2.1 Aura MLS

The Microwave Limb Sounder (MLS) is one of four instruments on NASA's Aura satellite, launched on 15 July 2004. The MLS antenna scans the atmospheric limb as Aura orbits the Earth in a near-polar sun-synchronous orbit, and the instrument measures thermal emission (day and night) in narrow spectral channels, via microwave radiometers operating at frequencies near 118, 190, 240, and 640 GHz, as well as a 2.5 THz module to measure OH. MLS (see Waters et al., 2006) has been providing a variety of daily vertical stratospheric temperature and composition profiles (∼3500 profiles per day per product), with some measurements extending down to the upper tropospheric region and some into the upper mesosphere or higher. For more information and access to the MLS data, the reader is referred to http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/MLS, last access: 4 April 2019; the current data version is labeled 4.2x (with x varying between 0 and 3, depending on the date). Data users interested in MLS data quality and characterization, estimated errors, and related information should consult Livesey et al. (2018), the latest update to the MLS data quality document (available from the MLS website at http://mls.jpl.nasa.gov, last access: 4 April 2019).

2.2 GOZCARDS

The data set considered here for longer-term model evaluation analyses is from GOZCARDS, a data record created using satellite-based Level 2 data as source data sets, which were merged into global monthly zonal means for O₃, H₂O, and HCl, going back in time before the (2004) launch of Aura. Readers are referred to the GOZCARDS description and highlights provided by Froidevaux et al. (2015). In brief, for O₃, the GOZCARDS version 1.01 (v1.01) data record starts in 1979 with solar occultation data from the first Stratospheric Aerosol and Gas Experiment (SAGE I) and continues with data from SAGE II, the Halogen Occultation Experiment (HALOE), the Upper Atmosphere Research Satellite (UARS) MLS, the Atmospheric Chemistry Experiment Fourier Transform Spectrometer (ACE-FTS, using solar occultation), and Aura MLS. Basically, the overlap time periods for different data sets are used to calculate offsets between zonal mean time series in 10^∘ wide latitude bins at each pressure level, and the data sets are adjusted to a reference value (SAGE II mean values, for O₃, or an average of satellite measurements for H₂O and HCl). Monthly standard deviations are also provided, along with other diagnostic quantities. GOZCARDS data extensions past 2012 were created simply by adding more recent MLS data, appropriately adjusted to account for zonal mean differences between versions, once the MLS O₃ v2.2 data became unavailable (for 2013 onward); ACE-FTS data were not included in these more recent years. GOZCARDS O₃ has been used for past O₃ trend assessments and in comparisons to other data records (e.g., WMO, 2014; Nair et al., 2015; Tummon et al., 2015; Harris et al., 2015; Ball et al., 2017, 2018).

For the GOZCARDS O₃ data discussed here, unless otherwise noted, we use GOZCARDS v2.20, a recent improvement and update to the original version. This data set was provided for an updated assessment of stratospheric ozone by Steinbrecht et al. (2017), as well as for the assessment activities of the Long-term Ozone Trends and Uncertainties in the Stratosphere (LOTUS) project and in preparation for the latest international report on the state of the ozone layer, led by the World Meteorological Organization (WMO). GOZCARDS ozone data updates are also used as part of the yearly “State of the Climate” stratospheric ozone-related summaries, produced for the Bulletin of the American Meteorological Society (BAMS). For GOZCARDS O₃ v2.20, the stratospheric retrieval pressure grid is twice as fine as for v1.01; there are now 12 regularly spaced levels per decade change in log of pressure. UARS MLS O₃ data were not included in v2.20, since these retrievals are not readily available on the finer vertical grid (although approximations such as interpolation could be used); also, there is no easy provision of UARS MLS retrieval uncertainties on a finer grid. The most significant change for the new merged O₃ is the effect of using the more robust version 7 data from SAGE II (Damadeo et al., 2013). Version 7 uses National Aeronautics and Space Administration (NASA) Global Modeling and Assimilation Office (GMAO) Modern-Era Retrospective Analysis for Research and Applications (MERRA) temperature (T) profile data (Rienecker et al., 2011) in the retrievals, and these values (rather than T from the National Centers for Environmental Prediction, NCEP) also have a significant impact on the conversion of SAGE II O₃ from its native density–altitude grid to the GOZCARDS mixing-ratio–pressure grid. Also, Aura MLS v4.2 O₃ data are now used (instead of MLS v2.2); HALOE v19 O₃ profiles are included after interpolation to the finer pressure grid before merging. As a result of these changes, we have observed closer agreement and larger correlation coefficients between the Stratospheric Water and Ozone Satellite Homogenized (SWOOSH) ozone data (Davis et al., 2016) and GOZCARDS v2.20 O₃ series than between SWOOSH and GOZCARDS v1.01 (SWOOSH also uses SAGE II v7 O₃ data). More details regarding the impact of GOZCARDS v2.20 O₃ on trends are provided in Sect. 5. We note that no ACE-FTS data were included in this newer version of GOZCARDS O₃.

4.1 Average abundances

We provide climatological latitude–pressure contour plots in the Supplement (Fig. S1 for O₃ and H₂O; Fig. S2 for HCl, HNO₃, and N₂O), showing Aura MLS and WACCM average distributions for 2005–2014. Since such plots do not easily allow one to quantify areas of model–data disagreement, we show in Fig. 1a and b the percent differences between WACCM and MLS climatologies; a positive value means that, on average, the model values exceed the data values. Figure 1c provides a comparison between the models (also as percent differences). We have crossed out the regions in Fig. 1a and b where there is little statistical significance in the model–data differences; more specifically, this is where the absolute differences are less than 1.5 times the systematic uncertainties in MLS abundances. These error estimates have been provided in MLS validation and error characterization work as “typical” (global) profiles versus pressure; such estimates for version 4 MLS data are provided by Livesey et al. (2018) and are typically 2σ estimates; thus, regions with the most significant differences (above the 3σ error level) are shown without crosses. The vertical profile of MLS O₃ systematic errors is given in the Supplement (Fig. S3). Past validation references for MLS O₃ include Jiang et al. (2007), Froidevaux et al. (2008a), and Livesey et al. (2008), as well as the more recent work covering many satellite instruments by Hubert et al. (2016). The original MLS data validation work for H₂O is from Read et al. (2007) and Lambert et al. (2007), who also described N₂O validation; MLS HNO₃ validation was provided by Santee et al. (2007). Based on Fig. 1, most of the model O₃ climatology falls within 5 % to 10 % of the data climatology, except in the upper troposphere and lower stratosphere (UTLS), where SD-WACCM O₃ values are even larger than those from FR-WACCM (Fig. 1c). Our work focuses on the stratosphere, but both FR-WACCM and SD-WACCM average O₃ values at low latitudes are lower than the observed means from 215 to 261 hPa (and these levels lie in the upper troposphere at these latitudes); in this region, there are known MLS positive biases versus tropical ozonesonde data, and this could account for part of the apparent model low bias. However, midlatitude O₃ from 100 to 215 hPa is biased high in SD-WACCM, with a difference to systematic error ratio larger than 2 to 3. An illustration of the more significant differences is given in Fig. 2 for O₃ data and model series at 215 hPa for 50–60^∘ N (which corresponds to lower-stratospheric values). This shows that both FR-WACCM and SD-WACCM values are larger than the data there, and more so for SD-WACCM, for which the overestimate can be larger than 50 %. In relation to this, Imai et al. (2013) showed that SD-WACCM O₃ values are also larger than Superconducting Submillimeter-Wave Limb-Emission Sounder (SMILES) O₃ in the lowest portion of the stratosphere (18–20 km). In Fig. 2, we also see that the model O₃ annual amplitude (AO) is larger than the observed amplitude; both models overestimate seasonal amplitudes by ∼60 %. We discuss the AO more generally in the next section. If anything, MLS O₃ at midlatitudes is slightly high (by ∼5 %) with respect to a multi-instrument mean based on a number of satellite data sets from the Stratosphere-Troposphere Processes And their Role in Climate (SPARC) Data Initiative (DI), as discussed by Tegtmeier et al. (2013), who also showed that the MLS O₃ seasonal cycle at midlatitudes and 200 hPa is in good agreement with the multi-instrument mean. The MLS measurements discussed here provide generally strongly peaked averaging kernels with a resolution of 2.5 to 4 km (see Livesey et al., 2018, for sample kernel plots). Smoothing the model profiles using the MLS averaging kernels (and a priori information) gives very little change (less than a few percent for the Fig. 2 example, and even less at higher altitudes) in O₃ abundances and seasonal cycles, and we see no real need to use such smoothing for our comparisons. The significant model–data differences in Figs. 1 and 2 are not caused by this sort of issue. The midlatitude O₃ differences mentioned above require more detailed investigations but appear to implicate (in part) a transport-related issue in these models.

Figure 2Monthly mean ozone time series at 215 hPa and the 55^∘ N latitude bin (for averages over 50–60^∘ N) from MLS, FR-WACCM, and SD-WACCM (see legend for color coding).

Download

Figure 3Same as Fig. 1 but for stratospheric and mesospheric water vapor.

Download

Figure 3 is the H₂O analog of Fig. 1 but for pressures reaching up to 0.01 hPa; again, a typical profile of MLS systematic error estimates (Livesey et al., 2018) was used to define regions with the most statistically significant model–data differences; there are almost no such regions in Fig.  3. We note that MLS H₂O v3 stratospheric data exhibit a slight high bias (of a few to 5 %) versus multi-instrument means, with a somewhat larger positive bias (of ∼10 %) in the lower mesosphere (Hegglin et al., 2013). Such biases are within the expected measurement systematic errors. MLS version 4 stratospheric H₂O data show essentially no systematic change versus version 3 (Livesey et al., 2018). FR-WACCM and SD-WACCM H₂O mean values are on the low side (by ∼5–15 %) relative to MLS H₂O in the upper stratosphere and in most of the mesosphere (see Fig. 3), implying that the models are in good agreement with the SPARC DI multi-instrument mean H₂O. There is independent evidence that MLS H₂O has a dry bias near the hygropause (at the low end of the vertical range shown in Fig. 3, where the bottom level is 150 hPa); this has been known for some time (Read et al., 2007; Vömel et al., 2007). This is also consistent with the existence of a model high bias relative to MLS near 150 hPa. The H₂O model–data comparisons are generally in agreement within the systematic errors, and the level of agreement is slightly better in the case of SD-WACCM.

Figure 4Same as Fig. 1 but for stratospheric HCl.

Download

For HCl, the climatological comparisons of Fig. 4 show that both models exhibit a small (5 %–10 %) low bias versus MLS HCl in much of the stratosphere, with a stronger negative model bias in the tropics between 100 and 150 hPa. The model–data relative biases in stratospheric HCl are generally within the MLS HCl systematic errors. MLS HCl is slightly on the high side of multi-instrument mean climatological results provided in the SPARC DI report (SPARC, 2017). The small negative model bias in the upper stratosphere could also arise from the lack of a sufficiently pronounced decrease in upper-stratospheric MLS HCl, as a result of the interruption in the main MLS HCl (band 13) data after early 2006 (Livesey et al., 2018). There is also a known positive bias in MLS tropical HCl at 150 hPa (Froidevaux et al., 2008b), so model underestimates in this region are not a sign of model weakness. We also note that both models exhibit a systematic difference versus LS HCl observations (with larger differences for SD-WACCM), as well as a downward sloping pattern (equator to pole) in the Southern Hemisphere (SH) and smaller mean differences (for SD-WACCM) in the Northern Hemisphere (NH).

Figure 5Same as Fig. 1 but for stratospheric N₂O.

Download

Figure 5 provides average comparisons for N₂O; this species is long-lived in the lower stratosphere, which means that good (or poor) model–data agreement in this region can confirm (or deny) accurate model representations of the dynamics. While mean lower-stratospheric SH N₂O values are larger for SD-WACCM than for FR-WACCM (Fig. 5c), the mean fit for SD-WACCM is not significantly better. The most significant climatological differences with respect to the error bars are in the upper stratosphere at low latitudes; in this region, SD-WACCM agrees somewhat better with MLS. However, this is also where the data values decline with height, towards the limits of the MLS sensitivity. Not too surprisingly, this is also where the SPARC DI results for N₂O show the largest scatter in terms of percent differences (often exceeding 10 %–20 %; see SPARC, 2017).

Figure 6Same as Fig. 1 but for stratospheric HNO₃.

Download

Finally, the comparisons for HNO₃ (Fig. 6) reveal very few areas of model–data disagreements outside the systematic uncertainties. However, there is a model underestimation by both WACCM versions, especially in the polar upper stratosphere; this will be discussed more later. At high latitudes in the lower stratosphere, the models tend to overestimate the data. The upper troposphere is where the satellite-based HNO₃ data have been validated the least, but there is some evidence for a high MLS bias in this region (see SPARC, 2017). While this might explain, qualitatively, why the models underestimate MLS tropical UT HNO₃ (in Fig. 6), more work is needed to better evaluate HNO₃ from models and satellite-derived data in this particular region.

It is also worth emphasizing differences between modeled and observed seasonal changes in the polar lower stratosphere over Antarctica. Indeed, we see in Fig. 7 that model HCl values at 46 hPa for 70–80^∘ S do not decline as fast in early winter as shown in the data, even though SD-WACCM tracks the interannual variability better than FR-WACCM does (see Fig. S4 for the relevant time series). Uncertainties regarding lower-stratospheric heterogeneous chemistry modeling for SD-WACCM at high latitudes in the polar winter and spring have been discussed by Solomon et al. (2015) for one specific year (2011), including most of the features shown in Fig. 7. For our broader period, we see that the average rate of HCl decline from May to July (dominated by nighttime conditions) is slower in both SD-WACCM and FR-WACCM than the corresponding mean HCl rate of change from MLS (Fig. 7a). Grooß et al. (2018) recently discussed this HCl model–data discrepancy for dark polar vortex conditions, in comparison to simulations using the Chemical Lagrangian Model of the Stratosphere (CLAMS), which shows even larger HCl discrepancies. These authors discuss possible mechanisms and uncertainties, and they argue that additional decomposition of condensed-phase HNO₃ might play a role, possibly via galactic cosmic ray impacts. Since this rapid decline in HCl occurs during polar night, this HCl issue does not lead to much difference in polar ozone loss rates, which only become significant during sunlit conditions (early spring). Figure 7 confirms that, on average, the SD-WACCM O₃ decline and rise match the data well. We also note that FR-WACCM shows smaller-than-observed declines in HNO₃ and H₂O, whereas SD-WACCM matches these observations much better. The temperature panel (Fig. 7e) gives a possible reason for these differences, as T from FR-WACCM is larger by a few degrees during the coldest phase than T from SD-WACCM (MERRA-based) and larger than the MLS-derived values. This would lead to less irreversible denitrification and dehydration. The nature of the Fig. 7 results is similar at other lower-stratospheric pressures, although there is some variability in the magnitude of the differences. Over the Arctic region (not shown here), temperature-related differences are not as large as over Antarctica, but similar model–data differences in the early winter rate of HCl decline exist there also (as mentioned by Grooß et al., 2018).

Figure 7Each of the panels shows average seasonal changes from 2005 to 2014 for the 70–80^∘ S region at 46 hPa. Data values (black) are from Aura MLS and model comparisons (FR-WACCM in blue, SD-WACCM in red) are provided for HCl (a), HNO₃ (b), H₂O (c), O₃ (d), temperature (e), and N₂O (f). For each month, the error bars represent twice the standard errors in the means, based on the set of 10-monthly averages (from 2005 through 2014).

Download

As an addendum regarding the evaluation of models in comparison to data sets, we provide in Appendix A1 the results of a model grading approach used before (e.g., Douglass et al., 1999; Waugh and Eyring, 2008). We find that this grading method often leads to low grades (see Fig. A1), if applied using systematic uncertainty estimates from the MLS team (Livesey et al., 2018). The model results, as good as they are in many respects, cannot always match the data closely enough, at least based on such grades, although this grading formulation can be reconsidered or slightly modified. For example, the multiplicative error factor (see Eq. A1) can be increased (from 2 to 4) to force these grades to span a more useful range (see Fig. A2); some similarities to the percent difference diagnostics used here are noted in Appendix A1.

Figure 8Amplitude of the stratospheric ozone annual cycle (a, c, e) and semiannual cycle (b, d, f) for Aura MLS (a, b), FR-WACCM (c, d), and SD-WACCM (e, f), based on fits to time series from 2005 through 2014.

Download

4.2 Annual and semiannual cycles

Figure 8 displays the amplitudes of annual and semiannual variations for MLS O₃ and WACCM runs for 2005–2014. These results come from a simple regression fit to the monthly mean series in each latitude–pressure bin. The primary time dependence of the fitted function is given by additive sine and cosine terms (with 12-month and 6-month periods), in addition to constant and linear trend terms; the AO and SAO amplitudes are given by the square root of the sum of the squares of the corresponding fitted coefficients. We see from Fig. 8 that the overall data and model patterns of AO and SAO variability are quite similar. The O₃ AO amplitudes peak at mid- to upper-stratospheric levels, with high-latitude variations also observed as a result of the effects of winter and spring polar chemistry and dynamics. The lower-stratospheric peak AO amplitudes are more prominent over the southern polar regions, where stronger O₃ depletion occurs on a seasonal basis. These MLS O₃ AO patterns are similar to those obtained by Schoeberl et al. (2008), using a much shorter period (September 2004 to December 2006); the same holds for other species (H₂O and HCl). The observed SAO amplitude for O₃ exhibits strong peaks in the upper stratosphere, both in the tropics and at high latitudes. The anti-correlation between O₃ and temperature as a result of temperature-dependent photochemical production and loss terms for O₃ has long been known to cause most of the O₃ variability in the upper stratosphere (Perliski et al., 1989). The AO and SAO amplitudes obtained in that and other past studies (e.g., Ray et al., 1994) are similar to the patterns shown here. If we look more closely (and based on AO amplitude ratio plots not shown here), there are often O₃ AO amplitudes 20 %–80 % larger than those derived from MLS data for both WACCM runs in the lower stratosphere (from 50 hPa at low latitudes to 215 hPa at high latitudes). Such model overestimates of the AO amplitude were shown in Fig. 2. Outside of this region, we observe somewhat closer fits to the MLS AO amplitudes for the SD-WACCM version, and both models track the data and each other well, with AO amplitudes typically within a range of ∼25 %. For the Antarctic lower stratosphere, the timing and magnitude of the seasonal recovery after the ozone hole plays a role, and we have observed that SD-WACCM generally fits the MLS data better than FR-WACCM does. In the tropical upper stratosphere, where the SAO is larger than the AO (see Fig. 8), the SD-WACCM results match the observed SAO amplitudes slightly better than those from FR-WACCM. Despite the existence of a few model–data differences, these AO and SAO amplitude comparisons, coupled with our examination of model/data amplitude ratio plots as well as the time series (which include the phase information), do not elicit major concerns regarding the model characterization of the primary processes expected to govern these modes of O₃ variability. The study of dynamical forcing mechanisms in relation to such modes continues to be an active area of research (for example see Ern et al., 2015; Smith et al., 2017).

For H₂O, a similar overview of the AO and SAO amplitudes is given in Fig. 9, which covers the range from 100 to 0.01 hPa. A peak in these amplitudes resulting from seasonal downward transport before the winter, followed by wintertime dehydration, is observed in the lower-stratospheric southern polar region; we note that SD-WACCM results match this feature better than FR-WACCM does. Other features include the Southern Hemisphere's upper-stratospheric AO peak in the extra-tropics. This has been seen by many satellite measurements (see Lossow et al., 2017a). Lossow et al. (2017b) explained this “island” feature in more detail; they argue that vertical advection tied to the upper branch of the Brewer–Dobson circulation largely explains the seasonal highs (lows) via downwelling (upwelling). They also show that an AO maximum is observed as well in other species in roughly the same region, including in N₂O MIPAS (Michelson Interferometer for Passive Atmospheric Sounding) data. We confirm this behavior (see also Fig. S5) from the N₂O AO amplitude feature observed in MLS data, as well as in the model runs (and more so in SD-WACCM). The derived AO and SAO amplitude patterns in H₂O from Lossow et al. (2017a) are consistent with what we find; this includes the peak values in the upper stratosphere and mesosphere, attributed to combined effects of photochemistry and vertical transport. For ozone, a dominant feature in the SAO amplitude exists in the tropical upper stratosphere; see Lossow et al. (2017a) for a brief review of past work explaining such dynamically driven features for H₂O. While there is generally a good level of model–data agreement in the main H₂O AO and SAO patterns, both WACCM comparisons tend to underestimate observed AO and SAO amplitudes in the lower stratosphere and overestimate AO amplitudes in the SH mesosphere, while slightly underestimating mesospheric SAO amplitudes in both polar regions. The largest amplitude differences reach a factor of 2, in places, for the lower-stratospheric model underestimates; however, we note that this is also the region where AO and SAO amplitudes are smallest (typically <0.1 ppmv).

Figure 9Same as Fig. 8 but for H₂O annual and semiannual cycles in the stratosphere and mesosphere.

Download

For N₂O, we already mentioned the existence of the upper-stratospheric AO peak (the “island” feature described by Lossow et al., 2017b) in the southern hemispheric extratropical region. This is similar to the H₂O AO amplitude maximum feature, but the N₂O seasonal variations are anti-correlated with H₂O, as demonstrated by Lossow et al. (2017b), using MIPAS data (and as is also apparent in MLS time series, not shown here). Furthermore, we observe in Fig. S5 a somewhat better match in the AO and SAO N₂O amplitude patterns for SD-WACCM than for FR-WACCM versus MLS, in particular for tropical to southern midlatitudes. This is also manifested in better time series fits for SD-WACCM, besides the closer match in average values; this is what one would generally expect from these two model versions. For HCl and HNO₃, the AO and SAO amplitudes are dominated by the variations at high latitudes (see Figs. S6 and S7). The model HCl upper-stratospheric AO and SAO amplitudes match up fairly well with the observed amplitudes, despite the aforementioned issues relating to MLS upper-stratospheric HCl trends. For HNO₃, the main AO and SAO model features follow the MLS patterns, although there is a model underestimation of the amplitudes in the upper-stratospheric polar regions, because these models do not properly capture the observed recurrences of enhanced HNO₃, as mentioned earlier (see also Sect. 5).

5.1 Anomaly time series: fits and variability

5.1.1 Fits

We wish to evaluate which of the two WACCM models provides a better match, or fit, to the temporal variations in observed deseasonalized anomalies. We would expect SD-WACCM to generally fit these anomalies better than FR-WACCM. We calculate a diagnostic of model fit to the data by using the rms differences between these deseasonalized model and data series and normalize by dividing this quantity by the rms of the data anomalies themselves. A diagnostic value much less than unity means that the match to the series is much smaller than the typical variability; this also implies a good fit to observed anomalies. In Appendix A2, we provide the mathematical expression for this “rms difference diagnostic”. A better model fit to the observations will be given by a smaller rms difference diagnostic value. We also calculate the (Pearson) correlation coefficients, R, between model and data anomalies, and we use R² as another measure of goodness of fit for the models. The first diagnostic is unitless and does not depart too much from the 0 to 1 range; R² is limited to the 0 to 1 range, with larger values indicating a higher degree of linear correlation. We also use these two diagnostics together, by calculating the ratio of R² over the rms difference diagnostic to obtain a “combined diagnostic”. This diagnostic could have a large value (a good model result) from both a large R² in the numerator, meaning a high correlation with observed anomalies, and a small rms difference in the denominator, implying a good fit. This tends to amplify differences between two model comparisons to the same data. An ideal model fit would correlate tightly in time to observed oscillations but also exhibit the right magnitude by “hugging” the anomaly series. Indeed, two model series could have oscillations in phase with data variations and the same R² values, but with different amplitudes and overall fits; conversely, two model series could have different R² values versus observations, if one is more out of phase than the other, but they could still produce similar rms difference fits.

Figure 10Latitude–pressure contours of diagnostics that show how well the deseasonalized anomalies of model ozone time series (FR-WACCM, a, c, e; SD-WACCM, b, d, f) compare to MLS O₃ anomaly series for 2005–2014. (a, b) show the rms difference diagnostic (see text) and (c, d) show R² values; small rms difference values represent a closer fit, while large R² values represent highly correlated results. (e, f) provide a combined diagnostic, namely the ratio of R² to the rms difference diagnostic from the top panels; larger values here represent a better result for comparisons to the observed time series.

Download

Figure 10 shows latitude–pressure contours of the above diagnostics for FR-WACCM and SD-WACCM O₃ anomalies in relation to MLS anomalies for 2005–2014. We see from Fig. 10a and b that SD-WACCM rms difference diagnostic values (in the 0.2 to 1 range in the stratosphere) are smaller than those from FR-WACCM (between about 0.8 and 1.2). R² values, between about 0.7 and 0.95 in most of the stratosphere, show that SD-WACCM correlates very well with the observations; R² is somewhat smaller in the UTLS. The FR-WACCM O₃ series correlate well with the data at low to midlatitudes for pressure levels between 70 and 7 hPa, which is also where the FR-WACCM rms difference diagnostic shows better performance and (as an explanation) where the dynamics are nudged (to tropical winds) in a similar way as for SD-WACCM. However, FR-WACCM shows almost zero correlation at high latitudes and in all of the uppermost stratosphere.

Figure 11Time series of monthly zonal mean O₃ mixing ratios at 2.2 hPa (a, b) and deseasonalized anomalies (c, d), with the 0–10^∘ N and 40–50^∘ N latitude bins on the left and right, respectively. The two model time series (FR-WACCM in blue and SD-WACCM in red) are compared to the MLS series (in black) for 2005–2014. Diagnostic values (see text for a description) are shown in parentheses in the bottom two panels (c, d), with the first number referring to FR-WACCM and the second number to SD-WACCM.

Download

Figure 11 shows sample O₃ series for the upper stratosphere (at 2.2 hPa) for 0–10^∘ N as well as 40–50^∘ N. Although the tropical series show good correlations for both models versus data, some of the details in the observed semiannual peaks and the interplay between the AO, SAO, and the quasi-biennial oscillation are better followed by the SD-WACCM curve. The differences in O₃ amplitude and phase are more clearly displayed in Fig. 11c, showing deseasonalized anomalies. Diagnostic values provided in this panel show that SD-WACCM performs better than FR-WACCM, with a much larger R² value and a smaller rms difference diagnostic value and, hence, a much better combined diagnostic value. The same comments apply to Fig. 11b and d, which showcase the NH midlatitudes at 2.2 hPa. In the high-latitude lower stratosphere, the poorer FR-WACCM results in Fig. 10 are generally caused by time series that are less in phase with polar winter and spring variations, as well as by more departures in variation magnitudes. Figure 10e and f amplify model differences, with combined diagnostic values mostly below 1 for FR-WACCM but between 1 and 3 in the SD-WACCM case. The more realistic dynamics in SD-WACCM allow for better SD-WACCM fits, as shown in Figs. 10 and 11. These plots also point to poor upper tropospheric results for both models (with SD-WACCM slightly better), although this is not the focus of this paper.

Figure 12Same as the Fig. 10 diagnostics but for H₂O up to 0.01 hPa.

Download

Figure 12 describes the same diagnostics as above but for model–data H₂O comparisons and a top level at 0.01 hPa. Again, the diagnostics of fit are usually much better for SD-WACCM, which yields R² values of 0.6 to 0.9 and rms difference diagnostic values below 1 for most of the stratosphere and lower mesosphere, and therefore better combined diagnostic results than FR-WACCM. The SD-WACCM diagnostics are poorest in the upper mesosphere and at high latitudes near 215 hPa. In the upper stratosphere and mesosphere, the better diagnostic results for SD-WACCM are seen in time series (not shown) as a better match versus the MLS H₂O anomalies in terms of the interannual variability at all latitudes, as well as for some seasonal peaks at high latitudes. We interpret this as the result of a better dynamical representation of the mesosphere for SD-WACCM. The high-quality representation of mesospheric composition by SD-WACCM is also demonstrated in comparisons to measurements of CO profiles above Kiruna, Sweden, by MLS and the Kiruna Microwave Radiometer (Ryan et al., 2018). For H₂O near 200 hPa, poor fits at high latitudes occur where MLS data exhibit a dry bias versus sonde and Atmospheric Infrared Sounder (AIRS) data, as discussed in MLS data documentation (Livesey et al., 2018), as well as by Vömel et al. (2007) and Davis et al. (2016). MLS H₂O is low by a factor of several here versus the WACCM runs (which show values of 10–60 ppmv). The data variability may also be affected by the dry bias retrieval (and oscillation) issues at the lowest altitudes at high latitudes, where observed anomalies are more poorly tracked by the models; a planned future update to the MLS H₂O retrievals will help to mitigate this discrepancy.

Figure S8 displays results similar to Fig. 10 but for stratospheric HCl. SD-WACCM HCl results versus MLS are superior to those from FR-WACCM and show high correlations (R²>0.7) in most of the lower stratosphere, with somewhat poorer results in the upper stratosphere, where the rms difference fits and the combined diagnostic are poor for both models. This upper-stratospheric issue is caused by a data problem in this region, where MLS HCl trends are known to be too flat (close to zero). Poorer correlations are observed for FR-WACCM at high latitudes, but even subtle differences in the timing (phase) can lead to poorer correlations. Poorer fits at 100 hPa in the deep tropics are related to a large underestimate of the data, which may be caused, in part, by a high bias from MLS (Froidevaux et al., 2008b) but more so, for R², by out-of-phase model variability.

For N₂O, we also observe in Fig. S9 (showing model comparisons to the 68 to 1 hPa observations from the 190 GHz MLS N₂O band for 2005–2014) that SD-WACCM fits the data better than FR-WACCM, in both the R² and the rms difference categories. FR-WACCM exhibits poor results in the upper stratosphere and at high latitudes in the lower stratosphere. Both models exhibit poorer rms fits and poorer correlations in the tropical lower stratosphere. Partly, this appears to be caused by a model underestimation of the MLS N₂O variability in this region, with some QBO phasing differences as well.

Figure 13HNO₃ monthly zonal mean mixing ratio time series (2005 through 2014) from MLS, FR-WACCM, and SD-WACCM for 3.2 hPa and 70–80^∘ S.

Download

HNO₃ results (see Fig. S10) show, again, better fits to stratospheric MLS data from SD-WACCM than from FR-WACCM and poor performance from FR-WACCM at high latitudes. Both models do poorly in the upper stratosphere, and Fig. 13 illustrates the magnitude of this discrepancy in the region (3.2 hPa and 70–80^∘ S) where it reaches its maximum. Since its launch, MLS has been observing very large values of HNO₃ in the upper stratosphere, mostly in the polar regions during winter. The WACCM runs used here do not include the necessary photochemical pathways, including the effects of energetic particle precipitation on ion chemistry in the upper atmosphere, to adequately represent such variations; implementation of the necessary missing chemical reactions has not made its way into most CCMs. The solution seems tied to ion cluster chemistry during energetic particle precipitation (EPP) events, which includes large solar proton events (SPEs) as well as more regular auroral activity. Direct high-altitude EPP effects enhance NO_x, which can propagate downward in polar winter and increase stratospheric NO_x and HNO₃ via this indirect effect and conversion of N₂O₅ on ion water clusters (Böhringer et al., 1983). Large polar enhancements in upper-stratospheric HNO₃ were observed by the MIPAS after SPE activity in 2003 (Orsolini et al., 2005; von Clarmann et al., 2005; Lopez-Puertas et al., 2005; Stiller et al., 2005). More complex modeling (e.g., Funke et al., 2011; Verronen et al., 2011; Kvissel et al., 2012; Andersson et al., 2016) has produced EPP-induced enhancements in high-latitude HNO₃, with related improvements in model–data comparisons into the mesosphere. Regarding low-latitude upper-stratospheric HNO₃, the poorer model fits seem to be caused at least in part by more noisy and variable MLS data, under low HNO₃ conditions. Finally, tropical MLS HNO₃ data at 147 and 215 hPa are not fit well by either model, as the data exhibit seasonal oscillations between 0.2 and 0.5 ppbv, whereas model values are smaller than 0.1 ppbv. There have been few tropical UT validation comparisons for HNO₃ (Santee et al., 2007), but in situ data from an airborne chemical ionization mass spectrometer have shown that UT HNO₃ tropical values are mostly below 0.1–0.2 ppbv (Popp et al., 2007, 2009).

We saw in Sect. 4 that there are good climatological comparisons between SD-WACCM and MLS variations over Antarctica during polar winter and spring, except for the rate of HCl decline in early winter; also, poorer results are obtained by FR-WACCM. We also find, not too surprisingly, that interannual differences in lower-stratospheric chemical evolution over Antarctica are not as faithfully reproduced by FR-WACCM as by SD-WACCM. This is shown by the anomaly time series comparisons of Fig. S11 for O₃ and temperature at 68 hPa and 70–80^∘ S, along with the associated model diagnostic values. These plots also show that springtime anomalies dominate the variability, with warmer-than-usual springs (in October in particular), such as in 2012 and 2013, leading to more positive ozone anomalies, i.e., less ozone depletion; conversely, years (2006, 2008, 2010, 2011) with colder-than-usual springs are correlated with negative ozone anomalies and more depleted conditions.

5.1.2 Variability

Given our expectations that SD-WACCM would match observed time series of multiple species better than FR-WACCM, and having demonstrated this in the previous section, we turn to what should be a more fair comparison between the two sets of model results, namely the variability aspect. We calculate the ratio of model to data interannual variability, as obtained from the root mean square values of deseasonalized monthly anomaly series, expressed as a percent of climatological means; a simple linear trend is first subtracted from the series, so that the variability comparisons remove any significant trend differences. We do this for the MLS data, but also for longer-term time series, using the GOZCARDS data. The models are sampled following the monthly sampling of the data sets (but not at the daily sampling level of detail); sampling plays a role for the longer-term (merged) GOZCARDS data, which are comprised of some unevenly sampled occultation data records (depending on latitude and pressure). Figure 14 compares the O₃ variability ratios (models versus data) using as a reference the MLS 2005–2014 data (Fig. 14a and b) and the GOZCARDS merged ozone (1992–2003) data (Fig. 14c and d). To first order, we observe similar patterns for both time periods. The SD-WACCM variability is generally within 10 %–20 % of the data variability (ratio values between 0.8 and 1.2). The FR-WACCM variability is somewhat smaller than the data variability in the polar regions and in the upper stratosphere. Bandoro et al. (2018) also found that the free-running version of WACCM displays smaller ozone variability in the upper stratosphere, both for shorter-term and longer-term variabilities, than the observed variability, based on the merged SWOOSH O₃ data record. In the lower stratosphere at low to midlatitudes, FR-WACCM exhibits slightly larger variability than the data, whereas SD-WACCM shows slightly smaller variability than the data; Bandoro et al. (2018) found that FR-WACCM slightly overestimates the decadal variability in this region. We also see in Fig. 14 that, at high latitudes, FR-WACCM underestimates the actual variability, whereas this is less of an issue for SD-WACCM, with its more realistic representation of the dynamics; as an example, refer to Fig. 13 for model and data anomalies at 68 hPa and 70–80^∘ S.

Figure 14Variability ratios (model results divided by data results) for stratospheric O₃, with FR-WACCM results on the left (a, c) and SD-WACCM on the right (b, d). Before calculating the ratios, the variability values are obtained as the root mean square of detrended deseasonalized monthly anomaly time series and expressed as a percentage of mean (climatological) abundances; (a, b) show comparisons to MLS data for 2005–2014, whereas (c, d) are for 1992–2003 comparisons to GOZCARDS.

Download

Figure 15Same as Fig. 14 but for ratios (model/data) of H₂O stratospheric and mesospheric variability for two different time periods.

Download

Figure 16Time series (1992–2014) at 100 hPa and 10–20^∘ S for temperature (b, d) and H₂O (a, c), with deseasonalized anomalies shown in (c, d). The temperature plots just show the two models (FR-WACCM in blue, SD-WACCM in red), whereas the H₂O series show the comparisons for the models versus GOZCARDS merged H₂O data (in black).

Download

A similar overview of model/data variability ratios is provided for H₂O in Fig. 15, for the mesosphere and the stratosphere. Here, while the variability from SD-WACCM is closer than that from FR-WACCM to the data variability during both periods, the tendency for both models is to underestimate the observed variability, with FR-WACCM showing a stronger underestimate in the upper stratosphere and mesosphere. Such an underestimate for FR-WACCM implies that a trend detection in the future will require more years of data, if H₂O continues to have larger variability than models. The FR-WACCM underestimate of the variability is sometimes by as much as a factor of 2, although it is more typically by ∼30 % (see Fig. 15). For a series with rms variability about the fit represented by σ_t, the number of years needed to statistically detect a trend is proportional to ${\mathit{\sigma }}_t^{\mathrm{2}/\mathrm{3}}$ (Weatherhead et al., 1998), and thus an increase of σ_t by factors of 1.3, 1.5, and 2.0, for example, will lead to an increase in the number of years for trend detection by factors close to 1.2, 1.3, and 1.6, respectively. In the tropical lower-stratospheric case, H₂O and temperature values and anomalies for 1992–2014 are shown for 100 hPa and 10–20^∘ S in Fig. 16. Again, we note the smaller-than-observed variability in model H₂O oscillations, with SD-WACCM tracking the data better. This correlates with the temperature series, where smaller variability is seen in FR-WACCM, in comparison to SD-WACCM (which follows MERRA temperatures); we also note that FR-WACCM temperatures are somewhat larger (by ∼1 K on average) than SD-WACCM temperatures in this region. This poorer tracking of cold point temperatures for FR-WACCM (Fig. 16) has likely implications for poorer stratospheric trend results for FR-WACCM H₂O as well, as we will actually observe in the trends section (Sect. 5.2). It is well known that stratospheric entry level H₂O is governed by temperatures near the tropopause cold trap; the monthly average variations shown here are similar to what has been shown in past H₂O work (e.g., Randel et al., 2004, 2006; Randel and Jensen, 2013). Brinkop et al. (2016) used model runs from both free-running and nudged simulations to analyze the impacts of different constraints, including sea surface temperatures (SSTs) and meteorological fields, on sudden drops in H₂O; they found that several of these factors play a role in the H₂O variations, including the timing of El Niño–Southern Oscillation (ENSO) and SST variability, the phasing with the QBO, cold point temperatures, and the correct dynamical model state. Many other analyses of the relation between entry level H₂O, tropopause temperatures, transport, and convection have been carried out previously (e.g., Holton and Gettelman, 2001; Jensen and Pfister, 2004; Fueglistaler and Haynes, 2005; Rosenlof and Reid, 2008; Read et al., 2008; Schoeberl et al., 2013). Our point here is that the WACCM H₂O anomaly series underestimate the observed variability. We note also that this model underestimate exists if we calculate relative variability using a maximum minus minimum range from yearly average anomalies rather than monthly averages. We provide a global view of lower-stratospheric variability differences (models versus data) in the anomaly series comparisons at 83 hPa for all latitude bins in Fig. S12. This also shows that the observed interannual changes in H₂O are better followed by SD-WACCM than by FR-WACCM, including the drop in H₂O after 2011 (see Urban et al., 2014). While lower-stratospheric H₂O variability is underestimated by SD-WACCM by ∼20 %, the correlation between SD-WACCM and observed anomalies is very good (as was shown in Fig. 12).

Figure 17Same as Fig. 14 but for ratios (model/data) of HCl stratospheric variability.

Download

For HCl, we show the (detrended) variability ratios in Fig. 17. The observed HCl variability is fairly well matched (within ∼20 %) by both models in the MLS time period, with an edge given to SD-WACCM. The observed variability is often underestimated (by ∼30 %) by both FR-WACCM and SD-WACCM in the earlier period (1992–2003). We believe that the HALOE sampling plays a role in this; i.e., even if we limit the model comparison (as we do here) to just the same months as when HALOE observations occurred, incomplete sampling in latitude and time can lead to differences versus a fully sampled model (see Toohey et al., 2013) and more so in the polar regions where the HCl variability is large. In the upper stratosphere, the variability ratios are comparable to or somewhat smaller than those in the middle stratosphere, and there is a 20 %–30 % underestimate of the observed variability, which is based on HALOE HCl observations for 1992–2003. There have been difficulties in fully understanding (or modeling) observed upper-stratospheric HCl variations before the declining phase that started after about 2000 (Waugh et al., 2001); see also Sect. 5. For 2005–2014, SD-WACCM actually matches the upper-stratospheric MLS variability fairly well, although these variability values are small.

We also show the ratios of model to data variability for stratospheric HNO₃ (2005–2014) and N₂O (2005–2012) in Fig. S13. We already discussed the issues with missing model chemistry for upper-stratospheric HNO₃ variability, as well as the low signal-to-noise issue for HNO₃ data at low latitudes in this region. There is reasonably good agreement in the HNO₃ variability between SD-WACCM and MLS for the lower to mid-stratosphere, while FR-WACCM generally overestimates the HNO₃ variability in this region. Figure S13 shows N₂O results down to 100 hPa. Here, MLS N₂O–640 data (from the 640 GHz radiometer) for 2005–2012 are used; these retrievals were curtailed in the first half of 2013 as a result of degradation in the 640 GHz radiometer signal chain. Based on results shown in SPARC (2017), there appears to be good agreement in the tropical interannual variability comparisons for N₂O at 100 hPa between MLS and other satellite-derived results. The lower-stratospheric N₂O time series behave more smoothly at low latitudes in the models than in the observations. The interannual variability in the MLS N₂O measurements there is somewhat smaller than the standard deviations in monthly mean N₂O values (of 20–30 ppbv). The MLS N₂O measurement noise itself for a monthly zonal mean (made up of about 5000–6000 profiles) should be less than 1 ppbv. Smoothing the model in the vertical domain to better match the MLS vertical resolution would not lead to a better fit to the observed variability. However, we should keep in mind that the MLS-derived N₂O variability is a small percentage (<3 %) of the monthly zonal mean N₂O abundances. In summary, SD-WACCM shows some underestimate of the observed lower-stratospheric tropical variability for all the species considered here, except for HNO₃; FR-WACCM does so also for three species (H₂O, HCl, and N₂O). It may be that some of the larger variability in the measurements arises from effects not tied just to MLS radiance noise issues, or from variability caused by the proximity to the tropopause for measurements with finite vertical resolution; WACCM could also be genuinely underestimating the actual atmospheric variability near the tropopause (for unknown reasons).

5.2 Trends

In this section, we discuss how the WACCM runs compare to stratospheric observations when it comes to trends, from fairly short-term trends (from the Aura MLS time period) to longer-term trends based on comparisons with O₃, H₂O, and HCl GOZCARDS data records (see Sect. 2 and Froidevaux et al., 2015). Trend analyses have their own complexities in terms of analysis methods and uncertainty estimates. For example, O₃ trend assessments have had to deal with trend estimates from different long-term data records, each with its own characteristics (Tummon et al., 2015; WMO, 2014; Harris et al., 2015; Steinbrecht et al., 2017; Ball et al., 2017, 2018). This kind of analysis is especially difficult when investigating trends from time series with high variability compared to the size of an underlying long-term change over time, which is certainly an issue for the lower stratosphere. Also, global modeling efforts have led to improved characterizations of the expected combined and separate impacts on ozone profiles of long-term changes in halogen source gases and greenhouse gases (WMO, 2014).

Here, we focus mostly on trend results from WACCM and observations, given the application of the same analysis methods for the different series. We have applied multiple (or multivariate) linear regression (MLR) to the time series of deseasonalized anomalies from the data, FR-WACCM, and SD-WACCM. In Appendix A3, we provide more details regarding the regression model, which includes commonly used additive functional terms, including constant, linear, and periodic terms, along with proxies describing well-known variations arising from the QBO and ENSO, as well as an 11-year solar cycle proxy term. Examples of observational time series from merged ozone observations for 1998 through 2014 are provided in Fig. A3, along with the fits to the series and the linear components (trends). We also discuss our methodology (a block bootstrap method) for trend error evaluations in Appendix A3; such an approach was used, for example, by Bourassa et al. (2014) for their trend analyses of ozone from the OSIRIS (Optical Spectrograph and Infrared Imager System) retrievals. We display the resulting trend error bars as 2σ values (which is very close to the 95 % bounds on the distribution of trend results). Such calculations often lead to significantly larger error bars than more standard methods or code (which neglect the autocorrelation of the residuals).

Figure 18Sample time series of deseasonalized ozone anomalies (%) from 1979 through 2014 from the GOZCARDS data record (version 2.20) compared to the corresponding model anomalies from FR-WACCM (blue) and SD-WACCM (red). Upper-stratospheric series at 3.2 hPa are shown in (a, c, e) and lower-stratospheric series at 68 hPa in (b, d, f); three latitude bins are displayed (30–60^∘ N, a, b; 20^∘ S–20^∘ N, c, d; and 30–60^∘ S, e, f).

Download

For ozone, we give an overview in Fig. 18 of percent deseasonalized anomaly time series for three latitude bins (northern midlatitudes, tropics, and southern midlatitudes) and two pressure levels (3.2 hPa for upper stratosphere, 68 hPa for lower stratosphere). The series were deseasonalized in 10^∘ latitude bins and then averaged. The GOZCARDS data used here (version 2.20) are an update to the original (version 1.01) record (Froidevaux et al., 2015), as mentioned in Sect. 2. Figure 18 shows generally good agreement between the various time series, although if one looks carefully, SD-WACCM is generally closer to the observational time series than FR-WACCM is, as one might expect from previous considerations of goodness of fit and variability; also, percent variability is larger in the lower stratosphere than in the upper stratosphere, thus rendering trend detection more difficult at lower altitudes. We compare in Fig. 19 the ozone profile trend results from MLS data alone, from 2005 through 2014, to those from FR-WACCM and SD-WACCM, for the three aforementioned latitude bins. We show the error bars as 2σ estimates, as this provides an easy way to visualize if there are significant differences between models and data, or between models. Furthermore, we have also shown (in grey) the data trend results for time series through the end of 2018, as a timely reference, but with no corresponding available model results. Trend results through 2018 also help to underscore how trend variability depends on the number of years used in the analyses. For ozone, this shows that there are certainly some altitude and latitude regions where the slightly longer-term trends differ from the shorter-term trends by as much as 0.2 % yr⁻¹ (a typical 2σ uncertainty), and especially at northern midlatitudes in the lowest altitude region, where a significant decrease in trend values is obtained (from 0.5 to 0.9 % yr⁻¹ to near zero for the longer-term trend). In contrast, the southern midlatitudes show that the longer-term trends are positive throughout the stratosphere, whereas slightly negative values were obtained for 2005–2014. These changes can be traced back to ozone variability in the observational record; this underscores that large dynamical variability in the lower stratosphere will continue to create significant variability in the trends. Indeed, the importance of meteorological variability was recently emphasized by Chipperfield et al. (2018), who included MLS data through the end of 2017 in their comparisons to simulations from a chemical transport model; they showed that the MLS data exhibited large increases in the SH lower stratosphere in 2017, which led to a positive tendency for 2005–2017 trends in that region, in contrast to slight declines (or near-zero trends) in the NH, in general agreement with the results shown here, which include one more year (2018) of MLS data. Moreover, Fig. 19 provides a robust indication from the MLS data alone that upper-stratospheric ozone values have been on the upswing in the past decade, at a rate of about 0.1 to 0.3 % yr⁻¹, depending on latitude–altitude region, with 2σ uncertainties of ∼0.2 % yr⁻¹. While these 2σ error bars (obtained using the bootstrap method mentioned earlier) are fairly large, there are several latitude regions and pressure levels with similar results, and this positive trend is a robust near-global upper-stratospheric result (with even smaller uncertainties). These results are broadly consistent with O₃ trends obtained by Steinbrecht et al. (2017), who use MLS as part of the longer-term merged data records, although they studied a longer time period (2000–2016). All things being equal, the errors in these trends should diminish as more years of data are added to the MLS O₃ record, which, for the middle and upper stratosphere, has been characterized as “very stable”, namely within 0.1 to 0.2 % yr⁻¹ versus sonde and lidar network ozone data (Hubert et al., 2016); it seems difficult to quantify “absolute stability” to much better than this, especially in the lower stratosphere. In the lower stratosphere, trend results are closer to zero, with larger variability and error bars (in % yr⁻¹), and unambiguous detection of post-1997 ozone trends in this region remains elusive (WMO, 2014; Harris et al., 2015). The 2005–2014 trends in Fig. 19 show good broad agreement between model and data, with a tendency for SD-WACCM to agree better than FR-WACCM with MLS, albeit not significantly so, given the size of the error bars, especially in the lower stratosphere. The lower-stratospheric tropical results from FR-WACCM are negative, in contrast to both the observations and SD-WACCM, but with large error bars. In the tropical upper stratosphere, both models exhibit a somewhat more positive trend than observed for this period, with FR-WACCM diverging more from the observations than SD-WACCM does. We do find it rather striking that, as a function of latitude (in 10^∘ wide bins) and pressure, the 2005–2014 SD-WACCM O₃ trends follow the MLS trend results quite well; this is clearly shown in Fig. 20 for central latitudes from 55^∘ S to 55^∘ N. As mentioned above (but not shown in Fig. 20), the agreement in these patterns is not as good for FR-WACCM.

Figure 19Ozone stratospheric trends for 2005 through 2014 obtained from monthly zonal mean data (version 4.2 Aura MLS) and models (FR-WACCM and SD-WACCM), after multiple linear regression analyses of deseasonalized anomaly time series, as described in the text. Each panel refers to results from different latitude band average series (see legend). The error bars are 2σ estimates based on bootstrap resampling results (see text). We also show the data trends for 2005–2018 (in grey) to provide an update on how the past 4 years change the (slightly longer-term) tendencies; to avoid more clutter in these plots, we have not overplotted the corresponding error bars, but these are only slightly smaller than the black observational error bars.

Download

Figure 20Ozone trends in different latitude bins for SD-WACCM (red) versus Aura MLS data (black) for 2005–2014. Closed and open circles are for northern and southern latitude bins, respectively. For clarity, error bars are omitted here, as these generally show that model–data trend differences are not significant for this time period.

Download

Figure 21Same as Fig. 19 for ozone trend comparisons, except for the use of longer-term GOZCARDS (version 2.20) ozone data records, with 1985–1997 shown in (a, c, e) and 1998–2014 in (b, d, f), along with the corresponding model results. We also show the updated ozone data trends with 1998–2018 results in grey (for b, d, f).

Download

For a consideration of longer time periods, we compare in Fig. 21 trends from the merged O₃ GOZCARDS record (version 2.20) to model results for two time periods: Fig. 21a, c, and e for 1985–1997 focus on the main declining ozone phase, while Fig. 21b, d, and f show the post-1998 early ozone recovery stages. The largest differences between the two GOZCARDS data versions occur in the tropical upper stratosphere for the declining ozone phase; Fig. S14 displays the tropical trend differences that we obtain for the same three periods as in Fig. 19. In agreement with this are the trend differences provided by Ball et al. (2017), who showed results for the original (version 1.01) GOZCARDS data and for SWOOSH. GOZCARDS version 2.20 data are now in better agreement with the merged SWOOSH O₃ product (as both use SAGE II version 7 data); also, Steinbrecht et al. (2017) showed that these two merged records lead to similar (post-2000) trend results. The improvements in GOZCARDS version 2.20 ozone are a result of the incorporation of the SAGE II v7 retrievals (Damadeo et al., 2013) and the use of the MERRA temperatures (used in v7) for the conversion from density–altitude to the GOZCARDS mixing-ratio–pressure grid. We note, however, that the lower-stratospheric region exhibits interannual variability that is several times larger than that in the upper stratosphere, as seen in Fig. A3 for tropical 1998–2014 data versus SD-WACCM. Even fairly subtle differences in time series over a few years can lead to a sign change in the trends, although there is no statistical significance in the resulting trend differences. For the pre-1998 period, Fig. 21 shows generally good agreement in the trends versus the models; FR-WACCM tends to fit the observed trends more poorly than SD-WACCM in the lower stratosphere, although the upper-stratospheric results at midlatitudes are somewhat poorer for SD-WACCM. The longer-term trends for 1998–2018 based on GOZCARDS data show slightly more positive trends than the 1998–2014 results, and the recent ozone changes that affected the results in Fig. 19 (with a 2005 starting year) are diluted somewhat for the period starting in 1998. The Southern Hemisphere midlatitudes exhibit near-zero or slightly positive trends over the whole stratosphere, for 1998–2018; we note that zero trends are included inside the corresponding error bars, although we did not add (grey) data error bars in these (already busy) plots. We note also that the upper-stratospheric early recovery (positive trends) is robust in the GOZCARDS data record, but the SD-WACCM trend results for the uppermost region tend to overestimate the observed trends and lie outside the error bars in a few cases. We show in Fig. S15 the O₃ anomaly series for 1998–2014 at 1 hPa for 30–60^∘ N, where SD-WACCM and GOZCARDS trends lie outside their 2σ error bars (Fig. 21); here, FR-WACCM happens to be in better agreement with the data. One aspect that could impact model–data differences is that the models are not sampled, here, following the sparser (occultation) viewing, neither in latitude nor in time (time within each month and local time also, since model values are 24 h averages). Also, some of the differences in the upper stratosphere might arise because the averaging of sunset and sunrise occultation data is not as robust for 1998–2004 as for pre-1998, when SAGE II was operating continuously in both modes (and O₃ varies more strongly with local time at 1 hPa than at lower altitudes). Also, HALOE had decreasing spatiotemporal coverage in later years. Denser spatial and temporal sampling is obtained for the MLS period, with very regular sampling; while small systematic (sampling-related) differences may affect the comparisons, such differences should be consistent from year to year, thus minimizing the impact on trend differences.

Our ozone trend results are largely consistent with other previous work (references mentioned above), which (for records including MLS) typically used merged O₃ from GOZCARDS or from SWOOSH (Davis et al., 2016). We find statistically significant trends (meaning that a zero trend is not included in the 2σ error bar range) mostly in the upper stratosphere, both pre-1997 and after 1998. While the GOZCARDS results point to some small O₃ decreases in the lower stratosphere post-1998, as obtained recently by the novel analyses of Ball et al. (2018), our study finds little statistical significance there and a fair level of sensitivity to the starting year or to the data sets used, with a swing to more positive (but marginally significant) results, if the starting year is 2005 (Fig. 19). Past work (e.g., Harris et al., 2015) has also shown sensitivity to start and end points; also, different regression analysis methods can lead to non-negligible differences (e.g., Nair et al., 2013; Kuttippurath et al., 2015). We also note that past analyses of lower-stratospheric tropical O₃ data have shown positive tendencies, based not just on satellite data as indicated here (with marginal significance for now), but also based on SCIAMACHY data (Gebhardt et al., 2014); in that work, a positive trend was also seen in averaged tropical ozonesonde data. The lower-stratospheric ozone trend issue will continue to require further data and additional studies, towards a longer-term result.

Figure 22Trends in three latitude bins for stratospheric and mesospheric H₂O from an analysis of the 2005–2014/2018 MLS data and the two WACCM models for 2005–2014/2018 (a, c, e) and the 1992-2010/2018 time period (b, d, f); only the observational trend results are extended through 2018 (grey points) for MLS (a, c, e) and for GOZCARDS (b, d, f).

Download

For H₂O, there have been conflicting past results on stratospheric trends, depending on whether one investigates sonde or satellite data (e.g., Oltmans et al., 2000; Scherer et al., 2008), and regarding mechanisms that could account for more than a few tenths of a percent per year increase in H₂O, as CH₄ increases do not appear to be large enough for this. Changes in cold point temperatures or in the circulation need to be invoked in order to account for significant decadal-scale trends in H₂O (e.g., Randel et al., 2000; Rohs et al., 2006; Tian and Chipperfield, 2006; Hegglin et al., 2014). Based on our analyses for 2005–2014, we find from Fig. 22 (a, c, and e) a positive H₂O trend both in the MLS data and in SD-WACCM, which tracks observations (versus latitude as well as pressure) better than FR-WACCM does. FR-WACCM exhibits systematically smaller H₂O trend values than both MLS and SD-WACCM at all pressures except near 100 hPa. The fact that FR-WACCM trend results for the 2005–2014 period are significantly smaller for the mid-stratosphere to lower mesosphere than the SD-WACCM and observed trends appears to mainly be a result of slightly different decadal variability in this run; we also see instances in longer time series (not shown) where FR-WACCM short-term changes appear to be larger than those from SD-WACCM. Finally, the FR-WACCM trends have smaller error bars, given the lower variability found in this model over the time period investigated here. With such lower variability, detection of a given trend would take less time than with the actual (observed) variability. The overlap for FR-WACCM and MLS is marginal in the upper stratosphere and lower mesosphere, where the impact on H₂O from CH₄ decomposition should be at its maximum, and cold point temperature variability issues are smaller than near the tropical tropopause; also, the solar cycle effects on the H₂O time series are largest in the upper mesosphere (see also Froidevaux et al., 2015), but our regression analysis includes a solar flux term (see Appendix A3) and largely takes this effect out. For the longer timescale provided on the right side of Fig. 22, for H₂O trends since 1992, we see that the upper-stratospheric and mesospheric trends are much closer to zero than for 2005–2014 or 2005–2018. We have used 2010 as the end year for the model comparisons here to minimize the impact of drifts in the MLS H₂O trend results (see the next paragraph for further details); trend results for 1992–2018 are provided as a reference but should be considered as an upper limit, given this drift issue for (roughly) the last 8 years of MLS data. An analysis of H₂O HALOE profiles and ground-based microwave profiles over Hawaii (Nedoluha et al., 2009) showed that changes in upper-stratospheric and mesospheric H₂O were indeed sensitive to the solar cycle but showed only negligible overall trends between 1992 and 2008. We can also see this evidence for smaller trends in the earlier portion of the H₂O record in time series (not shown here) from the GOZCARDS data set, with the stronger part of the trend coming after 2007 (as confirmed by the Fig. 22 results). The causes for CH₄ changes over the past few decades have been difficult to identify with confidence (see Feldman et al., 2018; Turner et al., 2017). Simple algebra indicates that such CH₄ changes can lead to only part (about half) of the H₂O increases reflected in the 2005–2014 trends, and there must be other reasons for these fairly large short-term trends. We believe that the significant decadal variability in H₂O, which arises from cold point temperature variability, propagated upward as a “tape recorder” signal, and QBO variability account for a large part of the large positive H₂O trends for 2005–2014. Indeed, Garcia et al. (2007) noted in their studies of WACCM trends that multiple decades are likely needed to enable detection of the underlying secular rise in stratospheric H₂O, given the variability arising from ENSO, cold point temperature changes, the QBO, and other factors. Also, a sudden drop in water vapor after 2000 can lead to a stronger post-2000 stratospheric H₂O trend, if one is considering a rather short time period. In terms of the lower-stratospheric longer-term trend results, Fig. 22 (b, d, and f) shows that there have been some decreases since 1992 (and with the 1992–2018 results being viewed as an upper limit). Based on time series (not shown here), the models underestimate these decreases and show less variability, in terms of significant drops in water vapor as well as smaller positive oscillations; other factors that could contribute to these differences include the poorer atmospheric sampling prior to the MLS data and potential systematic uncertainties in the merged data set (which is a more difficult hypothesis to test).

We should also note that water vapor drifts have been detected between coincident MLS and sonde H₂O data, mostly since about 2010, with MLS-derived trends being more positive than those from frost point profiles (Hurst et al., 2016). This relative drift (of as much as 0.5–1.6 % yr⁻¹ for 2010–2015, depending on altitude and location) could play a role in some discrepancies between model (SD-WACCM) and MLS trends for 2005–2014, and it does complicate such comparisons. While SD-WACCM results agree well with MLS in Fig. 22 (a, c, and e), they would become larger than MLS (adjusted) results if one were to subtract more than about 0.3 % yr⁻¹ from these MLS trends, although we should keep the error bars in mind. Possible causes for drifts between MLS and sonde H₂O data are being investigated, with only a small part of this discrepancy currently attributable to a known instrument degradation issue for MLS H₂O, which probably impacts other MLS data from the 190 GHz spectral region (notably N₂O). We see that tropical H₂O at 100 hPa (near the stratospheric water vapor entry level) does not display much of a trend. Previous studies of entry-level H₂O using satellite data have concluded that no significant long-term trend is discernible (Hegglin et al., 2014; Dessler et al., 2014). The former study led to slightly negative lower-stratospheric H₂O trends (although with no statistical significance). GOZCARDS H₂O data used here, however, do not include SAGE II data back to the late 1980s, as was the case in the work by Hegglin et al. (2014), who also obtained positive long-term (1980–2010) trends in the upper stratosphere from satellite-derived H₂O anomalies, merged using a global CCM as a transfer function. As found by others, especially when dealing with large decadal-type variability, the choice of start and end points, as well as the length of period studied, can significantly influence trend values, be it for H₂O or O₃. Lower-stratospheric H₂O trend detection is rendered difficult by such variability, including significant short-term changes (Randel et al., 2006; Hurst et al., 2011; Fueglistaler, 2012; Urban et al., 2014), as noted here also (see Fig. S12).

Figure 23Same as Fig. 19 but for lower-stratospheric MLS HCl data trends for 2005–2014/2018 (with grey points showing MLS results through 2018) and WACCM HCl trends for 2005–2014.

Download

Figure 24Time series (1992–2014) of lower-stratospheric HCl (ppbv) for the GOZCARDS HCl merged data record (black), as well as models (FR-WACCM in blue and SD-WACCM in red). Each panel is for a different pressure level and latitude bin, as labeled (15 hPa, a–c; 32 hPa, d–f; 68 hPa, g–i); the three latitude bins used in this work are 30–60^∘ S (a, d, g), 20^∘ S–20^∘ N (b, e, h), and 30–60^∘ N (c, f, i).

Download

For HCl, changes in stratospheric values have been nonlinear, with a rapid rise prior to 1998 and a slower decrease after 2004, as expected from time-shifted abundances of total surface chlorine (Froidevaux et al., 2015). Focusing first on 2005–2014/2018, we show model and data HCl trends for the lower stratosphere in Fig. 23. The agreement between SD-WACCM and MLS trends (for 2005–2014) is good, especially for 30–60^∘ S, although the error bars are fairly large. However, the (negative) HCl trend results from both models lie below the observed trends almost everywhere. The upper portion of this model–data bias follows what we observe also in the upper stratosphere (not shown), where MLS-derived HCl trends are too shallow compared to expectations (from model and surface-derived chlorine trends), whereas upper-stratospheric trends from the original MLS HCl product were more negative (see Froidevaux et al., 2006; Livesey at al., 2018). As a reminder, the MLS team recommends that data users not include upper-stratospheric MLS HCl data (post-2006) in any trend studies. For the lower stratosphere, where the HCl line is broader, there is less concern about the inability to track the HCl trend. Also, the near-zero drifts (drifts <0.1 % yr⁻¹) obtained between two separate MLS O₃ band retrievals (not shown), one from the same radiometer as HCl and one from the main (very stable) standard MLS O₃ product (see Hubert et al., 2016), provide some confidence regarding the stability of LS HCl trends. At low latitudes, MLS HCl shows a positive tendency (largest at 68 hPa, per Fig. 23), albeit with marginal significance. The vertical gradient in these trends is duplicated fairly well by the SD-WACCM results, although the model trends are usually less than those from MLS. Latitude–pressure trend variability, including positive tendencies, could be related to circulation changes, as implied by analyses of short-term increases in lower-stratospheric HCl seen in both ground-based and GOZCARDS data (Mahieu et al., 2014). For the most part, the MLS trend results for HCl do not change appreciably when we consider the longer period through 2018 (grey points in Fig. 23). Given the rapid rise in chlorine prior to 1998 and the nonlinear changes near the peak period, we show in Fig. 24 lower-stratospheric time series from GOZCARDS merged HCl (Froidevaux et al., 2015) and WACCM for 1992–2014. There is fairly good agreement in the nonlinear behavior observed in both data and model series. The scatter in HCl data decreases after 2005, and the earlier time series suffer from more inhomogeneous sampling, which may at least in part explain the larger scatter and model–data differences (there is no attempt here to sample models within each month like the data, and this would be difficult for a merged data set). There are also regions and periods of slow HCl increases in both data and models (Fig. 24), as well as hemispheric differences in short-term tendencies (Mahieu et al., 2014; Froidevaux et al., 2015). The HCl time series are tracked well by SD-WACCM, which generally matches the data better than FR-WACCM; this is consistent with the understanding that dynamically driven variations are better captured by the incorporation in SD-WACCM of realistic meteorological fields (MERRA). Stolarski et al. (2018) investigated the removal of dynamical variability from MLS lower-stratospheric HCl series by using MLS N₂O data as a fitting function in the regression; this led to HCl trends that match expectations based on changes in surface chlorine. The search for detailed explanations of such short-term increases and variability in lower-stratospheric HCl (and other composition changes) continues to be an interesting area of investigation.

In the upper stratosphere, it has been difficult to explain the details of observed HCl variations from 1998 to 2002, including the dip between these years (Waugh et al., 2001). We show in Fig. S16 near-global (60^∘ S–60^∘ N) GOZCARDS HCl series at 1 hPa. This shows a systematic model underestimate of HCl in the uppermost stratosphere; the model–data difference is much smaller at pressures closer to 5 hPa. While the systematic uncertainties (of ∼0.2 ppbv) in the data encompass the model values, the model total should be increased by the chlorine contribution from very short lived halogenated substances (VSLS) to the stratosphere; although this contribution is only of order 0.1 ppbv (Carpenter et al., 2014), recent evidence suggests that there could be a somewhat larger stratospheric chlorine contribution from VSLS (Oram et al., 2017). Nevertheless, the historical maximum for total tropospheric chlorine was about 3.65 ppbv (WMO, 2014), and this should be the maximum total chlorine expected in the uppermost stratosphere. While WACCM includes the proper abundance and evolution of chlorine source gases, maximum WACCM HCl in the upper stratosphere (and lower mesosphere) is just under 3.4 ppbv. It is also interesting that the gap between the models (both versions) and the data worsens from 1992 to 2000, with the HCl peak occurring later in the data (with a broader peak than in the models). After about 2002, the decrease in near-global HCl roughly follows the model decrease; additional years of HCl data from ACE-FTS should help refine this comparison. In terms of HCl trends, Hossaini et al. (2019) have recently shown that there are positive changes (by ∼15 %) in model upper-stratospheric HCl trends since 2000, i.e., the HCl decreases are smaller, if one takes into account the likely impact of changes in stratospheric chlorine from VSLS.

Figure 25Same as Fig. 23 but for N₂O (a, c, e) and HNO₃ (b, d, f) data and model trends in the lower stratosphere. The N₂O data results are from the N₂O–640 MLS product (retrieved from the 640 GHz radiometer band data), which was discontinued in 2013 because of an instrument issue affecting this band (see text), and these data and model trends apply to the 2005–2012 period. The HNO₃ trend results (data and models) are for 2005–2014, with additional data results (in grey) for 2005–2018.

Download

For N₂O and HNO₃, lower-stratospheric model trends are compared to the corresponding MLS data trends in Fig. 25. We note that the MLS standard product right after launch was N₂O–640 (retrieved from the 640 GHz radiometer data), but it was discontinued after mid-2013, as a result of a hardware degradation issue affecting that band (N₂O only). The current MLS product, N₂O–190, is retrieved from the 190 GHz band. Figure S17 provides evidence of negative drifts in lower-stratospheric N₂O–190, apparently accelerating in the last few years, since the SD-WACCM and actual N₂O values would be expected to continue to rise slowly after the end date on this plot, notably in the tropical lower stratosphere, where N₂O should follow tropospheric trends. Indeed, tropospheric N₂O has been increasing at a fairly steady rate of ∼0.26 % yr⁻¹ (WMO, 2014), consistent with the model N₂O and MLS N₂O–640 lower-stratospheric increases at low latitudes (see Fig. S17 and especially the tropical trends in Fig. 25 at 100 hPa). FR-WACCM N₂O trends show slightly poorer agreement than SD-WACCM versus N₂O–640, although this is not statistically significant. Tropical lower-stratospheric MLS N₂O–190 trends (not shown) are negative (albeit with error bars that encompass small positive trends) but show some differences versus expectations and the N₂O–640 results. As for HCl, interhemispheric differences in lower-stratospheric N₂O trends are interesting in terms of their implications for effects relating to transport (age of air) and changes in the circulation. At lower pressure values, the N₂O trends do not mirror the tropospheric N₂O trends (in % yr⁻¹), and other factors play a role (changes in circulation, QBO, N₂O photodissociation). The asymmetry in age of air obtained by analyses of (2002–2012) MIPAS SF₆ data (Haenel et al., 2015) could also be related to asymmetries in N₂O tendencies. They found relatively older air in the northern extra-tropics and younger air in the southern extra-tropics; this could also imply opposite trends for N₂O between southern and northern lower stratosphere. However, Bönisch et al. (2011) have pointed out that tracers (e.g., O₃ and N₂O) can be impacted differently by both vertical and quasi-horizontal transport effects, depending on their relative vertical and meridional gradients. Moreover, their work indicates that detailed attribution of tracer variations to structural changes in the Brewer–Dobson circulation is a complex matter, and short-term and longer-term changes may well have different characteristics. Our work mainly identifies similarities between some of the trend patterns versus pressure and latitude from SD-WACCM, in particular, and the observed trends, for a fairly short-term period. For the HNO₃ trends (Fig. 25), we see fairly good agreement between models and data for 2005–2014; latitudinal tendencies and interhemispheric differences therein are similar for model and data. The spatial gradients of these species are different in the lower-stratosphere (HCl and HNO₃ increase with height, in contrast to N₂O), and we see that the decreasing HCl trends for 2005–2014 at 30–60^∘ S (Fig. 23), in particular, are qualitatively similar to those from HNO₃ in this region. For lower-stratospheric HNO₃, there is an underlying trend part caused by the slow increases in N₂O, as we can observe in longer-term (1980 to present) model time series (not shown here), but more abundant volcanic aerosols also led to large HNO₃ changes, tied to heterogeneous chemistry, in the 1980–1992 time period. N₂O and H₂O (source gases for HNO₃) are significantly affected by the QBO, and there is a strong related variability in lower-stratospheric HNO₃. Observed short-term trend patterns in HCl, HNO₃, and N₂O are better captured by SD-WACCM, overall, than by FR-WACCM, as we show in Fig. S18 for the 2005–2010 period, which is relevant to the results of Mahieu et al. (2014), who emphasized short-term HCl increases during this time. We note the correlation in these short-term trend results for HNO₃ and HCl but an anti-correlation for N₂O versus HCl (and HNO₃). Figure 25 also shows that the inter-hemispheric asymmetry in trends for lower-stratospheric HNO₃ has persisted for the 2005–2018 period, although longer-term SD-WACCM results (not shown) indicate that this is not modeled as a stable pattern over several decades. Such short-term tendencies and correlations (or anti-correlations) were also noted by Douglass et al. (2017), who emphasized the results from various column time series, from both observational records and model simulations (including modeled age of air).