Stjepan IvatekŠahdan and Martina Tudor
Meteorological and Hydrological Service, Grič 3, HR-10000 Zagreb, Croatia
ABSTRACT
The article will present very high resolution dynamical adaptation of the wind field with hydrostatic version of the ALADIN model used in the operational suite, including a sensitivity study, imapct study and a MAP IOP 15 case study. Bura is a strong wind with severe gusts that often disturbs traffic on the eastern Adriatic coast. It occurs suddenly, has large spatial variability and depends on an upstream terrain configuration. The described approach is able to predict the occurrence, strength and spatial variability of the wind field in the mountainous area of the eastern Adriatic coast very well. Although this area is often affected by severe wind events (bura), measured wind data have poor coverage and representativeness. Good performance of the operational wind prediction encouraged its use in impact study to estimate the expected extreme wind speeds on the route of the new highway. In the course of this impact study, two sensitivity tests were carried out to explore the impact of the neglected processes. It was found that the impact is negligible in the case of severe bura event.
Keywords: bura, dynamical adaptation
1. Introduction
The main focus of this article will be to present the ability of the recently established operational forecast to predict the occurrence, strength and spatial variability of bura. There have been numerous theories developed and case studies related to bura wind (Brzović 1998/99, 1999,
Brzović and Jurčec, 1997, Ivančan-Picek and Glasnović 2000, Tutiš et al. 2000). For example,
Jurčec and Brzović, (1995) discuss 2 cases of bura and mention 3 mechanisms of bura; the fall wind, mountain waves and hydraulic. They had found that the extreme wind speeds were associated with the surface pressure perturbation and superadiabatic layers in the low troposphere, concluding that a non-hydrostatic fine mesh model with a realistic topography and sophisticated parameterizations of moist and turbulent processes. Numerical simulations of bura and the Adriatic cyclone associated to it were also performed (Brzović and Jurčec, 1997), including some sensitivity studies (Brzović, 1999).
Although the theoretical background and mechanisms that control bura development were explored in numerous studies, its strength, occurrence and spatial variability had not been operationally predicted. The theory suggested that a sophisticated model with very high resolution is necessary for this purpose. This would require a powerful computer.
Žagar and Rakovec (1999) have published results where ALADIN model was used for the very high resolution dynamical adaptation of the wind field to orography. Instead of the 48 hour evolution of the processes in the dynamical adaptation using the full model, they have reduced the number of the vertical levels in the upper troposphere, excluded moist processes and run the model for 30 min. This was enough for the wind field to adapt to high-resolution orography. It was expected that the method is performing better when the pressure gradient and orography are playing the most important role and when the flow is strong. It is computationally cheap, so it was introduced in the operational suite for 5 2-km resolution domains covering the mountainous parts of Croatia.
The operational forecast of the occurrence, strength and spatial variability of bura using the described method was a success. This suggested that not all of the above mentioned processes are important for successful prediction of bura. MAP IOP 15 case study and sensitivity studies have shown that this approach may be used to estimate the expected extreme wind speeds in areas where measurements are scarce.
Following this introduction, the operational suite is described in Section 2. The process of obtaining the high resolution wind forecast is explained in Section 3, followed by results of the operational suite, impact and sensitivity study presented in Section 4 and conclusions in Section 5.

2. The operational suite
ALADIN (Aire Limitee Adaptation Dynamique developement InterNational) is a limited-area model
(LAM) built on the basis of the global model IFS/ARPEGE (ARPEGE - Action de Recherche Petite
Echelle Grande Echelle, IFS - Integrated Forecast System).
ALADIN keeps the same vertical discretisation, grid point dynamics and physics as the global model
ARPEGE. ALADIN uses the spectral technique for the horizontal representation of fields. A double
Fourier representation is used limited by an elliptic truncation (Machenhauer and Haugen, 1987). A hybrid pressure-type

coordinate (Simmons and Burridge, 1981) on 37 model levels is used in the vertical with the finite difference method. Primitive prognostic equations are solved for the wind components, temperature, specific humidity and surface pressure using the two-time-level semiimplicit semi-lagrangian integration scheme.
The physical parameterizations package includes a vertical diffusion parameterization (Louis et al.,
1982) with shallow convection (Geleyn, 1987); convective and stratiform processes are treated separately by a Kessler-type large-scale precipitation scheme and a modified Kuo-type deep convection scheme. Radiation is parameterized according to Geleyn and Hollingsworth (1979) and
Ritter and Geleyn (1992). The transport of moisture and heat vertically in the soil are parameterized in two layers (Giard and Bazile, 2000). To ensure a continuous transition from large-scale to smallscale data, an intermediate zone is defined - the coupling zone - where a large-scale solution computed with the global ARPEGE, or bigger LAM model is mixed with the solution resulting from the ALADIN integration following the relaxation technique (Davies, 1976). The central zone represents the region of meteorological interest, where the forecast is fully adapted to small-scale conditions.
In the Croatian Meteorological Service ALADIN is running operationally for 00 and 12 UTC. First the full model 48 hour forecast is obtained using ALADIN on LACE domain that covers almost the whole Europe with 12 km resolution. The initial and boundary conditions are obtained from the analysis and forecasts of the global model ARPEGE run in Meteo-France, with DFI (digital filter initialization) on the analysis. Afterwards, output fields are dynamically adapted to the Croatian domain with 8-km resolution using the 48-hour integration of the ALADIN model with full physics package. The output wind field of the last is finally dynamically adapted to 2-km resolution using the following procedure.
3. The operational high-resolution dynamical adaptation
Dynamical adaptation is run sequentially for each output file. The procedure takes the output fields from the operational ALADIN with 8-km resolution, interpolates to the 2-km resolution, perturbs the fields and destroys the quasi-stationary state. The model fields adapt to high-resolution relief by running the numerical model from this initial state with the new resolution for a time necessary to achieve the quasi-steadiness (30-min, 60 sec time step). Part of the model physics describing the moist and radiation processes is excluded and the number of levels in the upper troposphere and stratosphere is reduced. This approach is not able to predict local thermal circulation or convection induced circulation. The method might also fail in the case of front propagation because the stationary state is not reached, and when large-scale forecast of the driving model fails. The method performs much better when the wind is strong enough to overcome the circulation induced by local thermal or convection induced circulation. It is especially useful when the main wind forcing is the pressure gradient over mountains, as it is the case with bura.
The model is run on 15 levels in the vertical (lowest is 17 m above the terrain), using vertical diffusion and GWD (gravity wave drag) parameterizations. The vertical resolution in the lower troposphere is the same for all domains. The operational suite takes about 20 minutes on a linux
PC for 48-hour forecast with 3h interval using the high-resolution dynamical adaptation for one domain (80x80 points).
Bura is a strong wind with severe gusts that often disturbs transport and other aspects of life on the eastern Adriatic coast. It is characterized by its sudden occurrence, spatial variability and dependence on the upstream terrain configuration. It occurs when the cold air mass is advected over the coastal mountains towards the sea, usually strongest downstream of the mountain passes,

and more severe and frequent during the cold part of the year. The cold air mass upstream of the mountain moves toward the warm air above the sea. The pressure gradient across the mountains increases. Usually there is precipitation on the upwind side, but the air remains colder and denser than the air above the sea and accelerates during its motion down the slope. Exploration of the influence of the mountain waves on the downslope winds has shown that the energy for the rapid acceleration of the low-level flow may be extracted from the kinetic energy of the mean flow. If the mountain is sufficiently high, the effects of breaking waves on bifurcation have increased importance.
3.1. Representation of the orography
Figure 1. Orography representation in the ALADIN/Croatia 8-km resolution domain (left), zoom of it to the area covered by the dynamical adaptation (center) and in the dynamical adaptation 2-km resolution domain (right), the scale is the same on all pictures. The black lines represent land/sea mask and political borders from the graphic software.
Terrain of the Croatian coast is represented as smoothed in LACE and HRv8 domains. For example
Zavižan station is placed at 1594 m height, altitude in LACE is 715 m, in HRv8 888 m and in Senj domain for dynamical adaptation 1512 m. Increase in resolution produces a large impact on the spatial variability of the surface wind field. New valleys open and peaks rise higher so the flow needs to adapt to this new features in the terrain.
Bura wind strongly depends on local upstream terrain configuration; it is strongest downstream of the low passes in the mountain range. In the neighboring areas in the lee of the tops, it is significantly weaker. The bura is the consequence of the synoptic forcing interaction with a mountain in the immediate vicinity of the sea. During winter, the sea is warmer than the ground, and air pressure is usually lower above the sea than above the ground. This produces a pressure gradient that can get very strong when the cold, high-pressure air mass is advected from the northeast. But the advection is blocked or significantly modified by the mountains. The pressure gradient wants to push the air further but the mountain blocks the flow allowing the air to find its way through the gaps in the mountain range. Since bura depends on the synoptic forcing that is modified (or shaped) by local terrain, it should be appropriate weather feature to see how dynamical adaptation works. Enger and Grisogono (1998) have shown that the enhanced pressure gradient due to the low pressure above the sea behind the orographic obstacle caused by the temperature difference between the sea and the land brings bura further from the coast.
4. Results of the operational dynamical adaptation
Results from the operational forecast using ALADIN model in extreme weather situations concerning strong flow across the Dinaric Alps will be presented. The performance of the operational high-resolution dynamical adaptation will be illustrated on one fresh case of bura, MAP
IOP 15 and an impact study. On Figure 2. total precipitation, surface wind and mean sea level pressure from the 8-km resolution operational forecast are shown. The zoom of the surface wind field on 5 domains used for the high-resolution dynamical adaptation is shown in Figure 3, and the surface wind field as predicted by the high-resolution dynamical adaptation is shown in Figure 4.
The weather situation is characterized by a wide cyclone in the southeastern edge of the Pannonian valley and a smaller cyclone above the Adriatic. On the eastern Adriatic coast there is a ridge in the pressure fields with strong gradients perpendicular to the coastline. Both low-pressure systems have a precipitation associated to them. Most of the inland part of the precipitation is snow.

Figure 2. Total precipitation (left) accumulated during previous 3 hours, and wind field 10 m agl
(right) from Croatian domain, 8-km resolution, 30 hours forecast, valid at 06 UTC, 5 th February
2003. The numerical value on the bottom presents the domain maximum. Wind vector is shown in every fourth grid-point.
The comparison of the wind fields obtained by 2-km resolution dynamical adaptation shows that the wind field during bura is much stronger in certain areas than predicted by an 8-km resolution forecast, and corresponds better to the measured data (Figure 9 and Figure 10). The data at the edges of the dynamical adaptation domain (the intermediate or coupling zone) remain the same as in the driving model, while the impact of the adaptation is strongest in the central part of the domain.
The increase in resolution produces a large impact on the spatial variability of the surface wind field.
The comparison of the wind fields obtained with 8-km and 2-km rezolutions shows that the main characteristics of the bura wind situation is already captured by the 8-km resolution forecast.
However, the maximum wind strength and spatial variability are underestimated.
Figure 3. Zoom of the wind field 10 m agl from Croatian domain on 5 dynamical adaptation domains, 8-km resolution, 30 hours forecast, valid at 06 UTC, 5 th February 2003.

Figure 4. Wind field 10 m agl from 5 dynamical adaptation domains, 2-km resolution, 30 hours forecast, valid at 06 UTC, 5 th February 2003.
4.1. MAP IOP 15
The bura occurs when there is a strong pressure gradient across the coastal mountains, connected with a cold, high-pressure air mass advection from the northeast and cyclone development over the sea. The weather situation is described according to the model output from the 00 UTC run on 6 th
November on the Croatian domain and the high-resolution dynamical adaptation of the wind field from the same model run. All fields are 42-hour forecasts valid at 18 UTC on 7 th November 1999.
The 850 hPa wind field (Figure 5) has a strong ageostrophic component. The flow is stronger as it approaches the coastline but then is suddenly slows at this height, while the flow close to the surface becomes much stronger. When the movement of the air is mostly downward it is also adiabatic. The vertical cross-section of the wind speed and potential temperature fields, in Figure 5, show the stable layer of air close to the ground, the less stable air above and the maximum vertical wind speed just below the ridge.
Prior to the visualization, the fields were interpolated from the model terrain-following levels to pressure levels. Comparison of the vertical cross-sections of the horizontal wind speed and direction for 8-km and 2-km resolution shown in Figure 6 shows that the wind maximum is much closer to the ground and to the mountain slope in the high resolution model. The arrows show the direction of the horizontal wind. Wind velocity changes significantly with height above the sea. This is first observed as the change in the direction, but as the flow close to the surface strengthens, it weakens on the upper levels. The energy of the flow is transported downwards and confined to a shallow layer resulting in a low-level jet by stable stratified air adjacent to the ground.

Figure 5. AT850 and wind field (top, left), RT500/1000 and vertically averaged relative moisture
(top, right), surface wind field and mean sea level pressure (bottom, left) and vertical cross-section of the vertical wind speed (omega - Pa/s) and potential temperature along lat=44.24 (bottom right).
Figure 6. Vertical cross-sections along lat=44.24 with 8-km resolution (top row) and 2-km resolution
(bottom row) for 7 th November 1999, 12 UTC (left) 18 UTC (center) and 8 th November 1999, 00

UTC (right). Arrows on all pictures represent the horizontal direction of the wind, and shaded areas correspond to the wind speed in (m/s).
The location of the Maslenica Bridge is very important for road traffic - with wind speeds of the magnitude shown the bridge gets closed. The ALADIN/HR forecast produces too weak winds for this bridge just downstream (in the case of bura ) the mountain pass on the Velebit Mountain.
Dynamical adaptation gives wind speeds that correspond much better to the measurements.
Figure 7: Wind measurement data, ALADIN Croatia forecasts and Dynamical adaptation of surface wind for Maslenica for the period from 6th to 11th November 1999, 00 UTC. The ALADIN Croatia forecast data are from ALADIN 48 h forecast 00 UTC runs for the 6th, 7th, 8th and 9th of November
1999 and DADA data are from the Dynamical adaptation of the forecasts on the Maslenica domain.
4.2. Impact study
In Croatia, there is a highway construction under way. The highway will connect inland part with the
Adriatic coast, it will cross the Velebit mountain and Velebit channel over the Maslenica bridge.
Bura is generated by the interaction of synoptic forcing with orography; the wind is strongest downstream of the lowest passes in the mountain range and most frequent and strongest on the
Northern Adriatic.
When the original impact study for the highway was done, the data was available from the automatic meteorological station on Maslenica Bridge. Therefore, ALADIN model was used to estimate the strength of the wind along the new road. Dynamical adaptation of the surface wind speed has already proven to be very good for Maslenica Bridge in strong bura situations on a study on MAP IOP 15 case. Now, data from few new automatic stations along the future highway path are available and it can be seen that the choice is justified.
After the construction of the road began, a few new automatic meteorological stations were set up.
2-km resolution dynamical adaptation forecast results were compared to the new measurement data. To check the impact of the excluded parts of the physics package for one case of strong bura with precipitation inland, operational wind speed forecasts were compared to the two alternatives;
- dynamical adaptation run using the full physical parameterization package and
- 48 hour forecast run with 2-km resolution, using the full physical parameterization package.

Although the operational set-up in general gave slightly higher wind speeds than the other two, the comparison to the measurements shows that the choice of the operational dynamical adaptation set-up for the estimation of the expected wind speed in extreme cases was good. Previous sensitivity study (Brzović and Jurčec, 1997) using the same numerical model has shown that the wind is weaker when the moist processes are omitted, while this one has the opposite result. But the moist processes were already accounted for by the driving model, including them and thermal processes in the high-resolution adaptation model run slows the process of adaptation.
Figure 8. Surface wind for the operational dynamical adaptation (top, left), dynamical adaptation using the whole physics package (top, centre) and with full forecast integration on the dynamical adaptation domain (top, right). Vertical wind cross-section along lat=44.24 for the operational dynamical adaptation (bottom, left), using the whole physics package (bottom, centre) and forecast with 2-km resolution (bottom, right). Arrows on all pictures represent the direction of the horizontal wind, and shaded areas correspond to the wind speed in (m/s). The vertical cross-section is along the lat=44.24, and the highway stretches along this cross-section from lat=15.52 to lat=15.65.
4.3. Comparison with measurement data
The first impact study was based on a few cases of bura when the wind measured on the Maslenica bridge was strong to severe and well predicted. New automatic stations were established in the autumn of 2002 and beginning of 2003. Measurements from those stations show that the choice of dynamical adaptation to 2-km resolution using operational ALADIN model was justified.
The measurement of wind speed and the results of the simulation with 8-km and 2-km grid models are compared with measured data for the eastern part of Adriatic Sea. The data are measured on anemometers on the Baričević Viaduct, exit from the Ledenik Tunel, Maslenica, Krk and Pag
Bridge. The instruments measure wind speed in the range from 0.2 to 70 m/s, with 0.1 m/s precision, and wind direction ranging from 0° to 360°, with 5° precision, in 1-sec intervals. The 10minute average wind speed and direction and the strongest wind gust in the 10-minute interval and its direction are stored. Only wind speed is used in the comparisons below. The model output is wind speed taken usually at the nearest point of the model.
Recent model data from new automatic stations have shown that the 2-km resolution dynamical adaptation corresponds well with the observed data. The use of dynamical adaptation with the
ALADIN model for the impact study was proven as a good choice and the use of ALADIN forecast operational products. In the Figure 9, 8-km and 2-km resolution forecast data for each operational model run (every day, 00 and 12 UTC run) are compared to the measurements, 10-min average and 10-min maximum wind speed. It can be seen when the model starts, surface wind speed in analysis is much lower than the measured one, because it is originally from the global analysis. The wind adapts to the 8-km orography during the first 3 hours of integration, but the measured strength

is not reached, because the 8-km resolution representation of orography is insufficient. Further dynamical adaptation of the wind field to 2-km resolution orography gives much better results. The obvious failure of predicting wind speed for the period from 6 th to 7 th of February 2003 is caused by the wrong large scale forecast, as can be seen in detail in Figure 10. Only one forecast run is shown, but the nonexistent increase in the wind speed on 6 th February persists.
Figure 9. Measured wind speed for the Baricevic Viaduct (left) and Ledenik Tunel (right) automatic stations and modeled data from the closest model point for February 2003. Measured 10 min average wind speed (dark blue), 10 min maximum (light blue), all model forecasts for February
2003 (00 and 12 UTC runs) with 8-km resolution (orange) and 2-km resolution operational dynamical adaptations (yellow). 2-km resolution predicts occurrence and strength of the 10-min average wind speed well. The 'bad forecast' for 6 th of February probably corresponds to bad large scale model forecast.

Figure 10. Measured wind speed for automatic stations and modeled data for the 00 UTC run on
4th Feb 2003 from the closest model point. Measured 10 min average wind speed (dark blue), 10 min maximum (light blue), 8-km resolution model forecast (red), 2-km resolution dynamical adaptations; operational (orange), using whole physics package (yellow) and 2-km resolution 48hour model integration (green).
5. Conclusions
The understanding of bura phenomenon was among the most important issues in the Croatian meteorology as this phenomenon has significant impact on local infrastructure. The prediction of severe bura events was based on the experience of the forecaster who would recognize a synoptic situation that would lead to severe bura event. Therefore, numerous cases were analyzed and theories developed. With the ALPEX and MAP experiments, the data were obtained that put these theories to tests. However, the occurrence, speed and spatial variability of bura wind haven't been predicted by an operational model.
The local version of the operational forecast using ALADIN model was established including the new approach of the high-resolution dynamical adaptation for the mountainous parts of Croatia.
Although the model did not include all the processes anticipated important for bura in previous studies, the forecast was found very successful. Usage of a hydrostatic model with very high resolution (that was not designed for this resolution) with reduced physical parameterization package in a mountainous area was not recommended, and the non-hydrostatic model was prescribed in the cases of big slopes. The driving model is also hydrostatic, so the non-hydrostatic effects are omitted completely. The effects of moisture and heating/cooling are already included in the input fields and do not require a high-resolution representation. They are important when the synoptic situation is developing - in the large-scale model. Local wind, however, is the result of the interaction of this large-scale pressure field and small-scale orography. The moment when bura will start and finish are already predicted by the 8-km resolution model, while the local wind speed is much better predicted by 2-km resolution dynamical adaptation.
L ITERATURE
Brzović, N.,1998/99: A case study of a cold air outbreak on 20th April 1997 over Croatia, Cro. Met.
J., 33/34, 1-18.
Brzović, N.,1999: Factors affecting the Adriatic cyclone and associated windstorms. Contr. Atmos.
Phys., 72, 51-65.
Brzović, N. and V. Jurčec, 1997: Numerical simulation of the Adriatic cyclone development,
Geofizika, 14, 29-46
Courtier P. and J.-F. Geleyn, 1988: A global spectral model with variable resolution – application to the shallow-water equations. Quart. J. Roy. Meteor. Soc., 114, 1321-1346.
Courtier P., C. Freydier, J.-F. Geleyn, F. Rabier and M. Rochas, 1991: The ARPEGE project at
METEO-FRANCE. Proceedings from te ECMWF workshop on numerical methods in atmospheric models. 9-13 September 1991, 2, 193-231.
Davies, H.C., 1976: A lateral boundary formulation for multi-level prediction models. Quart. J. Roy.
Meteor. Soc. 102, 405-418
Enger L. and B. Grisogono, 1998: The response of bora type flow to the sea surface temperature.
Q.J.R. Meteorol. Soc. 124, 1227-1244.
Geleyn J.-F., 1987: Use of a modified Richardson number for parameterizing the efect of shallow convection. In: Matsuno Z. (ed)., Short and medium range numerical weather prediction, special volume of J. Meteor. Soc. Japan, 141-149.
Geleyn J.-F. and A. Holligsworth, 1979: An economical analytical method for computation of the interaction between scattering and line absorption of radiation. Contr. Atmos. Phys. 52, 1-16
Ivančan-Picek B. and D. Glasnović, 2000: Capability of the ALADIN model to predict heavy precipitation and strong wind events during the MAP, Proceedings of the 9th International
ALADIN Workshop, Brussels, 6-8 March 2000.
Jurčec V. and N. Brzović, 1995: The Adriatic bora: special case studies, Geofizika, 12, 15-32

Louis J.-F., M. Tiedke, and J.-F. Geleyn, 1982: A short history of PBL parametrisation at ECMWF.
Proceedings from the ECMWF Workshop on Planetary Boundary Layer Parametrisation. 25-
27 November 1981, 59-79
Machenhauer B. and J.E. Haugen, 1987: Test of a spectral limited area shallow water model with time-dependent lateral boundary conditions and combined normal mode/semi-Lagrangian time integration schemes. Proceedings from the ECMWF Workshop on techniques for horizontal discretisation in numerical weather prediction models. 2-4 November 1987, 361-
377
Radnoti G., 1995: Comments on ‘A Spectral Limited-Area Formulation with Time-dependent
Boundary conditions to the Shallow-W ater Equations’. Mon. Wea. Rev. 123, 3122-3123.
Ritter,B. and J.-F.Geleyn, 1992: A comprehensive radiation scheme for numerical weather prediction models with potential applications in climate simulations, Mon.Wea.Rev. 120, 303-
325.
Schmidt, F., 1977: Variable fine mesh in a spectral global model. Beitr. Phys. Atmos., 50, 211-217
Simmons A. J. and Burridge D.M., 1981: An energy and angular momentum conserving vertical finite-difference scheme and hybrid vertical coordinate. Mon. Wea. Rev. 109, 758-766
Tudor M. and S. IvatekŠahdan, 2002: MAP IOP 15 case study. Cro. Met. Jour.
, 37 , 1-14.
Tutiš V., B. Ivančan-Picek and D. Glasnović, 2000: MAP SOP severe Adriatic bora event. 26th
International Conference on Alpine Meteorology, 11-15 September 2000., Innsbruck,
Oesterrechische Beitraege zu Meteorologie und Geophysik. Heft Nr.23/Publ.Nr.392. (CD
ROM)
Žagar, M. and J. Rakovec, 1999: Small-scale surface wind prediction using Dynamical adaptation.
Tellus , 51A , 489-504.