A model-based study of the dynamics of Arctic low-level jet events for the MOSAiC drift

High-resolution weather forecast and regional climate models (RCMs) are important tools to investigate mesoscale processes in data-sparse regions like the Arctic. While specially designed field experiments yield valuable insight into the occurrence of atmospheric mesoscale features, models can help to better understand the processes producing their 3-dimensional structure and temporal development. One of these features is the low-level jet (LLJ), which is a climatological important phenomenon in the stable boundary layer (SBL) of polar regions.

In situ observations of LLJs in polar regions have been performed using radiosondes, tethersondes, and aircraft. Using radiosondes and tethersondes for a 4-month drift in the Weddell Sea in austral fall and winter (Andreas et al., 2000) find the inertial oscillation as the most frequent process for LLJs. A study with high-resolution tethersonde soundings by Heinemann and Rose (1990) over the Antarctic Ronne ice shelf shows an example of an inertial oscillation LLJ. The classical inertial oscillation LLJ forms due to the diurnal cycle, when the atmospheric boundary layer (ABL) changes to an SBL during the evening. The stable stratification leads to a strong decrease of turbulence inducing a decoupling of the upper layer of the SBL as friction becomes negligible. The resulting imbalance of Coriolis and pressure gradient forces (see Equation S3 in the supplement) causes a supergeostrophic wind by an inertial oscillation (Blackadar, 1957; Thorpe and Guymer, 1977). The classical inertial oscillation requires a geostrophic wind, which is constant with height and in time. The period of the inertial oscillation is about 12 h for the central Arctic and the duration of supergeostrophic winds is limited to about 7 h. There is a turn in wind direction associated with a clockwise rotation of the ageostrophic wind.

Tethersonde soundings of the drifting station “Tara” over the central Arctic Ocean during April to August 2007 were used by Jakobson et al. (2013) to analyze LLJs. They concluded that most LLJs were baroclinic. This means that the wind speed gradient above the jet is partly caused by the gradient of the geostrophic wind (see Equation S4 in the supplement). Case studies of baroclinic LLJs in the marginal ice zone are shown by Guest et al. (2018) using radiosondes, who find that the jets were in quasi-geostrophic balance above the LLJ height and in quasi-frictional balance in the boundary layer below.

Aircraft studies of LLJs are limited to short periods but allow for a high spatial resolution and the measurement of additional quantities such as turbulent fluxes. The turbulence structure in strong katabatic LLJs exceeding 20 m/s over Greenland was investigated by Heinemann (2002). It is shown that the turbulent kinetic energy (TKE) and the magnitude of the turbulent fluxes show a strong decrease with height between the lowest flight level (30 m) and the LLJ height, which was typically between 100 m and 150 m.

Ground-based remote sensing methods for wind profiles enable the measurement of continuous profiles of the ABL with high spatial and temporal resolution and the study of LLJs for longer periods. A study for a full Arctic winter period using Sound Detection And Ranging (SODAR) measurements in the Laptev Sea (Heinemann et al., 2021a) showed that LLJs were present in 23% of all profiles. The mean jet speed was about 7 m/s and the mean height 240 m. They concluded that the main driving mechanism for LLJs was baroclinicity, and no inertial oscillations were found. SODAR measurements in the Siberian Arctic for 3 full years (2017–2020) by Heinemann et al. (2022a) showed LLJs associated with topographic channeling, but the SODAR measurements were mostly limited to wind speeds <12 m/s and captured often only the lower part of the LLJs, because the signal-to-noise ratio for SODAR data is poor for high winds. While the SODAR signal is based on backscattering at temperature inhomogeneities and therefore needs turbulence, measurements by wind lidar need backscattering by aerosols or cloud particles, which can be a problem for the clean polar ABL. Wind lidar observations of an LLJ over the sea ice of the Weddell Sea are presented by Zentek et al. (2018), who find that the LLJ was associated with a synoptic front.

While experimental studies are either point measurements (1D) or 2D/3D in the case of aircraft data, model-based studies can provide full 4D studies of LLJs and a climatology for a larger domain. Guest et al. (2018) used the Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS) model with 5 km resolution for LLJs at the sea ice edge and find that the jets had widths of 250–400 km and extended typically several hundred kilometers parallel to the ice edge. LLJs associated with gap flows were studied in Nares Strait by several model studies. Samelson and Barbour (2008) used the Polar MM5 (Fifth-generation Mesoscale Model) with a 6 km resolution for 2 years, but they focused on the 10 m wind rather than on the associated LLJs. A climatology of LLJs in Nares Strait for winters 1987–2016 was shown by Kohnemann and Heinemann (2021), who used the regional climate model Consortium for Small-scale Model—Climate Limited area Mode (CCLM) with 14 km resolution and found LLJs typically occurring at a height of 100–250 m with core wind speeds of up to 40 m/s in the daily mean. Topographically channeled winds in the Siberian Arctic were simulated with CCLM with 5 km resolution for 3 full years by Heinemann et al. (2022a). They show that LLJs were detected in 37% of all profiles and that 29% of all LLJs had jet speed ≥15 m/s, which occurred mainly during channeling events lasting at least 12 h.

A climatology of LLJs for the wintertime Arctic using reanalysis data with 30 km resolution (Bromwich et al., 2018) for the years 2000–2010 was presented by Tuononen et al. (2015). They found the highest frequency of LLJs being associated with katabatic winds with frequencies exceeding 80%. Baroclinically forced LLJs at the sea ice edge had frequencies around 40%. For the central Arctic fewer (20%) and weaker LLJs are found. In a recent study, López-García et al. (2022) used ERA5 reanalyses (Hersbach et al., 2020) for the years 2000–2010 for the climatology of LLJs over Arctic ocean areas and find a similar pattern for the LLJ distribution as Tuononen et al. (2015). An LLJ climatology based on CCLM simulations is shown by Heinemann and Zentek (2021) for the Antarctic for the years 2002–2016. They find LLJs that are most frequent in the katabatic wind regime over the ice sheet and in barrier wind regions. During winter, katabatic LLJs occur with frequencies of more than 70% in many areas with LLJ heights mostly below 200 m and speeds of typically 10–20 m/s. LLJs over the sea ice cover a broad range of speeds and heights, and frequencies are generally lower than 30%.

In the present article, we use simulations of the regional climate model CCLM (see Section 2.2) with a horizontal resolution of 14 km (Figure 1) as well as profiles from radiosondes and wind lidar data from the MOSAiC experiment during October 2019 to September 2020 for the study of LLJs. During this period, the German research vessel Polarstern (Knust, 2017) drifted with the ice in the central Arctic (Shupe et al., 2022). A study of LLJs from radiosonde data and ERA5 data for the MOSAiC site is shown in López-García et al. (2022). Their approach is based on data of operational radiosonde ascents (every 6 h) and the corresponding ERA5 wind profiles. They find LLJs with a mean annual frequency of occurrence of more than 40% at the MOSAiC site and a typical height at 120–180 m.

Here, we use the 1 h data of CCLM with a higher horizontal and vertical resolution compared to ERA5 and we focus on the dynamics of LLJ events. The data and the CCLM model are described in Section 2, where we also explain the methodology for the detection of LLJ profiles, LLJ events, and their dynamical characterization. In Section 3 we present first detailed case studies, then the statistics of LLJ profiles and events are analyzed, followed by the results for the LLJ characteristics. Section 4 contains the discussion; Section 5 gives a summary and conclusions.

During the MOSAiC experiment, comprehensive measurements of the atmospheric boundary layer were performed during a full year in the Arctic. The German research icebreaker Polarstern was the base of this experiment, and during the first phase of the experiment Polarstern drifted from the northern Laptev Sea with the transpolar drift almost to Svalbard (October 2019 to mid-May 2020; Figure 1). The drift was interrupted due to logistic reasons (with a stay near Longyearbyen, Svalbard, for June 4–8, 2020) and the necessity to relocate the drift position, and new drift phases took place from middle of June to end of July and end of August to end of September 2020 (see Shupe et al., 2022 for details).

An overview of measurements relevant for the present study is given in Table 1. Radiosondes (Maturilli et al., 2022) were launched from the helicopter deck of Polarstern every 6 h operationally and every 3 h during special observation periods (see Heinemann et al., 2023b for details). Radiosonde data are available as initial (level-2) data (Maturilli et al., 2021), which were used, for example, in the LLJ study of López-García et al. (2022) and as level-3 data (Maturilli et al., 2022) post-processed according to the Global Climate Observing System (GCOS) Reference Upper-Air Network (GRUAN) algorithm (Sommer et al., 2023). GRUAN provides uncertainties (which correspond to 2 times the standard deviation) of the radiosonde measurements for each data point individually, which are larger than the values given by the radiosonde manufacturer, particularly for wind speed (typically 3 m/s standard deviation for the lowest 500 m). A second difference between level-2 and level-3 data is the vertical smoothing of the wind data, which is necessary to filter out the pendulum motions of the radiosonde with a period of 15 s (Ingleby et al., 2022). These pendulum motions reduce the vertical resolution of the wind profile to at least 75 m. The level-3 data contain a lot of wind fluctuations (Heinemann et al., 2023b), while level-2 wind data are smoother. Ingleby et al. (2022) stated that the Vaisala processing for the level-2 data uses a similar smoothing as an older version of the GRUAN algorithm described by Dirksen et al. (2014), who use a low-pass filter of 40 s (200 m). The GRUAN algorithm used for the processing of the MOSAiC data (Sommer et al., 2023) used a smoothing with a kernel size of 30 s. In the present article, we generally used level-3 soundings for the comparison with model profiles for case studies, but also level-2 soundings for a comparison of LLJ statistics. For a comparison of LLJ statistics from CCLM and radiosonde data a Gaussian filter with a standard deviation of 50 m was applied to both radiosonde data sets. A further difference between level-2 and level-3 data is that level-2 data use the wind measured by the Polarstern meteorological system (Knust, 2017) at the lowest level.

Wind data from the Galion wind lidar (Brooks, 2023) were used with their native resolution of 5 min. The lidar data include an uncertainty estimate, and only data with a wind speed error less than 2 m/s are used. Wind data from the HALO wind lidar have been processed as in Zentek et al. (2018), and only data with a wind speed error less than 2 m/s are used. While wind profiles from the Galion lidar are available for all drift periods, the HALO wind profiles are available only in October 2019 and from March to September 2020, since the HALO was mainly used for different measurement programs in the other periods. However, HALO wind profiles are also available during transfer periods, where no Galion data are available.

The model used for this study is the nonhydrostatic regional climate model CCLM (Rockel et al., 2008; Steger and Bucchignani, 2020, available at https://clmcom.scrollhelp.site/clm-community). CCLM has been used for several studies of air/sea-ice/ocean interactions and boundary layer processes in polar regions (e.g., Bauer et al., 2013; Gutjahr et al., 2016; Kohnemann and Heinemann, 2021). Verification studies for CCLM for the MOSAiC period have been performed for near-surface quantities (Heinemann et al., 2022b) as well as for vertical profiles (Heinemann et al., 2023b). The latter study compares CCLM simulations with MOSAiC data of radiosondes, wind lidar, and retrievals of integrated water vapor and temperature from microwave radiometers. They find a wind speed bias of 0.3 m/s in the lower 200 m for the comparison of CCLM with radiosondes, and biases of ±0.1 m/s up to 2,000 m. Katabatic LLJs were simulated and evaluated for Greenland by Heinemann (2020), who show that CCLM shows a realistic representation of the daily temperature cycle and of temperature and wind profiles for the LLJs. Other studies of LLJs using CCLM including evaluations of the simulations have been mentioned in the Introduction.

In the present article, CCLM is used with a horizontal resolution of about 14 km for the whole Arctic (C15, Figure 1a) for the whole MOSAiC year (October 2019–September 2020), and for the winter period (November 2019–April 2020) with a sea ice data set including sea ice leads (see below).

Initial and boundary data are taken from ERA5 (Hersbach et al., 2020) with hourly resolution (Table 2). The model is used in a forecast mode (reinitialized daily at 18 UTC, spin-up time of 6 h) for October 2019 to September 2020 (the period of the MOSAiC experiment). No nudging is performed during the run. Model output is available every 1 h. In the vertical, the model extends up to 22 km with 60 vertical levels, 13 levels are below 500 m in order to obtain a high resolution of the boundary layer. The first model level is 5 m above the surface. Sea ice concentration as shown in Figure 1 is taken as daily data from an Advanced Microwave Scanning Radiometer 2 (AMSR2) data set with 6 km resolution (Spreen et al., 2008). The run using AMSR2 data is referred to as C15. In addition, daily SIC and information about sea ice leads are taken from Moderate Resolution Imaging Spectroradiometer (MODIS) data for November 2019 to April 2020 (Heinemann et al., 2022b), which is referred to as the C15MOD0 run. The sea ice lead fraction is computed for CCLM grid points from MODIS data with 1 km resolution (Willmes et al., 2023). The thin ice thickness in leads is computed by thermodynamic growth during the daily initialization of the sea ice model of CCLM. The C15MOD0 run has a smaller bias for the near-surface temperature compared to C15 for the whole winter at the MOSAiC site (Heinemann et al., 2022b). Sea ice thickness is prescribed daily from interpolated Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS) fields (Zhang and Rothrock, 2003). Topography data are taken from Hastings and Dunbar (1999). A full technical documentation of the CCLM model is given in Zentek (2019). Modifications were introduced in CCLM for the stable boundary layer, which improve the simulations of the surface inversion over ice surfaces (Heinemann, 2020; Zentek and Heinemann, 2020). In contrast to the ERA5 model, which uses a fixed sea ice thickness without snow layer, we implemented a 2-layer sea ice model with a variable snow layer in CCLM. We also use a tile approach for sea ice (Gutjahr et al., 2016; Heinemann et al., 2021b), which includes the parameterization of the subgrid-scale ice thickness (thin ice in leads and polynyas), a parameterization of the sea ice form drag (Lüpkes and Gryanik, 2015), and parameterizations for the roughness length of heat according to Andreas (1987). For closed pack ice we choose a value of 2.5 × 10⁻³ for the drag coefficient as proposed by Elvidge et al. (2016), which corresponds to a value for the roughness length of 3.3 mm. For details see Heinemann et al. (2022b).

The radiosonde data from Polarstern were assimilated in ERA5 data, which were used at the boundaries for CCLM. However, the CCLM domain is large enough to allow for developments of synoptic and mesoscale systems which were different from ERA5 (Heinemann et al., 2023b).

The methodology for detecting LLJ profiles follows the method used in Heinemann et al. (2022a) and Heinemann and Zentek (2021) with some modifications. Wind maxima were searched below 1,000 m for each profile. Then it is tested, if the maxima fulfill the LLJ criteria. Generally, the height of the lowest wind maximum detected as LLJ is taken as the jet height (h_j). In the case of multiple maxima, it is checked if other maxima (including the absolute maximum of the wind profile) are present within the vertical width of the lowest jet (see Table 3) so that they actually represent the same LLJ. Then the height of the strongest maximum is taken as jet height. For a few cases this results in jet heights larger than 1,000 m.

There are a lot of different LLJ criteria used in literature, which differ in the choice of an absolute criterion of difference between the wind maximum and a minimum above, a relative criterion of the increase of the wind for the maximum compared to the minima above and below, and the maximum height for the search of the minimum above the jet height. For the absolute criterion, a value of 2 m/s is a generally accepted value. For the relative criterion, an increase of 25% is often used (Tuononen et al., 2015; López-García et al., 2022). Heinemann and Zentek (2021) found a high sensitivity of the number of LLJs on the value of the relative criterion and concluded not to use it. We follow this conclusion and use only the absolute criterion requiring a wind decrease of 2 m/s above and below the jet height, and the decrease should be present for at least 2 model levels. The choice of the maximum height for the search of the wind minimum above the jet height is also critical for the LLJ detection. In order to detect only “well-shaped” LLJs with a clear wind anomaly we limit the search for the wind decrease above the jet to a variable search radius, which is taken as the minimum of 1.5 h_j and (0.2 h_j + 300 m) with a maximum search height of about 1,500 m (actually 1,510 m, since the nearest CCLM level is 1,509 m, see Table 2). This prevents the detection of very broad wind maxima. The methodology described above is applied to each CCLM profile. For each LLJ profile, a number of diagnostic quantities characterizing the LLJ are computed (Table 3). The geostrophic wind and wind speed advection were computed at each level from the pressure gradient and wind components of the 4 surrounding grid points by using centered differences (the period June 4–8, 2020, when Polarstern was near Svalbard, was excluded). The background of the diagnostic quantities is explained in more detail in the supplement.

LLJs over the sea ice in the central Arctic are mesoscale features lasting several hours. An LLJ event is defined as the period where the LLJs are detected in consecutive profiles (gaps of one data point in the hourly data are allowed). Events as defined in this article must have a duration of at least 6 h. In addition, the LLJ heights of consecutive profiles should be consistent. If jumps in height larger than ±300 m occur in consecutive profiles, this is considered to be the end of the event and the possible start of a new event. Some diagnostics of LLJ events are similar as for the LLJ profiles, but generally taking the mean of the quantities for profiles over the duration of the event. In addition, changes prior to the event and during the event (e.g., the rate of the change of wind direction during an event (dd_rot)), the maximum wind and wind speed percentiles are computed.

The events are characterized by using different dynamic quantities (see supplement and Table 4). The baroclinicity factor is the fraction of the wind decrease above the jet height by the geostrophic wind decrease averaged for the event. This is a measure of the geostrophic balance of the wind profile above the LLJ height. From the case studies and the statistics shown below, we selected a threshold of 30% to classify baroclinic events. The supergeostrophy factor quantifies how much the wind speed exceeds the geostrophic wind. The role of the wind speed advection for the formation and evolution of the jet is analyzed in 2 ways. Taking the budget equation for the wind speed (Equation S6 in the supplement), the change of the wind speed v_H with time can be written as:

The terms on the right side are the projections of the vectors of horizontal advection, ageostrophic forcing, and friction on the direction of the horizontal wind. We use these terms for the analysis of the jet core dynamics by taking averages over the vertical width of the jet (Table 3). The influence of the advection to the wind speed change of the jet core is described by the advection factor SF_Adv, which is the ratio of F_Adv and the wind speed change (see Table 4 and Equation S11 in the supplement). SF_Adv is evaluated separately for the period before the maximum of the event (intensification phase) and after the maximum of the event (decaying phase). The final value of SF_Adv is calculated as the maximum of both phases, which describes the potential contribution of the horizontal advection for the event. While the ageostrophic forcing can be computed from the model output, the vertical profiles of the friction vector are not in the model output and the friction F_R can only be computed as the residuum of the other terms.

The change of the wind speed gradient can be described by taking the vertical derivative of Equation 1 as shown in the supplement (Equation S7). The establishment of a negative wind gradient with height is necessary for the LLJ formation. The role of advection for the formation of the jet is studied by calculating the gradient of the advection above the jet level for the last 3 h before the start of the event. The wind gradient advection WG_Adv (Equation S10) describes the rate of the formation of the negative wind gradient above the jet level prior to the jet.

We use several criteria for the detection of potential inertial oscillations. The inertial oscillation is associated with a clockwise rotation of the ageostrophic wind, and thus the wind should turn accordingly. We take a positive rate of 2°/h as criterium for the wind direction change during the event. An LLJ caused by inertial oscillation must be supergeostrophic, and we choose a value of 20% for the maximum supergeostrophy factor of the event. In an idealized study, Thorpe and Guymer (1977) simulate a supergeostrophy of almost 50%. Some observations show even larger values for the supergeostrophy, for example, 80% (Malcher and Kraus, 1983), but generally values around 40% are observed (Baas et al., 2009; Baas et al., 2012). Since the inertial oscillation requires a decoupling of the jet layer and the surface, we introduced an additional criterium. The increase of low-level stability is one candidate, which was computed as the change of the gradient of the potential temperature in the lowest 100 m during the last 3 h prior to the start of the event is computed (chstab). The second candidate was to use the change in friction for the jet core between the start of the LLJ event and 6 h before the start (chfriction). We use the friction criterion with a value of 2.0 m/s/6 h as standard, since this includes also the decoupling not related to surface cooling. Note that the chosen value for the change in friction means that the wind speed of the jet core can accelerate up 2 m/s within 6 h due to the decrease in friction. Since the period of the inertial oscillation is close to 12 h for the latitudes of the MOSAiC drift, only events with a maximum duration of 8 h were considered as potential inertial LLJs. Since a pure inertial oscillation with a period of 12 h gets supergeostrophic for about 7 h, the chosen minimum time of 6 h for an event will likely exclude shorter inertial oscillation LLJs. Thus we tested also, whether potential inertial oscillations occur for events of 4 h duration. The criteria used for the characterization of LLJ events are listed in Table 4.

This dynamical characterization of events is new compared to previous studies. Jakobson et al. (2013) classified LLJs as generated by baroclinicity, inertial oscillations, or gusts. Their baroclinicity criterion is that the geostrophic wind speed is at least 2 m/s smaller at the height of the wind minimum above the jet than at the surface. This is similar to the thermal wind DV_g in Table 3, but we take 2 × h_j instead of the height of the wind minimum, which is generally higher. Jakobson et al. (2013) defined potential inertial oscillations jets as jets that have their core at levels where the gradient Richardson number is in the range of 0.2–0.7. As this classification is only based on a stability criterion, they state that this criterion is not a real proof that the LLJ is generated by inertial oscillation. In the present article, we account for the fact that an inertial oscillation must result in supergeostrophic wind speeds, a rotation of the wind and is associated with a decoupling at the jet level. The criterion of gusts was checked by Jakobson et al. (2013) comparing the adjacent profiles (within 1 h). If an LLJ was detected only in single profile, the LLJ was classified as generated by wind gusts. These LLJs in single profiles are not included in the events as defined in the present study.

A total of about 180 LLJ events were detected at the Polarstern site for the whole MOSAiC year (Section 3.2). Here we present some good examples of LLJ events, where wind lidar data are available. Additional cases are shown in the supplement.

The LLJ event on May 7–8, 2020 developed at the end of May 7 and had the highest intensity in the first half of May 8. Figure 2 shows cross-sections of the lowest 1,000 m for the wind speed and direction from C15 simulations and the Galion wind lidar. The simulations (Figure 2a) show an LLJ at about 300 m height with maximum speeds of about 20 m/s. The simulated wind direction (Figure 2c) shows a vertical shear from north-north-easterly flow at low levels to north-easterly flow at the LLJ height, while the flow is easterly above 400 m before and after the event. The wind lidar data (Figure 2b, d) show the full jet with a few missing data in the jet core and confirm the simulated structures and evolution.

The simulations of cross-sections for the potential temperature (Figure 3a) show a cooling in the lowest 300 m during the development of the LLJ at the end of May 7, but a warming in the same layer during the first half of May 8. This warming is even more pronounced above the LLJ, which is followed by a cooling during the decaying phase of the LLJ. The simulated development of the thermal structure is confirmed by the radiosonde data (Figure 3b), which are available every 6 h. The simulations allow to get more insight into quantities, which are not directly measured. The simulated TKE shows large values below the LLJ core (Figure 3c), which underlines the importance of LLJ for the turbulent transport and mixing. The geostrophic wind computed from pressure gradients of the simulation (Figure 3d) shows that the LLJ developed during a strong increase of the geostrophic wind. The geostrophic wind is largest at the surface and decreases with height, indicating the baroclinic nature of the LLJ.

A comparison of profiles from radiosonde and simulations is shown in Figure 4 for 05 UTC May 8, 2020, that is, for the maximum development of the LLJ. The radiosonde data show a well-mixed layer in the lowest 400 m and stable stratification above. This structure is represented well in the simulation, but the temperature is slightly higher. The wind direction shows the positive shear with height, which is also simulated well and fits to the expected wind direction change due to friction. The simulated wind speed profile shows a broad LLJ with the maximum at about 400 m, while the radiosonde data show the LLJ less clear because of wind fluctuations. The LLJ detection method would detect the LLJ height at the model level of 372 m in the C15 data, but at 600 m in the radiosonde data. The lower wind speeds of the radiosonde in the lowest 100 m are possibly a result of the impact of the ship’s superstructure on the wind field. The geostrophic wind is largest at the surface and decreases with height. The LLJ is subgeostrophic and the geostrophic wind explains only a part of the total wind decrease above the LLJ height. The supergeostrophy factor for this profile is −5%, while the baroclinicity factor is about 40%. This means that 40% of the wind decrease above the jet in this profile is explained by baroclinicity. According to the characteristics of LLJs described in Section 2 the whole event is largely baroclinic with an average baroclinicity factor of about 60%.

The simulations allow for the study of the LLJ event for a larger region around the MOSAiC site. Figure 5a shows the wind field at the LLJ height for a subsection of the model domain for 09 UTC May 8, 2020 (the distance of wind vectors is 70 km). A broad band of high wind speeds extends from the Kara Sea to Greenland, and Polarstern is located near a relative maximum of the wind speed exceeding 20 m/s and also close to a strong baroclinic zone. Figure S1 in the supplement shows the temporal development of the wind field. The wind maximum is advected toward Polarstern from 03 UTC to 09 UTC and merges with the high wind region east of Greenland at 12 UTC. The wind maximum at Polarstern is associated with a pronounced maximum in wind shear below (Figure 5b) and above (Figure 5c). The wind decrease above the jet has values of more than 6 m/s in a narrow band along the baroclinic zone. The LLJ with a wind decrease of more than 2 m/s has a mesoscale size of about 200 km across the baroclinic zone. About half of the wind decrease is caused by the decrease of the geostrophic wind above the jet (Figure 5d). For the latter, Figure S2 shows also the time series from 03 UTC to 12 UTC. The zone of negative vertical wind shear is advected with the front. This example underlines the importance of advection for the LLJ formation.

The LLJ event on March 27–28, 2020, is also shown as an example for the evaluation of CCLM using lidar data in Heinemann et al. (2023b), but here we study this case in more detail with respect to the structure and dynamics as in the case study for May 8, 2020. Figure 6 shows cross-sections for the wind speed and direction from C15MOD0 simulations and the Galion wind lidar. The simulations (Figure 6a) show an LLJ at 200–300 m height. It developed at the beginning of March 27 and reached maximum speeds of more than 20 m/s during the second half of March 27 and in the morning of March 28. It decayed during the afternoon of March 28, but had still maximum speeds around 15 m/s. Only the lower part of the LLJ was captured by the wind lidar (Figure 6b), since it was limited by the need for the presence of backscatter particles and cannot penetrate thick clouds. The wind direction showed a vertical shear from north-north-westerly flow at low levels to northerly flow at the LLJ height, but also a turning above and below the jet particularly on March 28.

The simulations of cross-sections for the potential temperature (Figure 7a) show a well-mixed lower ABL with a strong inversion at 300 m for the first half of March 27. During the development of the LLJ on the second half of March 27 a warming in the lower ABL and a rise of the inversion to about 700 m takes place, which is followed by a cooling of the lowest 400 m during the decay of the LLJ on March 28. The data of the radiosondes (Figure 7b) confirm the simulated development of the thermal structure, but the temporal resolution is too coarse to capture details of the development. The simulated TKE shows very large values of more than 1 m²/s² below the LLJ (Figure 7c). The geostrophic wind computed from pressure gradients of the simulation (Figure 7d) shows that the LLJ develops during a strong increase of the geostrophic wind. The geostrophic wind shows a large decrease with height, indicating the baroclinic nature of the LLJ. The highest geostrophic wind speeds occur during the second half of March 27, when the LLJ is most intense.

The radiosonde profile at about the time of the maximum wind (Figure 8) showed the strong inversion with its base at 300 m capping a well-mixed layer and the LLJ at about 300 m. The simulations show a slightly warmer lower ABL and slightly higher wind speeds. The directional shear between the near-surface level and the LLJ height is simulated well (about 25°), but an offset of the wind direction can be seen. The wind is almost geostrophic above 400 m (see also the wind direction), but at the LLJ height the wind is supergeostrophic. A large part of the wind decrease above the jet is caused by the baroclinic wind decrease. The supergeostrophy factor for this profile is about +15%, and the baroclinicity factor is about 55%. This means that more than 50% of the wind decrease above the jet in this profile is explained by baroclinicity. The whole event has an average baroclinicity factor of about 50%, and the maximum supergeostrophy factor is about 25%.

More insights into the dynamics of this event are obtained by the cross-sections of wind speed forcing terms (Equation 1). The wind speed advection (Figure 7e) shows negative wind speed advection above the jet and slightly positive values at the jet level and below the jet level for the intensification phase of the LLJ. The advection contributes to the wind decrease above the jet level during this phase. At the jet level strong negative advection occurred after the time of the maximum wind speed, which contributes to the decay of the LLJ. The ageostrophic forcing F_ag (Figure 7f) shows large positive values below and at the jet level, which are mainly caused by subgeostrophic winds due to friction and are largely counteracted by the friction forcing (F_R in Equation 1). Above the jet level, the ageostrophic forcing partly compensates the wind speed advection.

The wind field at the LLJ height for a subsection of the model domain for 18 UTC March 27, 2020 (Figure 9a) shows that Polarstern was located at the eastern edge of a broad band of high wind speeds with up to 30 m/s near the north-eastern part of Greenland, but also in a strong baroclinic zone. While the wind shear below the LLJ height is large for the whole band of high winds (Figure 9b), the wind decrease above the jet height has values of up to 10 m/s in a narrow band along the baroclinic zone (Figure 9c). Thus the LLJ is only present in this narrow frontal zone, and more than half of the wind decrease is caused by the decrease of the geostrophic wind above the jet (Figure 9d). Figure S3 in the supplement shows the temporal development of the wind field for 09–18 UTC. The frontal zone near Polarstern is almost stationary on March 27, 2020, but highest winds occurred at 18 UTC. The shear of the geostrophic wind above the LLJ (Figure S4 in the supplement) shows that a strong negative shear is mainly associated with the quasi-stationary front, which explains the long duration of the LLJ at the Polarstern site.

The LLJ event on September 22–23, 2020, was the strongest LLJ of the whole MOSAiC year. Figure 10 shows cross-sections for the wind speed and direction from C15 simulations and radiosonde data. The simulations (Figure 10a) show an LLJ at about 300 m height with a maximum speed of about 35 m/s. The LLJ started in the second half of September 22 and decayed during the first hours of September 23. At the jet level the wind turns by about 60° during the development (Figure 10c). This LLJ is an example of a very broad jet with a relative short duration. Starting at noon on September 22, there are 3-hourly radiosonde ascents available for this event. Since no data of the Galion wind lidar are available for this period, we use data of the HALO wind lidar. The Halo data have a lot of gaps, but they capture the period of the wind maximum. Both data sets confirm the simulations of the wind structure.

The simulations of cross-sections for the thermal structure (Figure 11a) show a surface inversion on September 22 prior to the LLJ development, which changed to a shallow neutral layer below 200 m during the LLJ. The development of the LLJ was associated with a warming in all layers up to 1,500 m, which is also seen in the radiosonde data (Figure 11b). The simulated TKE shows very large values of more than 1.5 m²/s² below the LLJ for a short period (Figure 11c). The geostrophic wind (Figure 11d) shows that the LLJ developed during a strong increase of the geostrophic wind with a large decrease with height. The wind speed advection F_Adv (Figure 11e) shows negative wind speed advection for all levels for the intensification phase of the LLJ, while the ageostrophic forcing F_ag (Figure 11f) shows positive values for all levels for this phase. The sum of F_Adv and F_ag can explain largely the wind speed increase being almost constant with height. At the jet level and above the jet level strong negative advection occurred after the time of the maximum wind speed, which contributed largely to the decay of the LLJ.

The radiosonde profile for 23 UTC September 22, 2020, (Figure 12) showed the well-mixed conditions in the lowest 200 m and a strong inversion between 200 m and 400 m. The comparison with the profile at 20 UTC shows the warming by 3–4°C at all levels. The simulated temperature profile shows good agreement above 400 m, but the warming is delayed in the lower ABL resulting in a stronger inversion compared to the radiosondes. However, the wind structure is simulated well. Above the LLJ height, the simulated wind is very close to the geostrophic wind. The baroclinicity factor for this profile is about 70%. The whole event is classified as baroclinic with an average baroclinicity factor of about 110%.

The wind field at the LLJ height for a subsection of the model domain for 23 UTC September 22, 2020, (Figure 13a) shows that Polarstern was located inside a broad band of very high wind speeds exceeding 30 m/s ahead of a warm front associated with a low with its center near Franz Josef Land. The wind shear below the LLJ height was very large for the Polarstern area (Figure 13b). The wind decrease above the jet height had values of up to 8 m/s in a narrow band along the warm front (Figure 13c), which had not reached the Polarstern position at 23 UTC. The wind decrease at the Polarstern position was only 2 m/s at that time, which was about the same as the geostrophic wind decrease (Figure 13d). Figure S5 in the supplement shows the temporal development of the wind field. The wind maximum is advected toward Polarstern from 15 UTC to 21 UTC and had passed Polarstern at 03 UTC September 23, 2020. The time series of the wind shear above the LLJ height (Figure S6) show how the band of very strong vertical wind shear was advected toward Polarstern and that the LLJ at Polarstern was part of a larger jet system associated with the front. The geostrophic wind shear above the jet height (Figure S7) was strongly negative at the warm front but was variable ahead of the front due to a complex temperature structure near the sea ice edge and over the sea ice (see also Figure 1).

Some more interesting LLJ events are shown in Figures S8–S12 in the supplement. The plots show the comparison of wind speed cross-sections from CCLM with radiosonde and wind lidar data. These cases have been selected with respect to a good coverage with lidar data.

Figure S8 shows an event with an LLJ with more than 20 m/s wind and lasting more than 12 h for October 7–8, 2019. The lidar data generally captured only the lower part of the LLJ, except for the maximum during the first hours of October 8, when the lidar data showed the full LLJ structure. The event occurred prior to the passage of a cold front. The comparison with the geostrophic wind shows that the overall temporal development of the wind field was strongly related to the geostrophic wind, but the wind decrease above the LLJ on October 8 cannot be explained by the baroclinicity, since the geostrophic wind increased with height, which results in a negative baroclinicity factor. Here a large contribution of the wind speed advection is found. The detection algorithm detects the LLJ event starting at the end of October 7, since the most wind maxima on October 7 were too broad and LLJs were only detected in single profiles.

The event on January 31, 2020 (Figure S9) is an example with an almost symmetric jet core. It occurred prior to the passage of a cold front. It was mainly controlled by the geostrophic wind during the intensification, but was supergeostrophic during the decaying phase, which was also shown in the measurements. The lidar and radiosonde data showed the LLJ with a larger vertical extent than the simulations, indicating a too early decay in the simulations. The baroclinicity factor for this event is about 80%.

A very long event occurred in the period February 19–21, 2020 (Figure S10) with a passage of a cold front after the event. In the first half of February 19, an LLJ developed at 400 m under geostrophic forcing, but then an LLJ at 300 m prevailed during the next 24 h. The decrease of the wind speed above the jet maximum cannot be explained with the geostrophic wind. From the mid of February 19 to mid of February 20 the wind minimum above the jet was at 500–600 m, which is also found in the lidar and radiosonde profiles. The baroclinicity factor for this event is only about 10%.

The event on May 13, 2020 (Figure S11) was embedded in a strong wind period associated with a warm front lasting for about 24 h, which was dominated by geostrophic forcing. The LLJ criteria are fulfilled only for the first half of May 13, 2020, where an LLJ with more than 30 m/s is detected at 500 m for 10 h. The baroclinicity factor for this event is about 35%.

The last case is an event that occurred on August 9, 2020, prior to the passage of a cold front (Figure S12). It was also associated with a period of strong geostrophic winds, but the geostrophic wind was constant or even showed an increase above the jet. However, there was strong positive wind speed advection at the jet core superimposed by weaker or even negative wind advection above the jet core. The lidar data showed the LLJ during the first half of August 9, 2020, while the simulations show the maximum speeds some hours later. A remarkable feature is the breakdown of the LLJ around 00 UTC August 10, 2020. Having in mind that the model restarts daily and that data at 23 UTC and 00 UTC are from different runs, there might be a discontinuity at 00 UTC from the model setup. However, the jump in wind speed around 00 UTC was also present in the observations.

The statistics for all and strong LLJ profiles for the whole year from C15 simulations are shown in Figure 14. LLJs were detected in 40% of all profiles. The average jet speed was 13 m/s, the maximum speed was 36 m/s, and the 75%tile 16 m/s. As strong LLJs have the largest impact on the ABL and transport processes, we define strong jets as jets with a speed of at least 15 m/s, which is approximately the fraction of the upper 25% of the jets. Strong LLJs occurred in 13% of the profiles and had a mean speed of 19 m/s (Figure 14b). The highest frequency for the jet height was found at 150–250 m for all LLJs (Figure 14a), while strong LLJs were most frequent at around 300 m (Figure 14a). Note that the height levels correspond to the model levels. The average jet heights were 330 m (all LLJs) and 370 m (strong LLJs), respectively. The static stability (gradient of the potential temperature) below the jet (Figure 14d) shows that stable stratification was present for almost all LLJ profiles. This holds also for strong LLJs, but the stability distribution is shifted to neutral values due to the stronger turbulent mixing. The directional shear between low levels and the jet core (Figure 14c) is in a range of 20–40° for all and strong LLJs, which can be expected due to Ekman dynamics.

The distribution for the supergeostrophy factor for all profiles (Figure 15a) is biased to positive values. Values exceeding 20% are smaller for strong LLJ profiles, and values exceeding 40% are almost missing for strong LLJ profiles. The distributions of the baroclinicity factor (Figure 15b) are similar for all and strong LLJ profiles and biased to the positive side.

The average of all LLJs profiles for the lowest 1,000 m is shown in Figure 16. The mean wind profile (Figure 16a) shows a pronounced maximum at about 250 m (note that the height of the mean wind speed maximum is not the same as the mean LLJ height). The mean directional shear profile (relative to the lowest level) shows the turning of the wind with height due to friction of about 30° (Figure 16b). A mean temperature profile was not computed because of the large seasonal variation of the temperature. Instead, the mean stability is described by the average of the potential temperature anomaly in the lowest 1,500 m for each level and for each profile. The profile of the mean potential temperature anomaly shows the stable stratification for the LLJs with an increase of the potential temperature by 8°C in the lowest 500 m. The mean profile of the geostrophic wind speed is also shown in Figure 16a. The mean geostrophic wind decreases with height, which corresponds to a mean baroclinic forcing for the LLJs, and it is larger than the mean wind above the mean LLJ height. The mean scaled wind profile (height scaled with the LLJ height and wind scaled with the LLJ speed) is shown in Figure 16d. In contrast to Figure 16a, the nose-like shape of the wind profile is present and the variability given by the 25%- and 75%-tiles is relatively small.

The mean profiles for strong LLJs for the lowest 1,000 m are shown in Figure 17. The mean wind maximum for strong LLJs was at about 300 m (Figure 17a), but the wind anomaly was less pronounced than for the mean of all LLJs. This was a result of the variability in LLJ height. Wind speeds above the LLJ height were much larger than for all LLJs, which indicates the much stronger synoptic forcing. The jet structure was clearer for the mean scaled wind profile (Figure 17b), which is similar to the scaled wind profile of all LLJs (Figure 16d). The profile of the mean potential temperature anomaly (Figure 17c) is similar to the profile for all LLJs but shows a weaker stable stratification in the lowest 100 m due to the increased turbulent mixing. This is illustrated by the profiles for the TKE (Figure 17d), which shows that the TKE for strong LLJs is about twice as large as for all LLJs. The TKE above the LLJ height is much lower than below. The profile for the TKE for situations without LLJs shows that the TKE is only half compared to all LLJs. The importance of LLJs for establishing a continuous turbulent state of the surface layer was recently shown by Liu et al. (2023) for the MOSAiC measurements of turbulent fluxes.

In order to evaluate the LLJ statistics from CCLM we extracted the CCLM profiles at the times of the radiosonde ascents (Table S1 in the supplementary material). The statistics of this subset are close to the results using the full 1 h data set with LLJs detected in 40% of the profiles and strong LLJs detected in 13% of the profiles. Using level-2 and level-3 radiosonde data smoothed by Gaussian filter with a standard deviation of 50 m yields 45%–46% LLJ profiles for both radiosonde data sets. Using level-2 data without smoothing results in an increase to 54%. The differences between different data sets get smaller, if weak LLJs are not considered. While the statistics for the jet speed of CCLM and the smoothed radiosoundings are in good agreement, CCLM seems to underestimate the LLJ height by about 100 m. This is partly due to the fact that 6% of the LLJs detected in the CCLM profiles are below 100 m height, while this is the case in less than 1% of the radiosonde profiles.

The statistics for all LLJ events for the whole year from C15 simulations (Figure 18) show the averages for each event. About 180 events were detected with a mean duration of 13 h (Figure 18c), but some events lasted 36 h and more. Note that the event classification already demands a minimum of 6 h. The accumulated duration of all events is 26% of the whole period, which is less than the frequency of LLJ profiles (40%). This means that 14% of the hourly LLJ profiles are not part of events. Strong events (maximum jet speed of the event at least 15 m/s) have a slightly longer duration. The accumulated duration of strong events is 14% of the whole period, which is about the same as the frequency of strong LLJ profiles. This means that almost all strong LLJ profiles are part of events. The distributions for the jet speed (Figure 18b) and static stability (Figure 18d) are similar to the statistics of LLJ profiles. The height distribution shows relatively less values for heights above 400 m compared to the LLJ profiles, which means that the LLJ profiles not included in events are mainly at higher levels.

Quantities used for the characterization of events are shown in Figures 19 and 20. The distribution for the supergeostrophy factor for all events has the maximum at +10%, and the distribution is slightly biased to positive values (Figure 19a). Values exceeding 30% are missing for strong events, that is, larger values of supergeostrophy are associated with weaker LLJ events. The baroclinicity factor shows a wide range of negative and positive values (Figure 19b) but is biased to the positive side for all and strong events. The wind decrease above the jet height is fully caused by baroclinity only for a minority of events. Apart from supergeostrophy, the rate of wind direction change during the event (Figure 19d) is one criterion for the presence of an inertial oscillation. The distributions for all and strong events are slightly biased toward a positive rate, which would also be expected for the passage of fronts. A large difference can be seen between all and strong events for the TKE below the LLJ (Figure 19c). While the frequency of TKE is highest for values less than 0.2 m²/s² for all events, strong events show the highest frequencies in the range 0.2–0.4 m²/s².

One precondition for inertial oscillations is a decoupling from the friction at the surface at the LLJ level before the LLJ event. For mid-latitude inertial oscillation LLJs the stabilization of the lower ABL is the typical process for the decoupling. Figure 20b shows the change of the gradient of the potential temperature in the lowest 100 m during the last 3 h prior to the start of the event. For most events, there is no change of the stability prior to the event, and stabilization and destabilization occur with about the same frequency. The advection of the wind speed gradient above the jet during the last 3 h prior to the start of the event is shown in Figure 20a. Since positive values of the wind gradient advection correspond to the formation of the jet structure, a contribution to the required wind speed anomaly of 2 m/s can be seen for many events, particularly for strong LLJ events. The wind speed advection of the jet core (vertical average over the jet width) for the intensification and decaying phases is shown in Figure 20c. For the majority of the events the advection is in the sense of the wind speed change, that is, positive for the intensification phase and negative for the decaying phase, which corresponds to a positive advection factor (SF_Adv). The overall SF_Adv for an event is calculated as the maximum of both phases.

The absolute frequency of LLJ events and LLJ profiles for different months is shown in Figure 21. For all events, the frequency was between 10 and 19 events per month, with less events occurring during summer. C15 and C15MOD0 yield similar results for the winter months. For strong events, only 2–3 events occurred in July/August, and highest frequencies of 8–11 per month were found in the period February to April. The seasonal distribution of all LLJ profiles is similar to the distribution for events, but March 2020 had the most LLJ profiles. This is even more pronounced for strong LLJ profiles. It should be noted that the LLJ of different months represent different regions due to the movement of the ship.

The result of the characteristics of LLJ events for the C15 run according to Table 4 is shown in Figure 22. For the default criteria shown in Table 4, the baroclinic type has a frequency of about 35% for all and strong LLJ events (Figure 22a). The statistics shown in Figure 22a are related to the wind gradient above the jet, that is, how the wind decrease is caused by the decrease of the geostrophic wind with height (at least 30%), by the vertical gradient of the wind advection before the start of the event (at least 1 m/s in 3 h), and by the potential contribution of the wind speed advection to the evolution of the jet core (at least 50%). Given the criteria of Table 4, there are only 1% of LLJ events potentially caused by inertial oscillation and no case is found for strong events (see discussion below). The criterium for wind gradient advection is fulfilled by about 40% and almost 50% of all and strong LLJ events, respectively. A contribution of the wind speed advection by at least 50% to the evolution of the jet core is found for about 50% of all and strong LLJ events. Since the characteristics of LLJ events are evaluated independently for each criterium, the events can be regarded as a combination of different types of forcing. Figure 22b shows the statistics for sensitivity to variations of thresholds. The fraction of baroclinic events decreases to about 25% if the threshold for the baroclinicity factor is taken as 50%. The change of the vertical wind gradient by advection 3 h before the start of the event exceeds 2 m/s for about 20% of the events. In about 30% of the events the wind speed advection is equal or larger than the wind speed change of the jet core.

In contrast to studies of LLJs using reanalyses (e.g., Tuononen et al., 2015; López-García et al., 2022) we use CCLM model forecasts of up to 30 h. The CCLM model has a higher horizontal and vertical resolution than ERA5 (lowest CCLM level at 5 m). CCLM has the capability to simulate the wind structure in the ABL with high accuracy (Heinemann et al., 2023b), which is also shown in the comparison to measurements for LLJ case studies. LLJs are generally mesoscale features lasting several hours and cover large areas. This holds for LLJs due to inertial oscillation (Bonner et al., 1968) as well as for baroclinic LLJs (Guest et al., 2018; Heinemann and Zentek, 2021). Therefore, we defined an LLJ event as an LLJ lasting at least 6 h and with a consistency of the LLJ heights of 1-hourly consecutive profiles. This excludes LLJs found only in single profiles or with large jumps in height in consecutive profiles, which is similar to the criterion of gusts used by Jakobson et al. (2013). In the present study, LLJs are detected in 40% of the hourly profiles, but 14% of the hourly profiles are not part of events. The fraction of LLJs detected only in single profiles is very small for strong LLJs.

The concept of LLJ events allows for a better characterization of LLJs based on dynamical criteria, since criteria based on the changes during or prior the event can be used. Here we propose a new method to characterize LLJ events in terms of baroclinicity, inertial oscillations, and advection. While Jakobson et al. (2013) use only a stability criterion to identify inertial oscillations jets, we require that the jet develops after a stabilization of the ABL in the lowest 100 m, reaches a supergeostrophic wind speed, and shows a clockwise rotation (see Table 4). Ideally, baroclinic LLJs should be subgeostrophic and the wind decrease above the jet height should result from the decrease of the geostrophic wind. The latter is described by a baroclinicity factor, which is required to be larger than 30%. Jakobson et al. (2013) use as baroclinicity criterion that the geostrophic wind speed is at least 2 m/s smaller at the height of the wind minimum above the jet than at the surface. Since the geostrophic wind speed decreases also between the surface and the jet height, this criterion of Jakobson et al. (2013) is not necessarily relevant for the wind decrease above the jet height. The influence of advection on LLJs is quantified in two ways. The role of advection for the generation of LLJs is evaluated by the wind gradient advection above the jet before the start of the LLJ event. The impact of advection on the evolution of the LLJ core is evaluated by the contribution of the advection to the wind speed change in the core separately for the periods before and after the wind maximum of the event.

The LLJs at the MOSAiC site were mainly associated with advection and baroclinicity. No typical inertial oscillation jets were found: around 10 LLJ events were supergeostrophic (>20%) and showed a positive rotation, but only 3 occurred after a stabilization of the lower ABL. A close inspection of these cases shows that the changing geostrophic wind was dominating the wind change. Since a passage of a front is also associated with a positive rotation, these events may be misclassified as inertial oscillation if no other criteria are used. As stated in Section 2, the chosen minimum time of 6 h for an event will likely exclude shorter inertial oscillation LLJs. If we choose 4 h as minimum of LLJ event duration, we find more and shorter events, but still very few potential inertial oscillation jets. Table S2 in the supplementary material shows different criteria and combinations of criteria for the detection of potential oscillation jets. A detailed discussion is given in the text to Table S2. In summary, only 1% of all LLJ events with at least 4 h duration are potential inertial oscillation events. Maybe there are inertial oscillations, but they do not fulfill the LLJ criteria.

The absence of the classical inertial oscillation LLJs is not surprising. The classical inertial oscillation occurs due to the diurnal cycle of solar heating, which is not present during winter in the High Arctic. During summer the daily cycle of the near-surface temperature is also very small over the melting sea ice in the Arctic. A weak diurnal solar cycle in the High Arctic is only present in spring and autumn (Persson et al., 2002). During autumn Polarstern was close to the North Pole where no diurnal cycle can be expected (Figure 1b). For spring 2020, there is a clear diurnal solar cycle, but a distinct daily temperature cycle near the surface is only present with about ±1.5°C for a few cloudless days (Heinemann et al., 2022b). Daily variations of the near-surface temperature are generally much larger but are caused by clouds and synoptic weather systems. A decoupling of the upper layer of the SBL during the polar night could also result from warm air advection or by changes from cloudy to clear conditions (Heinemann et al., 2021a), but other conditions for inertial oscillations (geostrophic wind/pressure gradient constant with height and time) seem not to occur for the MOSAiC at the Polarstern position.

The characteristics of LLJ events depend on the criteria used (Figure 22). Purely baroclinic jets as shown by Guest et al. (2018) for LLJs at the sea ice edge are found only for a few cases (see case study for September 22–23, 2020). In about 35% of the events at least 30% of wind decrease above the jet is balanced by the geostrophic wind gradient. In about 40% of the events the wind gradient advection above the jet plays a large role in the LLJ formation. The dynamics of the jet core is strongly influenced by the wind speed advection. For 50% of the LLJ events the advection explains 50% of the wind speed change for either before or after the wind maximum of the event. Overall, from the statistics and the detailed case studies, the large importance of advection is demonstrated. The ageostrophic forcing for the wind gradient above the jet and for the wind speed of the jet core including changes in the pressure gradient is also important. The quantification of the ageostrophic forcing on LLJ formation is difficult, since it is largely counteracted by friction particularly below the jet level and at the jet core. Data for the friction vector profiles were not available as model output, which would be needed to quantify the net impact of these 2 forcings.

The detection of LLJs depends on the definition of the maximum jet height and on the search height for the minimum above the jet, which is variable depending on the jet height in present article in order to detect LLJs with a well-defined wind speed anomaly. If the maximum jet height in our study is increased to 1,500 m (with a maximum search height at 2,300 m), only about 1% more LLJ profiles are found. When the variable search height is not used but a fixed search height with a value of 1,500 m is taken, the fraction of detected LLJ profiles increases to almost 60%, but only 2% more strong LLJ profiles are detected (see Figure S13 in the supplement). Thirteen more events and only 3 more strong events are detected. More LLJs are found at lower levels associated with a very weak wind decrease with height. A further extension of the search height as done in López-García et al. (2022), who used 4,000 m for the search limit, would probably yield even more LLJ profiles, but these would be also LLJs with a very weak wind gradient above the maximum and without a nose-like shape of the wind anomaly. López-García et al. (2022) show also that the number of the detected jets using ERA5 data depends strongly on the search height. They find LLJs in 54% and 49% of all Polarstern radiosonde (level-2 data) and ERA5 profiles, respectively. Their LLJ frequency from radiosondes is the same as in our study, if we use the level-2 data. We have shown that the use of smoothed radiosonde data reduces the LLJ frequency by about 10%, since mainly weak LLJs are filtered out. The differences of their ERA5 results to the present study for the number LLJ profiles can be partly explained with the different search method. The study of Tuononen et al. (2015) using Arctic System Reanalysis (ASR) data (Bromwich et al., 2018) for the years 2000–2010 found LLJ frequencies of around 40% and 20% at the sea ice edge and for the central Arctic, respectively. However, the ASR data had a very coarse vertical resolution compared to CCLM and may have underestimated the LLJ frequency. Lower values for the study of Tuononen et al. (2015) may also result from the fact that the number and intensity of cyclones were anomalously large for winter and spring during the MOSAiC year (Rinke et al., 2021), leading to more advective and baroclinic LLJs. However, the LLJ frequency of 40% found in the present study over sea ice is much smaller than katabatic LLJs over polar ice sheets, which occur with frequencies of more than 70% during winter (Tuononen et al., 2015; Heinemann and Zentek, 2021).

The LLJs detected only in a single profile can result from turbulence-generated wind maxima, particularly for observational data. The use of single profiles for LLJ detection yields an overestimation of the frequency of LLJ events, since only 65% of the LLJ profiles can be attributed to LLJ events in the present study, which have a fraction of 26% of the whole period.

An analysis of LLJ based on CCLM simulations for the whole MOSAiC year was performed. LLJs were detected using hourly model output. We define LLJ events as LLJs that last at least 6 h, since we regard LLJs as mesoscale features which should occur several hours in order to have an impact in terms of transport processes and on the boundary layer in terms of increase in vertical mixing. This requires that the LLJ is present at the same level for several hours. Detailed case studies of LLJ events are shown using wind lidar and radiosonde data as well as CCLM simulations. CCLM simulations were used to study the statistics of LLJs and LLJ events as well as the dynamics. A method to identify the dynamic characteristics for development of LLJ events is presented.

The main conclusions from this study are:

• Instead of single LLJ profiles the method of detecting events seems to be more reliable to detect LLJ cases in a more consistent way. The LLJs detected only in a single profile can result from turbulence-generated wind maxima, particularly for observational data. In our study, the use of single LLJ profiles yields an overestimation of the LLJ frequency compared to the time fraction of LLJ events. This overestimation is not present for strong LLJs.

• A new method is presented to characterize the physical background of the formation of LLJs. Most LLJ events during MOSAiC are influenced by baroclinicity and by advection. Inertial oscillation is not important for the LLJs at the MOSAiC site. The formation of LLJs is often associated with frontal zones.

• LLJs are generally embedded in larger jet regions extending for several hundreds of kilometers that are advected toward the MOSAiC site.

• A pronounced seasonal cycle is found for strong LLJs with a minimum during the summer months.

• The TKE in the lower ABL is about 4 times larger for strong LLJs compared to situations without LLJs, which underlines the impact of LLJs on turbulent processes in the ABL.

• The forecast of LLJs in polar regions can be important for logistic operations, particularly for aircrafts.

Radiosonde data (Maturilli et al., 2022, https://doi.org/10.1594/PANGAEA.943870 and Maturilli et al., 2021, https://doi.org/10.1594/PANGAEA.928656), Galion wind lidar profiles (Brooks, 2023, https://dx.doi.org/10.5285/86d4b9195b40469e920cb56044adb265), and data of the HALO wind lidar (Heinemann et al., 2023a, https://doi.org/10.1594/PANGAEA.962694) are available on MOSAiC data archives. Model data of atmospheric profiles are published on Zenodo (Heinemann, 2023, https://doi.org/10.5281/zenodo.7756964).

The supplemental files for this article can be found as follows:

Heinemann_LLJ_MOSAiC_Supplement_final.pdf.

Data used in this manuscript were produced as part of the international Multidisciplinary drifting Observatory for the Study of the Arctic Climate (MOSAiC) with the tag MOSAiC20192020 and the Project_ID: AWI_PS122_00. The authors thank all those who contributed to MOSAiC and made this endeavor possible (Nixdorf et al., 2021). The Galion lidar was provided by the UK National Centre for Atmospheric Science (NCAS) Atmospheric Measurement and Observing Facility (AMOF). They thank Ian Brooks (University of Leeds) for providing the Galion lidar data and AWI for providing the radiosonde data. They thank Andreas Preußer (AWI) for performing the HALO wind lidar measurements and Clemens Drüe (University of Trier) for processing the data. They thank the whole ATMOS team for maintaining the operations of the wind lidars. Thanks go to the CLM Community and the German Meteorological Service for providing the basic CCLM model. This work used resources of the Deutsches Klimarechenzentrum (DKRZ) granted by its Scientific Steering Committee (WLA) under project ID bb0474. Model data processing was done with Climate Data Operators (CDO) (https://doi.org/10.5281/zenodo.3539275) and using the R software.

This research was funded by the Federal Ministry of Education and Research (BMBF) under grant 03F0887A in the frame of the MOSAiC project “Modelling the impact of sea ice leads on the atmospheric boundary layer during MOSAiC (MISLAM).” The Halo wind lidar of the University of Trier was funded by the German Research Foundation (DFG) under grant INST 246/116-1 FUGG. The publication was funded by the Open Access Fund of the University of Trier and the German Research Foundation (DFG) within the Open Access Publishing funding program.

The authors have declared that no competing interests exist.

Contributed to conception and design: GH.

Contributed to acquisition of data: LS, RZ.

Contributed to analysis and interpretation of data: GH, RZ.

Drafted and/or revised the article: All authors.

Approved the submitted version for publication: All authors.

How to cite this article: Heinemann, G, Schefczyk, L, Zentek, R. 2024. A model-based study of the dynamics of Arctic low-level jet events for the MOSAiC drift. Elementa: Science of the Anthropocene 12(1). DOI: https://doi.org/10.1525/elementa.2023.00064

Domain Editor-in-Chief: Detlev Helmig, Boulder AIR LLC, Boulder, CO, USA

Knowledge Domain: Atmospheric Science

Part of an Elementa Special Feature: The Multidisciplinary Drifting Observatory for the Study of Arctic Climate (MOSAiC)