to decreasing abundance of NO2, such that the upper limit of NO2 measurements is usually considered to be at 45–50 km. The scatter of the measurements above 40 km indicates that the uppermost GOMOS NO2 measurements may be at this noise floor. Differences in the diurnal phase of NO2 cannot explain the scatter because the measurements were obtained at a virtually constant solar zenith angle. On 3 November the agreement between the NO2 profiles is better. The profiles are very similar in shape and differ in magnitude by less than 50% between 35 and 60 km. The proton forcing brings the model and measurements into a better agreement. This indicates to us that the modeling of SPE effects, at least the production of NOx, is generally correct. Also, the larger amount of NO2 at the higher altitudes seems to give more accurate measurements by GOMOS so that for special cases of enhanced production it is possible to extend the altitude range of NO2 measurements above 50 km.
 Nighttime ozone profiles (Figure 6, left) show generally a good agreement below 40 km. On the other hand, SIC seems to overestimate the amount of ozone between 40 and 80 km by up to 100% and underestimate it around the secondary maximum. The situation is not changed by the solar proton events, as the comparison of 3 November profiles shows. However, the comparison of relative change of ozone displays quite similar features, although there are quite substantial quantitative differences at many altitudes. A continuous depletion region is located between 30–70 km with the maximum change at 45–50 km. At the maximum the measurements show about twice as much depletion as the model. Both the model and measurements show an increase in ozone around 70 km and a depletion around 80 km, although the magnitudes again differ. Above 80 km the agreement between SIC and GOMOS is poor.
 Daytime ozone profiles (Figure 6, right) show best agreement between 60 and 70 km. Below 60 km the model overestimates the amount of ozone predicting 100% higher values between 35 and 50 km. A part of this difference is probably due to the negative bias in GOMOS daytime measurements (see section 4.2). The model shows again significantly lower values around the secondary maximum of ozone. On 3 November the situation is similar, although the comparison between 40 and 50 km is slightly worse, the model showing about 120% higher values. The profiles of relative change agree very well, also in the magnitude of depletion. The model successfully reproduces the main depletion region between 50 and 80 km and the maximum depletion is seen around 70 km in both modeling and measurements.
 The variation of the relative changes of ozone between 23 October and 5 November is presented in Figure 7 for three altitudes, 50, 60, and 70 km. The agreement between GOMOS measurements and SIC modeling is generally good, although at 50 km the initial effect of the 29 November event seems to be underestimated by the model. It is obvious that the SIC model is able to represent reasonably well the day-to-day changes in ozone due to the proton forcing. The diurnal variation of ozone depletion increases with increasing altitude because of increasing dependency on the HOx reactions and faster daytime production of Ox. At 50 km the variation is not significant while at 70 km the magnitude of the depletion can differ within a day by several tens percent, depending on the local time considered. The largest depletions occur during sunrise hours. Therefore an instrument measuring solely at that time would generally overestimate the magnitude of ozone depletion in the middle mesosphere [see also Aikin and Smith, 1999].
 We have studied the short-term effects of the October–November 2003 solar proton events on the middle atmosphere. Model results as well as VLF and GOMOS measurements indicate significant changes in the D-region ionosphere, enhancement in concentrations of odd hydrogen and odd nitrogen, and depletion of ozone in the middle atmosphere. The ozone depletion displays a clear diurnal cycle in the middle mesosphere as a result of being modulated by the HOx cycle but also due to faster ozone production, i.e., recovery, at higher altitudes.
 We have used the Sodankylä Ion Chemistry Model electron densities as an input to the LWPC model to model the behavior of the VLF signal from the NRK transmitter to Ny Ålesund and compared them with the equivalent observations. The results show a good agreement between the modeled and the measured VLF signal, thus indicating that SIC is correctly calculating the ionization and the electron densities. This suggests that taking into account ionization from only proton precipitation and secondary electrons but not from additional electron precipitation is a good assumption for the model SPE study.
 During the events, modeled HOx and NOx values are both enhanced drastically but display a quite different recovery. HOx show a rapid response to proton forcing below 80 km, which is most visible at night with order-of-magnitude enhancements, and a quick recovery after one quiet day. NOx is affected above 40 km sustaining its level after proton forcing and continuing to affect ozone beyond our model calculation [Seppälä et al., 2004]. The comparison of NO2 profiles shows agreement between the SIC calculation and GOMOS measurements, especially after the onset of the event, indicating a generally correct modeling of NOx production.
 Model results show that ozone is depleted between 40 and 85 km during the solar proton events. The largest relative changes, due to HOx increase, occur just after sunset on 29 October between 75 and 80 km, with a maximum depletion by 95%. The magnitude of the ozone depletion is dependent not only on the flux and energy of the protons but also on the diurnal cycle of HOx constituents. A relatively fast partial recovery of ozone occurs at 60–85 km after a reduction in the proton forcing. Around the stratopause a persistent 20% decrease due to NOx increase/persistence is seen after the largest event on 28–31 November, lasting until the end of our modeling calculation (6 November). These model results are in good agreement with ozone measurements of the GOMOS instrument.
 The work of PTV was partly supported by the Academy of Finland through the ANTARES space research programme. The work of AS was supported by the Academy of Finland (MAIST: Middle Atmosphere Interactions with Sun and Troposphere). CFE acknowledges financial support through the European Community's Human Potential Programme under contract HPRN-CT-2002-00216, CAL. PTV thanks Simon Chabrillat for providing the vertical diffusion code and Donal Murtagh for providing the transmission and O2 cross-section code for the Schumann-Runge bands.
 Arthur Richmond thanks Arthur C. Aikin and Miriam Sinnhuber for their assistance in evaluating this paper.