New insights into OH airglow modelling to derive night-time atomic oxygen and atomic hydrogen in the mesopause region

An OH airglow model was developed to derive night-t ime atomic oxygen (O( P)) and atomic hydrogen (H) from satellite OH airglow observations i the mesopause region (~75-100 km). The OH 10 airglow model is based on the zero dimensional box model CAABA/MECCA-3.72f and was empirically adjusted to fit four different OH airgl ow emissions observed by the satellite/instrument configuration TIMED/SABER at 2.0 μm and at 1.6 μm a s well as measurements by ENVISAT/SCIAMACHY of the transitions OH(6-2) and OH (3-1). Comparisons between the “Best fit model” obtained here and the satellite measurements suggest that deactivation of vibrationally excited 15 OH(ν) via OH(ν≥7)+O2 might favour relaxation to OH( ν’≤5)+O2 by multi-quantum quenching. It is further indicated that the deactivation pathway to OH(ν’=ν-5)+O2 dominates. The results also provide general support of the recently proposed mechanism OH(ν)+O(P)→OH(0≤ν’≤ν-5)+O(D) but suggest slower rates of OH( ν=7,6,5)+O(P). Additionally, deactivation to OH( ν’=ν-5)+O(D) might be preferred. The profiles of O( P) and H derived here are plausible between 80 km a nd 95 km. The values 20 of O(P) obtained in this study agree with the correspond ing TIMED/SABER values between 80 km and 85 km, but are larger from 85 to 95 km due to d ifferent relaxation assumptions of OH( ν)+O(P). The H profile found here is generally larger than T IMED/SABER H by about 30-35 % from 80 to 95 km, which might be attributed to too high O 3 night-time values.

Atomic oxygen in its ground state (O( 3 P)) and atomic hydrogen (H) strongly influence the energy budget in the mesopause region (~75-100 km) during day and night, and consequently affect atmospheric air temperature, wind, and wave propagation.Therefore, an improved knowledge of the abundance of O( 3 P) and H is of great importance when studying the mesopause region.At these altitudes, O( 3 P) has a direct impact on the heating rates by participating in several exothermic chemical 30 reactions (Mlynczak and Solomon, 1993, their Table 4).But O( 3 P) also contributes to radiative cooling by exciting CO 2 via collisions, leading to increased infrared emissions of CO 2 and partly opposing the O( 3 P) chemical heating effect.Night-time H plays a crucial role in the mesopause region due to the destruction of ozone (O 3 ) which is accompanied by the release of a considerable amount of heat (Mlynczak and Solomon, 1993).This chemical reaction additionally leads to the production of excited 35 hydroxyl radicals (OH(ν)) up to the vibrational level ν=9, causing the formation of OH emission layers in the atmosphere (Meinel bands;Meinel, 1950).
Direct measurements of O( 3 P) and H are relatively rare because as atomic species they do not have observable vibration-rotation spectra.Consequently, measuring these species in the mesopause region by remote sensing requires complex methods while in situ observations are rather expensive (e.g.40 Mlynczak et al., 2004;Sharp and Kita, 1987).Thus, there exists no global data set based on direct observations.As a consequence, an indirect method was introduced by Good (1976) to derive O( 3 P) and H during night, using OH airglow emissions.This approach was also adapted by Mlynczak et al. (2013;2014;2018) which derived a global data set of night-time O( 3 P) and H in the mesopause region from satellite observations of OH(ν).The method is based on the assumption of chemical steady state of O 3 45 and further depends on several radiative lifetimes, chemical reactions, and physical processes involving OH(ν).However, the corresponding total rate coefficients and branching ratios are still not sufficiently known, and thus present a large source of uncertainty in the derivation of O( 3 P) and H.
There are two major issues currently discussed in the literature which considerably affect the overall abundance of derived O( 3 P) and H.The first problem addresses the underlying deactivation schemes of 50 OH(ν) from the higher excited state ν to the lower excited state ν' in case of the sudden death approach, it is still unknown where such a huge amount of excess energy is transferred.The second crucial point comprises the deactivation scheme and the total rate of 55 OH(ν)+O( 3 P), including the new pathway OH(ν)+O( 3 P)→OH(0≤ν'≤ν-5)+O( 1 D) suggested by Sharma et al. (2015).
Over the last decades, several model studies attempted to fit OH airglow measurements, using different rates and schemes for the deactivation of OH(ν) by O 2 and by O( 3 P).And at least to our knowledge, there is no general agreement about which model is correct.The deactivation of OH(ν) by O 2 in many 60 models (e.g.von Savigny et al., 2012;Mlynczak et al., 2013;Grygalashvyly et al., 2014;Panka et al., 2017) is based on the model proposed by Adler-Golden (1997).It assumes a combination of multiquantum and single-quantum quenching and was derived from theoretical considerations and groundbased observations.Xu et al. (2012) investigated measurements of the Sounding of the Atmosphere using Broadband Emission Radiometry (SABER) instrument on board the NASA Thermosphere-65 Ionosphere-Mesosphere Energetics and Dynamics (TIMED) satellite of the OH airglow emissions at 2.0 µm and at 1.6 µm.Their results support the model of Adler-Golden (1997) but suggest slower total OH(ν)+O 2 rates.They further exclude the sudden death mechanism as a possible deactivation scheme.
There are also two theoretical studies (Shalashilin et al., 1995;Caridade et al., 2002) which investigated OH(ν) deactivation via O 2 , both supporting a combination of multi-quantum and single-quantum 70 quenching similar to the model of Adler-Golden (1997).
However, Russell and Lowe (2003) and Russell et al. (2005) analyzed OH(8-3) and O( 1 S) airglow emissions measured by the Wind Imaging Interferometer (WINDII) instrument on board the Upper Atmospheric Research Satellite (UARS).Both airglow emissions were used to individually derive O( 3 P) and the best agreement between these two O( 3 P) profiles was obtained when a sudden death scheme for 75 OH(ν)+O 2 quenching was applied.Kaufmann et al. (2008) investigated several OH airglow spectra between 1 µm and 1.75 µm measured by the Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY) instrument on board the Environmental Satellite (ENVISAT).They found best agreement between their model and the measured OH airglow spectra when a combination of sudden death and single-quantum quenching was used.80 Vibrationally dependent rates of OH(ν)+O( 3 P) were determined by Varandas (2004)  (2013), using quasi-classical trajectory calculations.Their results suggest that deactivation occurs via a chemical reaction as well as multi-quantum quenching.Kalogerakis et al. (2011) obtained a deactivation rate of OH( 9) +O( 3 P) from laboratory experiments which is several times larger than the rate from these calculations.But applying this fast quenching rate led to non-physical O( 3 P) values and associated 85 heating rates (Smith et al., 2010;Mlynczak et al., 2013).Thus, Sharma et al. (2015) proposed a new mechanism OH(ν)+O( 3 P)→OH(0≤ν'≤ν-5)+O( 1 D) to account for results from both theory and experiment.Very recent laser experiments and model studies support this new pathway while the exact values of the branching ratios and total loss rates are still not known (Kalogerakis et al., 2016;Panka et al., 2017).However, recently published results by Mlynczak et al. (2018) oppose this mechanism.They 90 also applied the new rate of Kalogerakis et al. (2011) for OH(9)+O( 3 P).But in order to get the annual energy budget into near balance, it was necessary to assume that at least OH(9)+O( 3 P) occurs via singlequantum relaxation.Additionally, the rate of OH(8)+O 2 had to be reduced and is considerably smaller than the value reported from Adler-Golden (1997).
In order to address the two major issues stated above, this paper is focused on the development of a zero 95 dimensional box model for atmospheric OH airglow with the intention to derive night-time O( 3 P) and H in the mesopause region.The model considers the formation of OH(ν) via H+O 3 and deactivation of OH(ν) due to spontaneous emission of photons, chemical reactions and physical collisions with atmospheric air compounds N 2 , O 2 , and O( 3 P).We used the indirect method introduced by Good (1976) and derived night-time O( 3 P) and H from TIMED/SABER OH emissions at ~2.0 µm, while also 100 considering the OH airglow observations from TIMED/SABER at ~1.6 µm as well as the OH(6-2) and OH(3-1) transitions measured by ENVISAT/SCIAMACHY.Further sensitivity runs were carried out to estimate the uncertainty on the derived values of O( 3 P) and H due to the different deactivation schemes, overall rate constants, and branching ratios.

ENVISAT/SCIAMACHY
The SCIAMCHY instrument (Bovensmann et al., 1999)  In this paper, we used SCIAMACHY level 1b data v7.04 to retrieve OH airglow volume emission rates (VERs) of the OH(3-1) and OH(6-2) bands in the wavelength ranges of 1515-1546 nm and 837.5-848 nm, respectively.The retrieval approach applied here is very similar to the one described in von Savigny et al. (2012).The retrieval does not cover the complete spectra of the OH(3-1) and OH(6-2) bands, and consequently a "correction factor" of 2.48 for OH(3-1) VER and 2.54 for OH(6-2) VER was added to 120 account for the entire band emissions at mesopause temperature.The data set further includes corrections for misalignments and other measurement errors (Gottwald et al., 2007).Investigations performed by Bramstedt et al. (2012) showed a drift of the SCIAMACHY tangent height of less than 20 m year -1 which is negligible for our study.
The uncertainties of the OH(3-1) VER and OH(6-2) VER retrievals from SCIAMACHY limb 125 observations correspond to the propagated uncertainties of the observed limb emission rate (LER) profiles.The latter are estimated from the LER values in the tangent height range between 110 km and 150 km, where the actual atmospheric emissions should be zero.The VER uncertainties are first determined for daily and zonally averaged data.The uncertainties used in this analysis correspond to the mean uncertainties averaged over all days with co-located SCIAMACHY and SABER observations.130

TIMED/SABER
The SABER instrument (Russell et al., 1999)  µm is only about a few percents (Xu et al., 2012;Mlynczak et al., 2013) and is neglected in this paper.140 In this study, we used the SABER Level 2A data v2.0 of the "unfiltered" OH VERs at 2.0 µm and at 1.6 µm, the air temperature and pressure, and the volume mixing ratios (VMRs) of O 3 (derived at 9.6 µm).
New night-time VMRs of O( 3 P) and H (Mlynczak et al., 2018) were used for comparison with the results derived from our model.The "unfilter" factor applied to OH VER adjusts the originally measured OH VER by the SABER instrument to the total VER emitted by OH in the corresponding 145 vibrational bands, while considering the shape, width, and transmission of the SABER broadband filters (Mlynczak et al., 2005).Outliers were excluded by screening the data as suggested by Mlynczak et al. (2013).The SABER data used here were further restricted to observations between 21 LT and 23 LT to approximately match the SCIAMACHY measurement time at ~22 LT.In order to be consistent with the naming of the SCIAMACHY OH airglow observations, the SABER OH airglow at 2.0 µm and at 1.6 150 µm are referred to as OH(9-7)+OH(8-6) and as OH(5-3)+OH(4-2) throughout the paper.

Method
In order to minimize issues between SABER and SCIAMACHY due to different measurement characteristics, we focused on the latitude range from 0° to 10° N, which was covered by both instruments throughout the entire year.A broader latitude band is not recommended because SABER 155 and SCIAMACHY do not uniformly cover the same latitudes, leading to disagreements between the real latitude of the observations and the nominal latitude of the interval.The accepted profiles of both instruments within the chosen latitude interval were averaged to zonal mean nightly mean values.All The approach to derive O( 3 P) and H applied here was developed by Good (1976) and is described in detail in Mlynczak et al. (2013).Thus, we only give a brief summary here.The measured SABER OH(9-7)+OH(8-6) VER (photons cm -3 s -1 ) is given by Eq. ( 1): where k 1 is the rate constant of the chemical reaction H+O 3 , representing direct production.The 165 function G (Eq. ( 2)) comprises all relevant production and loss processes of OH( 9 The subscripts ν and ν' (ν'<ν) are the vibrational states of OH before and after the corresponding process.The terms f v are the nascent distributions and describe the production efficiency of OH(ν) via 170 the reaction H+O 3 .Total radiative loss due to spontaneous emissions is considered by the Einstein coefficients A ν (s -1 ) which are the inverse radiative life times of OH(ν).The total loss rate C ν (s -1 ) is the sum of loss due to collisions with the air compounds (N 2 , O 2 , O( 3 P)), including chemical reactions and physical quenching.The terms A νν' and C νν' represent the specific state-to-state transitions.
In the second step, chemical equilibrium of O 3 during night is assumed as follows: 175 Finally, rewriting Eq. ( 1) enables the derivation of H while O( 3 P) is calculated by substituting Eq. ( 3) in 180 Eq. ( 1) and rewriting the resulting term as follows: by the input parameters and therefore identical in every model run.However, the goal of this paper is to develop a model which does not only fit OH(9-7)+OH(8-6) VER observations but also reproduces the three other airglow measurements OH(6-2) VER, OH(5-3)+OH(4-2) VER, and OH(3-1) VER.We have to further point out, that the relation between O( 3 P) and OH(9-7)+OH(8-6) VER is not linear since the function G also depends on O( 3 P), as represented by the terms C ν and C νν' .In fact, Eq. ( 4b) is a 195 quadratic expression with respect to O( 3 P) but treated here as a linear one, making no substantial differences for small O( 3 P).Nevertheless, this issue is addressed in detail in Sect.3.4.

The OH airglow Base model
The model used in this study is based on the atmospheric chemistry box model Module Efficiently Calculating the Chemistry of the Atmosphere/Chemistry As A Box model Application 200 (MECCA/CAABA-3.72f;Sander et al., 2011).The box model calculates the temporal evolution of chemical species inside a single air parcel of a certain pressure and temperature, making the model very suited for sensitivity studies.The CAABA/MECCA standard model was extended by several chemical reactions and physical quenching processes involving OH(ν) which are described in this section.The model was run until it reaches steady-state, defined by the agreement between the measured and 205 modelled OH(9-7)+OH(8-6) VER.
The OH airglow model described in this section is referred to as "Base model" because it is the starting point of our model studies.But we have to point out that there is no such a thing as a commonly accepted OH airglow base model in the literature.The Base model takes into account all major formation and loss processes of OH(ν) (Table 1) which are commonly used in other models in the 210 Propulsion Laboratory (JPL) report 18 (Burkholder et al., 2015).The reaction H+O 3 populates OH up to the vibrational level ν≤9 and the nascent distribution of OH(ν) was taken from Adler-Golden (1997).
The spontaneous emissions are given by the Einstein coefficients at 200 K (Xu et al., 2012).
Quenching of OH(ν) by O 2 is based on the values reported by Adler-Golden (1997, their Table 3) which comprise a combination of multi-quantum and single-quantum quenching.However, Adler-Golden (1997) applied a factor of ~1.5 to account for mesopause temperature based on comparisons between laboratory measurements at room temperature of OH(8)+O 2 and the corresponding rate inferred from 225 OH(8-3) rocket observations in the mesopause region.But later experiments reported by Lacousiere et al. (2003) and calculations by Caridade et al. (2002) suggest smaller values.The latter study further indicates that the temperature dependence decreases for lower vibrational levels and becomes negligible for OH(ν≤4).Consequently, the rates presented in Adler-Golden (1997) were scaled to room temperature measurements (ν=1-6 Dodd et al., 1991;ν=7 Knutsen et al., 1996;ν=8 Dyer et al., 1997;230 ν=9 Kalogerakis et al., 2011), and afterwards a factor of 1.1 for OH(ν≥6) and 1.05 for OH(5) was added.
The removal of OH(ν) via collisions with O( 3 P) is included by using a combination of multi-quantum quenching (Caridade et al., 2013, their Table 1) and chemical reactions (Varandas, 2004).The rates were obtained from quasi-classical trajectory calculations at 210 K, approximately matching mesopause 235 temperature.The model results of OH(6-2) VER and OH(3-1) VER are a 4 km running average to take the averaging 240 kernels of SCIAMACHY measurements into account.The Base model approximately matches the general shape of the measured profiles but overestimates the three OH airglow measurements at the altitude of maximum VER.A closer look at the relative differences shows that the ratio model/observation at the altitude of maximum VER is about 2.0, 1.2, and 1.3 for OH(6-2), OH(5-3)+OH(4-2), and OH(3-1), respectively.Furthermore, these ratios increase with decreasing altitude, 245 indicating that the overestimation of the Base model might be associated with O 2 quenching.

Results and discussions
The differences between Base model and observations are quite substantial in case of OH(6-2) VER.This implies a general problem of the rates or schemes included in the Base model, requiring a detailed error analysis.The focus was set on potential error sources of OH(6-2) VER because the relative differences between model and measurements are largest compared to the other two OH transitions, and 250 secondly because changes of OH(6) will affect the lower vibrational levels, but not vice versa.

Potential error sources of OH(6-2) VER in the Base model
Based on the results presented in Fig. 1, the potential error source has to have an effect on the entire height interval and must have a stronger impact on OH(6-2) compared to the other two OH transitions.
We further focus on quantities with large uncertainties.For the latter reason, temperature is excluded as 255 possible source because to account for a reduction of OH(6-2) VER by a factor of 2, temperature must be increased by more than 20 K (not shown here).Such a large error is very unlikely considering that a zonal mean climatology (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011) is used here.
Since the overestimation of the Base model is especially large for OH(6-2) VER, an impact of the Einstein coefficient of the corresponding transition must be considered.Regarding this aspect, we have to 260 point out that studies based on HITRAN 2004 data set should be viewed more critically, because of erroneous OH transition probabilities.The Einstein coefficients used in this study were recently recalculated (Xu et al., 2012, their Table A1) and correspond to a temperature of 200 K, which is very close to mesopause temperature.Furthermore, these Einstein coefficients are consistent with the values of the HITRAN 2008 data set (Rothman et al., 2009), Turnbull and Lowe (1989) (Mlynczak et al., 2018).Therefore, an error of a factor of 2 for the transition of OH(6-2) is rather unlikely.Consequently, we exclude the Einstein coefficients as a potential fundamental error source.
The nascent distribution of the excited OH states of the chemical reaction H+O 3 was observed in several 270 studies and all of them agree that OH(ν) is primarily formed in the vibrational levels ν=8 and ν=9 (e.g.Charters et al., 1971;Streit and Johnston, 1976;Ohoyama et al., 1985;Klenerman and Smith, 1987).
The values used in the Base model were taken from Adler-Golden (1997) which are based on measurements reported by Charters et al. (1971) and agree with values obtained by Klenerman and Smith (1987) and Streit and Johnston (1976).The values found by Ohoyama et al. (1985) show some 275 differences, but according to Klenerman and Smith (1987), their results are fundamentally flawed.This also affects the nascent distribution used by Mlynczak and Solomon (1993) which is an average of Charters et al. (1971), Ohoyama et al. (1985), and Klenerman and Smith (1987).
Therefore, we think that our nascent distribution used here is likely not a serious error source.However, minor errors might be introduced by extrapolating the nascent distribution to lower vibrational levels as 280 it was done for the values used in our study (Adler-Golden, 1997).It is also possible that part of the nascent value of OH( 6) is not due to direct production via H+O 3 but results from contributions of OH(ν≥7).In order to test the potential impact of the OH(6) nascent value on OH(6-2) VER, we assumed an extreme scenario by reducing the OH(6) nascent value from 0.03 to zero.But the corresponding results of OH(6-2) VER of the Base model run (not shown here) are only about 15 % 285 lower compared to the values presented in Fig. 1.Further sensitivity runs also showed that an increase of the ratio f 9 /f 8 is associated with a decrease of modelled OH(6-2) VER but even the extreme case of f 9 =1 and f 8 =0 could not account for a factor of 2. Note that changes of the overall rate constant of H+O 3 affect all considered OH transitions in a similar way, Thus, we conclude that direct production of OH(ν) is unlikely to be the reason for the overestimation of OH(6-2) VER by the Base model.290 The physical removal of OH(ν) by N 2 is included as single-quantum relaxation which is supported by theoretical studies (Shalashilin et al., 1992;Adler-Golden, 1997).Assuming a sudden death scheme with the same overall deactivation rates resulted in a decrease of simulated OH(6-2) VER by less than 10 % at the altitude of maximum VER.The total deactivation rate for OH(9) used here is about 1.5 times higher than the one suggested by Adler-Golden (1997) but the difference between the corresponding 295 model OH(6-2) VERs is negligible (<1 %).There are two studies reporting temperature dependence of N 2 quenching (Shalashilin et al., 1992;Burtt and Sharma, 2008), both agreeing with measurements at room temperature.However, the calculations of the former study imply slower quenching rates at mesopause temperature compared to their respective values at room temperature whereas the latter publication indicates the opposite behaviour, reporting a ratio between the rate at 200 K and 300 K of 300 approximately 1.7 for OH(ν=8) and 1.3 for OH(ν=9).These factors are generally supported by López-Puertas et al. ( 2004) which applied an empirically determined factor of 1.4 to the rates of Adler-Golden (1997) to account for mesopause temperature.Since the temperature dependence is still uncertain, we tested both possibilities.We increased and decreased the overall OH(ν)+N 2 quenching rates by a factor of 1.5 which led to higher or lower OH(6-2) VERs by about 5 %.Therefore, N 2 is too inefficient as a 305 OH(ν) quenching partner to cause differences of OH(6-2) VER of a factor of 2.
The overall rate and exact pathways of OH(ν)+O( 3 P) are also still not known well enough but O( 3 P) has nearly no influence on OH(ν) at altitudes below 85 km.It therefore cannot be the only reason for the differences presented in Fig. 1.Consequently, deactivation by O 2 is the only remaining candidate which has a crucial influence on OH(ν) throughout the entire height interval.Therefore, we will first focus on 310 OH(ν)+O 2 (Sect.3.2) before investigating a potential influence of O( 3 P) on OH(ν) in Sect.3.3.

Deactivation of OH(ν ν ν ν) by O 2
The overestimation of OH(6-2) VER by the Base model can be generally corrected either by slower rates of OH(9,8,7)+O 2 or by a faster rate of OH(6)+O 2 .The overall deactivation of OH(9) was measured by Chalamala and Copeland (1993) and they recommended a value of 2.1 × 10 -11 cm 3 s -1 .This result was 315 later confirmed by Kalogerakis et al. (2011), reporting a rate of 2.2 × 10 -11 cm 3 s -1 .The rates for OH(8,7,6)+O 2 are each based on a single study only (ν=8 Dyer et al., 1997;ν=7 Knutsen et al., 1996;ν=6 Dodd et al., 1991).But at least to our knowledge, there are no signs that the rates of OH(9,8,7,6)+O 2 are fundamentally flawed.In order to test the impact of the individual rates on OH(6-2) VER, we carried out sensitivity runs by varying the overall rates within their recommended 2σ errors.320 Thus, we reduced the values of OH(9,8,7)+O 2 to 16× 10 -12 cm 3 s -1 , 7× 10 -12 cm 3 s -1 , and 5× 10 -12 cm 3 s -1 , respectively, while the rate of OH(6)+O 2 was increased to 4.5× 10 -12 cm 3 s -1 .But even under this favoured condition, the Base model output of OH(6-2) VER decreased only by a factor of 1.5, still not close to the required difference of a factor of 2. Additionally, the assumed scenario is rather unlikely since the overall rates were obtained by independent studies.325 The possibility of a systematic offset of OH(ν≤6)+O 2 rates, which are based on the single study (Dodd et al., 1991), is also excluded because of the very good agreement of this OH(2)+O 2 rate with the value obtained by Rensberger et al. (1989).Furthermore, when we increased the OH(ν≤6)+O 2 rates by a factor of 3, the Base model approximately fits OH(6-2) VER and OH(3-1) VER but underestimates OH(5-3)+OH(4-2) VER by more than 30 %. Temperature dependence also affects the O 2 deactivation 330 rates used here.But the factor to account for mesopause region temperature is suggested to be lower than 1.3 (Lacousiere et al., 2003;Cadidade et al., 2002) which has a weaker impact on OH(6-2) VER than the scenarios considered above.
Consequently, when applying the standard deactivations rates and schemes found in the literature, neither errors of the overall rates nor uncertainties of the temperature dependence can give a reasonable 335 explanation of the overestimation of OH(6-2) VER Base model output shown in Fig. 1a.Since the overall rates were actually measured while the deactivation schemes are solely based on theoretical considerations, it is more convincing that the potential error source probably lies within OH(ν)+O 2 deactivation scheme rather than in the deactivation rates.
In order to considerably reduce OH(6-2) VER, we assumed an extreme scenario and substituted the 340 multi-quantum relaxation (OH(ν)+O 2 →OH(ν'<ν)+O 2 ) in the Base model by a sudden death (OH(ν)+O 2 →OH+O 2 ) approach.This new model is referred to as "O 2 SD model" and the corresponding results are displayed in Fig. 2 as red lines, showing that the simulated OH(6-2) VER matches the observations within the error bars below 85 km.The model still overestimates the measurements in the altitude region above which might be related to O( 3 P) quenching (see Sect. 3.3).345 The O 2 SD model output for the other two OH transitions (Fig. 2b-c) is clearly too low, implying that OH(ν)+O 2 quenching cannot occur via sudden death alone.We also conclude that the contribution of higher excited states OH(ν≥7) to OH(6) must be negligible or even zero and these higher states are suggested to primarily populate lower vibrational levels OH(ν≤5).Therefore, OH(ν)+O 2 has to occur via multi-quantum quenching because in case of single-quantum deactivation the contribution of 350 OH(ν≥7) to OH( 6) is considerably larger than zero.
According to Finlayson-Pitts and Kleindienst (1981), OH(ν) might be relaxing to ν'=ν-5 while the excess energy is transferred to form O 2 (b 1 Σ).This vibration-to-electronic energy transfer was also mentioned by Anlauf et al. (1968) and is supported by the close energy match of the transition from OH(9) to OH(4) and from O 2 (X 3 Σ) to O 2 (b 1 Σ) of about 36.6 kcal mol -1 and 37.5 kcal mol -1 , respectively.355 Although there is no experimental support of this deactivation pathway, this approach gives a reasonable explanation for the observed pattern in our study and OH(ν) as a potential source of excited O 2 was discussed in Howell et al. (1990) and Murtagh et al. (1990).However, evaluating whether the product is really O 2 (b 1 Σ) or another excited O 2 state is beyond the scope of this study.Thus, we concluded that deactivation of OH(ν) by O 2 has to satisfy the following condition: 360 while we further assume that the pathway is the preferred deactivation channel.
If R10a is integrated in the model (Fig. 2b-c, O 2 ν-5 model), the corresponding model output at altitudes <90 km is only about 10 % below the observations of OH(5-3)+OH(4-2) VER and approximately matches OH(3-1) VER measurements within the error bars.The underestimation of the OH(5-3)+OH(4-370 2) VER measurements by the model could be attributed to minor errors of the OH(ν)+O 2 overall rates in combination with a slightly different OH(ν) branching of H+O 3 .Therefore, we cannot completely rule out R10a as a possible solution, even if there are still some differences between the modelled and the observed OH VER.Including R10b in the model (Fig. 2b-c, O  The results shown in Fig. 2 suggest that the OH airglow model is not able to reproduce the three OH airglow observations when sudden death or simplified multi-quantum schemes for OH(ν)+O 2 are applied.But the O 2 ν-5 model output is quite close to the measurements, suggesting that R9 might be the dominating deactivation channel within a multi-quantum relaxation scheme in accordance with R8. 380 We therefore included these two conditions in the so-called "O 2 best fit model" and the results are displayed in Fig. 3.The corresponding branching ratios for the individual pathways are summarized in Table 2.
The simulated OH airglow fits well with the three OH airglow observations within the error bars below 85 km.In the altitude region above 85 km, it is seen that the model still overestimates OH(6-2) VER 385 while OH(3-1) VER is indicated to be slightly underestimated.Furthermore, this pattern is not seen in OH(5-3)+OH(4-2) VER and therefore could be attributed to deviations due to the different satellite/instrument configurations between TIMED/SABER and ENVISAT/SCIAMACHY.But since this behaviour only occurs in the upper part of the vertical profiles and is not seen throughout the entire height interval, it is more likely related to O( 3 P) quenching.390
We adapted R11 in the "O 2 best fit O( 3 P) ν-5 model" in such a way that the product is OH(ν'=ν-5)+O( 1 D) and the results obtained are displayed as blue lines in Fig. 4. Comparisons for OH(6-2) VER 400 in Fig. 4a show an underestimation of the model at altitudes >85 km.A sensitivity study was carried out which implied that the impact of OH(9,8,7)+O( 3 P) on OH(6-2) VER is negligible.This seems reasonable because these three upper states only indirectly influence OH(6-2) via R11.Consequently, our analysis suggests a lower value of k 11 (6) and best agreement between model output and OH(6-2) VER observations was obtained for an overall rate of approximately 0.8 × 10 -10 cm 3 s -1 .405 In case of OH(5-3)+OH(4-2) VER, presented in Fig. 4b, the new approach leads to a weak underestimation of the observations by the model in the altitude region above 85 km, even if OH(9)+O( 3 P) of R11 solely populates OH(4).The model results are most sensitive to k 11 (5), and therefore this rate might be too high.Considering our best fit value obtained for k 11 (6), it is indicated that k 11 (ν) decreases with decreasing vibrational level.This assumption is supported by the overall rate 410 of OH( 7)+O( 3 P)→OH(ν')+O( 1 D) at mesopause temperature which is suggested to be on the order of 0.9-1.6× 10 -10 cm 3 s -1 (Thiebaud et al., 2010;Varandas, 2004).Thus, an upper limit of k 11 (5)<k 11 ( 6) is recommended and the actual rate coefficient has to balance the direct contribution of OH(9) to OH(4) via R11.Investigating another scenario of k 11 (5) being zero showed that the branching of OH(9) to OH(4) has to be at least about 0.6 which corresponds to a rate of a ~1.4 × 10 -10 cm 3 s -1 .415 It is seen in Fig. 4c that observations and O 2 best fit O( 3 P) ν-5 model output of OH(3-1) VER are in agreement within the corresponding measurement errors but the model values seem to be slightly too low at heights >85 km.In this altitude region, simulated OH(3-1) VER is most influenced by OH(9,8)+O( 3 P) of R11 because both vibrational levels can directly populate OH(3).However, not much is known about the individual branching ratios of R11 except that OH(9)+O( 3 P)→OH(3)+O( 1 D) is an 420 important deactivation channel but not necessarily the dominating one (Kalogerakis et al., 2016).This agrees with our results presented here because the O 2 best fit O( 3 P) ν-5 model only considers a contribution of OH(8) to OH(3) and the underestimation indicated in Fig. 4c could be attributed to the missing channel OH(9)+O( 3 P)→OH(3)+O( 1 D).The conclusions drawn from comparisons between three different airglow observations and our model studies with respect to OH(ν)+O( 3 P) quenching are 425 summarized in Table 3.
Finally, all these findings presented in Table 2 and 3 were adapted in the "Best fit model" (Fig. 4, red lines), resulting in an overall agreement between model output and measurements within the corresponding errors.Note that k 11 (7) used here is the average of the lower and upper limits derived from Thiebaud et al. (2010) and Varandas (2004) which is unlikely to be seriously in error.Furthermore, it is indicated that the value of k 11 (8) might be lower than 2.3 × 10 -10 cm 3 s -1 which can be assumed based on the other rates of k 11 (ν).But we did not find any study reporting an observed k 11 (8) rate, and consequently we did not change k 11 (8).Besides, we have to point out that lowering k 11 (8) does not affect the general conclusions drawn in this section.
The empirically determined solution presented here implies that the contribution of OH(9) to OH(8) via 435 quenching with O( 3 P) is close to zero (see Table 1 and this section).In contrast, the model described in Mlynczak et al. (2018) assumes single-quantum relaxation (OH(9)+O( 3 P)→OH( 8)+O( 3 P)) to get the global annual energy budget into near balance.But applying this approach in our OH model (same total rate of 3 × 10 -10 cm 3 s -1 and varying the rates for OH(ν≤8)+O( 3 P)) leads to a considerable overestimation of OH(6-2) VER.Additionally, the shape of simulated OH(5-3)+OH(4-2) VER slightly mismatches the 440 observed OH(5-3)+OH(4-2) VER above 90 km (not shown here).Based on these sensitivity runs, we conclude that at least part of OH( 9)+O( 3 P) channel has to be deactivated via multi-quantum quenching.This is supported by the results presented by Panka et al. (2017) which adjusted an OH airglow model to fit night-time CO 2 (ν 3 ) emissions at 4.3 µm.However, this study reported empirically determined rates for OH(5≤ν≤8)+O( 3 P) generally higher than the rates obtained in this work.But these differences 445 might be attributed to their faster values of OH(ν)+O 2 because they seem to have falsely assumed that the rates of Adler-Golden (1997) do not take mesopause temperature into account.Thus, we think that their rates of OH(ν)+O 2 are too high, at least by a factor of ~1.5.Since they performed an empirical study, it is not possible to estimate how much this issue affects the rates of OH(5≤ν≤8)+O( 3 P).But we know from our work that higher rates of OH(ν)+O 2 lead to higher values of OH(6-2) VER, OH(5-450 2)+OH(4-2) VER, and OH(3-1) VER which can be generally balanced by higher rates of OH(5≤ν≤8)+O( 3 P).Considering our comparisons with these two studies, we think that the rates of OH(ν)+O( 3 P) should be investigated in more detail in future studies as this rate has a huge impact on derived values of O( 3 P) (Panka et al., 2018).This might be partly caused by too high O 3 night-time values, as suggested by Mlynczak et al. (2018).

Derived profiles of O( 3 P) and H 455
Similar to the comparisons with O( 3 P), Best fit model H results also shows unexpected patterns above 95 km.
The quality of the derived profiles is primarily affected by three different uncertainty sources.The first 470 source includes uncertainties due to the rates of chemical and physical processes considered in the Best fit model.We assessed the 1σ uncertainty by assuming uncorrelated input parameters.Adler-Golden (1997) did not state any uncertainties for f 9 and f 8 but these values should be similar to the uncertainty of f 8 derived by Klenerman and Smith (1987).Therefore, we applied an uncertainty of 0.03 for f 9 and f 8 .
In case of the Einstein coefficient, we adapted an uncertainty of 10 % as suggested by Mlynczak et al. 475 (2013).All the other 1σ uncertainties of the input parameters were taken from their respective studies.Otherwise, it should be viewed more critically.This was done for each altitude and we found that the O( 3 P) and H profiles presented in Fig. 5 are plausible in the altitude region <95 km.In combination with 520 the estimation of chemical equilibrium of O 3 , we think that the O( 3 P) and H derived by the Best fit model provides reasonable results between 80 km and 95 km.Note that these altitude limits do not affect the results with respect to OH(ν)+O 2 and OH(ν)+O( 3 P) presented in the Sect.3.2 and 3.3.
was an 8-channel spectrometer on board ENVISAT, providing atmospheric OH airglow emission measurements between ~220 nm and ~2380 nm.ENVISAT was launched into a polar and sun-synchronous orbit and crossed the equator at ~10 LT 110 and ~22 LT.The ENVISAT mission started in March 2002 and SCIAMACHY was nearly continuously operating until the end of the mission in April 2012 caused by a spacecraft failure.The SCIAMACHY instrument performed measurements in different observations modes, including night-time (~22 LT) limb scans over the tangent altitude range ~75-150 km.These measurements are only available throughout the year at latitudes between the equator and 30° N. 115 on board the TIMED satellite has been nearly continuously operating since January 2002, collecting over 98 % of all possible data.The instrument scans Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-755Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 6 August 2018 c Author(s) 2018.CC BY 4.0 License. the atmosphere from the surface up to altitudes of ~400 km while providing a vertical resolution of about 2 km throughout the entire height interval.Due to the geometry of the satellite orbit and the 135 regular yaw manoeuvres every ~60-65 days, SABER only provides complete coverage of the latitude range between ~55° S and ~55° N. The SABER instrument measures the OH VERs at ~2.0 µm and at ~1.6 µm which approximately corresponds to the transitions of OH(9-7)+OH(8-6) and OH(5-3)+OH(4-2), respectively.The contribution of OH(7-5) to OH VER at 2.0 µm and of OH(3-1) to OH VER at 1.6 these zonal mean nightly means from January 2003 to December 2011 were used to calculate a climatology, including only days on which both SCIAMACHY and SABER data are available.160 Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-755Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 6 August 2018 c Author(s) 2018.CC BY 4.0 License.
meaning that O 3 loss due to H and O( 3 P) (left side) is balanced by the three-body-reaction O( 3 P)+O 2 +M (right side).Here, k 2 and k 3 are the corresponding rate constants of O( 3 P)+O 3 and O( 3 P)+O 2 +M, respectively, while M is the total number density of the air.

)
Air temperature and air pressure from SABER were used to calculate M as well as the number densities of O 2 (VMR of 0.21), N 2 (VMR of 0.78), and SABER O 3 via the ideal gas law.The chemical reaction rates and physical quenching processes involved are described in Sect.2.3.The values of O( 3 P) and H were individually derived for each altitude.Finally, the obtained vertical profiles of O( 3 P) and H were used to initialize the OH airglow model (see Sect. 2.3).It is apparent from Eq. (4a-b) that any changes applied to the input parameters (G, O 2 , O 3 , M, k 1 , k 2 , k 3 ) are balanced by the derived values of O( 3 P) and H.In contrast, OH(9-7)+OH(8-6) VER is not affected 190 literature and are assumed not to be seriously in error.The model comprises the production of OH(ν) via the chemical reaction H+O 3 as well as the deactivation due to spontaneous emission and the removal Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-755Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 6 August 2018 c Author(s) 2018.CC BY 4.0 License.physical quenching and chemical reactions with N 2 , O 2 , and O( 3 P).The chemical reactions H+O 3 , O( 3 P)+O 3 , and O( 3 P)+O 2 +M were already included in the CAABA/MECCA standard model and their corresponding rates were taken from the latest Jet 215

Figure 5
Figure 5 displays the vertical profiles of O( 3 P) and H obtained by the Best fit model in comparison withthe results derived from SABER OH(9-7)+OH(8-6) VER only(Mlynczak et al., 2018).The O( 3 P) Fig. 5.The error bars of SABER O( 3 P) and H were adapted from the corresponding publication.In case of the Best fit model O( 3 P) profile, the 1σ uncertainty varies between 20 % and 30 %, depending 480 on altitude.The individual contributions of the input parameters to the total 1σ uncertainty are considerably different.Einstein coefficients and nascent distribution each account for about 5 % throughout the entire height interval.The influence of the collision rates is lower than 6 % and gradually decreases to zero with increasing altitude.In contrast, the chemical reaction rates k 2 and k 3 account for ), and consequently errors in OH(5) and OH(9) might be compensated by errors in OH(4) and OH(8) or Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-755Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 6 August 2018 c Author(s) 2018.CC BY 4.0 License.vice versa.Finally, we have to stress that we performed an empirical model study and the total rates and deactivation channels suggested here heavily depend on the OH transitions considered.Including 540 additional OH transitions might result in other values and deactivation schemes.But these issues will only be solved eventually when future laboratory experiments provide the corresponding OH(ν)+O 2 and OH(ν)+O( 3 P) relaxation rates.Justified by chemical equilibrium of O 3 and a nearly linear relation between O( 3 P) and OH(9-7)+OH(8-6) VER, we conclude that the O( 3 P) and H profiles derived by the Best fit model are plausible in the 545 altitude range from 80 km to 95 km.The corresponding 1σ uncertainty due to uncertainties of chemical reactions and physical processes varies between 20 % and 30 %, depending on altitude.

Table 1 .
Physical processes and chemical reactions included in the Base model