2.1 GOMOS on Envisat

The GOMOS spectrometer is one of nine instruments aboard Envisat. It is designed to monitor ozone profiles and other trace species using stellar occultation and atmospheric transmission measurements in limb-viewing mode (ESA, 2010). Envisat follows a Sun-synchronous orbit with an Equator crossing time (descending node) of 22:00 LT (Gottwald et al., 2011). The operation period of GOMOS dates from April 2002 to April 2012. However, there was an instrument malfunction in summer 2005, resulting in a data gap of nearly 3 months. The GOMOS instrument delivers one vertical profile of measurements for each occultation, and the altitude coverage spans from 5 to 150 km, with a vertical sampling rate of better than 2 km (Kyrölä et al., 2012). It has four spectral channels in the ultraviolet to near-infrared spectral range. The spectrometer B2 (SPB2), which provides the data used in this work, covers 925–955 nm with a spectral resolution of 0.13 nm at full width half maximum (FWHM) and a sampling step of 0.056 nm (Massimo Cardaci and Lannone, 2012). The GOMOS detector has three parallel bands. The central band probes the star spectra and the upper/lower bands record the atmospheric background radiation as calibration information, as indicated in Fig. 1. The altitudes of tangent points observed by three bands differ roughly by 1.7 km. OH and O₂ A band are regularly detected in the upper/lower bands, together with auroral emissions and the stray light scattered by particles or molecules in the atmosphere. This dataset, which is used in our analysis, is archived in the level-1b limb dataset but not directly utilized in the operational level-2 data retrieval routines. The first analyses of the extracted OH and O₂ A-band nightglow measurements from these background datasets were reported by Bellisario et al. (2014).

Figure 2Latitudinal distribution of resampled GOMOS data available from 2002 to 2012. Color coding indicates the number of selected profiles for each monthly and zonally averaged 10^∘ latitude bin.

Download

Figure 3Monthly averaged spectrum from GOMOS (black solid line) for February 2004 at 40–50^∘ N and at a tangent altitude of 89.5 km. Strong emission lines from the OH(8–4) band are annotated with the branch and rotational quantum numbers. The wavelength range from 930 to 935 nm is selected and used in the retrieval. The corresponding SCIAMACHY data (red dashed line) are also given here for comparison.

Download

2.2 Data selection and resampling

The GOMOS data were processed with the processor version 6.01–2012. The resulting level-1b limb products have already been geolocated and calibrated (Massimo Cardaci and Lannone, 2012). The signal-to-noise ratio (SNR) of single spectra is of the order of 1, and the averaging of data is required for further processing of the data.

Figure 4(a) GOMOS monthly zonal mean spectra of OH(8–4) emissions at tangent altitudes as given in the figure legend for October 2003 at 0–10^∘ S and at a local time of 22:00–00:00 LT. (b) The spectrally integrated radiance over 930–935 nm versus tangent altitude for the same conditions. The error bars indicate the measurement noise for integrated radiance (see text).

Download

The raw data from level-1b limb products are first filtered with the corresponding auxiliary “quality flag” and “product confidence data”(PCD), which indicate the presence of bad pixels, saturation, cosmic rays, modulation, dark current, flat-field or vignetting correction, with only data in the normal status being kept. This is then followed by a geolocation-related selection, in which the data with ray-tracing errors are eliminated, and their star IDs and geolocation errors are restricted to within an acceptable range, as recommended by Dehn (2012). The stray light entering the field of view (FOV) of the instrument affects the illumination of the spectrometer and enhances the background noise. The stray light is characterized by the illumination flags and solar zenith angle (SZA) of satellite and tangent points, which are geometrically computed. The illumination conditions of GOMOS measurements are categorized into five flags (Kyrölä et al., 2010; van Gijsel et al., 2010), and the “bright limb” flag thereof is excluded in this work. SZA > 108^∘ is also applied as selection criteria. Near-infrared aurora at wavelengths of around 939 and 947 nm, originating from the atomic nitrogen (N I) emissions and ${\mathrm{N}}_{\mathrm{2}}^{+}$ Meinel (2–1) band (Baker et al., 1977) are in the spectral range of SPB2. Observations in polar regions are therefore not considered in our analysis.

Due to the nature of the stellar occultation observations, the tangent points of single vertical profiles diverge significantly and are not stationary in latitude–longitude locations. In the level-2 product, they are characterized by the obliquity (Kyrölä et al., 2010), which is not available in the level-1b data. Therefore, in this work, the latitude spread of tangent points is used instead and profiles with >4^∘ deviation in tangent point latitudes are disregarded to ensure that every selected profile spans a geographical area of within ±5^∘ latitude.

The archived level-1b data are signals recorded by the detector, which must be dynamically decoded to electrons and then converted to a physical unit of flux with wavelength-specific radiometric calibration factors. The star is a point source, and part of the stellar light is spread to the lower and upper bands, which is supposed to be totally imaged in the central band in an ideal case. Considering the contamination of star leakage and residual stray light, which are assumed to be constant with altitude, the averaged spectra from above 110 km are subtracted from each profile as background radiation. No airglow emissions are found above the region of 110 km in the GOMOS measurements. The subtraction is then followed by the individual “base” removal at each altitude layer, in which this “base” offset is the mean of residual noise of the emission lines. The processed data are resampled into monthly and zonally averaged 10^∘ latitude bins with a fixed altitude grid of 3 km to enhance the spectra SNR and improve retrieval quality. The number of profiles selected for one sample bin (shown in Fig. 2) is around 100 to 300. In order to eliminate the effect of random and systematic noise as well as outliers while retaining as many profiles as possible, the largest and smallest 1 % values are disregarded from the measurements at each sample bin.

Figure 5(a) Simulated spectra (fit, solid line) and measurements (raw, dashed line) of GOMOS monthly zonal mean measurements of OH(8–4) airglow emissions at tangent altitudes, as given in the figure legend for August 2003 at 30–40^∘ S and a local time of 22:00–00:00 LT. (b) The spectrally integrated radiance over 930–935 nm versus tangent altitude for the same conditions.

Download

Figure 6Atomic oxygen abundances, derived from GOMOS monthly zonal mean measurements of OH(8–4) airglow emissions for February 2006 at 10–20^∘ N. The error bar represents the statistical uncertainty coming from the measurement noise. It increases towards higher altitudes as a consequence of the corresponding SNR being lower.

Download

Barrot et al. (2003) reported high pixel response non-uniformity (PRNU) variation of around 12 % in spectrometer B (SPB). As shown in Fig. 3, in the spectral range of our interest (SPB2), we found the GOMOS data show a good agreement with the SCIAMACHY data at the spectral range of 930–935 nm, whereas the GOMOS radiances at the wavelength range of 935–955 nm are always 25 %–30 % lower compared to the SCIAMACHY measurements, which is not understood (Erkki Kyrölä, personal communication, 2019). Therefore, of the entire spectral range, only the wavelength region of 930–935 nm is utilized in the retrieval to derive the atomic oxygen abundances. It includes a number of emission lines from OH(v= 8–4) band, which originates from the radiative transitions of OH($v^{\prime }=\mathrm{8}⟶v^{\prime \prime }=\mathrm{4}$). The dominant emission lines are mainly in the R branch with a rotational state quantum number of $K^{\prime \prime }=\mathrm{1}$, 2 and 3.

The quality of the reprocessed spectra is evaluated by calculating the standard deviation (SD) of averaged spectra for each sample bin, supplemented by the SNR analysis. The calculations show that the mean SD for a typical sample bin in autumn at midlatitudes is around 2–4 × 10⁹ photons s⁻¹ cm⁻² nm⁻¹ sr⁻¹ and that SNR increases to more than 10 at peak altitudes and to around 3–5 at lower altitudes. A typical profile of processed hydroxyl spectra and integrated radiance is illustrated in Fig. 4. Three lines are clearly visible in the spectra (Fig. 4a), while the emission peak layer appears at the tangent altitude of around 85 km, according to Fig. 4b. The error bars in Fig. 4b indicate the measurement noise for integrated radiance. The measurement noise is calculated from the standard deviation of the residual noise in the spectral range in between of the emission lines and assumed to be the same for all wavelengths, as the intensities of remaining weak emission lines from high rotational levels in the spectral region are by several orders of magnitude lower and therefore negligible. For the integrated radiance, the measurement noise is increased by a factor of $\sqrt{N}$, and N refers to the number of integrated wavelength points.

4.1 Atomic oxygen abundances

Applying the global fitting method to GOMOS level-1b limb products, a globally distributed time series [O] dataset is derived, along with other quantities. Shown in Fig. 5a is a typical profile of the fitted spectra compared with the measurements. In general, simulations and measurements are in good agreement. The spectrally integrated radiances in Fig. 5b also show consistency. The derived oxygen densities are within an altitude range of 80 to 100 km, covering the period from May 2002 to December 2011 and spanning local times from 22:00 to 00:00 LT. A typical atomic oxygen profile is shown in Fig. 6 with a maximum concentration of about 3.5 × 10¹¹ atoms cm⁻³ at 95 km. Above the maximum, there is a downward flux of atomic oxygen by diffusive transport (Swenson et al., 2018). Turbulence associated with gravity wave breaking, along with damped waves or tides, is the dynamic process that contributes to this diffusive transport (Smith et al., 1987; Li et al., 2005). Below the maximum, there is a rapid decrease in atomic oxygen density, which is mainly due to the vertical transport and chemical losses. At an altitude of around 85 km, atomic oxygen density already declines by 1 order of magnitude to 10¹⁰ atoms cm⁻³. The typical value of simultaneously retrieved spectral resolution is around 0.48 nm, and no wavelength shift is found.

Figure 7(a) The averaging kernel and (b) the vertical resolution of the retrieval for February 2006 at 10–20^∘ N and a local time of 22:00–00:00 LT. The vertical resolution is obtained from the distribution of each row in the averaging kernel by calculating the corresponding FWHM.

Download

4.2 Error analysis

The total uncertainty of the derived atomic oxygen densities not only depends on the measurement noise but also on the smoothing error as well as on uncertainties in forward model parameters and the background atmosphere input. The largest source of uncertainties is found in the forward model parameters. The influence of these uncertainties on the results is assessed through error propagation, by the perturbation of forward model parameters. The chemical reaction rate coefficient k₁ has an uncertainty of around 20 %, contributing around 15 % uncertainty below 90 km and around 20 % at 95 km in derived abundances. k₃ introduces an increasing uncertainty of up to 6 % at 95 km. The nascent branching factor (e.g., f₈, f₉) explains the distribution ratio of excited hydroxyl radicals OH^* of different vibrational levels. f₈ has a linear influence on the uncertainty of the results; a perturbation of 10 % on its values results in a similar retrieval uncertainty. The errors of Einstein coefficients correspond to an uncertainty of around 7 % in the results. The uncertainty in the quenching coefficient $k_{{\mathrm{N}}_{\mathrm{2}}\left(\mathrm{8}\right)}$ of OH^* radicals with nitrogen molecules introduces a uncertainty of 14 % at 80 km, which decreases to 5 % at 95 km, and the uncertainty in the rate coefficient for quenching by molecular oxygen $k_{{\mathrm{O}}_{\mathrm{2}}\left(\mathrm{8}\right)}$ corresponds to an uncertainty of 5 % at 85 km and 2 % at 95 km. The influences of other model parameters are on the order of 1 %–2 % or less. SABER temperature uncertainties are the predominant factor influencing the retrieval results in the background atmosphere. The uncertainties are around 5.5 K at 80 km and increase to 13 K at 90 km (Dawkins et al., 2018). Through error propagation calculation, this could lead to an uncertainty of 5 % below 90 km and up to 20 % above 95 km, taking into account the compensation effects of total density changes following the hydrostatic equilibrium (Zhu and Kaufmann, 2018).

At the altitude of 80–100 km, the effects of the smoothing error and measurement noise on the uncertainty are on the order of 0.5 % and 5 %, respectively. It is due to a properly chosen regularization in the retrieval procedure that the a priori information is negligible in the retrieval results. As part of a more in-depth look into the retrieval results, the averaging kernel and vertical resolution are investigated, as shown in Fig. 7. The summed-up averaging kernels for each row in the altitude region of interest (80–100 km) are equal to 1, indicating that the measurements instead of the a priori information contribute to nearly all of the retrieval result. The peaks of averaging kernels are found at the tangent altitudes and the corresponding vertical resolution for each altitude is around 3 km, which is close to the vertical spacing of the limb measurements. Since the sum of the averaging kernels is also near 1, the a priori influence is generally low.

Figure 8Latitude–altitude distribution of the zonal mean atomic oxygen density for 2007. Each row represents approximately a season. The data are linearly interpolated into a 1 km altitude grid for better illustration. The numbers in the subplots indicate the month of the year.

Download

4.3 Spatial and temporal analysis

Atomic oxygen reveals a two-cell structure near 95 km at midlatitudes, which is most pronounced during the equinox seasons (Fig. 8). The smallest values appear over the equatorial region and the largest values are at midlatitudes. As already mentioned and discussed by Smith et al. (2010) and Xu et al. (2010), the latitudinal distribution structure of atomic oxygen is influenced by tides. The vertical transport of air caused by tides leads to a vertical displacement of atomic oxygen. At a local time of almost midnight (the mean local time of the GOMOS measurements is around 23:00 LT, 22:00 to 00:00 LT), the atomic oxygen displacement by tides at the mesopause is upward at the Equator (resulting in an [O] decrease) and downward in subtropical latitudes (resulting in an [O] enrichment).

Figure 9Temporal evolution of the vertical distribution of monthly zonal mean atomic oxygen densities for 20–40^∘ N (a) and 0–10^∘ N (b). The data are linearly interpolated into a 1 km altitude grid.

Download

In Fig. 9, a vertical distribution comparison of derived densities from 2002 to 2011 over the midlatitude and equatorial regions is shown. Both the annual oscillation (AO) and semiannual oscillation (SAO) can be seen from the temporal evolution of middle and lower latitudes. The SAO reaches its maximum at equinox seasons, which is related to the semiannual variation of the atmospheric tide amplitudes (Smith et al., 2010).

Figure 10Multiple linear regression analysis of vertically integrated, monthly mean atomic oxygen densities of 80–97 km for 20–30^∘ N from 2002 to 2011. (a) The raw and fitted data are shown along with the baseline plus the solar components in the multiple linear regression results. The solar min and solar max values denote the fitted atomic oxygen column densities solely from the solar cycle component, under the solar minimum and solar maximum conditions, respectively. (b) SAO and (c) AO parts are also illustrated with corresponding parameters. A_SAO, P_SAO, A_AO and P_AO are the amplitudes and phase shifts of SAO and AO, respectively. The quantities of solar min, solar max, baseline and amplitudes are in units of 10¹² atoms cm⁻³. The phase shifts are in units of months. The gap present in the SAO is caused by the data discontinuity.

Download

A multiple linear regression analysis is applied to quantitatively analyze the longtime variations of the GOMOS [O] dataset. The monthly mean column density integrated from 80 to 97 km for 20–30^∘ N is analyzed by harmonic fitting, which includes components such as the solar cycle effect, SAO, AO and QBO (quasi-biennial oscillation), and baseline.

The variable t represents the month since 2002 and the column density is fitted by amplitudes (A, atoms cm⁻³) and phase shifts (P, months) of SAO (period of 6 months), AO (period of 12 months) and QBO (period of 27.5 months), complemented by the amplitude (A_solar, atoms cm⁻³ sfu⁻¹) and shift of the 11-year solar cycle effect, as well as a baseline. The coefficient I_solar is the solar radio flux proxy (F_10.7 cm, in units of sfu) taken from Tapping (2013). The typical mesospheric QBO (MQBO) period is about 27.5 months by investigating mesospheric zonal wind measurements (Ratnam et al., 2008; de Wit et al., 2013; Malhotra et al., 2016). The baseline is given as the averaged value of the monthly mean column densities along the time series. The non-linear least squares fitting method (Levenberg–Marquardt algorithm) is applied to derive these fitting parameters, as described in detail by Kaufmann et al. (2013) and Zhu et al. (2015).

In Fig. 10, the raw data and the fitting results are illustrated in panel (a). Besides, the baseline plus the solar terms are also shown in the plot. The solar min and solar max values denote the fitted atomic oxygen column densities solely from the solar cycle component, under the solar minimum and solar maximum conditions, respectively. The SAO and AO components from the harmonic fitting are given in Fig. 10b, c, respectively. The [O] longtime variations are well characterized by the fit. The 11-year solar cycle effect is captured, in which the atomic oxygen density is 17 % smaller in 2008/2009 (minimum of solar cycles 23/24) than in 2002 (near solar maximum conditions of solar cycle 23), due to different radiative forcing conditions during the solar cycles. This agrees with model investigations and experimental results, which are normally in a range of around 10 % to 30 % (Schmidt et al., 2006; Marsh et al., 2007; Kaufmann et al., 2014; Zhu et al., 2015). A significant semiannual oscillation is observed, reaching a maximum in equinox seasons, which is in agreement with the analysis above for Fig. 9, and the amplitude is about 18 % (with respect to the baseline). The annual oscillation has an amplitude of 10 %, with the maximum being reached near summer solstices and the minimum near winter solstices. These results are consistent with the analyses of Zhu et al. (2015) and Lednyts'kyy et al. (2017), which reported SAO amplitudes on the order of 15 % and 12 %, AO amplitude of 11 % and 7 %, respectively. The QBO amplitude is on the order of 2 %. The multiple linear fitting analyses on other latitudinal bands and altitudes also show a similar solar cycle effect as well as AO and SAO variations, as some examples are summarized in Table 2.

Table 2Summary of multiple linear regression analysis results of monthly mean atomic oxygen column densities integrated over 80–97 km for 20–30^∘ N, 0–10^∘ N and 20–30^∘ S from 2002 to 2011. The quantities are in units of 10¹² atoms cm⁻³. The solar min and solar max values denote the fitted atomic oxygen column densities solely from the solar cycle component, under the solar minimum and solar maximum conditions, respectively. A_SAO, A_AO and A_QBO are the amplitudes of SAO, AO and QBO, respectively.

Download Print Version | Download XLSX

Figure 11Temporal evolution of radiance differences (in percentage) between GOMOS and SCIAMACHY at a tangent altitude of 86.5 km. The radiance is integrated over the wavelength of 930–935 nm. Negative numbers indicate that SCIAMACHY radiances are larger than those of GOMOS.

Download

It could be considered to add an additional slope term in the harmonic fitting (Eq. 1) as well. In that case, the agreement between measurements and the fit increases marginally by about 2 % with an additional slope term. But the fitting parameters are not independent any longer, because a strong correlation between the slope, the baseline and the solar terms is found, which was not the case before. This indicates that the inversion problem (to obtain the fitting parameters) is now underdetermined. As an alternative approach, the solar (F_10.7) fitting parameter could be replaced by the slope term. In this case, the residual increases by about 5 % and the fitting parameters are not correlated (except for the offset and slope terms). From a mathematical point of view, this is an alternative to the original fit (with solar but without slope terms). For this setup, the slope is $-\mathrm{0.0002}\times {\mathrm{10}}^{\mathrm{12}}$ cm⁻³ month⁻¹, which means that there is virtually no trend apparent in the data. This can be explained if the change over time is considered as a combination of two linear trends, with a negative slope in the declining phase of the solar cycle and a positive slope in the following inclining phase. This hypothesis can be underpinned by looking at a subset of the time series, covering the time period from 2002 to 2009, only (roughly solar maximum to solar minimum). The slope for this period is about −3 % per year, indicating a linear decrease of atomic oxygen by 21 % for the given period. If the F_10.7 dependency is considered instead, a similar drop is modeled if a solar term with an amplitude of 0.0025×10¹² cm⁻³ sfu⁻¹ is used. This value is similar to 0.002×10¹² cm⁻³ sfu⁻¹, which is the value obtained when the total time series is considered. This line of arguments indicates that there is more likely a solar F_10.7 dependency apparent in the data than a plain linear dependency.

Figure 12(a) SCIAMACHY (solid line) and GOMOS (dashed line) observations of monthly zonal mean OH(8–4) airglow emissions at the tangent altitudes, as given in the figure legend for April 2004 at 20–30^∘ N and a local time of 22:00–00:00 LT. (b) The spectrally integrated radiance over 930–935 nm versus tangent altitude for the same conditions. The error bars are measurement noise, computed as in Fig. 4.

Download

5.1 Comparison with SCIAMACHY data

The SCIAMACHY instrument, another limb sounder aboard the Envisat satellite, observed OH emissions at various wavelengths from visible to infrared emissions (Bovensmann et al., 1999; Kaufmann et al., 2008). This provides us with the best opportunity for a comprehensive joint investigation of the GOMOS [O] dataset, as SCIAMACHY covers exactly the same OH(8–4) band wavelengths as GOMOS. Two more datasets of [O] derived from SCIAMACHY green-line emissions (Kaufmann et al., 2014; Zhu et al., 2015) and OH(9–6) band airglow (Zhu and Kaufmann, 2018) are currently available.

Figure 13Latitude–altitude distribution of percentage differences between zonal mean atomic oxygen densities derived from GOMOS and SCIAMACHY OH(8–4) airglow emissions for 2007. Each row represents approximately a season. Negative numbers indicate that SCIAMACHY abundances are larger than those obtained from GOMOS. The data are linearly interpolated into a 1 km altitude grid. The numbers in the subplots indicate the month of the year.

Download

Figure 14Latitude–altitude distribution of percentage differences between zonal mean atomic oxygen densities derived from GOMOS OH(8–4) and SCIAMACHY OH(9–6) airglow emissions for 2007. The SCIAMACHY OH(9–6) are taken from Zhu and Kaufmann (2018). This figure is plotted in a similar way to Fig. 13. Negative numbers indicate that SCIAMACHY OH(9–6) atomic oxygen abundances are larger than the GOMOS OH(8–4) abundances.

Download

SCIAMACHY performed the OH airglow measurements in dark limb-viewing mode in the flight direction, with the recorded spectra always near a local solar time of 22:00 LT and a fixed altitude grid of 3.3 km. The OH(8–4) band observation is located in channel 5 with a spectral resolution of 0.54 nm. SCIAMACHY data version 8–2016 is adopted in this work. A continuous observation was performed during the entire lifetime of Envisat. The number of recorded profiles in one sample bin was around 100–300 before 2005 and significantly increased to 400–600 because of a change in instrument operations. SNRs of single profiles are normally on the order of 6 at peak altitudes and decrease to 1 at lower altitudes. After monthly zonal averaging, SNRs increase by 1 order of magnitude, and the mean noise level is around 0.6×10⁹ photons s⁻¹ cm⁻² nm⁻¹ sr⁻¹.

Theoretically, the SCIAMACHY and GOMOS measurements should be identical in the same wavelength range. In practice, however, due to effects of various factors, such as instrument characteristics, radiometric calibration and fields of view, they do not fully conform with each other in terms of absolute radiance or instrument line shapes. In this study, the two data products are found to be consistent in terms of absolute radiance within ±20 % in the peak emissions layer (shown in Fig. 11), after monthly zonal averaging. Particularly, the difference of the GOMOS data with the SCIAMACHY data is gradually becoming positive from negative over time, and the potential source for the drift could be a degradation of the GOMOS or SCIAMACHY instruments (Bramstedt et al., 2009), which is not fully corrected or overcompensated in the level-0 calibration and the change of the system sensitivities over time. One specific example of spectra comparison is given in Fig. 12. The emission radiances from two data products are similar, but the GOMOS spectra are more noisy.

The same retrieval procedure is applied to the SCIAMACHY data. The differences between the atomic oxygen abundances from the two instruments are illustrated in Fig. 13. There are no major systematic discrepancies and they agree within a ±20 % difference in most latitude–altitude bins as expected from the differences of the corresponding radiances. The GOMOS data are found to be over 20 % lower in the Northern Hemisphere in February and also in tropical regions in March, May and September. GOMOS values appear to be 20 % larger at low altitudes of around 80 km in some scattered bins. In general, these two atomic oxygen datasets derived from OH(8–4) airglow emissions agree with each other within the combined uncertainties in the context of absolute abundances.

Figure 15Comparison of monthly zonal mean atomic oxygen densities derived from hydroxyl airglow emissions observed by the GOMOS and SCIAMACHY instruments in various latitude bins for different months. SCIA-OH(9–6) represents the atomic oxygen dataset derived from the SCIAMACHY OH(9–6) band by Zhu and Kaufmann (2018); SCIA-OH(8–4) is the dataset from the SCIAMACHY OH(8–4) band; and GOMOS-OH(8–4) is from the GOMOS measurements of the OH(8–4) band.

Download

Figure 16Comparison of atomic oxygen densities derived from various instruments and measurement techniques averaged for 20–40^∘ N (a) and 40–60^∘ N (b) in autumn (September, October and November) 2005. SCIA-O(¹S) is the atomic oxygen data derived from SCIAMACHY green-line emissions (Kaufmann et al., 2014; Zhu et al., 2015); SABER-OH refers to the atomic oxygen datasets derived by Mlynczak et al. (2018) (M18), and Panka et al. (2018) (P18) from SABER hydroxyl airglow emissions. The WINDII dataset is obtained from WINDII combined hydroxyl and green-line observations, 1993 (Russell and Lowe, 2003; Russell et al., 2005), while the OSIRIS dataset is derived from OSIRIS O₂ A-band measurements (Sheese et al., 2011). In situ data are obtained from rocket-borne experiments with mass spectrometers, conducted at different local times at 37–40^∘ N, from 1972 to 1976 (Offermann and Grossmann, 1973; Trinks et al., 1978; Offermann et al., 1981).

Download

Similarly, a latitude–altitude comparison of the GOMOS data with atomic oxygen obtained from SCIAMACHY OH(9–6) emissions (Zhu and Kaufmann, 2018) is given in Fig. 14 for 2007. In general, these two datasets agree with each other, but the GOMOS OH(8–4) dataset is found to be around 10 %–20 % lower than the SCIAMACHY OH(9–6) dataset in most latitude bins, especially in the altitude region of 85–95 km. The difference between the two datasets becomes more than 20 % at some data points near the Equator in March, May and September. Combing the derived results from SCIAMACHY OH(8–4), an intercomparison of the three datasets is given in Fig. 15 for different latitudinal and seasonal conditions. The absolute abundances of the three datasets are in the same order of magnitude and they agree with each other at the altitude region of interest of 80–95 km. Specifically, atomic oxygen abundances derived from OH(8–4) emissions by both instruments are found to be 10 %–20 % lower than those derived from OH(9–6) at around 90 km. This might be explained by a slight underestimation of the quenching of OH(v=9) to OH(v=8) by O₂, an overestimation of the deactivation of OH(v=8) due to collisions with atomic or molecular oxygen, or the over-/underestimation of the branching factors f₉ and f₈ in the OH airglow model.

5.2 Comparison with other datasets

There are a number of O₂ and O excited state emissions which can also be used to derive atomic oxygen. This includes O(¹S) green-line and O₂ A-band emissions. Their modeling is mostly independent from the calculation of OH(v) emissions, although some processes have to be considered in all models. Rocket-borne in situ measurements of atomic oxygen are the most independent from methods based on nightglow. Mostly performed in the 1970s (e.g., Dickinson et al., 1980; Sharp, 1980; Offermann et al., 1981), the measurements are very rare and selective in terms of the local time and location. Figure 16 gives an impression of how the various datasets of atomic oxygen available in the literature fit to each other. All the sets are selected using a similar local time of around 22:00–23:00 LT, with the exception of OSIRIS (18:30 LT) and in situ data with diverse local times at midnight or in the afternoon.

The datasets agree within their combined uncertainties in most cases. The absolute abundances are typically 4–6 × 10¹¹ atoms cm⁻³ above 90 km and decrease with descending altitudes by 1 order of magnitude (at around 80 km) for midlatitudes in autumn. GOMOS data are around 10 % lower than SCIAMACHY-O(¹S) and both SABER-OH datasets at 90–95 km but remain in good agreement with these datasets at lower altitudes below 90 km. The OSIRIS dataset appears as the lower bound of the values above 90 km, as it is always the lowest in this region, while it becomes relatively large below 90 km. The WINDII dataset is around 10 % lower than the GOMOS-OH data at an altitude of 87–92 km, but they generally fit to each other. In situ data scatter in a large variation, which might be caused by the diurnal tides (local time differences), and the GOMOS-OH dataset is still located in its overall range of spread.