Comparison of a photochemical model with observations of mesospheric hydroxyl and ozone


  • This article was originally published 16 January 2013. Subsequently it was discovered that M. G. Mlynczak was inadvertently omitted from the authorship and his affiliation was misapplied to the existing authors. This amended version has been created to correct the errors and represents the version of record. The publisher regrets the error.


[1] We present a comparison of a photochemical model with mesospheric hydroxyl (OH) data from the Spatial Heterodyne Imager for Mesospheric Radicals (SHIMMER) and mesospheric ozone (O3) data from the Sounding of the Atmosphere with Broadband Emission Radiometry (SABER). Although SHIMMER and SABER do not measure the atmosphere coincidently, by sampling the photochemical model at the appropriate local time of each measurement, an effective concurrent test of mesospheric odd oxygen and odd hydrogen theory can be achieved. Consistent with previous, more limited analyses of SHIMMER data, we find no evidence of a systematic model overprediction of mesospheric OH. However, at 80 km, the standard chemical scheme shows a model deficit in the morning  hours and a dramatic model excess in the late afternoon. Using a higher rate coefficient for the H + O2 + M → HO2 + M reaction ameliorates this problem. Such a higher rate is consistent with the only reported laboratory measurements at the low temperatures appropriate to the mesosphere. Regarding the SABER ozone, the model significantly underpredicts the data. Some of this could be explained by a previously reported, systematic high bias to the SABER ozone, and comparisons of our model with Microwave Limb Sounder data support that suggestion. Nonetheless, a persistent model ozone deficit remains. Since the model agreement with SHIMMER OH is generally very good, this model ozone deficit is unlikely to be due to a mischaracterization of mesospheric HOx.

1 Introduction

[2] As discussed by Siskind et al., [1995], one of the classic problems in aeronomy deals with the interaction of mesospheric odd oxygen (Ox = O + O3) and odd hydrogen (HOx = H + OH + HO2). Historically, this topic has been of interest for two reasons. The first is that despite the relative simplicity of the mesospheric ozone budget as compared with the stratosphere, initial model-data comparisons showed dramatic underpredictions of the early models of mesospheric ozone compared with the first observations [e.g. Solomon et al., 1983; Rusch and Eckman, 1985]. The second, which has received new attention of late [Merkel et al., 2011 and references therein] is that mesospheric ozone has long been believed to be a useful indicator of the response of the atmosphere to variable solar ultraviolet radiation.

[3] Regarding the model underprediction of mesospheric ozone, the speculation by Rusch and Eckman, [1985] was that this was due to a model overprediction of HOx, the radical family that dominates the mesospheric ozone loss. Indeed, this suspicion appeared to be borne out as several key HOx reaction rates were revised in the direction of lowering calculated HOx and thus increasing calculated ozone [Natarajan and Callis, 1989; Eluszkiewicz and Allen, 1993]. Despite these laboratory revisions, data from the Middle Atmosphere High Resolution Spectrograph Investigation (MAHRSI) still suggested that models overpredicted mesospheric odd hydrogen [Summers et al., 1997]. Recently presented new hydroxyl (OH) data have, however, suggested otherwise. Thus, first results from the Spatial Heterodyne Imager for Mesospheric Radicals (SHIMMER) [Englert et al., 2008] and the Aura Microwave Limb Sounder (MLS) [Pickett et al., 2008] found good agreement between their data and models. Englert et al. [2010] presented arguments that the MAHRSI data may have had an unidentified calibration problem which led to the somewhat lower values of hydroxyl abundance first reported by Summers et al. [1997] and concluded that there is no model overprediction of mesospheric odd hydrogen. Canty et al. [2006] came to a similar conclusion for altitudes below 60 km. The SHIMMER-based conclusion was limited to a few isolated studies at a single local time where Aura/MLS data are available and also did not include a concomitant comparison with mesospheric ozone. In addition to Aura/MLS, another global hydroxyl dataset has been provided by the Optical Spectrograph and Infra-red Imager System (OSIRIS) on board the Odin satellite [Gattinger et al., 2006]. Like Aura, ODIN is in a sun-synchronous orbit that is locked in local time, although because the orbit of ODIN is often near the dawn and dusk terminators, Gattinger et al. [2008] were able to make some unique measurements of the so-called OH " sunrise flash" .

[4] There have been a number of measurements of mesospheric ozone since the pioneering studies by the Solar Mesosphere Explorer [Rusch et al 1983, Thomas et al., 1984]. The Upper Atmospheric Research Satellite (UARS) included three experiments, the Halogen Occultation Experiment (HALOE) [Russell et al., 1993], the Microwave Limb Sounder (MLS), and the High Resolution Doppler Imager (HRDI) [Marsh et al., 2002]. While the HALOE data are only obtained at sunrise and sunset, both MLS [Ricaud et al., 1996] and HRDI provided diurnally resolved data. Currently the MLS on Aura [Froidevaux et al., 2008] measures ozone simultaneously with its OH measurements. Finally, while the MLS data are at fixed local times, the Sounding of the Atmosphere with Broadband Emission Radiometry (SABER) instrument on the TIMED satellite measures ozone at a variety of local times [Huang et al., 2008].

[5] Despite the existance of several mesospheric ozone datasets which resolve the diurnal cycle (UARS/MLS, UARS/HRDI and SABER), there have been relatively few detailed comparisons of these datasets with diurnal photochemical models. Ricaud et al. [1996] did compare UARS/MLS data with a model; however, their comparison did not extend above 70 km. Recently, Dikty et al. [2010] used a three-dimensional model to examine the broad global picture of ozone diurnal variations seen in SABER. However, there have been absolutely no coordinated model-data comparisons of both ozone and hydroxyl together. In the present study, we therefore aim to fill two gaps left from the previous work. First, we will provide the first multiyear analysis of the mesospheric OH diurnal cycle using SHIMMER data. Second, we will combine this with an analysis of the diurnal variation from SABER ozone taken concurrently with SHIMMER.

2 Approach to Model-measurement Comparisons

[6] In this section, we review the photochemical model and how we compare it to both SABER and SHIMMER.

2.1 Overview of the Photochemical Model

[7] Our basic approach is to compare a one-dimensional diurnal photochemical model (CHEM1D) with both SABER and SHIMMER data. Our altitude range will be from 55 to 80 km which is in the region where ozone is under photchemical control, thus justifying a simpler one dimensional modeling approach, and where the HOx cycle is the dominant catalytic loss process [Brassuer and Solomon, 1986]. CHEM1D has been used previously in our studies of HOx [Englert et al., 2010; Summers et al., 1997]. For its current application, we have made three small changes. First, we have upgraded the reaction rate coefficients to the JPL11 recommendation [Sander et al., 2011]. Specifically, two recommendations pertinent to mesospheric HOx-Ox chemistry have changed: the first is that the recommended formulation for the O + OH reaction rate increased the rate by about 10% at mesospheric temperatures, and the second increased the O(1D) + H2 reaction rate by about 8% at mesospheric temperatures. The other change is that we previously used a value of 3.42 × 1011 cm-2 s-1 for the solar Lyman alpha flux; however, an examination of the LASP Iteractive Solar Irradiance Datacenter (LISIRD) at suggests that it is slightly too low. For the two periods we focus on here, April–September 2007 and 2009, a value of 3.60 × 1011 is a better estimate and that is what we have adopted. Taken together, the net effect of these changes is to reduce the calculated OH by about 6% as compared with our calculations in Englert et al. [2010]. While this difference is small, since the calculated OH was already about 10% lower than the observations in the Englert et al. [2010] study, the new rates have the effect of making this difference more apparent.

[8] In addition to reaction rates and the solar flux, CHEM1D requires the neutral atmospheric temperature and water vapor profiles as input. We take these from the analyzed fields provided by the NOGAPS-ALPHA data assimilation system, discussed by Eckermann et al [2009]. The initial summer 2007 analysis described in that paper has been extended and discussed by McCormack et al. [2010] and Siskind et al. [2011]. Since the water vapor input to the data assimilation comes from the Aura/MLS, in practice, these inputs are similar to those used by Englert et al. [2010]. Our use of a model that is constrained in this fashion removes much of the uncertainties related to the background composition of the atmosphere.

[9] We concentrated our comparison on the highest Northern latitudes observed by SHIMMER, the band from 50–57°N. This latitude band is convenient for our purposes because the high latitudes in the mesosphere experience the widest variation of water vapor in the middle atmosphere, thus offering a test of the photochemistry under the widest range of conditions. Because special satellite operations interrupted the time series in 2008, our two best years were 2007 and 2009. Thus, Figure 1 shows the range of background water vapor variation for the period of analysis. The figure shows the NOGAPS-ALPHA water vapor mixing ratio for 50–57°N from April to September for both 2007 and 2009 for 0.02 hPa (about 75 km). The figure shows that H2O increases by nearly a factor of two from April to mid-summer. The H2O variation is asymmetric about the summer solstice (marked by the dotted line on day 170). Thus, for the drivers of OH chemistry, namely, solar flux and water vapor, we can delineate three approximate combinations of solar zenith angle and H2O. The first period, up to about day 150, is characterized by relatively low H2O and higher solar zenith angles. The period centered about the solstice has high H2O abundance (> 6 ppmv) and the lowest solar zenith angles. The period from late July (day 200) to mid September (day 260) still has high H2O abundances but higher solar zenith angles. As we will discuss below, this qualitative division of our period of study into three periods is particularly useful because the SHIMMER local time precession is such that one complete diurnal cycle is sampled in each of these periods.

Figure 1.

Zonal mean H2O for 2007 and 2009 at p = 0.02 hPa (about 75 km), for latitudes from 50–57°N from the NOGAPS-ALPHA analysis. The vertical dotted line is a fiducial to mark the solstice.

[10] Our specific approach is to run CHEM1D at 10 day intervals from days 150 to 260 in 2007 (day 150 is the start of the NOGAPS-ALPHA analysis) and from days 120 to 260 in 2009. We then interpolate the model to the specific date and local time combination appropriate for either SHIMMER and SABER. This is illustrated in the next two sections.

2.2 The SHIMMER OH Observations

[11] As described by Englert et al [2010], SHIMMER was the primary payload of STPSat-1 and was launched into a 35.4° inclination orbit in March 2007. SHIMMER imaged the limb, and STPSat-1 executed two yaw maneuvers each year to ensure that SHIMMER observed the summer hemisphere up to a latitude of about 57°. As shown in Figure 1 of Englert et al. [2010], this sampling pattern meant that the largest number of samples was acquired near the “turnaround” latitude of the orbit, and it is this latitude band (50–57°) that we focus on. Also, we found that the satellite tended to operate more reliably during the northward yaw, so we will focus on those periods for 2007 and 2009.

[12] Since SHIMMER takes a limb image every 20 seconds, this means that in a given day, many hundreds of images were acquired in the 50–57°N latitude band when pointed to the north. However, one complication with using this latitude band is that, as can be seen in Figure 1 of the study by Englert et al. [2010], the SHIMMER sampling pattern can cover up from 7 to 8 hours of local time. To better resolve the details of the local time OH variation, we separated the limb images into two bins, one for the ascending part of the orbit and one for the descending part. All the limb images in each bin are averaged together prior to performing the inversion of the OH radiances. We thus have effectively two independent measurements of OH at different local times for each day, each the result of averaging well over 100 individual limb images. Despite this separation into ascending and descending bins, each average still represents an average over as much as 4 hours of local time. A second complication with interpreting our results is that the subsequent inversion of these averaged profiles is purely one dimensional, i.e., we assume spherical symmetry about the tangent point. The impact of the 3–4 hour local time smearing is ameliorated by a concomitant smoothing of our photochemical model calculations by 3.5 hours. During the middle of the day, OH varies monotonically, and this smoothing has little effect; however, at sunset, our model shows large variations, and the local time smoothing is both more important and adds an element of uncertainty to our results. The impact of the spherical symmetry assumption is less easily handled and, in the absence of detailed retrieval simulations, has to be considered as a source of error that is not well quantified. It may be an additional contributing factor to the disagreements we see between our calculations and the observations near sunset, as we will show.

[13] Figure 2 shows the resulting averaged local time sampling of SHIMMER, for both the ascending and descending portions of the orbit, as a function of day of 2009, superimposed upon contours of photochemical model output at 80 km. Despite our separation of the limb images in this manner, our 3–4 hour local time resolution along the orbital track is still not good enough to resolve the details of the OH diurnal variation immediately at sunrise or sunset, as for example, the study by Gattinger et al. [2008]. This limits our ability to tightly constrain eddy diffusion effects which are most apparent in the first few minutes after sunrise.

Figure 2.

Contours of calculated OH at (a) 64 km and (b) 80 km from 15 photochemical diurnal models, run at 10 day intervals for 2009. The calculations have been smoothed by a 3.5 hour boxcar to account for the local time smearing inherent in the SHIMMER observational geometry (see text). Superimposed upon the model contours are the local time pattern of the SHIMMER sampling. The SHIMMER local times are an average for the latitude band 50–57°N. The latitude for models was 54°N.

[14] Figure 2 shows that from days 120 to 260, SHIMMER processed through three diurnal cycles. Compared with Figure 1, we can roughly map these three diurnal cycles with the three combinations of solar zenith angle and water vapor density discussed above. The summer solstice is marked by a dot-dashed line on day 171 to indicate the date of maximum photodissociation of H2O, the source of OH. As is particularly evident in Figure 2b, the calculated OH peaks after the summer solstice reflecting the effect of increasing H2O into the summer that was shown in Figure 1. As we will see in the SHIMMER data, the OH in August (days 210–240) is consistently higher than that in May (days 120–150). By sampling the model according to the local times shown in Figure 2, we can obtain a one-to-one comparison of SHIMMER with the model as a function of local time and day of year. Figure 2 also shows some differences between the diurnal variation of OH at 64 km compared with 80 km. Specifically, at 64 km, the OH variation is symmetric about noon, whereas at 80 km, the theoretical OH is asymmetric about noon. There is an afternoon buildup of OH such that at 1900–2000 hours, the OH is often higher than at any other time during the day. The tendency for calculated HOx abundances near 80 km to maximize in the afternoon was first discussed by Prather [1981] and was first demonstrated by Allen et al., [1984, cf. their Figure 9c].

2.3 The SABER Ozone Observations

[15] The sampling pattern of the SABER ozone dataset is different from SHIMMER. First, TIMED is in a different orbit than STPSat-1 was. As discussed, for example, by Zhang et al. [2006], TIMED is in a 74° inclination orbit and executes 180° yaw maneuvers every 60 days. The net effect is that SABER observes from 53° in one hemisphere to near the pole in the other. Fortunately, this means that for the latitude band we are considering, there is continuous coverage, although the daily local time coverage is completely different than SHIMMER and simultaneous observations by the two experiments of the same volume of space for the same local time almost never occur.

[16] Second, unlike with the SHIMMER OH measurements which relied upon solar fluorescence, SABER measures ozone both day and night. By day, there are two measurements, the thermal emission at 9.6 μm which extends from the lower stratosphere to 100 km and the 1.27 μm which results from O3 photodissociation and covers the mesosphere. The 9.6 μm data are discussed by Rong et al. [2009] and Smith et al. [2008], the 1.27 μm emission and its use to derive ozone is covered by Mlynczak, 2007; however, the 1.27 μm ozone has not yet been the subject of a validation paper. At night, there is only the 9.6 μm data. Figure 3 summarizes the sampling of SABER for the same period in 2009 and latitude band as in Figure 2 and is superimposed upon the results of the photochemical model runs. Unlike Figure 2, we now have both daytime and nighttime sampling, so the vertical axis is different. Also, the changes in sampling due to the 60 day yaw cycle are apparent. Thus, before mid-May and then again after mid-July, SABER only observes up to about 52°N. This means that there are two local times associated with the ascending and descending portions of the orbit that are close together, much as with the SHIMMER ascending and descending samples. From mid-May to mid-July, SABER observes to 83°N, and there are clearly defined nighttime and daytime samples that are separated by about 12 hours in local time (e.g., Huang et al., 2008, Figure 3).

Figure 3.

Contours of calculated ozone at (a) 64 km and (b) 80 km from the same models and latitude as shown in Figure 2. Superimposed upon the model contours are the averaged local time pattern of the SABER ozone sampling for latitude band 50–57°N.

[17] At 64 km, Figure 3 shows a clear diurnal variation with O3 daytime values near 2 × 109 cm-3 and nighttime values about twice as large. This diurnal variation is consistent with the recombination of atomic oxygen after sunset and has been documented for some time [cf. Connor et al., 1994]. A weak seasonal variation is also evident with the daytime values less than 2 × 109 cm-3 at the beginning and end of the period and greater values in between. At 80 km, the picture is more complex. Here the model exhibits a distinct AM enhancement which is most pronounced in late spring. This greater complexity of the ozone diurnal variation in the upper mesosphere was pointed out in the model studies by Allen et al. [1984] and Prather [1981], and its manifestation in SABER was discussed by Huang et al. [2008]. However, there has been little quantitative data/model intercomparison of these variations.

3 Comparison of Standard Model With Data

[18] The previous section illustrated the two different diurnal sampling patterns for SHIMMER OH and SABER O3. By sampling the same photochemical calculation at the appropriate local time for each instrument, we can use the model as a check of the photochemical consistency of both SHIMMER and SABER with standard theory.

[19] To compare the model with SHIMMER, we convolved the model with a 4 km FWHM triangular function, corresponding to the SHIMMER altitude resolution. Figure 4 shows the results of comparing the SHIMMER OH with the model for both ascending and descending parts of the orbit for 2007 and 2009 at 64 km. Figure 5 shows the same comparison at 70 km, and Figure 6 shows the same for 80 km. These altitudes span the most usable range of the SHIMMER data. While data was acquired over a larger altitude range, data below 64 km have large errors due to scratches incurred during vibration testing on certain parts of the grating [Englert et al., 2010], and data above 82 km are often missing due to low OH abundances. All three figures show the pattern of the SHIMMER diurnal sampling. Three complete cycles are observed for each year and for ascending and descending nodes. These three cycles correspond to the three combinations of solar zenith angle and H2O discussed with Figure 1.

Figure 4.

Comparison of SHIIMMER OH (black stars) with the photochemical model (red diamonds) at 64 km for ascending (left column) and descending (right column) nodes of the SHIMMER orbit for latitude band is 50–57°N (see Figure 2) for 2009 (top row) and 2007 (bottom row).

Figure 5.

Same as Figure 4 but for 70 km.

Figure 6.

Same as Figure 4 but for 80 km. Here we show a fiducial indicating the day of year when local solar times of 2 pm were sampled by SHIMMER. Days after this fiducial sample earlier local times.

[20] The effect of the seasonal increase in H2O shown in Figure 1 and the resultant displacement to late summer of the occurrence of peak OH are easily apparent in Figure 6 (OH at 80 km), provided one compares similar local times. For example, on day 225 in 2009, the OH in the ascending part of the orbit at 2 pm local time is almost three times greater than at the same local time in early May (about day 125), even though the penetration of solar Lyman alpha into the mesosphere (which produces OH from H2O dissociation) should be greater in early May because it is nearer the solstice than in mid-August.

[21] At 64 km, Figure 4 shows excellent agreement between the model and the observations. For the AM comparisons (refer back to Figure 2), the model and data symbols often completely overlap. In the afternoon, the data are often slightly larger than the model, by about 10%–20%. At 70 km (Figure 5), the agreement is fair; the data are more consistently about 20% greater than the model whereas our 1-σ error estimates are about 10% [Englert et al., 2010]. It is interesting in that historically the issue was a photochemical model overestimate of mesospheric OH; as we have noted [Englert et al., 2010], this is certainly no longer the case. At 80 km, some larger differences are apparent. Here the shape of the diurnal curve is in much poorer agreement with the observations. Most notable is a large afternoon model excess that is not observed. The model also generally underpredicts the morning SHIMMER data as well. Nonetheless, both SHIMMER and the model show the afternoon buildup which was first predicted by Prather [1981].

[22] As we noted above, we can compare the same model calculations to the SABER ozone data. Figures 7-9 present the results for 60, 70, and 80 km. Since ozone has such a large diurnal cycle in the mesosphere, it is convenient to separate each comparison according to day or night for both 2007 and 2009. We defined night as solar zenith angles greater than 102°. Day is solar zenith angles less than 90°. This definition excludes some twilight periods, for example, right near solstice when SABER observed only near 0400 LT. Also, the daytime comparison shows both the SABER 1.27 μm ozone and the 9.6 μm data. Most notable in these figures is that the model generally falls well below the observations; however, there are variations in this pattern. Thus, at 60 km, the model deficit is most consistently about 50%–70% for all local times. At 70 km, the difference is less, particularly for the 1.27 μm ozone data which are consistently less than the 9.6 μm data. At 80 km, the picture is more complicated. Here, the nightime model values generally fall dramatically below the observation, by factors of 2–3 (except near day 240). By contrast during the day, the model is often in good agreement with the observations, and for the period from days 140–160 which sample the morning hours from 8 to 10 LT, the model exceeds the observations. Possible explanations for these disagreements are discussed in the next section.

Figure 7.

Model comparison with SABER ozone (pluses, 1.27 μm data; stars, 9.6 μm data) and model (red diamonds) at 60 km. The model values are taken from the sampling pattern presented in Figure 3. Day is defined as solar zenith angles less than 90°. Night is defined as solar zenith angles greater than 102° (see text).

Figure 8.

Model comparison with SABER ozone in the same format as Figure 7 but for 70 km.

Figure 9.

Model comparison with SABER ozone in the same format as Figure 7 but for 80 km.

4 Discussion

[23] The previous section highlighted some areas of dramatic disagreement between the standard model and both SHIMMER and SABER. Perhaps the most dramatic is the very large underprediction by the model of the SABER ozone. We suggest that a large part of this, at least in the lower mesosphere, is because the SABER ozone has a high bias. To test this suggestion, we present two grand averages of the model/ozone data comparison. Figure 10 shows this for SABER, and Figure 11 shows this for the Aura/MLS data [Froidevaux et al., 2008; we use Version 2.2]. To produce Figure 11, we sampled our model at 1 p.m. and 1 a.m., corresponding to the local times of the MLS measurements and compared it with a grand average of MLS data taken from days 120 to 260 at 55 N for the 2 years. Since the vertical resolution of MLS is not better than 5 km (as compared with SABER which is less than 2 km [Rong et al., 2009]), we only plot five points over the 55–75 km altitude range. We did not plot the MLS day time value at 75 km because even with substantial averaging, the value is little different from the accuracy presented by Froidevaux et al., [2008, Table 2]. At night, ozone abundance at 75 km is much greater and well in excess of the MLS error. Note that, as discussed by several authors [Huang et al., 2008; Dikty et al., 2010] and as seen in Figure 3, SABER never samples local times near 0100 or 1300. Thus, to intercompare SABER and MLS, we need to either develop an empirical diurnal fit to SABER as was done by Huang et al. [2008] or use the same photochemical model with sampling appropriate to each observation, as we do here.

Figure 10.

Grand average of photochemical models (dashed curves) and SABER ozone data (solid curves). Day and night are defined as in Figures 7-9. The SABER vertical resolution is better than 2 km (Rong et al., 2009).

Figure 11.

Same as Figure 10 but for MLS ozone data. Daytime is defined as 0100 LT; night is defined as 1300 LT.

[24] A comparison of Figures 10 and 11 shows that over altitudes between 55 and 70 km, the photochemical model agrees much better with MLS than with SABER. At night in the 65–70 km region, the model-data difference is less than 10%. This is because the MLS ozone is consistently 20%–50% lower than the SABER data. We are not the first to suggest that SABER mesospheric ozone is higher than other observations. Huang et al. [2008] did an indirect comparison of SABER ozone with the shuttle-borne Cryogenic Infrared Spectrometers and Telescopes for the Atmosphere (CRISTA) and UARS/MLS (see their Figure 2) and found that SABER ozone was consistently higher over the range from 55 to 75 km. In addition, in a more careful one-to-one validation study, Rong et al. [2009] found that SABER O3 was higher than all the correlative datasets with which they compared. At certain times of year, specifically northern summer, for certain datasets such as MLS, SABER ozone was over 40% higher. Thus, our analysis is focused on that specific period where the SABER bias was found to be the largest. In short, our result here is consistent with the above studies. Interestingly, the comparison with MLS yields a lower mesospheric model ozone shortfall of 10%–30% which is close to that which we have published previously in comparison with LIMS, ground-based microwave [e.g., Siskind et al., 1995] and HALOE ozone [Siskind et al., 1998; Siskind et al., 2003]. At 80 km, the situation is not clear. On the one hand, Smith et al. [2008] compared nighttime SABER ozone with GOMOS, and while SABER was higher at 76 km, at 80 km, the two datasets agreed well. However, SABER O3 was higher than ground-based microwave at 80 km [Muscari et al. 2012]. Validation of SABER above 75 km is critical to quantify the possible increase in the model deficit.

[25] The second glaring discrepancy with the model is with the SHIMMER diurnal variation at the highest altitudes where the model persistently underestimates the morning and early afternoon OH and overestimates the late afternoon/evening OH. At these altitudes, near 80 km, well over 95% of the odd hydrogen is in the form of atomic hydrogen (H) during the day. As discussed by Allen et al. [1984], the dominant source of odd hydrogen is the photolysis of H2O, and the dominant sink is the reaction of H with HO2. At sunset, the H recombines with O2 or O3 to form HO2 or OH. Other reactions of OH and HO2 with atomic oxygen partition the three odd hydrogen species. While there are thus only about five chemical reactions, plus H2O photolysis, to consider, the interaction between them and with odd oxygen is highly nonlinear and complex.

[26] After some amount of trial and error, we found that the calculated morning/afternoon OH at 80 km can be increased by increasing the rate coefficient for

display math(1)

[27] Increasing the rate coefficient (k1) of the above reaction will produce more HO2 which then reacts according to

display math(2)

to produce more OH at the expense of H. It also decreases the odd oxygen which acts to suppress OH. With the reduction of atomic oxygen, we get a faster buildup of OH during the day. At sunset, there is a slightly smaller buildup of OH because of the lower H and O3.

[28] To increase k1 we simply substituted the value from Wong and Davis [1974] for the recommended expression from the JPL11 compilation [Sander et al., 2011]. The Wong and Davis reference is one of only 2 out of the 11 references considered by Sander et al. [2011] to have actually measured k1 at temperatures below 298 K (room temperature). The other is Kurylo [1972]. Both these references give larger values for k1 than the more recent references. Thus, for M = N2, Wong and Davis give an expression of

display math(3)

where T is the ambient temperature. By contrast, for the low pressure limiting rate constant, JPL11 recommends

display math(4)

[29] At room temperature, the Wong and Davis expression yields 6.0 × 10-32 cm6 sec-1 which is 36% larger than the JPL recommendation. Furthermore, the exponential form used by Wong and Davis to fit to their low temperature results shows a somewhat different temperature dependence than obtained using the power law expression given by JPL11. Thus, at 170K (a typical temperature at 80 km in our analysis), the rate obtained with Wong and Davis [1974], 1.4 × 10-31 cm6 sec-1, is a factor of 1.52 greater than the 9.2 × 10-32 cm6 sec-1 value obtained with the JPL11 expression.

[30] The effect of using the Wong and Davis formulation for (1) on the overall model-SHIMMER OH comparison at 80 km is shown in Figure 12 (compare with Figure 6). In general, the agreement throughout most of the day is significantly improved. At sunset, the model excess is reduced compared with standard chemistry because the faster rate of (1) diverts H from reacting according to

display math(5)

however, a model OH excess persists near sunset. This requires some further discussion.

Figure 12.

Model comparison with SHIMMER OH at 80 km similar to Figure 6; however, here the model uses the formula for the rate coefficient of reaction (1) by Wong and Davis [1974].

[31] The large increase in model OH in the later afternoon and evening at 80 km is because the odd oxygen and odd hydrogen diurnal variations are somewhat out of phase. In the morning, the faster photolysis of O2 leads to an increase in odd oxygen at the expense of odd hydrogen. By afternoon, the photolysis of H2O begins producing odd hydrogen which then suppresses the odd oxygen. This is discussed by Allen et al. [1984], and indeed, the OH variation in their Figure 9c shows a peak OH value of well over 1 × 107 cm-3, similar to what we show. One way to reduce the afternoon/evening buildup would be to lower the H2O that is used as input to the photochemical model, and indeed, this was discussed in the context of the earliest mesospheric photochemical models [Prather, 1981]. As we noted earlier, our water vapor is derived from the assimilation of MLS data. Lambert et al. [2007] suggest that the error in MLS could increase up to 1 ppmv near 80 km; however, in practice, Rong et al. [2010] found much better agreement, typically much less than 0.5 ppmv between H2O from the AIM Solar Occultation for Ice Experiment (SOFIE) and MLS at these altitudes in the Northern Hemisphere (see their Figure 5). By contrast, to change model OH by 50% requires an approximately equivalent H2O change [cf. Summers et al., 2001]. Furthermore, reducing H2O would only reduce the evening OH; it would not increase the morning OH which we require.

[32] There is no indication of a large jump in the OH in the raw spectra, and background effects are small at sunset such that a systematic OH increase of 30%–50% would be easily detected. Rather, as noted in Section 2.2, the fact that we are measuring across many degrees of longitude near the terminator and using a retrieval that assumes spherical symmetry should both tend to smooth out any large variations or localized enhancements that might be present. Furthermore, the time constants for odd hydrogen and odd oxygen chemistry are rapidly becoming long at 80 km, up to a large fraction of 1 day. Given the rapid changes suggested by our model at these local times, we might be particularly sensitive to transport. As discussed by Ko and Sze, 1984, where large horizontal gradients are present, advection by the zonal wind can be important near the terminator.

[33] To document the effect of changing the rate coefficient for k1 on the altitude profile, Figure 13 presents a comparison of the SHIMMER OH at two specific local times with 1-σ error bars. These error bars essentially represent our estimate of the SHIMMER systematic uncertainty as discussed by Englert et al. [2010]. The figure shows that the model fit to the morning data above 75 km is significantly improved, and the fit to the evening model enhancement is slightly improved. Note that the profile is roughly constant with altitude in the evening. From this, we rule out pointing uncertainties as source for the problem. Finally, changing k1 has only a small effect on the model below 75 km. As seen in Figure 5, the data exceed the model by just over 1 σ near 70 km. It is unclear how significant it is, given that several small changes to reaction rates between JPL11 and the previous compilation worsened the model-measurement agreement; we might imagine that there may be several small changes which would improve the comparison, but we have not evaluated this. Certainly, any increase in model OH to better match SHIMMER at 70 km would worsen the ozone deficit problem we see.

Figure 13.

Comparison of observed OH for the morning (top panel) and the evening (bottom panel) with the model for two assumed values for k1 (solid: from JPL11; dashed: using the formula by Wong and Davis [1974]; see text).

[34] Finally, regarding the ozone deficit, the slightly slower rate of (3) leads to a small increase in night time ozone, which helps a little with the model ozone deficit. This is shown in Figure 14 which shows the ratio of the grand average of the ozone calculations as in, for example, Figure 10, of the standard model with the model using the Wong and Davis rate coefficient. The new model ozone is also lower in the daytime; however, if we recall Figure 9, we see that during the AM, model ozone at 80 km was often larger than the observations. While it certainly does not solve the discrepancy, it goes in the right direction.

Figure 14.

Effect of using the new, faster rate for H + O2 + M → HO2 + M rate coefficient on calculated ozone. The figure shows the ratio of two models (new/old) where each model is a grand average of all the calculation illustrated in the previous figures.

5 Summary and conclusions

[35] We have compared a photochemical model against two years of SHIMMER OH measurements. Altogether, the dataset we have analyzed consists of several thousand limb images. The main conclusion is that below 75 km, the current recommended reaction rate coefficients give a very good description of the observations. Consistent with Englert et al. [2010], we find no evidence for any persistent overprediction of mesospheric OH which had been reported in the 1990s. If anything, the latest model slightly underpredicts the observations at 70 km. At the highest altitudes, there are discrepancies with the standard model, and we have highlighted the importance of a faster reaction rate for H + O2 + M (reaction (1) above) as a way to improve the calculated diurnal OH variation. Clearly, this reaction should be restudied at cold temperatures, below 200 K, with modern techniques.

[36] More problematical is the comparison of the model with observed ozone. In the lower mesosphere, below 70 km, we continue to report a photochemical model deficit relative to observations. The precise magnitude of this deficit depends upon the accuracy of various ozone datasets. As has been pointed out [Huang et al., 2008], there are large systematic uncertainties in mesospheric ozone measurements, and this complicates a precise model-measurement comparison. It is likely that the model ozone deficit is in the 10%–50% range which, as noted by Mlynczak et al. [2000], is much smaller than the first analyses in the 1980s [Rusch and Eckman, 1985]. However, as we have shown, a model ozone deficit appears to be remarkably persistent across a diverse range of observations. It is not unique to our model; Dikty et al. [2010] show the SABER data to be generally in excess of their model as well. Perhaps, therefore, the main result of our work is that the persistence of this model data discrepancy difference lends increased credence to that result. Furthermore, given that the model data OH agreement is generally so good, it is difficult, if not impossible to “point one's finger” at HOx chemistry as a cause of the ozone model-data disagreement.

[37] There are several areas for further work. First, given the apparent existence of a mesospheric model ozone deficit that is not connected with HOx chemistry, we may want to revisit some of the suggestions of alternative sources of mesospheric ozone. For example, Slanger et al. [1988] suggested enhanced odd oxygen production by vibrationally excited O2, and there have also been laboratory reports of so-called “triple dissociation” of ozone [Stranges et al., 1995; Turnipseed et al., 1991; Taherian and Slanger, 1985] where three atomic oxygen atoms can be produced by dissociating ozone. It may be relevant in this regard that we found HALOE ozone to be surprisingly insensitive to the variations in H2O seen by HALOE [Siskind et al., 2003], again suggesting the involvement of unidentified non-HOx mechanisms in ozone chemistry. Second, regarding the behavior of OH near the terminator, additional SHIMMER data at tropical latitudes is available. Here the viewing geometry is oriented at a smaller angle with respect to the terminator, thus affording less smoothing in local time. Finally, future work should extend the model-ozone comparisons to above 80 km at night. Here HOx is no longer a dominant catalytic loss process; furthermore, because odd oxygen is long lived, a multidimensional model is required.


[38] This work was supported by a grant from the NASA's Geospace SR&T program. SHIMMER was a joint program between NRL and the DoD Space Test Program. Support was provided by the Office of Naval Research, the DoD Space Test Program and the NASA Heliophysics Division. The NOGAPS-ALPHA analysis, which was used to provide the inputs to the photochemical model, was supported by the Office of Naval Research and the NASA AIM Explorer Science Project. We also thank the SABER team and J. M. Russell, PI, for helpful conversations about their data, the MLS team for the access to their data, and T. Slanger for the references on alternative ozone sources.