Figure 1 illustrates the unique situation of Andenes, Norway (69.3^∘ N, 16.0^∘ E), where all currently available wind measurement techniques covering the gap region in the USLM are concentrated. Wind radiometer, lidar and meteor radar are all contributing to the aforementioned ARISE project (Blanc et al., 2018).

A crucial aspect of intercomparisons of atmospheric data is to account for the different temporal and spatial sampling of the used instruments and models. Radiometer winds are retrieved on pressure levels while the lidar operates on a geometrical height grid; therefore all data are transformed to pressure coordinates according to the CIRA86 climatology (Fleming et al., 1990) for the respective day.

In the sake of conciseness we mainly present averages over several days of measurements in this section. For the interested reader the wind profiles from the radiometer, the lidar and the models for each individual day and night of observation are reprinted in the Figs. S1 to S4.

The longest uninterrupted lidar measurement of the intercomparison campaign took place from 3 to 11 February 2017 and lasted for 187 h. The time series of the lidar and radiometer zonal and meridional wind profiles are shown in Figs. 2 and 3, respectively.

Figure 2Time series of zonal wind during 187.1 h of coincident lidar and radiometer measurements recorded at Andenes from 3 to 11 February 2017. For better visibility the lidar data are binned to 1 km vertical resolution while the temporal resolution is 5 min. The trustworthy altitude range of the wind radiometer data, according to the definition given in Sect. 2.1, is marked by the horizontal dark grey lines. Data outside this range should not be considered as they may substantially be affected by a priori assumptions. All times are expressed in coordinated universal time (UTC) with the ticks at 00:00.

Download

Figure 3As Fig. 2 but for the meridional wind component.

Download

These observations cover a particularly dynamic time period in the vicinity of a minor sudden stratospheric warming. For instance the zonal wind reverts its direction from −40 to 40 m s⁻¹ within a few days. Obviously the lidar time series feature structures of atmospheric waves which cannot be resolved by the microwave radiometer. Beyond this difference of resolution the time series from the two instruments correspond very well.

In Fig. 2 the strong westward winds on 3 February, the pronounced decrease in the wind velocities at stratopause level in the evening of 4 February along with the wind direction being inverted to eastward above 0.3 and below 3 hPa are all captured by both instruments. The same is true for the following westward acceleration on 5 and 6 February as well as the inversion of wind direction to eastward with two distinct wind speed maxima in the night of 8/9 and 10/11 February. It should be noted that also the altitudes and the timing of the features correspond very well. The most notable difference is that the radiometer wind at stratopause level on 4 February stays slightly negative whereas the lidar reaches a zero zonal wind situation, which can mostly be attributed to the temporal averaging by WIRA that includes negative winds from before and after this event.

Similar considerations apply to the meridional wind in Fig. 3. Weak northward winds at the beginning of the time series are followed by a strong increase in wind speed between 4 and 7 February in both the radiometer and the lidar observations. The subsequent decrease and reversal to southward direction in the stratosphere during 8 February, another increase in northward wind and finally the reversal to substantial southward winds below 0.3 hPa in the night from 10 to 11 February are also recorded by both independent instruments.

For more quantitative analyses the average daylight and nighttime wind profiles of this measurement period are presented in Figs. 4 and 5, respectively. The wind at each level has been averaged with the weights being the integration times Δt_i of data sampled at this level during measurement i, i.e. $u_{\text{avg}}=\sum (\mathrm{\Delta }{t}_i\cdot u_i)/\sum \mathrm{\Delta }{t}_i$. The errors of these average winds were accordingly calculated using Gaussian error propagation: ${\mathit{\sigma }}_{\text{avg}}=\sqrt{\sum (\mathrm{\Delta }{t}_i\cdot {\mathit{\sigma }}_i{)}^{\mathrm{2}}}/\sum \mathrm{\Delta }{t}_i$.

Figure 4Mean daytime zonal and meridional wind profiles for the observation period 3–11 February 2017 measured by the wind radiometer WIRA and the RMR lidar in comparison with models, reanalyses and geostrophic wind calculations from MLS geopotential heights. All middle-atmospheric comparison data have been convolved with WIRA's averaging kernels according to Eq. (2) and cut at the limits of the trustworthy altitude range of the coincident radiometer observation. The discontinuities in the profiles visible at the uppermost altitudes originate from the averaging over observations with slightly different altitude coverage. At the uppermost altitudes the simultaneous observations from the meteor radar (not convolved) are shown. The random errors of all measurement data are denoted by dotted lines. Only data from times when lidar and radiometer were both operated in their respective daylight mode have been considered in this plot.

Download

Figure 5As Fig. 4 but for data from the time period when both instruments were operated in their respective night mode.

Download

Generally daylight observations are more challenging for the lidar because of potential uncertainties related to the positioning of the additional Fabry–Pérot Etalon, whereas the nighttime radiometer wind measurements were known to be biased before the upgrade of the retrieval by Rüfenacht and Kämpfer (2017). Only time periods when lidar and radiometer are both in their respective daylight or nighttime configurations have been considered in the following analyses in order to independently validate both modes of each instrument.

For the daylight periods (Fig. 4) both observations and all models are in very close agreement at all altitudes and for both wind components. They all lie well within the random errors, which are more than 1 order of magnitude smaller than the fluctuations of the atmospheric wind during this time. Only the meridional component of the geostrophic wind calculations shows an offset. This is, however, not surprising when considering the coarse resolution of this data set in combination with a very dynamic period with pronounced wind gradients around Andenes.

During nighttime (Fig. 5), up to 0.3 hPa, all data sources agree within WIRA's observation errors or are very close. Above, substantial differences among the various data sources are visible. Notably the radiometer and lidar zonal winds agree throughout the entire altitude range and the ECMWF forecasts are close to the observations. In contrast, MERRA2 is slightly offset by 7 m s⁻¹ and SD-WACCM by more than 15 m s⁻¹. For the meridional component, lidar and radiometer span the spread of offsets above 0.3 hPa, which can reach up to 20 m s⁻¹. The model winds are scattered in between with ECMWF closest to the lidar, SD-WACCM closest to the radiometer and MERRA2 equally differing from both observed wind profiles. The lowermost meteor radar observations tend to indicate a preference for the low wind speeds measured by the wind radiometer. The reason for the disagreement between the meridional winds measured by the radiometer and the lidar at high altitudes could not be definitely identified. Although the radiometer and the meteor radar cover an observation volume of significantly larger extent than the lidar, the discrepancy can most probably not be attributed to the different spatial sampling (see Fig. S6). Nevertheless, the substantial spread among the models and reanalyses indicate a rather heterogeneous atmosphere. A differing temporal evolution of the sensitivities of the lidar and the radiometer to these high altitudes might explain the dissent. Such effects could be introduced by temporally evolving cirrus or polar stratospheric clouds altering the transmission of the lidar signal or by variations of the mesospheric ozone concentration modulating the strength of the microwave emissions. In any case, as the zonal wind measurements are in very close agreement it is not believed that the high-altitude differences in meridional wind observations during these times indicate a fundamental instrumental problem.

The mean differences to the wind radiometer profiles over all measurements made during the entire 11-month intercomparison campaign are shown in Figs. 6 and 7.

Figure 6Mean difference of all convolved daytime zonal and meridional wind profiles from models, reanalyses, MLS geostrophic winds and the RMR lidar to the observations from WIRA recorded during the entire intercomparison campaign. The bands of random errors of the radiometer and the lidar observations are denoted by blue and red dotted lines, respectively. The discontinuities in the profiles visible at the upper and lowermost altitudes originate from the averaging over observations with different altitude coverage. Only data from the time period when both instruments were operated in their respective daylight mode have been considered.

Download

Figure 7As Fig. 6 but for data from the time period when both instruments were operated in their respective night mode.

Download

In total 348 and 326 h of coincident daylight as well as 169 and 158 h of nighttime zonal and meridional wind observations meeting the > 5 h criterion have been made by the lidar and the radiometer, respectively. The slightly reduced measurement time for the meridional component is due to the downtime of one power laser caused by its flash lamps reaching the end of their life cycle in October 2016. For an overview of the temporal distribution of lidar observations along with the dynamical situation around these recordings, the interested reader is referred to Figs. 9 and 10 in Sect. 6, where days with lidar measurements are marked by green dots.

The daylight zonal winds of all models agree within the radiometer's random errors (Fig. 6). The lidar and the radiometer are in close correspondence up to 0.3 hPa, above the lidar has recorded more eastward winds with offsets of up to 10 m s⁻¹ also slightly disagreeing with the models. One should, however, note the increasing uncertainty at these altitudes so that the error bands of WIRA and the RMR lidar almost overlap over the entire sensitive altitude range. The accordance for the daylight meridional component of the different data sources is very good at all altitudes.

A similar picture with again very close correspondence among the comparison data at all altitudes manifests for the nighttime zonal winds (Fig. 7). In contrast, the additional 54 h contributing to this plot can obviously not eliminate the previously discussed meridional wind biases from the 104 h of the long February observation shown in Fig. 5.

Despite the generally very good long-term agreement it should be noted that on shorter timescales the measurements may disagree with the models (see also Figs. S1 to S4). A particularly illustrative example of this situation is presented in Fig. 8. It shows the zonal wind profiles for the night from 4 to 5 February 2017, i.e. during the maximum of the minor stratospheric warming. Clearly the lidar and radiometer observations agree within their random errors.

Figure 8Mean profile of zonal wind for the part of the night from 4 to 5 February 2017 when both measurement systems were operated in night mode. For reasons of clarity only profiles convolved according to Eq. (2) are shown here, the same plot including the unconvolved wind profiles can be found in the Fig. S3 (second row, fourth column). The blue and red dotted lines represent the error bands of the radiometer and lidar observations.

Download

In contrast, all model and reanalysis data correspond well to the observations in the stratosphere but lie significantly outside the error bands above 0.3 hPa, apparently failing to correctly represent the extent of the mesospheric wind shear. With an offset of 17 m s⁻¹ at 0.1 hPa MERRA2 is by far the closest to the observations while the offsets for ECMWF forecasts and SD-WACCM exceed 35 m s⁻¹. This could be an indication that the assimilation of mesospheric data from MLS drives the model closer to reality. The geostrophic wind computed from MLS, disagreeing with the observations over the entire altitude range, is due to the very localised effects in space and time during this minor warming being averaged out when considering the 5^∘ latitude, 30^∘ longitude and 24 h time window for these calculations.

Lidar observations are limited to clear sky conditions. Moreover particularly short lidar observations have not been considered in the observational intercomparisons for two main reasons: some minimal integration time is needed for guaranteeing a sufficiently homogeneous wind field of the horizontal area sampled by the instruments and for the wind radiometer to deliver stable wind retrievals. Therefore the results in Sect. 5 only cover a comparatively small subset of the observations collected by WIRA.

In the present section we aim to exploit the entire data set obtained by the wind radiometer at ALOMAR during the study period, which almost covers a full annual cycle, and compare it with meteor radar, model and geostrophic wind data.

The time series of zonal and meridional wind profiles are shown in Figs. 9 and 10 along with bimonthly average profiles of this data set in Figs. 11 and 12.

Figure 9Time series of continuous or near-continuous zonal wind data at Andenes for the time period between 1 August 2016 and 30 June 2017. The green dots in the uppermost panel mark the dates where wind measurements from the RMR lidar are available. The uppermost panel shows wind radiometer data below the horizontal black line and meteor radar data above. The dark grey lines in the radiometer data denote the altitude limits within which WIRA data are trustworthy according to the conditions stated in Sect. 2.1. Radiometer data beyond this range are noticeably influenced by a priori assumptions should not be used for comparisons, e.g. with meteor radar observations. At the time of writing ERA5 was only available until 31 December 2016 (black vertical line), and therefore ECMWF forecast data are plotted for 2017.

Download

Figure 10As Fig. 9 but for the meridional wind component.

Download

Figure 11Bimonthly average profiles of zonal wind at Andenes from reanalyses and models convolved according to Eq. (2) in comparison with observations from the wind radiometer WIRA and its random error (dashed). ERA5 data were only available for the first two panels at the time of writing. At the uppermost altitudes the raw, i.e. unconvolved, meteor radar wind profiles are shown. Due to the temporal variations of the upper altitude limit of the radiometer observations visible in Figs. 9 and 10 the sampling period of the meteor radar average wind can be rather different from the highest levels of middle-atmospheric data especially for the summer half-year, i.e. in panels (a), (e) and (f).

Download

Figure 12As Fig. 11 but for the meridional wind component.

Download

The extent of the wind radiometer's error bars in Figs. 11 and 12 depends on the number of contributing measurement cycles at the respective altitude and on the signal-to-noise ratio of the wind signature in the recorded spectra. The latter is mainly determined by the opacity of the troposphere at the observation frequency, which is influenced by its variable liquid water and water vapour content. Similarly, the measurement conditions influence the upper limit of WIRA's trustworthy altitude range. In Figs. 11 and 12 USLM data at each altitude are only considered when the radiometer observations are judged trustworthy at this level. This guarantees that all USLM average profiles are based on simultaneous observations and data. As this approach is not possible for the non-overlapping altitude range of the meteor radar, its profiles are averages over all days. This may lead to slightly different temporal sampling between the USLM and the meteor radar data for the tree panels of the summer half-year when WIRA's uppermost trustworthy altitude is not constantly adjacent to the 0.02 hPa line (see Figs. 9 and 10). Moreover, it should be noted that in contrast to the USLM observations meteor radar winds are never convolved with WIRA's averaging kernels according to Eq. (2).

The ECMWF panel in Figs. 9 and 10 is shared by ERA5 before 31 December 2016 and forecast data afterwards. For a full intercomparison ECMWF's ERA5, its forecast and its operational analysis winds between 1 August and 31 December 2016 are reprinted in the Fig S5, which confirms that they follow a very similar pattern. Notable differences between ECMWF forecast and ERA5 are, however, found in the zonal and meridional winds above 0.2 hPa for October–November 2016 (Figs. 11 and 12), which according to Fig. S5 are not related to the change of model cycle in the ECMWF forecasts on 22 November (before this date ECMWF forecast and ERA5 both ran on 41r2) but rather to ERA5 systematically featuring lower absolute zonal and meridional wind speeds at these altitudes.

The fundamental pattern of the time series in Figs. 9 and 10 is the same for all model data and the radiometer observations. Similarly, in cases where the uppermost altitude of trustworthy wind radiometer data and the lowermost level of meteor radar observations are adjacent, the handover of the wind profile between these instruments is remarkably smooth without major jumps in the wind profiles. This behaviour is especially well illustrated for the rapidly changing meridional winds but it can also be discerned for the zonal component.

Despite the good overall agreement of the middle-atmospheric data sources, some differences can be distinguished.

For instance, the mesospheric westward wind above 0.2 hPa is substantially weaker in the ERA5 reanalysis and the ECMWF forecasts in comparison to WIRA, MERRA2 and the geostrophic wind computations from MLS in August 2016, which translates to a substantial bias in the upper part of Fig. 11a. SD-WACCM also features comparable wind speeds as the radiometer observations in August, but the strong mesospheric eastward winds around 28 August and 6 September, which could be artefacts, drive the bimonthly average towards zero. Therefrom one might conclude that the advantages of high-altitude input data (MERRA2 and MLS geostrophic wind) or high model tops (SD-WACCM) drive these models closer to the observations than ECMWF.

In the same panel stronger eastward winds of WIRA with respect to all comparison data can be distinguished below 0.4 hPa, where they lie slightly outside the error bands. This is most probably related to the strong eastward winds measured by the radiometer on several consecutive days at the end of August, which are not seen in this strength by any other source of comparison data. The reason for this difference could not be established and unluckily there are no lidar measurements at this time which could provide independent observational evidence.

SD-WACCM features substantially lower zonal wind speeds compared to all data sets and especially to the observations for October–November and December–January above 0.1 hPa. For December–January this tendency seems to be confirmed by the meteor radar wind whereas these are almost equally offset from the high-altitude radiometer and SD-WACCM data for October–November.

Finally, for February–March 2017 ECMWF forecasts, SD-WACCM and radiometer measurements are in close correspondence while MLS and the geostrophic winds have offsets of up to 7 m s⁻¹ around 0.1 hPa. Besides these few exceptions the agreement of the zonal wind from the different data sources in Fig. 11 is very good and the comparison data often lie within the error bars of the radiometer.

Regarding the meridional component, winds from all data sources agree with each other and are generally within the observation errors (Fig. 12) for almost all bimonthly average periods. The only notable exceptions occur for SD-WACCM in April–May 2017 with offsets of up to 10 m s⁻¹ and, more pronounced, in February–March 2017, when all comparison data suggest higher winds than observed by the radiometer above 0.2 hPa. Again, the spread between the model data is considerable, with offsets between WIRA and ECMWF extending from 7 m s⁻¹ at 0.1 hPa up to 12 m s⁻¹ at 0.3 hPa, while the difference between the observations and SD-WACCM remains below 3 m s⁻¹ at all altitudes. In both panels the transition from the radiometer to the meteor radar wind profile is almost perfectly smooth justifying some trust in the radiometer observations. The difference pattern for February–March 2017 is obviously related to what has been observed during the lidar intercomparisons in Fig. 7. When focusing on this period in the time series in Fig. 10 it appears that the two episodes of strong northward wind at the beginning and end of February extend to much higher altitudes in ECMWF than in WIRA data. A similar but reduced tendency is visible for MERRA2. However, the meteor radar observations during these periods seem to confirm the lower velocity meridional winds seen by the radiometer at high altitudes by also showing low adjacent velocities above 0.02 hPa.

In addition to the previously discussed long-term comparisons, interesting short-term events can be distinguished from the time series. One example shall briefly be discussed here: on 15 and 16 January 2017 a reversal of the zonal wind direction in the mesosphere is clearly visible in the microwave radiometer observations in Fig. 9 while the stratospheric winds remain at high eastward velocities. It appears that the transition from the WIRA to the meteor radar profiles is smooth also on these days with the meteor radar winds at the lowest levels also being reverted to westward direction. Moreover, a lidar observation in the night from 14 to 15 January corresponded very closely to the radiometer profile (see Fig. S3, second line/second column). Thus, there are good reasons to think of this inversion as a true atmospheric feature rather than a measurement artefact. MERRA2 and the geostrophic wind calculations from the MLS GPH feature a similar pattern as WIRA with a clear reversal of the zonal wind direction in the mesosphere. This change is not captured to its full extent by ECMWF's forecasts and SD-WACCM, which only show a reduction in mesospheric wind speeds. Hence the feature is present in all data sets that either are direct observations or use observations from mesospheric altitudes (MLS) as base for the calculations or as assimilation data. The fact that it is not seen in ECMWF, which only assimilates a few infrared temperature data, and SD-WACCM, which is completely free-running in the mesosphere, may be interpreted as an indication that this effect is not captured by the model physics and solely exists when real observations are considered.

ERA5 as well as forecasts and operational analyses from ECMWF are available at https://www.ecmwf.int/en/forecasts/datasets (ECMWF, 2017b), while MERRA2 reanalysis data can be obtained from https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/data_access (NASA, 2018). The temperature profiles from MLS used for the calculation of the geostrophic wind field are available at https://mls.jpl.nasa.gov (NASA, 2017). The lidar, wind radiometer and meteor radar observations as well as the computations of geostrophic wind and the SD-WACCM simulations can be made available upon request.

The supplement related to this article is available online at: https://doi.org/10.5194/amt-11-1971-2018-supplement.

The authors declare that they have no conflict of interest.