The core unit of a VAHCOLI 1 comprises all the hardware necessary for fully autonomous measurements in a rugged, climate-controlled enclosure with a volume of roughly 1 m3. Figure  illustrates how this enclosure is built in modular layers that are 3D printed in-house, using large-scale FFF (fused-filament fabrication) printers. Likewise, the majority of the internal structural components are printed in-house. Layer 1 contains housekeeping hardware, power supplies for the system, a battery to buffer short power grid issues, and a custom industrial computer, which controls the whole system with three high-performance measurement cards. Layer 2 contains parts of the optical set-up, like the seeding laser, the diode pump, and a fibre resonator for reference, as well as the compact detection bench with its two avalanche photodiode (APD) detectors and stabilised filters. Layer 3 houses the emission part, with the biggest part being the diode-pumped Alexandrite ring laser, which was specially designed in close collaboration with the Fraunhofer Institute for Laser Technology in Aachen . The telescope cylinder, for the vertically facing 50 cm mirror telescope, passes through all layers.

Based on this hardware, the lidar can do high-resolution spectroscopy of the backscattered signal. For this reason, the frequency of the power laser is quickly and precisely scanned over two highly stabilised filters, a broadband solid etalon with a full width at half maximum (FWHM) close to the Doppler width of the atmospheric molecular line, and a narrowband confocal etalon (FWHM 7.5 MHz). Of the two APDs, one records the signal, which is transmitted through the planar and confocal etalon (APD1), and the other one records the signal, which is transmitted through the planar etalon but is reflected on the confocal etalon (APD2). The resulting spectra contain a mixture of Mie and Rayleigh backscatter on APD2 and mostly Mie backscatter on APD1, with a little bit (roughly 1/60th) of Rayleigh backscatter as background. We derive atmospheric properties like wind and aerosol content from these measured spectra. A more detailed description of the core unit and the measurement technique can be found in , , and .

In order to expand the capabilities of the core unit to measurements of horizontal winds, the system was upgraded with multiple fields of view (MFOV upgrade). While other Doppler lidars that measure at stratospheric altitudes use one telescope for each wind component, VAHCOLI 1 has four tilted telescopes that allow it to measure each horizontal component with two opposing fields of view . This allows the direct observation of gradients in the measured atmospheric parameters by the comparison of two fields of view.

The general optical set-up of the MFOV upgrade can be divided into a two-axes galvanometer scanner (inside the core unit) and an L-shaped enclosure holding the four additional telescopes, which is mounted to the walls of the core unit. In the core unit of VAHCOLI 1, the galvanometer scanner enables rapid switching between the five fields of view in shorter than 1 ms by rapidly rotating its two mirrors to preprogrammed positions.

The following specifications were met in the design of the four additional telescopes: (1) compact and relatively light design to keep the dimensions of the upgrade small and forces on the core unit low, (2) reasonable large aperture to measure up to the stratosphere, and (3) a cost-effective solution to allow for network construction as envisaged in VAHCOLI. Since no available telescopes on the market met these requirements, we developed them to the specifications we needed. We designed diffraction-limited lens telescopes with a clear aperture of the front lens of 200 mm, an overall length smaller than 600 mm, and a weight lighter than 5 kg. We use commercial off-the-shelf and 3D-printed components, allowing for fast and cost-effective integration of the MFOV upgrade. A challenge we faced in the telescope design was caused by the requirement to operate the system during night and day. This presented the risk of the telescopes pointing directly at the Sun, resulting in considerable thermal load. To overcome this obstacle, we implemented a robust cooling system and carefully selected materials that ensures that the telescopes are resilient in such circumstances. Four of these telescopes are held in the 3D-printed enclosure of the MFOV upgrade at a 30° off-zenith angle facing north, south, east, and west. The pointing precision of the telescopes is limited only by the precision of the 3D printer, since the enclosure and telescope mount are printed as one big part. From the dimensional accuracy and surface quality our 3D printers produce, we estimate the angular error introduced by the 3D-printed construction to be less than 2 mrad. A tilt sensor inside the core unit measures the level of the whole unit with a precision better than 0.1°, and an integrated camera can be used to additionally verify the instrument orientation using the stars with a precision better than 0.2°.

The MFOV upgrade drastically increases the capabilities, while the hardware upgrades were manageable. The high-speed two-axes scanner and the four extra fields of view, however, required significant upgrades to the control software.

To acquire the measurements presented in this publication, the VAHCOLI 1 instrument was placed outside on a paved and fenced-in area, with markings on the ground indicating true north direction. No further preparations of the site were required. The orientation of the system in the true north direction was verified by targeting Polaris. During the measurement, the local weather conditions were monitored by a weather station situated roughly 50 m away from the instrument. Over the course of the measurement, the air pressure increased from 1000 to 1024 hPa, the ground temperature was between -2 and -8 °C, and the relative humidity was high, with values ranging between 75 % and 100 %.

At the beginning of the measurement, the signal strength of all four oblique fields of view was about a factor of 3 to 4 less than that of the vertical field of view (see Fig. ). This is expected due to the smaller size of the lens telescopes. Over the course of the measurement, the signal in the meridional fields of view stayed almost constant, whereas the zonal fields of view experienced a sharp loss in signal strength. The signal decreased due to icing and condensation occurring on the two zonal-pointing telescopes, caused by the cold temperatures and high relative humidity. An internal heating element was switched on for de-icing at around 12:00 UT on 17 December, leading to an increase in the signal of both telescopes. The signal in the westward-pointing telescope only recovered to 13 % of its initial value. As a result, our westward zonal winds have a lower top altitude than our eastward zonal winds, and both have a significantly reduced performance with more data gaps compared to the meridional telescopes. During this measurement, the planar etalon discussed in Sect.  was not installed, leading to a slight increase in the daylight background. Despite this, no significant performance drops connected to this daylight background were observed.

Except for the icing problem, the system and the MFOV upgrade performed well, with no major problems or interruptions, even under these harsh winter conditions close to the Baltic Sea. After this initial measurement campaign, the VAHCOLI 1 system remained in operation and, as of June 2023, has accumulated more than 600 h of measurements, of which 85 h were done in MFOV mode.

VAHCOLI is a frequency-scanning lidar, which measures the spectra of the backscattered signal . For the measurements presented in this publication, the Mie channel of the system has been used. Mie scattering provides the best wind measurements up to 25 km, due to its very narrow spectrum, but is limited in the lower stratosphere by the decreasing concentration of aerosols. Typically above 25 km, the aerosol density drastically drops, giving an upper altitude limit for the use of this method.

For data processing the individual pulse raw data, produced by the system with a time resolution of 2 ms and an electronic altitude resolution of 1 m, are first converted to backscatter spectra with an integration time of 2 min and a height resolution of 1 m for each individual field of view. The laser pulse itself has a length of approximately 165 m (full width at half maximum) in the atmosphere. In order to improve the statistics for the analysis, these spectra are then integrated in time and altitude with a floating window to the wanted final resolutions of the data. Unless stated otherwise, an integration window of 2 h and 1000 m is used, which is moved in 2 min and 100 m steps. Winds are derived from the Mie backscatter spectra by fitting a Voigt function to each height channel, with an example of this fit shown in Fig. . The laser's spectral shape was determined to be a Voigt function through lab measurements, which will be detailed in an upcoming publication. The detection filter employed for Mie scattering is a confocal etalon, which has a Lorentzian filter curve. The atmospheric broadening, corresponding to wind fluctuations, is assumed to follow a Gaussian distribution. Hence, the convolution of a Voigt function, a Lorentzian function, and a Gaussian function yields another Voigt function.

From the Doppler shift, Δf, given by fitting a Voigt function to the measured spectra, the line-of-sight wind can be derived by rearranging the Doppler shift equation to vLOS=Δf/(2f0)⋅c with c the speed of light and f0 the fundamental frequency (e.g. the laser frequency). Since we measure both the vertical and four off-zenith line-of-sight winds simultaneously, the vertical component can be immediately removed to derive horizontal wind components. The horizontal winds must be divided by sin⁡(30°) to compensate for the off-zenith tilt. It is important to note that this technique is based on a relative measurement of the Doppler shift. This is done by comparing the measured backscatter spectrum to the reference spectrum within the system, as described by . Due to the design of the instrument as a resonance lidar, the exact laser wavelength of 769.898 nm is known from the usage of potassium saturation spectroscopy as an absolute wavelength reference, even though the Mie–Doppler measurements do not strictly require this precise knowledge of it.

The error estimate for the LOS winds (σΔv) is derived from the error estimate of the Doppler shift (σΔf), using the propagation of uncertainties (σΔv=σΔf⋅c/(2f0)). σΔf is a product of the fit routine, which uses the least square fitting of the spectrum, with the standard deviation of each fit parameter being the square root of the corresponding diagonal element of the covariance matrix . This requires an estimate of the error in the photon counts (n) in each frequency bin of the spectrum. Here we assume a Poisson distribution with a width of σn=n. The precision of the result of the fit benefits from the very narrow spectral shape of the received Mie signal and its high-edge steepness, which facilitates precise determination of the Doppler shift. Together with the suppression of the background Rayleigh signal, the number of needed photons is significantly reduced to achieve acceptable wind errors compared to traditional DWL systems . Figure  shows this connection between the measured spectrum of the Mie scattering, the calculated LOS wind, and the error estimate. A change in the peak position in the measured spectrum results in a change in the wind. Black lines in the background of the spectrum correspond to the presence of cirrus clouds, which significantly increase the signal to noise ratio of the spectrum and thus reduce the calculated error in the measurement. This leads to a double distribution of the wind error visible in the bottom right of Fig. , with one peak around 0.15 m s-1 produced by measurements with cirrus clouds and one peak around 1.15 m s-1 from measurements without cirrus clouds. Figure  shows an example of each of these cases. The clear-air case shows a low signal-to-noise (SNR) ratio of 4.2, resulting in an error estimate of the Doppler shift of 1.6 MHz and thus a wind error of 0.6 m s-1. In the case of a cirrus cloud, the SNR drastically increases by more than an order of magnitude, resulting in an error estimate of the Doppler shift of 0.2 MHz and thus a wind error of less than 0.1 m s-1. In cirrus clouds and other strong aerosol loads, the integration in time and altitude can be significantly reduced while maintaining wind errors below 1 m s-1.

In this work, we utilised the ECMWF winds in two distinct approaches. First, we directly employed the ECMWF winds interpolated to the instrument's location (see Figs. , , and ). Second, to achieve a more comprehensive comparison with the individual fields of view, we interpolated and sampled the ECMWF winds along simulated beams (see Figs. , , and ). This interpolation and sampling process is very important due to the increasing positional difference between the beams at higher measurement altitudes. At a height of 20 km, the separation between two opposite fields of view reaches 23 km, exceeding the ECMWF horizontal resolution of about 9 km. In both cases, an observational filter is applied to match the lidar altitude and time resolution.

The Aeolus data have been reprocessed using the baseline 2B15 and regridded to an altitude resolution of 1 km, as in . Only Rayleigh clear winds are used, due to the scarcity of Mie cloudy winds. It is worth noting here that Aeolus winds and ECMWF winds are not really independent, since Aeolus data were assimilated into the ECMWF model.

The horizontal winds from all four off-zenith lidar directions are shown in Fig. , as well as the corresponding wind time series from the ECMWF Integrated Forecast System (ECMWF IFS) interpolated to the location of the lidar. In the meridional wind, a drastic change during the measurement is observed, which corresponds to small-scale wind variations associated with the edge of a developing high-pressure region, which is visible in weather maps of this time period. While the meridional winds above 5 km are between 10 and 20 m s-1 (northward) at the beginning of the measurement, a wind reversal in the tropopause region is observed, with the winds exceeding -40 m s-1 (southward).

In the zonal winds, the previously discussed signal losses are clearly visible as gaps in the top measurement height. Despite the gaps, it was still possible to retrieve zonal wind up to 15 km​​​​​​​ or higher using the eastward-facing telescope during the majority of the measurement. Here the wind peak in the tropopause region is also visible at the end of the measurement, with winds reaching 40 m s-1.

A first qualitative comparison to the winds from ECMWF IFS shows the same features in zonal and meridional winds with similar amplitudes. A more quantitative comparison can be found in Sect. .

The vertical winds measured are shown in Fig. . Overall, the measured vertical winds show absolute wind speeds of 1 m s-1 or slower, with a clear separation in the behaviour between the troposphere and the stratosphere. While above 11 km the winds fluctuate around zero, presumably due to the passage of gravity waves, below a constant background flow of descending air can be observed. A possible influence of the horizontal winds due to pointing errors will be discussed later in Sect. . When comparing the measured vertical winds with vertical winds from ECMWF, differences are immediately apparent. ECMWF underestimates the vertical wind fluctuations by approximately an order of magnitude, which is especially visible in the stratosphere. Additionally, ECMWF does not show the same clear separation between the troposphere and stratosphere.

The comparison of the mean winds during the measurement with the time-averaged ECMWF data over the 48 h of the measurement (see Fig. ) shows the good agreement of the ECMWF data with the measurement in the meridional direction.

The qualitative comparison presented in Fig.  shows that the wind pattern observed is similar to the ECMWF zonal wind pattern. In the vertical winds, the behaviour discussed in Sect.  is clearly visible, with downward winds in the lower part of the measurements. The comparison of the meridional winds measured along the north and south field of view (see Fig. , top left) shows a systematic difference between the two fields of view, with the winds in the north field of view being larger than in the south field of view. This effect increases with altitude. The column mean of the time-averaged difference is 1.32 m s-1, with a standard deviation of 0.88 m s-1. The same analysis of the interpolated and beam-sampled ECMWF winds shows a similar trend, but much less pronounced (see Fig. , top right), with a column mean of the time-averaged difference of 0.33 m s-1​​​​​​​ and a standard deviation of 0.35 m s-1.

The direct comparison of the time-averaged field of view difference in Fig.  highlights this. In both the lidar measurements and the ECMWF data, differences in similar altitude dependency are visible, but the difference between the north and south field of view is much more pronounced in the lidar measurements.

The comparison of the meridional winds from the lidar measurement and ECMWF along the individual fields of view (see Fig. , bottom) paints a similar picture. Along the north field of view, the meridional winds measured with the lidar are slightly higher on average than the winds in ECMWF but show excellent agreement in general. The column mean of the difference between lidar and ECMWF winds in the north field of view is 0.30 m s-1, with a standard deviation of 0.33 m s-1. Along the south field of view, the lidar-measured winds are on average lower than the ECMWF winds, with a column average of -0.93 m s-1 and a standard deviation of 0.73 m s-1, and differences are more pronounced in the time-resolved heatmap. Especially in the stratosphere, the differences show an altitude dependence associated with the increase in horizontal distance between the fields of view with altitude. Around the tropopause region, significant deviations are also observable.

For an additional comparison to a Doppler wind lidar, we used measurements from the Aeolus satellite. The closest overpass occurred on 16 December 2022 at 16:40:55 UT (over the coordinates 54.615° N, 12.1785° E), approximately 64 km from the VAHCOLI test site. For the comparison, the VAHCOLI measurements were downsampled with an altitude resolution of 1 km by 2 h integration time to match the altitude resolution of Aeolus. The measured wind components were combined and projected to the horizontal line of sight (HLOS) of Aeolus. The distance between the Aeolus position and the location of VAHCOLI 1 has been compensated by a time lag of 96 min, based on the observed drift of wind features through the multiple field of view of the VAHCOLI instrument. The comparison (see Fig. ) shows good agreement between the projection of the Aeolus HLOS measurement and the projection of the VAHCOLI measurements. Using the 16:40:55 UT profile from Aeolus for a statistical analysis of the deviation like before results in a standard deviation of 3.31 m s-1 and a mean deviation of -0.12 m s-1, with the biggest differences occurring in the stratosphere. The average error estimate of 1.3 m s-1 for the projected VAHCOLI wind is a factor of 2.5 lower than the average error estimate of 3.3 m s-1 for the Aeolus HLOS wind.