4.1 Comparison OHP lidar and SABER

From 2002 to 2018 there were 1100 coincident measurements of sufficient quality between OHP lidars and SABER. Figure 3a shows the monthly median temperature differences between the lidar and SABER, while Fig. 3b shows the mean seasonal temperature bias with altitude.

Figure 3a contains the monthly median temperature differences between an OHP lidar temperature profile and a SABER temperature profile. After 2010 there are several time periods during which the Lidar Température et Aérosol (LTA) was not in routine operation or was in the process of being upgraded. To fill in these data gaps we have used temperature profiles derived from the ozone differential absorption lidar (DIAL), also referred to as Lidar Ozone Stratosphérique (LiO₃S), which is described and validated for temperature in Wing et al. (2018). Given that the main scientific interest of LiO₃S is stratospheric ozone, the noise floor of the raw lidar signal occurs at a lower altitude than for LTA for similar vertical integration. To produce temperature profiles which extend into the mesosphere, we use a coarser vertical resolution and a minimum altitude of 30 km, and often stop the temperature profile below 80 km if the temperature error becomes excessive.

Figure 2Example of co-located temperature profiles from the OHP lidar (green), SABER (blue), MLS (red), and MSIS (black).

Download

Figure 3The 16-year systematic comparison of OHP lidars and SABER temperatures. The monthly median temperature differences between the lidar and SABER are shown in panel (a). Red indicates that the lidar is warmer than SABER and blue that the lidar is colder. There are 1100 nights of coincident measurements in the colour plot. Panel (b) is a seasonal ensemble of lidar minus SABER temperature differences. The summer (May, June, July, August) ensemble in red includes 306 nights of coincident measurements, and the winter (November, December, January, February) ensemble in blue includes 397 nights of coincident measurements. Shaded errors represent 1 and 2 standard deviations.

Download

Figure 3a shows a relative warm bias for the lidars with respect to SABER above 70 km. Discrepancies in this region are likely due to lidar initialization errors and background uncertainty, which we have attempted to minimize in the companion publication (Wing et al., 2018). There is also an evident seasonal relative warm bias in the winter stratosphere between 30 and 50 km – a region where lidar uncertainties in both altitude and temperature are well described (Leblanc et al., 2016a, b). Figure 3b shows a very distinctive “S” shape of the bias in both the winter and summer ensembles, which is indicative of a vertical offset between the lidar and satellite measurements. The basic S-shaped bias was identified in studies of synthetic lidar data as being due to vertical offsets between lidar instruments (Leblanc et al., 1998). Unfortunately, this offset is neither constant from night to night, nor constant with altitude, as evidenced by the elongated and distorted nature of the S shape.

If we bin all the temperature differences by month we can clearly see that there is a winter stratospheric warm bias below 45 km and a pronounced summer cold bias in the mesosphere between 50 and 70 km, as shown in Fig. 4.

4.2 Comparison OHP lidar and MLS

From 2004 to 2018 there were 1741 coincident measurements of sufficient quality between OHP lidars and MLS. Figure 5a shows the monthly median temperature differences between the lidar and MLS, while Fig. 5b shows the mean seasonal temperature bias with altitude.

As was the case with the lidar–SABER comparison, in Fig. 5a, we see a lidar warm bias above 70 km and a strong winter stratospheric warm bias near 45 km. In this comparison the stratospheric warm bias appears to have a downward phase migration as the winter progresses. In the corresponding panel, Fig. 5b, we see very pronounced summer time systematic differences which alternate from warm to cold throughout the stratosphere and mesosphere. The winter ensemble shows a very large lidar warm bias near the stratopause.

Figure 4Monthly median temperature difference between lidar and SABER temperature measurements. Red indicates regions where the lidar measures warmer temperatures than SABER and blue regions where the lidar measures colder temperatures than SABER.

Download

Following the same procedure of binning lidar–MLS temperature differences by month, we see a very pronounced downward phase progression of the winter stratospheric warm bias from 45 km in January descending down to 40 km in February and March. Additionally, there is an evident layered cold bias in the summer stratosphere and mesosphere. The three layers appear near 37, 53, and 68 km in Fig. 6.

Figure 5The 14-year systematic comparison of OHP lidars and MLS temperatures. The monthly median temperature differences between the lidars and MLS are shown in panel (a). There are 1741 nights of coincident measurements. Panel (b) is a seasonal ensemble of lidar minus MLS temperature differences. The summer (May, June, July, August) ensemble in red includes 554 nights of coincident measurements and the winter (November, December, January, February) ensemble in blue includes 653 nights of coincident measurements. Shaded errors represent 1 and 2 standard deviations.

Download

5.1 Method to determine the vertical offset between measurements

Matching the two temperature profiles exactly in amplitude and altitude requires a unique altitude-dependent correction factor for each comparison. However, we can make a rough estimate of the average vertical offset between the two measurements by focusing on the region of the stratopause which generally has a defined altitude and a clear structure. We used a simple least-squares method to best estimate the vertical offset that would minimize the temperature differences between the lidar measurement and the satellite measurement. Two examples of this offset calculation for SABER are shown in Fig. 7 and two examples for MLS are shown in Fig. 8. The examples in these figures show nights during which the lidar and satellite temperatures are in good agreement or can be brought into good agreement by applying a small vertical displacement. However, it is important to note that there are examples of lidar–satellite temperature measurements which cannot be brought into good agreement with small vertical displacements. Two such examples can be found in Fig. 9. These examples of poor agreement are almost exclusively found in winter on nights during which the stratopause is greatly disturbed.

Figure 9Two examples of poor matches between lidar and satellite temperature profiles (MLS, a; SABER, b). These mismatches mainly occur between late November and early April on nights during which the stratosphere was disturbed and experiencing a warming.

Download

5.2 Trends in vertical offset between lidar and satellites

We calculated an offset for every coincident measurement between the lidars and SABER and the lidars and MLS. The monthly average of this altitude offset value is represented in Fig. 10 as a green line for years during which the comparisons were primarily between LTA and the satellites and as a blue line for years during which LiO₃S temperatures were used. The green and blue shaded regions are the respective standard deviations. Given the reduced vertical resolution of the temperature profiles from LiO₃S, the least-squares minimized correction for stratopause height is less sensitive to small- and medium-scale fluctuations in the temperature profiles, such as the triple peak structure seen in Fig. 7b. As a result, comparisons between LiO₃S and both satellites (blue curve in Fig. 10) tend toward the mean altitude displacement. This effect is more pronounced when comparing with SABER, which has a finer vertical resolution, than when comparing with MLS which has a coarser vertical resolution. There is a clear, but imperfect, seasonality to these altitude displacements.

Superimposing the traces shown in Fig. 10 onto the colour plots in Figs. 3 and 5 shows a clear correlation between lidar–satellite temperature anomalies and mean monthly altitude displacement between the lidar and satellite temperature profiles, as shown in Fig. 11.

Figure 10Panel (a) features the monthly average displacement of SABER measurements with respect to the OHP lidars (green for LTA and blue for LiO₃S). The standard deviation is given as the shaded area. The mean offset (magenta) is 1446 m with a standard error of 49 m. Panel (b) shows the same analysis with the monthly average MLS displacement. The mean value is 911 m with a standard error of 90 m.

Download

7.1 The need for vertical altitude correction of satellite data

Improved observations of stratospheric and mesospheric temperature profiles and dynamical phenomena are required to advance our understanding of the middle atmosphere. The process of ground to satellite measurement comparison and validation is a vital ongoing scientific activity. By comparing long-term, stable, continuous, high-quality temperature measurements, such as those made by the lidars at OHP, to other data sets we can help to identify potential issues with calibration or retrieval algorithms.

We have presented individual cases in Figs. 7 and 8 in which both MLS and SABER temperature profiles benefited from a slight vertical displacement based on lidar-derived stratopause height. While this scalar adjustment does not correct for non-linear distortions in the altitude vector, it can significantly reduce the magnitude of the temperature bias in the stratosphere and lower mesosphere, as seen in Fig. 14a and b. This technique does not seem to work well when the stratopause is highly disturbed, as can be seen in the two wintertime examples in Fig. 9. The implications of satellite underestimation of sudden stratospheric warming events is of particular concern for reanalysis projects attempting to model middle atmosphere dynamics. However, using lidar data to supplement the satellite record, these fast dynamical processes can be better resolved.

An additional point about vertical resolutions should be made before further interpreting the results of our lidar–satellite temperature comparison. As was noted in Sect. 2 both MLS and SABER measure radiances on a pressure grid with reported vertical averaging kernels, whereas the lidar measures temperature on a fixed geometric altitude grid. SABER has a relatively small and uniform vertical resolution with a FWHM of 2 km, and MLS has a somewhat lower vertical resolution in the middle atmosphere, which increases with height, reaching a FWHM of 15 km at the top of our comparison region at 80 km. To make a fair comparison we have reduced the lidar vertical resolution to accommodate the satellites. There are considerations to keep in mind when comparing the results of the lidar–SABER and lidar–MLS temperature biases. The first is the effect that vertical resolution has on the temperature difference profile. SABER has a smaller FWHM and is therefore much more likely to reproduce the small-scale variations seen in the lidar data. This means that sharp features like the stratopause can easily be used as a tool for detecting possible bias. In contrast, MLS has a much wider FWHM, which means that we expect much less vertical fidelity in the lidar comparison. The lower vertical resolution should act to smooth out the temperature peaks and valleys in a way that may not correspond to the higher resolution ground-based measurements.

7.2 Temperature biases between OHP lidar and SABER

In the companion publication (Wing et al., 2018) we attempted to reduce the magnitude of the initialization-induced lidar warm bias, which is often reported above 70 km. We have reduced the bias by up to 5 K near 85 km and nearly 20 K at 90 km. Some residual systematic warm bias still remains between the lidar satellite comparisons in this publication.

The average 9.9±9.7 K bias at 80 km reported by Dawkins et al. (2018) using nine different metal layer resonance lidars compares favourably to our ensemble bias of 5 K at 80 km Fig. 14b. Given that the resonance lidars do not initialize their temperatures using the same inversion algorithm as the Rayleigh lidars, and that the resonance lidars have a minimum uncertainty near 85 km, perhaps our Rayleigh temperatures are not as influenced by our choice of the a priori density as we initially thought. Further work needs to be done on the topic of initialization-related bias to fully address the effects of noise and a priori choice on high-altitude Rayleigh lidar retrievals. However, we are encouraged by our results and cannot discount the possibility that some of the remaining temperature difference is due to errors or bias in the satellite altitudes.

When considering the residual temperature differences between the OHP lidars and SABER after the altitude correction based on lidar-derived stratopause height, we can see that much of the seasonally varying bias in the stratosphere and mesosphere has been reduced. We are still left with a general summertime cold bias over most of the atmospheric column, except near 45 km, which now achieves a maximum of −4 K in the June mesosphere. We cannot explain this bias from the perspective of the lidar data as nothing in our range resolution changes, our data acquisition cadence and measurement duration are very similar (Wing et al., 2018), and we are well into the linear region of lidar count rates and are not influenced by our a priori or saturated count rates. It is possible that there could be a tidal contribution as summer time lidar measurements start a bit later than wintertime measurements due to a shorter astronomical night. However, given that our criteria for coincidence were chosen to minimize the effects of the first few tidal harmonics, this seems unlikely. It is also possible that there is a seasonally dependent bias in the choice of a priori estimates used in the satellite retrieval of the geopotential height, which could influence the satellite altitude vector.

The cold bias seen below 30 km is most likely due to possible contamination in the lidar data from aerosols and saturation in the low-gain Rayleigh channel. Current OHP lidar measurements use Raman scatter data to correct for these effects and produce temperature profiles down to 5 km. However, these Raman data are not available for the entire 2002 to 2018 analysis period so we have opted not to include them in this work.

7.3 Temperature biases between OHP lidar and MLS

As with the comparison between the lidar and SABER, the lidar and MLS comparison has a pronounced warm bias above 70 km, which is in keeping with previous studies. However, the magnitude and extent of this warm bias in MLS are much more pronounced than in the SABER comparison plot. Much of this difference is due to the reduced vertical resolution of MLS at these high altitudes. This holds true particularly when comparing lower vertical resolution lidar data to MLS.

The lidar MLS comparison has a wintertime stratospheric warm bias which is not much reduced by simply shifting the location of the MLS stratopause (Fig. 14e). We have reduced the magnitude of the difference by 4 K, but the stratopause altitude correction was markedly less successful than in the case with SABER. It is almost universally the case that sudden stratospheric warmings seen by the lidar are missed or smoothed over in the corresponding MLS measurement. Figure 9a is very much a typical comparison for periods when the stratosphere is highly disturbed. There is a limit to how much can be done to improve the lidar–MLS comparison using a simple scalar correction.

The vertical structure which dominates much of the middle portion of the lidar MLS comparison is also difficult to account for. The structure is particularly evident in Fig. 14c and is nearly insensitive to our applied altitude correction. There is nothing in the lidar technique that could explain this pattern. A similar horizontal banding pattern is seen in the comparison of MLS to the European Centre for Medium-range Weather Forecasts (ECMWF) assimilation in the MLS geopotential validation paper (Schwartz et al., 2008). The effect is most likely an artefact introduced in some stage of the satellite retrieval. Studies like ours provide a perfect opportunity to incorporate lidar information into the satellite retrieval and improve the satellite data products. Given the confidence we have in the fixed width and amplitude of the vertical kernels in the lidar measurement, a lidar altitude and temperature vector could be used to recalculate the MLS geopotential and temperature profiles to help identify the source of this artefact.

It is also important to acknowledge that simply correcting for stratopause height offset was counterproductive for our lidar–MLS comparisons above 50 km, as seen in Fig. 14a. It is likely that any potential lidar-derived correction for MLS will be more complex than a simple scalar offset. Such a correction may even have different functional forms in the stratosphere and mesosphere.

7.4 Comparison with previous work

We can compare our results to a few of the studies involving Rayleigh lidar cited in the Introduction; however, it is important to note a few caveats. The methodologies of the different studies vary significantly and a direct comparison between our work and previous studies is fraught with confounding variables, differences in sampling size, and statistics. As a brief reminder, our work was done in a 30^∘ by 8^∘ geographic box centred on the OHP and had a temporal coincidence window of 12 h. We then made nightly average temperature differences between the lidar and satellites and presented the monthly medians of those temperature differences over the span of each satellite (2002–present for SABER and 2004–present for MLS).

The two studies by Sivakumar (Sivakumar et al., 2011; Siva Kumar et al., 2003) made use of monthly averaged lidar and SABER temperatures and found semi-annual temperature anomalies of 3–5 K in the stratopause region during February–March and September–October, which was attributed to averaging and possibly the annual oscillation. We have shown a similar temperature difference in the stratopause of 4 K in the summer before altitude corrections, reduced to zero bias after altitude correction, and 6 K in the winter before altitude corrections, reduced to 1 K after altitude correction. We have correlated this temperature bias directly to a vertical displacement of the satellite altitude with respect to the lidar altitude and not to the annual oscillation. Further work must be done to explore the possibility of North Atlantic Oscillation or annual oscillation effects, but a quick correlation of relative vertical displacement is seen in Fig. 10 and a monthly average AO phase shows an R-squared value of only 0.04 for SABER and 0.03 for MLS. There are isolated periods of up to a year for which it seems like the correlations are significant; however, it is clear that over a period of nearly a decade, the AO phase and wintertime stratospheric temperature anomalies are not correlated.

The study by Yue et al. (2014) compared 120 nights of lidar temperatures to SABER monthly average temperatures and found wintertime temperature anomalies in the stratopause region of 5 K. The stratospheric warm bias was attributed to tides; however, this explanation cannot explain the seasonal nature of this bias found in this work, nor can it explain why a simple vertical displacement of the satellite stratopause height offers a suitable correction. Subsequent to this work, Hauchecorne et al. (2018) published a temperature data set derived from GOMOS stellar occultation measurements, which were validated using Rayleigh lidar. In this paper, the tidal contribution to the lidar–satellite temperature bias in the stratosphere is estimated to be less than 2 K, based on tidal characteristics extracted from the Global Scale Wave Model for tides.

The same 2–5 K stratopause temperature bias was found by Dou et al. (2009). This study used 2332 nights of lidar data from six different NDACC lidars and compared them with the zonal mean temperatures from SABER. The bias was attributed to possible tidal aliasing or aerosol contamination. In Part 1 of this paper we note that contamination by aerosols is not significant above 30 or 35 km and that aerosols cannot be responsible for lidar temperature bias in the stratopause. As was stated in the previous paragraph, Hauchecorne et al. (2018) estimate an upper limit on the effect of tidal aliasing for a small geographic area of less than 2 K. It is possible that tides could play a larger role when comparing measurements from a localized lidar ground site to a zonal average temperature from a satellite. The altitude corrections shown in the present paper account for all 4 K of temperature bias in summer and 5 of 6 K of bias during winter. This type of correction may account fully for the stratopause temperature bias reported by Dou et al. (2009).

In summary, previous studies have shown a consistent systematic temperature bias of 2 to 5 K between different independent lidars and SABER near the stratopause. Tides, aerosols, planetary oscillations, and satellite calibration errors were all given as possible sources of error when accounting for this discrepancy. Our study has shown that the temperature bias has a clear seasonality, with the largest temperature differences occurring in the winter months. We have also shown that a simple vertical correction of the satellite temperatures based on the height of the stratopause, as measured by the lidar, significantly reduces the bias.