Complex bora flow in the lee of Southern Velebit

Transcription

1 Quarterly Journal of the Royal Meteorological Society Q. J. R. Meteorol. Soc. 138: , July 2012 B Complex bora flow in the lee of Southern Velebit Ivana Stiperski, a * Branka Ivančan-Picek, a Vanda Grubišić b and Alica Bajić a a Meteorological and Hydrological Service, Zagreb, Croatia b University of Vienna, Vienna, Austria *Correspondence to: I. Stiperski, Institute of Meteorology und Geophysics, Innrain 52f, 6020 Innsbruck, Austria. ivana.stiperski@uibk.ac.at The complexity of bora flow in the lee of Southern Velebit, Croatia, is investigated by means of high-resolution numerical simulations carried out with the US Naval Research Laboratory s (NRL) Coupled Ocean Atmosphere Mesoscale Prediction System (COAMPS) model. The aim of this study is to identify reasons for the strong spatial variability of bora in the wider Zadar region and the uncharacteristically weak bora in the city of Zadar. The primary focus is put on the severe bora episode of 20 December 2004 during which a sodar (sonic detection and ranging) system was operated at Zemunik airport. Numerical results are verified against the available surface and upper-air observations. Upstream conditions governing dynamics of this particular bora event are characterized by the presence of a mean-state critical level at 3 4 km above mean sea-level (AMSL) defining the upstream bora layer. Low-level airflow over Zadar peninsula during this event displays marked temporal variability. During the morning, a highly three-dimensional and unsteady wake (i.e. region of reduced wind speed) is established over Zadar peninsula, downstream of a hydraulic jump that forms in the lee of, but not collocated with, highest portions of the Southern Velebit terrain. During its most extensive phase the wake scales the entire length of Zadar peninsula in the horizontal and extends up to the critical level in the vertical. A wake vortex develops as the flow reverses along the wake centreline. In the evening hours, an undular bore forms in the same location, with large-amplitude waves on top of it inducing boundary layer separation and rotors. Wake flow is affected both by changes in the upstream atmospheric structure and diurnal boundary-layer evolution. Sensitivity experiments show that the influence of Velebit s height is particularly strong, governing the onset and strength of bora. The terrain of Zadar peninsula exerts a strong influence on the characteristics of the developed bora flow, particularly when non-hydrostatic effects are significant. Boundary layer separation is particularly sensitive to downstream orography. Copyright c 2012 Royal Meteorological Society Key Words: atmospheric rotors; boundary layer separation; critical level; secondary orography; undular bore; wake; wake vortex; wave breaking Received 17 March 2010; Revised 21 December 2011; Accepted 17 January 2011; Published online in Wiley Online Library 6 March 2012 Citation: Stiperski I, Ivančan-Picek B, Grubišić V, Bajić A Complex bora flow in the lee of Southern Velebit. Q. J. R. Meteorol. Soc. 138: DOI: /qj.1901 Copyright c 2012 Royal Meteorological Society

2 Complex Bora Flow in the Lee of Southern Velebit Introduction Bora, a gusty north-easterly downslope windstorm, occurs frequently along the eastern Adriatic coast and is particularly severe in the lee of the Velebit mountain range (Figure 1(a)), the steepest and most pronounced mountain range within the Dinaric Alps in Croatia. Several factors contribute to Velebit being the location of frequent severe bora outbreaks: its orientation perpendicular to the northeasterly winds, particularly steep leeward slopes and several prominent mountain passes. The northern Adriatic severe bora is more frequent than its more complex southern counterpart (e.g., Jurčec and Visković, 1994; Ivančan-Picek and Tutiš, 1996; Horvath et al., 2009) and has been studied more extensively (e.g., Smith, 1987; Klemp and Durran, 1987; Bajić, 1991; Grubišić, 2004; Belušić et al., 2007; Gohm et al., 2008). Insight into the severe bora dynamics in the northern Adriatic region has been greatly facilitated by field campaigns, especially ALPEX (Alpine Experiment; e.g., Smith, 1987; Tutiš and Ivančan-Picek, 1991; Ivančan-Picek and Tutiš, 1995) and MAP (Mesoscale Alpine Programme; e.g., Grubišić, 2004; Volkert et al., 2007). A comprehensive review of recent advances in understanding bora dynamics is given in Grisogono and Belušić (2009). Wave breaking is identified as the primary mechanism for severe bora formation (e.g. Klemp and Durran, 1987; Grubišić, 2004; Jiang and Doyle, 2005; Gohm et al., 2008). A temperature inversion and/or a synoptic-scale critical level upstream, commonly present in the bora profiles (Glasnović and Jurčec, 1990), contribute to wave breaking and thus facilitate the formation of a wave-induced critical level downstream of orography. The near-surface flow becomes dynamically decoupled from the flow aloft, which can lead to a supercritical transition and formation of a hydraulic jump in the lee of higher terrain (Grisogono and Belušić, 2009). This process is primarily hydrostatic (Durran, 1986; Klemp and Durran, 1987) and two-dimensional shallow water theory (Smith, 1985, 1987) describes the basic northern Adriatic bora characteristics well (e.g., Bajić, 1991; Jurčec and Glasnović, 1991; Grubišić, 2004). It fails, however, in accounting for three-dimensional and smaller scale processes such as wakes and gap jets (e.g., Jiang and Doyle, 2005; Gohm and Mayr, 2005; Gohm et al., 2008), atmospheric rotors (e.g., Grubišić andorlić, 2007), potential vorticity (PV) banners (e.g., Grubišić, 2004) and pulsations (Belušić et al., 2007). Non-hydrostatic effects are, generally, of tertiary importance for overall severe bora structure (e.g., Klemp and Durran, 1987; Blockley and Lyons, 1994; Grisogono and Belušić, 2009), even though trapped lee waves have been known to induce severe downslope windstorms (Zängl and Hornsteiner, 2007). The importance of non-hydrostatic trapped lee waves, that can coexist with bora-type hydraulic flows (Gohm et al., 2008), is primarily due to their role in rotor formation (e.g., Doyle and Durran, 2002). Limited observations and numerical simulations have shown the existence of bora-related rotors along the northern Adriatic coast (Belušić et al., 2007; Grubišić and Orlić, 2007; Gohm et al., 2008; Prtenjak and Belušić, 2009; Prtenjak et al., 2010). Although severe bora is typically associated with hydraulic jump-like flow, most of the studies of bora-related rotors to date focus on trapped lee-wave rotors (Type 1 rotors according to Hertenstein and Kuettner, 2005). Belušić et al. (2007) simulate the formation of a lee-wave rotor in connection with the appearance of an upper tropospheric cross-barrier jet stream and the reduction in upstream inversion strength (cf. Vosper, 2004). The lee-wave rotor in Gohm et al. (2008) develops downstream of a mountain gap underneath a trapped lee wave, forming alongside a propagating hydraulic jump. In Prtenjak and Belušić (2009) and Prtenjak et al. (2010) rotor development is associated with a hydraulic jump-like flow feature; however, the turbulenceintensity and the vertical extent of the rotor circulation are not consistent with the characteristics of a hydraulic jump rotor (Type 2 as defined by Hertenstein and Kuettner, 2005). Along-coast variations in bora strength, corresponding to the alternation of gap jets and wakes, are clearly visible in Spaceborne Synthetic Aperture Radar (SAR) images of several bora episodes (Alpers et al., 2009; Signell et al., 2010). Although this is a semi-permanent feature of bora flow, strongly correlated to the upstream orographic characteristics such as mountain gaps and peaks (e.g., Grubišić, 2004; Jiang and Doyle, 2005; Belušić and Klaić, 2006), there appears to be significant variability in the strength of the wake for different bora episodes. In particular, the region in the lee of Southern Velebit is characterized by intense bora winds with significant spatio-temporal variability of wind speed over a relatively small area (e.g., Grubišić, 2004; BelušićandKlaić, 2006). The state of the atmospheric boundary layer exerts a strong influence on rotor characteristics (e.g., Doyle and Durran, 2002; Jiang et al., 2007), as well as bora flow in general (Enger and Grisogono, 1998). Southern Velebit is the highest part of the entire Velebit range. It is separated from Northern Velebit by the deep Oštarijska Vrata pass (Figure 1(b)) and its orientation is perpendicular to the predominant bora direction. In contrast to Northern Velebit, whose cross-section is highly asymmetric, Southern Velebit is generally more symmetric but also narrower, with steep slopes. The highest bora wind gust ever recorded in Croatia (69 m s 1 ) was measured in the lee of the southern tip of Velebit, under the influence of what could be considered the Velebit tip jet (e.g. Doyle and Shapiro, 1999) or Drage jet as termed in Signell et al. (2010). Located only a short distance away (Figure 1(b)), in the lee of one of the highest Velebit peaks (1757 m), lies the area of the city of Zadar (ZD) characterized by considerably weaker bora winds compared with the surrounding area. This phenomenon will be referred to as the Zadar calm. The sheltered nature of Zadar in cases with bora flow emerges in both the standard wind measurements at the Zadar synoptic station (Kraljev et al., 2005; Bajić et al., 2007) as well as from the intercomparison of data from automatic weather stations (AWS) at several locations within the broader Zadar area (Figure 1(b)) operated during a special observations period in Ten-minute-averaged wind speeds reachedupto30ms 1 on the Velebit slopes within the period examined, with gusts up to 54.9 m s 1. At the same time the measurements at Zadar AWS (ZD) were twice as low, with 10 min averaged wind speed maximum equal to 12 m s 1 and a maximum gust of 27.3 m s 1 (Figure 2). The wind roses obtained from AWS measurements show that the strongest winds on the southern Velebit slopes (BA, BO) and at Pag Bridge (PB) correspond to the bora direction (north to northeast), whereas in Zadar the east and southeast directions prevail (Figure 3). The latter is connected with the southerly jugo winds (Adriatic sirocco), channelled by the mountains surrounding the Adriatic Sea. The largest

3 1492 I. Stiperski et al. (a) (b) Figure 1. (a) Terrain height (shaded) of the domain 4 covering Croatia. White marks the sea. Location of Zagreb (ZG), Zadar (ZD) and Velebit are indicated as well as domains 5 and 6. (b) Close up of the terrain height (shaded) of the target area of Zadar peninsula and Southern Velebit (S. Velebit) with locations of automatic weather stations at Barićević (BA), Božići (BO), Obrovac (OB), Pag Bridge (PB), Zadar (ZD) and Zadar Zemunik airport (ZZ). W1 W2 and R1 R2 indicate the location of two vertical cross-sections shown in Figures 12, 15 and 18. Oštarijska Vrata pass (O. Vrata) and Bukovica hill are also indicated. Figure 2. Time series of (a) monthly mean wind speed and (b) maximum wind gust in the given month at the automatic weather stations ZD, PB, BA and BO (for station location see Figure 1(b)) during the period. average wind speed corresponding to bora direction at BA and BO (Figure 3) is a consequence of high frequency of wind speeds exceeding 40 m s 1. Relative frequency of wind speed greater than 10 m s 1 changes from 20% close to the Velebit slopes to 1% at the Zadar station. The difference in wind regimes in Zadar and its surroundings is most obvious during January and February 2003, characterized by frequent and strong bora outbreaks (Figure 2). The area of Zadar city also coincides with a region of low wind-speeds in SAR images of several bora episodes (Alpers et al., 2009; Signell et al., 2010) and also emerges in the mean of all the events studied (Signell et al., 2010). In this study we investigate small-scale characteristics and spatial variability of severe bora in the wider Zadar area, located in the lee of Southern Velebit, by means of numerical modelling. Our aim is to identify the reasons for the Zadar calm and to understand the complex nature of the bora flow in this region. Particular stress is placed on the formation

4 Complex Bora Flow in the Lee of Southern Velebit 1493 results that are further discussed in section 6 in relation to previously published studies. Section 7 provides the summary and conclusions. 2. Synoptic event overview Figure 3. (Left) Relative frequency of wind direction and (right) average wind speed for each wind direction for locations in Figure 1(b) during the special observation period in winter of atmospheric rotors that are of great importance owing to the proximity of the Zadar-Zemunik (ZZ) airport. The primary focus of this study is the severe bora episode on 20 December 2004 (hereafter, 20Dec04), during the period of winter/spring 2004/2005 when a sodar (sonic detection and ranging) system was operated at the airport. Of all the bora episodes that occurred within this period, during 20Dec04 the largest difference between bora gusts at Zadar AWS (7 m s 1 ) and at nearby stations (e.g. Pag Bridge; 35 m s 1 ) was measured. A second bora event on 25 January 2005 (hereafter, 25Jan05) is studied as a verification of the results obtained. Synoptic conditions leading to the development of both bora episodes are described in section 2. Section 3 provides details of the numerical model used, whereas in section 4 observations are presented and model performance is verified against them. Section 5 presents the numerical The 20Dec04 severe bora episode studied in this paper is part of a longer severe bora outbreak that started on the morning of 20 December and lasted until 24 December The episode was associated with a shallow surface lee cyclone (extending up to 850 hpa) in the Gulf of Genoa and the build-up of a high-pressure ridge over the British Isles (Figure 4(a) and (b)). This caused low-level northerly cold air advection and the establishment of a southwest-directed pressure gradient force over the Dinaric Alps, leading to the onset of bora at 03 UTC at Obrovac and 05 UTC in Zadar when the wind direction consistently turned to northeasterly. The maximum measured mean wind speed at the Zadar synoptic meteorological station during 20Dec04 was 7.2 m s 1. At higher levels a strong upper-level trough extended over most of Europe and was responsible for westerly higher level winds over Croatia (Figure 4(a)). The shift in wind direction with height indicates the existence of a mean state critical level that defines the bora layer. During the course of the episode, the central European high-pressure system strengthened and evolved into a low-level anticyclone centred over Germany, whereas the lee cyclone weakened and moved southward along the Mediterranean coast of Italy. With the deepening of the northerly flow the height of the critical level increased. This bora event can be classified as a shallow cyclonic bora (e.g. Defant, 1951; Pandžić and Likso, 2005; Horvath et al., 2008) due to the influence of the Genoa lee cyclone and the presence of a critical level restricting the cross-barrier flow to km AMSL (above mean sea level). The 25Jan05 bora episode (Figure 4(c) and (d)) bears many similarities with the 20Dec04 episode. Developing as a result of a cyclone in the Gulf of Genoa and can also be classified as a shallow cyclonic bora. The mean state critical level was located at km AMSL, separating the northeast low-level flow from the southwesterly flow aloft in connection to a deep upper-level cyclone covering most of Europe. This bora episode falls into a period of prolonged severe bora outbreak lasting from 24 January to 30 January The maximum in the mean wind speed measured during 25Jan05 at Zadar synoptic station was 12 m s Numerical model The three-dimensional real-data numerical simulations of the two bora episodes were carried out using the atmospheric component of the Naval Research Laboratory s (NRL) Coupled Ocean Atmosphere Mesoscale Prediction System (COAMPS; Hodur, 1997). COAMPS is a fully compressible, non-hydrostatic limited area prognostic model with multiple nesting capabilities. In this study a one-way nesting technique is applied on six nested domains, with horizontal resolutions of 81 km, 27 km, 9 km, 3 km, 1 km (domain 5) and 333 m (domain 6), where the coarsest domain (domain 1) covers most of Europe (Figure 4(a)). Domain 5 covers the entire Velebit range whereas the target area (Southern Velebit and Zadar peninsula) is contained within the innermost domain (domain 6) covered by 199 by 202 grid points. Domains 4 to 6 are shown in Figure 1(a). In

5 1494 I. Stiperski et al. (a) (b) (c) (d) Figure 4. The COAMPS analysis for (top) 00 UTC 20 December 2004 and (bottom) 00 UTC 25 January 2005: (a) and (c) geopotential height at 300 hpa in 20 gpm intervals and (b) and (d) horizontal wind vectors and sea-level pressure at 2.5 hpa intervals. Reference wind vectors are shown below each panel. The boxes in (a) indicate domains 2, 3 and 4. the vertical 60 unevenly spaced terrain-following σ z levels are used. The grid spacing is smallest near the surface (10 m) and stretches to 3.75 km at the model top, placed at 34.8 km. A viscous layer is applied in the uppermost 13.3 km. The lowest model level is placed at 15 m, which is where the surface data presented here are taken from. The physical parameterizations used in this study include short- and long-wave radiation (Harshvardhan et al., 1987), surface fluxes (Louis, 1979; Louis et al., 1982), cloud microphysics (Rutledge and Hobbs, 1983) and convection parameterization (Kain and Fritsch, 1990, 1993). The use of convection parameterization in three innermost domains affects the flow in the area of interest to a limited degree; qualitatively the flow regimes are not affected by the use of convection parameterization, with wake development in the first part of the episode and undular bore in the second part. Quantitatively the wind speed in Zadar is on average 6% lower when no convection parameterization is applied. The wake in the first part of the episode is on average 12% stronger with convection parameterization, whereas the undular bore rotors are almost 40% weaker than in the run without convection parameterization. The depth of the undular rotor flow is not affected. Even though moist processes typical for cyclonic bora can generally be taken to be only of secondary importance in cases of severe bora such as the one studied here (Ivatek-Šahdan and Tudor, 2004), the influence on the non-hydrostatic regime appears to be more significant. The subgrid-scale turbulent mixing parameterization is based on a prognostic equation for the turbulent kinetic energy (Mellor and Yamada, 1982) in the vertical and the deformation-k closure scheme for the horizontal mixing (Smagorinsky, 1963) with a stability-dependent mixing length formulated following Thompson and Burk (1991). Fourth order horizontal diffusion and Robert time filtering are applied. The lateral and initial boundary conditions are obtained from the Naval Operational Global Atmospheric Prediction System (NOGAPS) forecast fields updated at a 6 h interval. A 6 h spin-up run for the 20Dec04 episode was initialized at 18 UTC on 19 December For the 25Jan05 episode the spin up run was 12 h long and initialized at 12 UTC on 24 January The subsequent simulation runs for both episodes were initialized at 00 UTC of the respective day and performed for 24 h. Blending of the spin-up COAMPS forecast, NOGAPS analysis and available operational upperair soundings and surface data created the initial fields for these 24 h runs (cf. Grubišić and Billings, 2007). Model orography in the baseline runs is based on the 1 km US Geological Survey terrain dataset (Figure 1). Apart from the baseline simulations, sensitivity experiments were also performed. For the 20Dec04 bora episode the influence of Southern Velebit height (gv) and the Zadar peninsula terrain (nzt) were examined. In the gv experiment Southern Velebit, as a separate part of the Velebit range, was replaced by a gradually falling extension of Northern Velebit (Figure 5(a)). In the nzt experiment the orography of the Zadar peninsula was flattened by setting its height to sea level while retaining the land surface characteristics (Figure 5(b)). For the 25Jan05 event only the gv experiment was conducted.

6 Complex Bora Flow in the Lee of Southern Velebit 1495 (a) (b) Figure 5. Model orography (shaded) within domain 5 for two sensitivity experiments (a) gv and (b) nzt. White denotes sea-level. 4. Measurements and model verification 4.1. Upstream sounding The radio-sounding station in Zagreb (ZG, Figure 1(a)) is the closest operational station upwind of the Southern Velebit target area. It is not immediately clear that the Zagreb sounding is representative of the undisturbed upstream conditions for bora in the lee of Southern Velebit. Backward trajectories from the baseline simulations, initiated at 12 UTC on 20Dec04 from Zadar and going back 12 h in time, are shown in Figure 6. The upstream flow in the layer between 1000 m (taken to be the height of the flow undisturbed by orography upstream of Southern Velebit) and the critical level originates in relative proximity to Zagreb, veering towards the west at higher levels. Comparison of model soundings in this region to the one at Zagreb shows no significant differences, especially in thermodynamic characteristics, confirming the same air mass (not shown). Zagreb radio-sounding data will therefore be utilized in determining the upwind conditions leading to the onset and evolution of bora episodes. The sounding on 20Dec04 at 00 UTC (Figure 7) represents the bora breakthrough environment. It is characterized by a stabily stratified layer extending up to 2500 m and a northnortheast low-level jet with a maximum (14.5 m s 1 ) at 700 m. The low level jet results from flow splitting around the Alps (not shown). The location of the minimum wind speed above the jet coincides with a 160 m deep isothermal layer with a base at 2300 m. A weak (1 K) 100 m deep inversion is located at 3200 m and is close to the height of the mean state critical level. The flow is westerly above. The 12 UTC sounding describes the upstream profile that corresponds to mature bora. A convective mixed layer (CML) had formed and at 1100 m was separated from the free atmosphere by an isothermal layer 100 m deep. Wind speed within the CML had decreased compared with the 00 UTC sounding, whereas the low-level jet maximum was found at the top of the CML. A 160 m deep inversion (1.3 K) was located at 2700 m. Due to the easterly movement of the upper-level trough and the strengthening of the central European ridge, the depth of northeasterly winds and the critical level height had increased to 4400 m AMSL. Figure 6. Horizontal backward trajectories originating from Zadar (ZD) at 12 UTC on 20 December 2004 and going 12 h backwards in time. Solid lines with symbols indicate the elevations at which the trajectories were taken: 1000 m, 1500 m, 2000 m and 2500 m AMSL. Wind vectors are at 400 m level AMSL. Contours correspond to terrain height in 200 m intervals. The model reproduces the general characteristics of the upstream flow well, particularly at upper levels. The inversions in the potential temperature profile are too weak and shallow to be resolved and are consequently seen as layers of increased stability. Wind direction and the location of the critical level are in good agreement with the measurements, particularly at 12 UTC. The differences in wind speed are greater: at 12 UTC the model overestimates the strength of the low level jet while underestimating the height of the jet maximum. These differences can be attributed to the fact that in contrast to vertical profiles in the model, radio-soundings represent a trajectory through space and

7 1496 I. Stiperski et al. Figure 7. Comparison between (black line) observed and (grey line) simulated vertical profiles of (a) and (d) potential temperature, (b) and (e) wind speed and (c) and (f) wind direction at upstream location of Zagreb on 20 December 2004 at 00 UTC (a) (c) and 12 UTC (d) (f). time, where wind speed is locally more variable than other air mass variables. The fact that the inversions are shallow and weak suggests that the mean state critical level plays a more significant role in the bora dynamics than the inversion, therefore the overall agreement between the measurements and the model results can be considered satisfactory, and the usage of the model results in studying the bora dynamics is justified. The upstream profiles at 00 UTC and 12 UTC on 25Jan2005 (not shown) are both characterized by a near constant tropospheric static stability, apart from the development of 700 m deep CML during daytime. No inversions exist in the profile. A persistent low-level jet characterizes the wind profile. The jet maximum is located at 1000 m height and its strength increases over time (15 to 18 m s 1 ) leading to a decrease in flow non-linearity. The mean state critical level is located at m AMSL. The upstream profile changes relatively little over time (not shown) compared with 20Dec04 and is reproduced well by the model Downstream atmospheric structure The orographic influences on bora flow in the far lee of Southern Velebit are depicted in the radio-sounding taken daily at Zemunik airport (ZZ; Figure 1(b)). The 12 UTC sounding shows the thermodynamic structure of a mature bora in the lee of Southern Velebit on 20Dec04 (Figure 8). A very shallow, neutrally stratified CML has evolved over Zemunik airport. The stably stratified bora layer extends up to 1200 m, topped by a 0.6 K 90 m deep inversion. The bora layer is defined well by constant stability and a weak northeasterly low-level jet (10 m s 1 ). Above the bora layer two well-mixed layers with weak winds, separated by a 0.7 K inversion 70 m deep, could indicate wave breaking regions. The 1.5 K inversion 180 m deep at 2500 m corresponds to the 2700 m inversion in the Zagreb sounding and separates the well-mixed layers corresponding to the northerly flow from the stably stratified westerly flow above. The low level jet is confined to a very shallow layer in comparison to the upstream Zagreb sounding. The wind speed is significantly lower than in Zagreb and does not exceed 10 m s 1 in the lower 6 km. Low wind speeds signify that Zemunik airport is not under the direct influence of supercritical bora flow originating from the slopes of Southern Velebit. Differences between the observed and modelled profiles, particularly wind speed, are more significant than in Zagreb. In the model the low-level jet speed is overestimated by as much as 50%, although the near-surface vertical wind shear is resolved well. Westerly flow in the layer between 2500 and 4000 m is significantly underestimated. Inversions in the potential temperature profile are resolved as layers with increased stability. The pronounced spatio-temporal variability of bora in the Zemunik area, as will be seen in section 5.1, could in part account for the differences in the measured and simulated bora strength and direction.

8 Complex Bora Flow in the Lee of Southern Velebit 1497 Figure 8. As in Figure 7 but at Zadar Zemunik on 20 December Automatic weather stations Spatial variability of near-surface flow is documented well by the wind data collected at three automatic weather stations (AWS) within the target area of Southern Velebit (Figure 1(b)): Pag Bridge (PB) located directly in the lee of Southern Velebit under the influence of bora shooting flow, Obrovac (OB) in the lee of southern end of Velebit under the influence of Velebit tip jet, and the city of Zadar (ZD). The data represent 10 min averages (Figure 9). Since both Pag Bridge and Obrovac are located in close proximity to the Velebit slopes and are under its direct influence, measured wind speeds (Figure 9(a) (d)) correspond to severe bora with average sustained wind speed exceeding 19 m s 1 and gusts reaching 35 m s 1 on 20Dec04 and 40 m s 1 on 25Jan05. Conversely, Zadar is located farther downstream and shows distinctly weaker winds (Figure 9(a) and (b)), with gusts below 10 m s 1 on 20Dec04 and on 25Jan05 hardly exceeding 10 m s 1. The measurements at Zadar also show significant wind speed and directional variability compared with the other two AWS. A northeasterly direction prevails at Zadar during the entire 20Dec04 episode, and still there are periods with easterly flow in the morning hours as well as southeasterly flow towards the end of the episode, indicating flow reversal. The model captures the magnitude of bora winds, wind direction and bora onset at Obrovac and Zadar well. At Pag Bridge bora breakthrough occurs earlier than measured and lasts longer while the wind direction is more northeasterly. The reason for this discrepancy could be the micro-location of the Pag Bridge AWS, situated in a very narrow channel that is difficult to resolve correctly by model orography Sodar measurements Sodar measurements at Zemunik airport (cf. Figure 1(b)) allow an insight into the vertical structure and temporal variability of bora in the lee of Southern Velebit (Jeričević et al., 2005). The data obtained between 50 and 300 m AGL (above ground level) will be considered here due to limitations of the sodar system and its vertical reach. The temporal resolution is 10 min. Figure 10 shows the evolution of lower tropospheric winds at Zemunik airport. The flow is, for the most part, characterized by relatively weak winds (barely exceeding 15 m s 1 ) of variable, predominantly northeasterly, direction. Periods of high wind speeds mostly correspond to an easterly direction near the surface and northeasterly aloft, and exchange with periods of lower winds in approximately 6 h intervals (Figure 10(a)). In the period between 12 and 17 UTC there are indications of near-surface flow reversal with southerly winds reaching up to 100 m (Figure 10(b)). During the latter part of the bora episode (after 20 UTC) southeasterly winds prevail within a layer extending up to 220 m. This is in agreement with the surface data from Zadar AWS (Figure 9). The near-surface wind speeds decrease during the afternoon hours of 20Dec04 due to CML evolution that causes dissipation of the downgoing wave by boundary layer turbulence (Grisogono and Belušić, 2009) and general decrease in the wave response (Gohm and Mayr, 2005). Vertical wind speed (Figure 10(c)) exhibits significant temporal variability with the period of 50 min dominating the vertical velocity power spectrum. The motion is predominantly upward in this layer. Strongest upward motions coincide with the minimum in horizontal wind speed. The model captures general flow behaviour at Zemunik airport. The alternation of periods with high and low wind speeds and the vertical wind speed variability with a 50 min period are reproduced well. Still, vertical wind speeds are underestimated in the model due to numerical diffusion, whereas the horizontal wind speed is overestimated. Timing of the wind speed maxima differs and the wind direction is more northerly. These differences can be explained by highly variable flow in this region, similar to the sounding data. 5. Complex bora flow Horizontal cross-sections of near-surface bora flow and flow at 500 m AMSL in the lee of Southern Velebit are shown in Figure 11. Flow in the lee of Southern Velebit is highly three-dimensional. On the upstream side the incoming flow is blocked by high terrain. Over the steep lee slopes near-surface high-speed shooting flow develops. Its downstream extent shows significant cross-flow variability. A pronounced unsteady wake is visible over Zadar peninsula, downstream of the highest part of Southern Velebit terrain. The wake is surrounded by two jets, one emanating from the Oštarijska Vrata pass (see Figure 1(b)) at the northern end of the domain whereas the second is a broad region of high wind speed that constitutes the tip jet emanating from

9 1498 I. Stiperski et al. Figure 9. Time series of (black) measured and (grey) modelled (solid line) 10 min wind speed, (thin line) maximum measured wind speed and (dots) wind direction at (a) and (b) Obrovac (OB), (c) and (d) Pag Bridge (PB) and (e) and (f) Zadar (ZD), for (a), (c) and (e) 20 December 2004 and (b), (d) and (f) 25 January 2005 bora episodes. Modelled data correspond to the innermost domain. Figure 10. Time height diagrams of (a) and (d) wind speed (m s 1 ), (b) and (e) horizontal wind direction (deg) and (c) and (f) vertical wind speed (m s 1 ) derived from (a) (c) the Zemunik airport sodar data and (d) (e) domain 6 in the baseline simulation during 20 December 2004.

10 Complex Bora Flow in the Lee of Southern Velebit 1499 (a) (b) Figure 11. The COAMPS simulated cross-barrier horizontal wind speed (m s 1 ; greyscale) and horizontal wind vectors at: (a) lowest model level and (b) 500 m AMSL within domain 5 of the baseline simulation at 10 UTC on 20 December Orography contours in 250 m intervals (grey) where 20 m contour marks the coastline (black). In (b) white signifies terrain higher than 500 m. the southern end of Velebit (Drage jet in Signell et al., 2010). The Velebit tip jet is more clearly discernable at higher levels (Figure 11(b)) where its structure is less affected by the lower terrain of Zadar peninsula, particularly the 640 m high Bukovica hill south of Obrovac (see Figure 1(b)). A region of reduced wind speed also forms within the valley upstream of Bukovica hill, to which we return in section Zadar wake evolution The 20Dec04 bora episode begins with the breaking of a hydrostatic vertically propagating internal wave over Southern Velebit in the early morning hours of 20Dec04 (not shown). The flow becomes supercritical and in the lee of higher portions of Southern Velebit a hydraulic jump forms (Smith, 1985) accompanied by significant turbulence. Initially the jump propagates downstream away from the Velebit slopes but the location of the flow separation point soon becomes stationary at the northeastern part of the Zadar peninsula (Figure 12(a)). Downstream of the hydraulic jump a wake forms over Zadar peninsula due to dissipative Bernoulli loss within the jump (cf. Pan and Smith, 1999; Grubišić, 2004; Jiang and Doyle, 2005). The wake scales the entire length of the peninsula and islands, beyond which bora strengthens again over the Adriatic (Figures 11(a) and 13(a)) due to higher sea-surface temperature (Enger and Grisogono, 1998). The location of the absolute minimum in the crossbarrier wind speed within the wake does not coincide with the highest peaks in the range, contrary to expectations and results from other wake flows along the Adriatic coast (cf. Grubišić, 2004; Jiang and Doyle, 2005; Gohm et al., 2008). Rather, a jet, separating the wake into two regions (north and south), forms in the lee of the highest, narrowest and steepest part of Southern Velebit (Figure 13(a)). The jet is a result of flow originating from the complex threedimensional concave terrain of these highest peaks, the configuration of which disrupts hydraulic jump formation, whereas a deep local canyon channels the flow (e.g. Reed, 1981). The direction of the jet is at first modulated by the direction of the canyon (N), however, farther downstream the flow changes into the prevalent northeasterly direction and extends far over Zadar peninsula. The wake is strongest and covers the largest region in the period between 9 and 12 UTC. During this time reversed flow (Schär and Smith, 1993; Epifanio and Durran, 2002) develops in the northern wake centreline (Figures 12(a) and 13(a)) reaching up to 1400 m AMSL and exceeding -8 m s 1. Reversed flow reaches almost as far as Zadar where the AWS shows a change in wind direction (Figure 9(a)). A weaker and shallower southern wake region influences the location of Zemunik airport where the 12 UTC sounding shows comparatively weak winds (Figure 8) and near-surface sodar data correspond to southerly winds, at a somewhat later time however (Figure 10(b)). The lesser extent of the southern wake region is due to complex interactions between the hydraulic jump in the lee of Velebit, the southern Velebit tip jet and the complex orography of Zadar peninsula. The wake is dynamically unstable (Schär and Smith, 1993) as seen by upstream propagation of the point of flow separation determining the upwind extent of the wake, and irregular vortex shedding over the sea (not shown). In the afternoon hours the wake weakens and is flushed downstream over the islands by the increasing strength of the shooting flow in connection with weakening of the hydraulic jump (Figure 12(d)). This is instigated by decreasing lower level stability upstream that leads to a decrease in nonlinearity (see Figure 7) and by diurnal planetary boundary layer (PBL) growth over Zadar peninsula (cf. Gohm and Mayr, 2005). Surface friction and low terrain still causes the locations of Zadar and Zemunik airport to be under the influence of winds weaker than those in close proximity to Southern Velebit, as evidenced by both the AWS (Figure 9) and sodar data (Figure 10). The sensitivity experiments confirm that, apart from the upstream thermodynamic structure, the height of Southern Velebit is the primary controlling parameter determining wake characteristics and flow steadiness. In the gv experiment (Figures 12(b,e) and 13(b)) the flow no longer blocked by Southern Velebit crosses the mountain in a deep and broad jet that extends along the entire coastal area in the Southern Velebit lee and is responsible for

11 1500 I. Stiperski et al. (a) (b) (c) (d) (e) (f) Figure 12. Vertical cross-sections of (shaded) cross-barrier wind speed (m s 1 ), (black solid line) potential temperature and vertical wind vectors parallel to the cross-section along W1 W2 (see Figure 1(b)) within the innermost domain of the (a) and (d) baseline, (b) and (e) gv and (c) and (f) ntz simulation at (a) (c) 10 UTC and (b) (d) 15 UTC on 20 December Negative wind speeds indicate reversed flow. high near-surface wind speeds. Both the tip jet and the wake disappear. The high-speed shooting flow in the gv experiment is also significantly steadier than in the baseline simulation (Figure 12(b) and (e)) stressing the importance of non-linearity as the source of unsteadiness. The highspeed flow extends far downstream from the mountains with the near-surface speed higher than in the baseline run (Figure 14) but still almost twice as weak as that closer to the Velebit slopes or at greater heights (cf. Figure 9). This reduction in wind speed is due to surface roughness of the smaller-scale Zadar peninsula terrain, which lifts the high speed flow. Flattening of the Zadar peninsula most significantly influences wake persistence (Figure 12(c) and (f)); the wake in the lee of Southern Velebit is present in the nzt experiment during the entire episode. The wake and reversed flow in the wake centreline are generally more intense than in the baseline simulation due to a stronger hydraulic jump. The point of flow separation from the surface in the hydraulic jump is located farther downstream in the nzt experiment (Figure 13(c)) and the surface flow is more turbulent (not shown). The wind speed at Zadar shows more variability than the baseline simulation (Figure 14), with flow reversal reaching Zadar several times during the afternoon. A wake over Zadar peninsula forms during the morning hours of the 25Jan05 episode as well (Figure 13(d)). In the mode similar to the 20Dec04 episode, the wake is split into two parts by the jet emanating from the canyon. Transitory flow reversal develops in the northern, stronger wake region in the beginning of the episode, close to the point of flow separation from the surface in the hydraulic jump, but is not as persistent as in the 20Dec04 episode Undular bore rotor Towards the end of the 20Dec04 episode (after 20 UTC) the wake over Zadar peninsula starts to be re-established. In contrast to the first part of the episode, low wind speeds at this time are the result of boundary layer separation (cf. Gohm and Mayr, 2005) in an undular bore (Figures 15 and 16). This is a non-hydrostatic phenomenon (Blockley and Lyons, 1994) closely related to wave breaking in environments without an upstream inversion (cf. Figure 17(a)) or to trapped wave formation on top of a hydraulic jump in the presence of an inversion (Jiang et al., 2007). The flow can be visually classified as a mixing jump regime according to Jiang et al. (2007) (their figure 15(d)). Undular bore is unsteady and rotors propagate upstream (Smith and Skyllingstad, 2009). The formation of undular bore is instigated by low level wave breaking that is a result of increasing flow nonlinearity due to decreasing crest level wind speed (Figure 17(b)) and weaker turbulent diffusion in the nocturnal PBL. The depth of the layer with positive vertical wind shear (S > 2.5 m s 1 km 1 ) increases and causes a strong decrease in the Scorer parameter in the lowest 500 m (Figure 17(c)) that supports low-level wave trapping. Wave amplitudes increase over time so that around 20 UTC the boundary layer is able to separate under the wave crests and shallow rotors form at three locations over Zadar peninsula (Figures 15(a) and 16(a)) parallel to the orientation of Southern Velebit. These undular jump rotors show significant differences compared with most other bora rotors studied (Gohm and Mayr, 2005; Belušić et al., 2007; Gohm et al., 2008). There is, however, some resemblance to those examined by

12 Complex Bora Flow in the Lee of Southern Velebit 1501 (a) (b) (c) (d) Figure 13. Simulated surface cross-barrier near-surface horizontal wind speed (m s 1 ; shaded) and wind vectors within the innermost domain of the (a) baseline, (b) gv and (c) nzt simulations at 10 UTC on 20 December 2004 and (d) baseline simulation at 1 UTC on 25 January The 10 m AMSL orography contour (black) marks the coastline. Orography contours (grey) are drawn for 50, 100, 200 m AMSL and subsequently in 200 m intervals. Figure 14. Time evolution of cross-barrier wind speed (U cross )atzadar for baseline, gv and nzt simulations. Prtenjak and Belušić (2009) and Prtenjak et al. (2010); in their study rotors develop in connection with a hydraulic jump but are not as vertically developed as suggested by Hertenstein and Kuettner (2005). In contrast to Prtenjak and Belušić (2009) and Prtenjak et al. (2010) rotors over Zadar peninsula, despite being shallow, are accompanied by significant turbulence (turbulent kinetic energy, TKE, exceeding 15 m 2 s 2 ) in the order of that in the undular bore rotor in Jiang et al. (2007). The longest lasting rotor (near-surface reversed wind speed exceeding 3 ms 1 for over 2 h) is found at the location where the high-speed flow separates from the surface in the hydraulic jump. The strongest rotor, on the other hand, develops underneath the second wave crest at 23 UTC, with surface reversed flow speed exceeding 4ms 1. Both of these rotors extend up to 250 m and are located upstream of Zadar. After 22 UTC, however, a relatively weak and shallow rotor also develops at Zadar. Its existence can be inferred from Zadar AWS (cf. Figure 9) showing a change in wind direction from northeast to south southsoutheasterly around the time of the rotor occurrence in the simulation. The fact that the second rotor is stronger than the first suggests constructive interference of waves riding on top of the undular bore by the low Zadar peninsula orography (cf. Grubišić and Stiperski, 2009; Stiperski and Grubišić, 2011). In the nzt sensitivity test, in which this orography is flat, the hydraulic jump is stronger and the largest amplitude undulation is the one closest to the hydraulic jump (Figures 15(b) and 16(b)). Reverse flow is stronger than in the baseline simulation ( 5ms 1 at 23 UTC) and extends higher (300 m). Wavelength of the undulations is reduced, whereas strong turbulent diffusion within the first rotor causes significant attenuation of secondary undulations and consequently weaker rotors. Reverse flow does not reach Zadar (Figure 16(b)). This undular jump rotor bears more resemblance to that analysed in Prtenjak and Belušić (2009), however, the turbulence (TKE > 13 m 2 s 1 ) again greatly exceeds theirs. In the gv experiment, undular bore develops earlier than in the baseline simulations (around 15 UTC) (Figure 15(c)). Due to lower mountain height the flow in the hydraulic jump is significantly weaker. Waves forming on top of it have longer wavelengths but smaller amplitudes that, due to stronger surface winds, are unable to induce boundary layer

13 1502 I. Stiperski et al. (a) (b) (c) Figure 15. Same as in Figure 12 but at 23 UTC on 20 December 2004 for (a) baseline, (b) nzt and (c) gv simulations. (a) (b) Figure 16. Same as Figure 13 but zoomed in on the Zadar peninsula at 23 UTC on 20 December 2004 for (a) baseline and (b) nzt simulation. City of Zadar (ZD) and Zadar Zemunik airport (ZZ) are indicated. separation. Near-surface turbulence is generally stronger than in the baseline simulation due to strong near-surface vertical wind shear; however, the TKE related to waves on top of the undular bore is much weaker and waves are therefore less attenuated downstream. On 25Jan05 a weak short-lived rotor also develops over Zadar peninsula as a result of an internal hydraulic jump in the beginning of the study period. However, this hydraulic jump rotor is highly transitory and develops without undulations or waves forming on top of it (not shown). All the examined rotors form late in the evening or during the night when surface cooling increases the stratification and strengthens the undulating jet capping the rotors and therefore facilitates boundary layer separation, in agreement with Jiang et al. (2007) and Smith and Skyllingstad (2009) Lee wave rotor Rotor circulations over Zadar peninsula, described in the previous section, are transient in nature. At the southern end of the valley upwind of the Bukovica hill (see Figure 1(b) transect R1 R2) a more persistent rotor circulation develops. This rotor forms under the direct influence of the Velebit tip jet, and is present between 8 and 10 UTC and from 19 UTC to the end of the episode examined. These are the periods when weaker upstream winds act in synergy with a stable and shallow nocturnal PBL to promote boundary layer separation (Doyle and Durran, 2002; Jiang et al., 2007). The periods during which the rotor is absent from the valley are characterized by flow channelling through the valley and coincide with the growth of CML. Vertical cross-sections along the R1 R2 transect (see Figure 1(b)) are presented in Figure 18. A rotor forms in connection with a large amplitude lee wave developing as the high speed shooting flow descends from the southern edge of Velebit only to re-ascend over the valley. The boundary layer separates from the surface close to the bottom of the valley where the high positive horizontal vorticity sheet is advected into the wave crest. The rotor is particularly strong towards the end of the episode; it is stronger and reaches higher than rotors over Zadar peninsula. The strongest near-surface reversed flow speed exceeded -8 m s 1 and the rotor extended up to 800 m AMSL, accompanied by severe turbulence, with TKE exceeding 20 m 2 s 2 at the top of the rotor at 22 UTC. Rotor flow shows significant sensitivity to the height of both the upstream and downstream orography. Reduction in Velebit height in the gv sensitivity experiment (Figure 18(b)) has a complex influence on the lee wave rotor; decreased mountain height reduces non-linearity to a lesser degree than the disappearance of the tip jet (see section 5.1) increases

14 Complex Bora Flow in the Lee of Southern Velebit 1503 Figure 17. Simulated (a) Brunt-Väisälä frequency (N), (b) cross-barrier wind speed (U) and (c) Scorer parameter (l) at upstream locationof Zagreb on 20 December 2004 at times from 12 to 21 UTC. it. The cumulative effect is larger amplitude waves that lead to a stronger rotor but reaching lower than in the baseline run. The rotor is present in the valley throughout most of the episode. The influence of diurnal PBL growth, however, causes strongest rotor circulation during night-time, as in the baseline run. The point of boundary layer separation is also affected by the change in upstream flow. In the initial part of the episode (until 13 UTC) the point of boundary layer separation is located close to valley bottom. In the later part of the episode, however, the boundary layer separation point moves upstream towards the hilltop so that the rotor scales the entire valley bottom (Figure 18(b)). In the nzt run, in which the entire Zadar peninsula orography and therefore also Bukovica hill is removed from the simulation, the boundary layer does not separate from the surface and rotors do not form (Figure 18(c)). Rather, high speed flow continues to sweep far downstream from the mountain. This clearly underlines the importance of downstream orography on boundary layer separation (cf. Tampieri and Hunt, 1985; Grisogono and Belušić, 2009) in a mode contrary to that for trapped waves in Stiperski and Grubišić (2011), according to which downstream orography does not facilitate boundary layer separation in a valley between two ridges where it would not occur for a single ridge. Difference in flow regimes (hydraulic type flow vs. trapped waves) could explain the discrepancy between the results presented here and those of Stiperski and Grubišić (2011). A rotor develops at the same location during the 25Jan05 episode as well, but only in the initial stages of the event (prior to 9 UTC). In the later part of the episode, increase in flow linearity caused by stronger cross-barrier flow prevents boundary layer separation. Reducing the height of Velebit (gv) again leads to the development of a rotor stronger and longer lasting than in the baseline run but still not extending throughout the event. 6. Discussion The general characteristics of bora flow during the 20Dec04 and 25Jan05 episodes, presented here, are in good agreement with the jet and wake structure of the northern Adriatic bora documented well in SAR images of Alpers et al. (2009) and Signell et al. (2010). The low wind speeds in the lee of Southern Velebit are visible in all their examined episodes, associated with wake formation over Zadar peninsula that is responsible for the Zadar calm. The magnitude of the wake, however, varies significantly between bora events (Signell et al., 2010) and appears to be more sensitive to the upstream forcing than the wakes over the Northern Velebit region; this is evidenced also by the transient nature of the wake on 20Dec04. The Zadar wake is not as persistent as other northern Adriatic wakes due to the fact that it is occurring over land and is therefore sensitive to frictional effects as well as local small-scale orography. Diurnal boundary layer evolution causes the wake to be the least pronounced during the afternoon hours (Smith et al., 1997). Weak winds and reverse flow can develop over Zadar peninsula in relation to two different flow regimes, both influenced by wave breaking. In the first part of the 20Dec04 episode strong wave breaking close to the mountain top causes supercritical transition and formation of a hydraulic