Assessing the sensitivity of the hydroxyl radical to model biases in composition and temperature using a single-column photochemical model for Lauder, New Zealand

Assessing the sensitivity of the hydroxyl radical to model biases in composition and temperature using a single-column photochemical model for Lauder, New Zealand Laura López-Comí1,2, Olaf Morgenstern1,*, Guang Zeng1,*, Sarah L. Masters2, Richard R. Querel1, and Gerald E. Nedoluha3 1National Institute of Water and Atmospheric Research (NIWA), Lauder, New Zealand 2Department of Chemistry, University of Canterbury, Christchurch, New Zealand 3United States Naval Research Laboratory, Washington, DC, United States *now at NIWA, Wellington, New Zealand Correspondence to: O. Morgenstern (olaf.morgenstern@niwa.co.nz)

OH to a variety of environmental factors mean that there is considerable disagreement among global chemistry-transport and chemistry-climate models regarding the global OH abundance; this is often 60 expressed in terms of the CH 4 lifetime (e.g., Stevenson et al., 2006;Naik et al., 2013;Voulgarakis et al., 2013). Several model studies have examined changes in OH abundance and the CH 4 lifetime since pre-industrial times. Chemistry-transport models (which use off-line, precalculated meteorology) generally simulate decreases in OH and increases in the CH 4 lifetime, ranging from 6% to 25% during the 21 st century (Thompson, 1992;Lelieveld et al., 1998;Wild and Palmer, 2008). These results differ from those produced by chemistry-climate models which account for changes in both emissions and climate (Stevenson et al., 2000;Johnson et al., 2001;Shindell et al., 2006;Zeng et al., 2010;John et al., 2012). All of them project a reduction in the CH 4 lifetime and an increase in OH.
In particular, Shindell et al. (2006) and Zeng et al. (2010) obtain a ∼10% decrease in the CH 4 lifetime using different emission scenarios in their simulations. More recent and comprehensive studies 70 compare present-day and future results for OH and the CH 4 lifetime among models participating in the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP, Naik et al., 2013;Voulgarakis et al., 2013). Naik et al. (2013) analyse the evolution of the CH 4 lifetime and OH in ACCMIP models since preindustrial times . They point out large variations in the sign and magnitude of OH changes (from −12.7% to 14.6%) amongst ACCMIP models, reflecting 75 uncertainties in natural CO, NO x , and NMVOC emissions as well as roles of the diverse chemical mechanisms included in the models. For present-day (year 2000) simulations of OH and the CH 4 lifetime, Voulgarakis et al. (2013) suggest that diversity in photolysis schemes and NMVOC emissions might cause large variations in simulated OH and the CH 4 lifetime. Trends in OH between 2000 and 2100 are mainly attributed to stratospheric O 3 changes and trends in modelled temperature 80 fields.
(2013) underestimate the CH 3 CCl 3 lifetime (and thus overestimate OH) by 5% to 10% relative to observations. CH 3 CCl 3 is controlled under the Montreal Protocol, meaning its abundance in the atmosphere is approaching the detection limit and it will no longer be a useful constraint on OH in decades to come. 95 A further indirect method to address OH is to measure 14 CO. Manning et al. (2005) find some considerable variability but no long-term trend using this method. According to Krol et al. (2008), this method is considerably more sensitive to high-latitude than low-latitude OH, in contrast to the CH 3 CCl 3 method which is mostly sensitive to tropical OH.
Therefore, a step forward in addressing the uncertainty in modelling OH in global models is to 100 quantitatively assess the contributions of biases in long-lived species that are central to OH. This sometimes involves juxtaposing global models to local-scale (box or single-column) models constrained as much as possible by observations and incorporating only fast photochemical processes.
For example, Emmerson et al. (2005Emmerson et al. ( , 2007) develop a box model to assess the sensitivity of OH and HO 2 to biases in long-lived species, and compare the model results to observations. However, their 105 analyses only pertain to polluted environments not representative of much of the global atmosphere and only take in episodic and surface measurements. Single-column models have been applied to modelling the atmospheric boundary layer (Mihailovic et al., 2005;Cuxart et al., 2006), diabatic processes (Randall et al., 2003;Bergman and Sardeshmukh, 2004), clouds and aerosols (Kylling et al., 2005;Lebassi-Habtezion and Caldwell, 2015;Dal Gesso et al., 2015), the impacts of GHGs 110 on climate change (Vupputuri et al., 1995), and the chemistry of halogen compounds (Piot and von Glasow, 2008;Joyce et al., 2014). Tropospheric OH chemistry of the remote atmosphere has not been assessed in a single-column model framework before.
In the present paper, we introduce and evaluate a single-column model (SCM) constrained with available long-term observations at Lauder, New Zealand (45 • S, 170 • E, 370 m above sea level), 115 to investigate how chemistry-climate model biases in long-lived chemical species and temperature affect OH. Lauder is known for its clean air (Stedman and McEwan, 1983;McKenzie et al., 2008) and large diversity of available measurements (it is part of the Network for the Detection of Atmospheric Composition Change (NDACC), Badosa et al., 2007;McKenzie et al., 2008;WMO, 2011).
Observations made at Lauder include UV radiation and surface, profile, and/or total columns of O 3 120 and several other species. The O 3 , H 2 O, and temperature records produced by ozone sondes cover 1986 to the present. Lauder therefore is ideal for this kind of study. The SCM is built around a medium-complexity stratosphere-troposphere chemistry scheme. The model is forced with Lauder observations and/or output from a chemistry-climate model that uses the same scheme (see below).
In Section 2, we describe the set-up of the SCM, the construction of time series of key species and 125 meteorological parameters that drive the SCM, and the simulations. In Section 3, we present results of simulated OH concentrations and trends from the SCM and analyse the sensitivity of OH to various forcings. Conclusions are gathered in Section 4. The single-column photochemical model (SCM) is a stand-alone version of the stratosphere-troposphere chemistry mechanism used by the National Institute of Water and Atmospheric Research -United Kingdom Chemistry and Aerosol (NIWA-UKCA) model, which comprises gas-phase photochemical reactions relevant to the troposphere and stratosphere (Morgenstern et al., , 2013Telford et al., 2013;O'Connor et al., 2014;Morgenstern et al., 2016). For consistency with NIWA-135 UKCA, the SCM uses the same chemical mechanism. Had we used a more complex mechanism (which the SCM approach lends itself to), then a direct comparison with the NIWA-UKCA output would no longer be possible, and also the results would be less relevant to other global CCMs characterized by relatively simple chemical mechanisms. The 60 vertical levels of the SCM are the same as in NIWA-UKCA, extending to 84 km. We do not use horizontal interpolation and take profiles of at-140 mospheric properties from the gridpoint closest to Lauder (45 • S,168.75 • E). Unlike NIWA-UKCA, the SCM excludes all non-chemistry processes, such as transport, dynamics, the boundary-layer scheme, radiation, emissions, etc. This means the model is only suitable for assessing fast photochemistry. Forcing data for the SCM are mostly interpolated from 10-daily instantaneous outputs from a NIWA-UKCA simulation (see below), except for those fields for which observational data 145 are used. Morgenstern et al. (2013) andO'Connor et al. (2014) describe the chemistry scheme included in the SCM. The SCM includes an isoprene oxidation scheme (Pöschl et al., 2000;Zeng et al., 2008;Morgenstern et al., 2016) not included in the NIWA-UKCA model version used by Morgenstern et al. (2013). In addition to CH 4 and CO, the model includes a number of primary non-methane 150 volatile organic compound (NMVOC) source gases, i.e. ethane (C 2 H 6 ), propane (C 3 H 6 ), acetone (CH 3 COCH 3 ), formaldehyde (HCHO), acetaldehyde (CH 3 CHO), and isoprene (C 5 H 8 ). As noted above, emission and deposition of species are not considered in the SCM. The SCM includes a comprehensive formulation of stratospheric chemistry ) comprising bromine and chlorine chemistry and heterogeneous processes on liquid sulfate aerosols. Overall, the model 155 represents 86 chemical species and 291 reactions including 59 photolysis and 5 heteorogeneous reactions. The FAST-JX interactive photolysis scheme (Neu et al., 2007;Telford et al., 2013) has been implemented in the SCM; this scheme solves a radiative transfer equation accounting for absorption by ozone. The chemical integration is organised through a self-contained atmospheric chemistry package (Carver et al., 1997), and the differential equations describing chemical kinetics are solved 160 using a Newton-Raphson solver ). O 3 profiles used here are a combination of ozonesonde time series (from the surface to 25 km, Bodeker et al., 1998) combined with the Microwave Ozone Profiler Instrument 1 (MOPI1) time series for altitudes above 25 km (Boyd et al., 2007;Nedoluha et al., 2015), covering 1994 to 2010 (Fig.   1a). The ozone sondes have been launched approximately weekly; this defines the temporal cover-175 age of the forcing data used in the SCM calculations. Microwave measurements used here come as several profiles a day at a variable spacing; we interpolate them to the ozone sonde launch times.

Construction of vertical profiles of forcing species and meteorological parameters
Any missing data (during the two periods when the microwave instrument was out of service) are filled using a Fourier series gap-filling method. We compare the two datasets in the height region usefully covered by both (20 to 30 km). The differences between the two measurements range be-180 tween −2% and +6%, and a mean bias that is less than 5%. O 3 profiles are linearly interpolated onto the SCM's grid. Total column ozone calculated by integrating the observed O 3 profiles is also compared to total-column O 3 measured by the Lauder Dobson instrument; the difference is about 5% on average (López Comí, 2016). Lauder ozone measurements have been used in various World Meteorological Organization (WMO) ozone assessments (e.g., WMO, 2011).

185
H 2 O profiles have been constructed using the weekly radiosonde measurements of H 2 O vapour below 8 km (the same soundings that also measure ozone) and NIWA-UKCA model output data above. For validation, we use the monthly National Atmospheric and Oceanic Administration (NOAA) Frost Point Hygrometer (FPH) H 2 O vapour measurements (Vömel et al., 2007;Hall et al., 2016) which start in 2003. FPHs are more accurate compared to radiosonde hygrometers, particu-190 larly for stratospheric conditions. However, due to the later start of the FPH time series and the lower measurement frequency, radiosonde measurements are used in this study. The comparison of FPH and radiosonde H 2 O reveals differences that are generally less than ±5% in the lower and middle troposphere but generally increase in and above the tropopause region (∼ 11 km, López Comí, 2016).
The radiosonde hygrometers have some known problems with measuring low humidity (Miloshe-195 vich et al., 2001). This is reflected in the large differences observed particularly at these altitudes (up to 30%), and to a lesser degree, below them (Fig. 2a). In a comparison of NIWA-UKCA output with FPH H 2 O, larger discrepancies are found throughout the whole troposphere and tropopause region ( Fig. 2b), as can be expected from a low-resolution model unconstrained by observations and subject to problems with modelling H 2 O. Given the consistency of FPH and radiosonde H 2 O below 8 km 200 found before, here we use radiosonde data up to 8 km of altitude merged, in the absence of a more suitable dataset, with NIWA-UKCA output above that.
We use surface in situ measurements from Cape Grim, Tasmania (Cunnold et al., 2002) to rescale NIWA-UKCA model profiles, producing CH 4 profiles that coincide with the ozone sonde launches.
The NIWA-UKCA model simulation had been constrained with historical global-mean surface CH 4 205 values, resulting in an overestimation relative to the Cape Grim data by ∼ 2% (not shown), and both data show a ∼ 5% increase in CH 4 at the surface over the period between 1994 and 2010. Cape Grim CH 4 is a good surrogate for the Lauder measurements because CH 4 is a long-lived, well-mixed atmospheric constituent.
The time series of CO profile over the period of 1994-2010 has been constructed using the NIWA-210 UKCA CO profiles, rescaled such that the total columns match those obtained from the mid-infrared Fourier Transform Spectrometer (FTS) at Lauder (Rinsland et al., 1998;Zeng et al., 2012;Morgenstern et al., 2012). Gaps in the total-column FTS series, such as the period between 1994 and 1996 when the FTS measurements had not started yet, are filled using a regression fit accounting for the mean annual cycle (modelled as a 6-term harmonic series) and the linear trend.

215
The time series of temperature profiles are constructed following the same procedure as used in the construction of O 3 profiles, comprising the radiosonde temperature profiles (from the surface to 25 km) merged with NCEP/NCAR reanalyses (Kalnay et al., 1996) temperatures used in the retrieval of MOPI1 ozone (above 25 km) for the period of 1994-2010. From near the stratopause upwards the NCEP/NCAR temperatures are merged with a mesospheric climatology based on local LIDAR 220 measurements. There are some warm anomalies occurring in the data at 40-60 km during winter months (e.g. in 1996); these may reflect planetary wave breaking in the upper stratosphere.

Simulations
We perform SCM simulations covering the period of 1994-2010, summarized in Table 1 fields (Rayner et al., 2003). The surface emissions of primary species are as described in Eyring et al. (2013), ozone-depleting substances (ODSs) follow the A1 scenario of the World Meteorological Organisation (WMO) Report (WMO, 2011), and surface (or bulk, in the case of CO 2 ) abundances of greenhouse gases (GHGs) follow the "historical" Intergovernmental Panel on Climate Change (IPCC) scenario of global-mean GHG mixing ratios (Meinshausen et al., 2011).
In a "reference" simulation of the SCM all forcings are taken from this REF-C1 simulation of NIWA-UKCA. Alternatively, in sensitivity simulations O 3 , H 2 O, CH 4 , CO, and temperature, or all of these simultaneously, are replaced with the time series of the profiles that are constructed using long-term observational data as described above. For species other than those 5 fields, in all cases 240 we use NIWA-UKCA REF-C1 forcings. We evaluate the SCM only for those times, spaced roughly weekly, for which ozone sonde data are available. With the exceptions of those simulations assessing cloud influences, simulations are conducted assuming clear-sky conditions.

OH sensitivity to correcting chemistry-climate model biases
In this section, we present sensitivity studies to assess the contribution of biases in known factors 245 (O 3 , H 2 O, CH 4 , CO, and temperature) affecting OH photochemistry at Lauder. The response of OH to changes in each forcing is assessed individually and also in combination.

OH sensitivity to O 3 biases
Three sensitivity simulations are conducted to quantify the impact of O 3 biases (defined as differences between observed O 3 and NIWA-UKCA simulated O 3 ) on OH at Lauder.

250
As discussed above, the rate of production P of HO x via O( 1 D) + H 2 O can be expressed as The results of these three sensitivity runs are displayed in Fig. 3. As expected, the pattern of 265 O 3 differences between observed O 3 and modelled O 3 (Fig. 3a) is reflected in the pattern of OH differences produced by the SCM, considering only the "kinetics" effect and assuming no changes in the photolysis rates ( Fig. 3b), with increases of ozone in spring and decreases in autumn, relative to the reference simulation, resulting in changes of the same sign in OH. However, there is a height dependence to this relationship. In summer and autumn, O 3 biases range between −5% and −45%, 270 meaning that the reference simulation overestimates the observations. Such a bias in O 3 results in up to 12% reductions in OH for these seasons when the bias is corrected. In spring between 2 and 6 km, observed O 3 is larger than in the reference simulation by up to 10% at 4 km in October.
Consequently, this results in an increase of OH at around the same altitudes and times of up to 5%.
Regarding the sensitivity simulation considering the photolysis effects, j O1D exhibits differences 275 relative to the reference simulation ranging from ∼ 14% to ∼ 30%. The corrections are positive everywhere, in accordance with the overestimation of TCO in the NIWA-UKCA model with respect to observations (Morgenstern et al., 2013;Stone et al., 2016). In accordance with eq. 1, such an intensification of j O1D causes OH to increase (Fig. 3c). The relative OH response is approximately 50% of the j O1D relative difference. However, Figs. 3(c) and (d) suggest that the magnitudes of the kinetics and the photolysis effects, for the O 3 bias found at Lauder, are comparable, but the seasonalities differ. For example, the kinetics effect maximizes in spring at 5% and minimizes in summer/early autumn at −15% (in the upper troposphere) whereas the photolysis effect on OH maximizes in summer at 16 to 20% and minimizes in spring (figs. 3b and 3d).
OH resulting from the combined kinetics and photolysis effects is displayed in Fig. 3(e). OH 285 responds approximately linearly to the two effects combined, compared to the sum of their individual impacts ( Fig. 3f), despite some small differences between Fig. 3(e) and (f).
Next, we examine the relationship between slant column of O 3 (SCO), j O1D , and OH. Fig. 4(a) shows that there is an approximately exponential relationship between j O1D and the SCO at 6 km of altitude (this effect also exists at other altitudes). The small curvature may be the result of inac-290 curately diagnosing the SCO (ignoring the curvature of the Earth). Another reason could be that the cross section of O 3 is wavelength dependent, and consequently the actinic flux spectrum moves towards longer wavelengths with increasing SCO. Under Lambert-Beer's Law, a perfectly exponential relationship would be expected for a monochromatic UV light source and an isothermal atmosphere.
j O1D and the OH concentration exhibit an approximately linear relationship ( eq. 1, fig. 4b). Com-295 bining these results, we derive an approximately exponential relationship between the SCO and the OH concentration ( fig. 4c). The fit parameters are stated in fig. 4. Due to the compact relationship between j O1D and the SCO, and to account for the curvature, we fit a quadratic relationship between the SCO and log(j O1D ).
To determine a simple coefficient that describes the quantitative contribution of O 3 to OH, a linear 300 regression between differences in OH and O 3 relative to the reference was conducted through the following expression (note that this equation is also used to derive the linear contributions of the other key species to OH chemistry at Lauder): The regression coefficients are depicted in Fig. 5. Reverting to infinitesimal notation, we note that The sensitivity coefficients of OH to the kinetics and photolysis effects of O 3 are shown in Fig.   5(a). Coefficient A 1 , which represent the kinetics effect, ranges from 0 to 0.25 (meaning the relative response of OH is up to a quarter of the relative difference in O 3 ). The sensitivity to photolysis (A 1 ) 315 is > 0.5 throughout much of the troposphere (meaning the relative response in OH is over half the relative change in j O1D ).  A 2 ), with high sensitivity in the lower and free troposphere and reduced sensitivity in the tropopause region.

OH
It is known that large uncertainties are associated with H 2 O vapour measurements. To illustrate this, we repeat the above simulation but now using European Centre for Medium-Range Weather The effect of CH 4 changes on OH is displayed in Fig. 7 (a,c,e). The CH 4 biases are generally small, up to only ∼ 2%, and are assumed to be vertically uniform, with some seasonal variations. Decreases in CH 4 lead to increases in OH due to reduced loss of HO x by CH 4 + OH. The response of OH to CH 4 changes maximizes at 0.6% around 2 km, and decreases at higher altitudes. The seasonal variation of the OH response to CH 4 biases maximizes in March/April (Fig. 7c), which coincides 340 with the maximum absolute bias in CH 4 (Fig. 7a) in the same months. The sensitivity coefficient describing the dependence of OH to CH 4 changes (denoted as A 4 in Fig. 5c) ranges from −0.17 at the surface to −0.32 at ∼ 2 km of altitude, and then decreases to −0.15 at 10 km.
The CO bias and the resulting differences in OH are displayed in Fig. 7(b,d,f). The relative difference of OH with respect to the reference simulation is less than ±5% for all seasons (Fig. 7d), 345 showing that decreases in CO generally lead to increases in OH through the reduced loss of OH through OH + CO. Note that during austral spring NIWA-UKCA overestimates CO, presumably due to exaggerated tropical biomass burning in the model which causes CO biases of up to 10% is the result of CO differences being close to zero).
The sensitivities of OH to CH 4 and CO show comparable values at the surface, but the OH sensitivity to CO increases with height whereas its sensitivity to CH 4 decreases. Note that the CH 4 + OH reaction rate is strongly temperature dependent, which may contribute to the lower sensitivity of OH to CH 4 changes at altitude than to CO. However, further investigation will need to investigate how 355 these ratios change in different chemical regimes, and to assess whether the relative sensitivity of OH to CO and to CH 4 are specific to the clean SH environment.

OH sensitivity to temperature biases
To assess the effects of changes in temperature on OH, we apply the same procedure as for O 3 , for which the effects of temperature have been decomposed into kinetics and photolysis effects. We 360 perform three simulations: In the first simulation, we only apply temperature changes to chemical kinetics, keeping all photolysis rates fixed (noting that most uni-, bi-, and termolecular reaction rates are temperature dependent). In the second simulation, we only consider the photolysis effect, which arises mainly because the cross section of O 3 , the primary UV absorber, is temperature dependent.
The impact of temperature on OH via ozone photolysis again occurs via two different mechanisms: At Lauder, the reference simulation is generally cold-biased (i.e., the temperature correction is positive; fig. 8a). This is particularly the case in the lowest 2 km and throughout the troposphere in the autumn-winter season. The kinetics effect leads to a reduction of OH by up to 2% (fig. 8b).
O( 1 D) + H 2 O and the quenching reactions (eq. 1) are not or weakly temperature dependent, making 375 CH 4 + OH (which is much more sensitive to temperature) the leading factor in causing this small OH reduction. The rate coefficient for this reaction in NIWA-UKCA and the SCM is k OH+CH4 = 1.85·10 −12 exp(−1690K/T ); at 290 K the sensitivity of k OH+CH4 to temperature changes evaluates to about 2%/K. However, OH is well buffered by other reactions, so its sensitivity is considerably smaller than that. The photolysis effect is often somewhat larger than the kinetics effect but peaks in 380 spring ( fig. 8c). This translates into a slight OH reduction comparable in magnitude to the kinetics effect ( fig. 8d). Both effects add nearly linearly in the combined simulation ( fig. 8e,f).
We calculate sensitivity coefficients A 6 and A 6 that define the OH responses to both effects ( fig. 5 e,f). Coefficient A 6 represents the kinetics effect and varies from 0 to −1.75 (i.e., in absolute terms, the relative OH response can be larger than the relative difference in T ). The sensitivity coefficient 385 that describes the sensitivity of OH to changes in photolysis (A 6 ) ranges from 0.6 at the surface to 0 at 10 km of altitude. Figure 5 (e,f) shows sensitivity coefficients for both effects (A 6 and A 6 ). OH changes due to both effects are small (up to 2.5%) and comparable in magnitude.
Several sensitivity studies have been conducted previously to elucidate the impact of temperature on OH (Stevenson et al., 2000;Wild, 2007;O'Connor et al., 2009). None of these studies separately

Linearity of OH sensitivity to biases in all forcings
Here, we assess the effect of changing all forcings (O 3 , H 2 O, CH 4 , CO, and temperature) simultane-400 ously on OH at Lauder. Fig. 9(a) shows the responses of OH to changing all forcings. A comparison with fig. 6 suggests that H 2 O changes dominate the total response of OH to changes in these forcings. At Lauder, NIWA-UKCA is too moist (relative to radiosonde water vapour); this translates into a large OH overestimation of up to ∼ 40% in the reference simulation (Fig. 9a). This is consistent with the underestimated CH 4 lifetime by the NIWA-UKCA model (Morgenstern et al., 2013;405 Telford et al., 2013), assuming that the NIWA-UKCA model is too moist also in other regions. (In the NIWA-UKCA reference simulation used here, the global CH 4 lifetime, disregarding dry deposition, is 7.2 years, whereas a recent best estimate is 9.8 years, with an uncertainty range of 7.6 -14 years (SPARC, 2013)). In general, in the SCM OH responds approximately linearly to the combined changes in major forcings that play an important role in OH chemistry (Fig. 9).
To examine the linearity of OH responses to simultaneous changes in key forcings defined in this study, the combination of all individual contributions, i.e. O 3 (kinetics and photolysis effects), H 2 O, CH 4 , CO, and temperature (kinetics and photolysis effects) to OH, was compared to the OH response to all forcings combined simulation in the SCM through Eq. (4):   Fig. 9(c) however also suggests that there are some 425 notable non-linearity in the chemistry of the troposphere at Lauder. Chemical feedbacks between the impacts of correcting water vapour and ozone may contribute to this non-linearity; for example, a change in the water vapour abundance may impact the sensitivity of OH to changing O 3 .

Trends in OH
We examine variability and trends in OH using the SCM simulation including all key forcings sepa-430 rately for different altitude bins. The results (Fig. 10) indicate that there are no significant long-term trends in OH throughout the troposphere for the period of the simulation (1994-2010) We find trends of −2.1±4.8% at 0-2.5 km, 0.9±2.3% at 2.5-5 km, 2.6±3.5% at 5-7.5 km, and 3.6±4.1% at 7.5-10 km over the period of 1994-2010), but there is evidence of interannual variations at all altitudes (e.g., Manning et al., 2005;Montzka et al., 2011).

435
In addition, we explore variability and trends in the OH column at Lauder to be compared with other estimates of global OH. As expected from the results of OH trends at different altitude bins, we find no significant long-term trend in the OH column (0.5 ± 1.3%) (Fig. 11) Fig. 12(a,c,e) shows the response of j O1D and OH to the presence of the ICs. j O1D and OH are generally reduced below the ICs, relative to the cloud-free situation. The maximum reduction in OH is 10 to 15% in winter below 2 km, coinciding with the maximum reduction in j O1D . There are increases in both fields (up to ∼ 8%) above the ICs in austral spring, associated with the seasonal peak in IC occurrence at the same time. In general, j O1D and OH impacts vary strongly with season, 455 with the maximum reduction occurring in winter close to the surface, and the maximum increase in spring above the ICs.
LWCs are mostly present between 1 and 4 km with the seasonal peak in austral spring ( fig. 12b).
Similarly to ICs, j O1D and OH are enhanced above and throughout much of the cloud layer, and reduced in the lowest 1 km above the surface ( fig. 12e,g). The enhancement in j O1D and OH peaks at 12% between 2 and 4 km of altitude, coinciding with the spring maximum in liquid water content at 1-2 km. Conversely, the reduction in j O1D and OH with respect to the clear-sky condition is ∼ 10% and is produced below the clouds.
The simulation with the combined effect of ICs and LWCs (LICs) produces a reduction in j O1D and OH that ranges between 0% and 20% below the transition of ICs to LWCs at around 2 km, since

465
LWCs are as much as twice as optically dense as ICs ( fig. 12g). An enhancement is produced above this altitude of up to 18%. The magnitudes of changes in j O1D and OH are similar when either ICs or LWCs are considered in the SCM. Furthermore, their effects add up slightly less than linearly when both are present in the simulations ( fig. 12h).
The results shown here indicate that lower clouds generally produce an enhancement in j O1D 470 (Fig. 12d), but higher clouds generally produce a reduction in j O1D in the free troposphere ( Fig.   12b; Tang et al., 2003;Tie et al., 2003;Liu et al., 2009). Furthermore, the vertically and seasonally averaged enhancement and reduction in j O1D are about 2% and 6% respectively for the LWC clouds, similar to the response for the ICs condition; this suggests that the cloud vertical distribution has a bigger effect on photolysis than the change in cloud water content (Tie et al., 2003).

Conclusions
The sensitivity of the OH abundance at Lauder to NIWA-UKCA model biases in key forcing vari- vapour is relatively large in the lower troposphere but decreases with altitude. Assuming this moist bias is not restricted to Lauder (which we do not assess here), this is thus a leading explanation for 500 NIWA-UKCA to produce an underestimated CH 4 lifetime (Morgenstern et al., 2013;Telford et al., 2013), relative to literature estimates (Naik et al., 2013;Voulgarakis et al., 2013;SPARC, 2013).
The bias in modelled CH 4 is small since surface CH 4 in the SCM reference simulation is constrained to follow globally averaged surface observations. The Southern Hemisphere generally has a slightly smaller CH 4 burden than the North. Correcting the resulting positive bias at Lauder causes 505 increases in OH throughout the troposphere, with a seasonal peak in March/April. OH is most sensitive to CH 4 changes in winter, though. In the analysis of the OH sensitivity to CH 4 , the impact of subsequent changes in CH 4 oxidation products which also affect OH could not be addressed within the constraints of an SCM. Inclusion of this effect could change the sensitivity coefficient for CH 4 (Spivakovsky et al., 2000).

510
Except for October-December, NIWA-UKCA has a tendency to underestimate CO. As with CH 4 , the sensitivity of OH to changes in CO is negative throughout the troposphere, reflecting that CO + OH is an important sink for OH.
We show that OH responds linearly to temperature biases. These effects cause a reduction in OH due to the strong dependence of OH + CH 4 on temperature (eq. 1). However, the impact of column CO exist (e.g., Pan et al., 1995;Morgenstern et al., 2012). However, in polluted regions, such as in much of the Northern Hemisphere, NO x and NMVOC levels are elevated relative to Lauder and affect in situ ozone production. This means that these constituents might need to be bias-corrected if the SCM is applied in such regions. This might affect the suitability of our approach under these conditions.

530
Having determined the contributions of the major forcings to the chemistry of OH at Lauder under clear-sky conditions, a step forward would be to assess the impact of clouds on photolysis and thus OH, which could be substantial. Due to a lack of suitable observations to constrain the SCM model with cloud profiles at Lauder, we only assessed how the presence of modelled cloud affects OH, relative to the clear-sky situation. The results show that OH response to cloud strongly depends on 535 the vertical distribution of the clouds, not just the total amount. Both liquid-and ice clouds lead to increases in OH above and to some extent inside the cloud, particularly in the spring season when this effect maximizes. Considering that clouds are amongst the most difficult aspects of the climate system to model adequately, we stipulate that observational profiles of cloud properties would be highly desirable to use for a future continuation of this line of research.

540
In summary, we conclude that at Lauder, OH modelled in NIWA-UKCA is most sensitive to issues with representing water vapour and ozone. This points to the need to improve representations of the hydrological cycle and of tropospheric and stratospheric ozone chemistry in NIWA-UKCA and possibly other, similar chemistry-climate models. Water vapour is coupled to clouds in NIWA-UKCA; it is well known that clouds are difficult to represent adequately in global low-resolution 545 climate models. The biases in ozone may well be partly caused by the moist bias in NIWA-UKCA; this is a subject of ongoing research.
Progress with the simulation of the hydrological cycle in present-generation Earth System Models should improve the simulated water vapour product. Simulating an accurate hydrological cycle has been a long-standing issue in climate models, and progess has been slow. If errors in the simulation of moisture cannot be avoided, perhaps their impact on OH can be corrected for using an approach similar to that which we have presented but using global water vapour measurements.
Such a "correction" of modelled OH might result in a reduction in the inter-model spread of the OH abundance and consequently a more accurate quantification of the methane lifetime. For this, tropical radiosonde data would be particularly valuable -most OH is located in the tropics (SPARC, 555 2013). A similar approach could be used to account for the influence of errors in ozone, although tropospheric in situ ozone measurements may be too sparse to allow for a sufficient characterization of the error in models.

Author contributions
O. Morgenstern devised the original idea. L. López Comí wrote the model, conducted the simula-560 tions, performed the data analysis, and led the writing of the paper, with support from S. Masters, O.
Morgenstern, and G. Zeng. G. Nedoluha contributed the microwave ozone data to the research; R.
Querel contributed the ozone sonde data. All authors contributed to the writing of the manuscript.
Acknowledgements. All data used in this paper can be obtained from the contact author. This work has been supported by NIWA as part of its Government-funded, core research from New Zealand's Ministry of Busi-565 ness, Innovation, and Employment (MBIE). We would like to thank the Lauder team for providing most of the measurements used here. We particularly thank Dan Smale for his help with various aspects of this work. We acknowledge NOAA for the FPH data. We acknowledge ECMWF for provision of the ERA-Interim data and NIWA-UKCA data for other species and temperature.

Reference
NIWA-UKCA data for all species and temperature