Source of the 2007 Ist meteotsunami (Adriatic Sea)
Abstract
[1] Široka Bay on the island of Ist in the northern Adriatic was struck by a severe inundation around 1530 UTC on 22 August 2007. The maximum wave height reached 4 m, and substantial material damage was induced. This paper investigates the generating force of the inundation. Data analysis indicates that the inundation was provoked through a double resonance mechanism, with initiation by a pronounced atmospheric disturbance that generated open ocean waves through the Proudman resonance. The waves were then further amplified inside the affected bay. The air pressure disturbance, which propagated at a speed of about 21–24 m s−1 over the northern Adriatic shelf, was investigated through both data analysis and atmospheric numerical modeling. Apparently, the air pressure disturbance was a surface manifestation of a ducted gravity wave, which propagated through the atmosphere from the Apennines to the eastern Adriatic coast over the northern Adriatic. According to the model simulation, the wave was trapped in a stable layer adjacent to the ground, which was capped by a dynamically unstable layer with wind speeds of about 22 m s−1. This made the unstable layer also a critical layer for gravity waves propagating with the same speed and allowed for their ducting in the layer below.
1. Introduction
[2] Extraordinary tsunami-like sea level oscillations with crest-to-trough heights of around 4 m inundated Široka Bay on the island of Ist around 1530 UTC on 22 August 2007, causing significant damage and injuring one tourist. This extreme event began when a sudden ebb dried out the shallow harbor, leaving most of the ships aground. A few minutes later, water rushed back in, inundating basements, houses and much of the local village and carrying with it previously stranded ships, some of which landed ashore. The peak of the major inundation wave stopped at a height of about 2 m above sea level, a value accurately determined by the water marks evident in a large number of photos and videos. (Wind-driven surface waves were not present at that time and therefore did not affect the water marks). A thunderstorm cloud passed above the nearby area shortly before the inundation, but besides that, no other significant weather phenomenon was witnessed.
[3] Waves of this kind, apparently “coming from nowhere,” are known to sporadically hit some harbors and bays of the Adriatic Sea. As eyewitnesses recall, Široka Bay on the island of Ist was struck by a similar inundation in the early autumn of 1984. Other Adriatic events include the famous flood of Vela Luka in the summer of 1978, when sea level oscillations reached wave heights of around 6 m [Hodžić, 1980], and the somewhat less destructive flood of Stari Grad Bay in the summer of 2003, which led to a maximum sea surface elevation of around 1.3 m [Vilibić et al., 2004]. Both floods have been thoroughly studied, and it has been concluded that they were caused by pronounced air pressure disturbances propagating above the middle Adriatic area [Orlić, 1980; Vilibić et al., 2004], a phenomenon that has been reproduced by a numerical model.
[4] The following generation mechanism was found to be plausible: a traveling air pressure disturbance characterized by a sudden air pressure change resonantly generates a traveling disturbance in the ocean, provided that the air pressure disturbance's speed of propagation v equals the speed of barotropic waves in the ocean c = 
, where g is the gravity acceleration and h is the local depth. Then the generated barotropic ocean waves further propagate and strengthen, constantly gaining energy from the air pressure disturbance as it passes above the sea. Upon reaching a coastal area, the incoming ocean waves induce eigenoscillations of stricken bays/harbors, provided that the waves have enhanced energy content with periods matching those of the normal modes of the bays/harbors. The eigenoscillations are particularly vigorous in the bays with large amplification factors. The first of the described mechanisms is known as the Proudman resonance [Proudman, 1929] and the latter as the harbor resonance [Raichlen, 1966]. Additionally, waves can be further amplified by topographic features in the affected area, causing them to reach destructive amplitudes.
[5] A similar generation mechanism is typically used to explain air pressure-related extreme sea level oscillations observed in other parts of the global ocean, including “rissagas” on the Balearic Islands [Monserrat et al., 2006; Vilibić et al., 2008], “abiki” waves in Nagasaki Bay [Hibiya and Kajiura, 1982] and marubbio/milghuba waves on the coasts of Sicily and Malta [Candela-Perez et al., 1999; Drago, 2008]. Because of their meteorological origin and the fact that they occur in the same frequency bands as tsunamis, events of this kind are often called meteotsunamis [Nomitsu, 1935; Monserrat et al., 2006].
[6] Until now, the response of the sea to the atmospheric forcing has been studied in detail, but the atmospheric forcing itself has been somewhat neglected. Two main questions are of particular interest: (1) what is the source of the air pressure disturbances and (2) what is their propagation mechanism? Knowing that high-frequency air pressure oscillations are short-lived [e. g., Lindzen and Tung, 1976], the latter question may be reformulated to: how do they manage to maintain quasi-persistent shapes over a considerably large area (a prerequisite for generation of long ocean waves through the Proudman resonance)? According to both data analysis and numerical modeling results, propagation of long-lived air pressure disturbances related to gravity waves, which will be of primary interest in this paper, is mainly due to the combined effect of wave duct and wave-Conditional Instability of Second Kind (CISK) mechanisms [e.g., Powers and Reed, 1993; Zhang et al., 2003]. Here, the wave-CISK mechanism represents coupling between a gravity wave and convection: wave-associated convergence forces moist convection, and moist convection provides energy for the gravity wave. The wave duct mechanism represents propagation of the waves in a stable layer adjacent to the ground that is capped by a dynamically unstable steering layer, which prevents vertical energy leakage from the wave. During specific events, one mechanism can be more pronounced than the other. Belušić et al. [2007] successfully modeled the atmospheric counterpart of the 2003 meteotsunami of Stari Grad Bay with the MM5 numerical model (Pennsylvania State University/National Center for Atmospheric Research fifth-generation Mesoscale Model [Grell et al., 1995]). They showed that this event was provoked by an atmospheric gravity wave that sustained and recovered its energy through the wave-CISK mechanism. On the other hand, “rissagas” of the Balaeric Islands are frequently found to be caused by ducted gravity waves or by ducted convection jumps [Monserrat and Thorpe, 1992; Jansà et al., 2007]. As for the question of their origin, the MM5 numerical simulation revealed that the gravity wave responsible for the 2003 meteotsunami of Stari Grad was caused by orographic excitation [Belušić et al., 2007], while, on the basis of measurements, shear instability is generally believed to be the source of the gravity waves provoking the Balearic Islands “rissagas” [Monserrat and Thorpe, 1992]. However, because of the lack of data in the latter case, other sources of wave activity such as geostrophic adjustment, convective activity or wave-wave interaction remain plausible (for the list of gravity wave sources, see, e.g., Fritts and Alexander [2003] and references therein). Consequently, to find the source of a specific atmospheric pressure disturbance and to fully understand its propagation mechanism, one cannot rely solely on available data but must engage in numerical modeling.
[7] In order to comprehend the atmospheric forcing presumably responsible for the Široka Bay inundation on 22 August 2007, we first analyzed available meteorological data and subsequently used the state-of-the-art Weather Research and Forecasting-Advanced Research WRF (WRF-ARW) model in order to numerically simulate the atmospheric component of the event. Through intensive data analysis, we first explored whether the Široka Bay inundation was indeed provoked by a propagating air pressure disturbance. Second, we tried to understand that pressure disturbance and in particular its propagation mechanism. Better understanding of the atmospheric conditions that have the potential to generate a meteotsunami, as well as further endeavors in numerical modeling of mesoscale atmospheric processes, including air pressure disturbances, can greatly increase our knowledge of meteotsunamis and hopefully allow for the forecasting of such events in the future.
2. Materials and Methods
2.1. Data Acquisition and Analysis
[8] In order to trace the atmospheric counterpart of the extreme inundation wave that struck Široka Bay on the island of Ist on 22 August 2007, meteorological data from several stations in the eastern part of the Adriatic Sea were utilized (Figure 1). Digital air pressure records sampled at 10 min intervals with ±0.05 hPa accuracy were available for the Rab and Zadar meteorological stations. However, the sampling interval of this data was highly inadequate for capturing extreme air pressure oscillations, which often occur at periods much shorter than 10 min. In fact, it was shown by Vilibić et al. [2008] that observed pressure tendency can occasionally be up to four times larger at 30 s than at 10 min sampling intervals, and the pressure tendency instead of pressure itself is responsible for the generation of a meteotsunami. To avoid the problem of under-sampling, we used available analog air pressure charts. These were recorded by microbarographs located at the Pula, Pazin, Rijeka, Rijeka airport, Senj, Zavižan, Rab, Mali Lošinj and Zadar meteorological stations. Microbarograph weekly charts (21–27 August 2007) were digitized at 2 min resolution. To verify the quality of the digitization, the acquired air pressure data were verified with the 10 min digital records available for the stations Rab and Zadar.
[9] Upper air soundings were carried out daily at the Zadar station at 0000 and 1200 UTC. Satellite data were obtained from the geostationary satellite Meteosat 9. Horizontal fields of mean sea level pressure, temperature, winds and humidity at 850 and 700 hPa were taken from the European Centre for Medium-Range Weather Forecasts analyses.
[10] We determined the propagation speed and direction of the air pressure disturbances using two methods: (1) the isochronal analysis method developed by Orlić [1980] and (2) the cross-correlation function analysis method developed by Monserrat and Thorpe [1992]. The first method assumes that a pressure disturbance is a plain wave, propagating over a given area with constant speed and direction. Given this assumption, the speed and direction of the pressure disturbance can be determined from its arrival times at different meteorological stations. For more details on the method see Appendix A.
[11] The isochronal analysis method relies on the assumption that the pressure disturbances are not dispersive. To test whether this hypothesis was true for the investigated event, we applied another method to the pressure time series. This method (cross-correlation function analysis), as given by Monserrat and Thorpe [1992], allows different frequency bands to have different speeds, wavelengths and directions of propagation. The method examines the relationship between the pressure time series measured at any three meteorological stations deployed as a triangle. Although we had data from nine meteorological stations and therefore 84 possible combinations of three station sets, we considered only five of those sets. Most of the other sets (74) were removed from the calculation because they contained at least one station where no significant pressure oscillations were observed. An additional five station combinations were left out because they formed extremely scalene triangles, which facilitate determination of only the component of the speed in the direction of the longest side of the triangle. We applied elliptical (Cauer) band-pass filters [e.g., Schlichthärle, 2000] on the time series from the chosen sets of stations for 1200 to 1800 UTC on 22 August 2007. We obtained five frequency intervals each with 0.5 h−1 width spanning frequencies from 0.5 to 3.0 h−1. The limitations on frequency bands were imposed by the distance between meteorological stations, as too large a phase difference between perturbations registered at two sites could lead to spurious results, and too small a phase difference would not necessarily be distinguishable. We then used these filtered time series to obtain speeds and directions of propagation for each of the five selected frequency intervals. A more detailed description of the procedure is given in Appendix A.
2.2. WRF Model Setup
[12] We utilized version 2.2.1 of the WRF-ARW model [e.g., Skamarock et al., 2005]. WRF-ARW is a three-dimensional, nonhydrostatic, fully compressible primitive equation atmospheric model designed for simulations of atmospheric processes on scales ranging from meters to thousands of kilometers. The model simulation was performed from 1200 UTC on 21 August to 1800 UTC on 22 August 2007, with the initial and boundary conditions obtained from the European Centre for Medium-Range Weather Forecasts analyses available at 6-hour intervals. Three two-way nested domains were used with horizontal resolutions of 18, 6 and 2 km with 100, 217 and 364 grid points in both horizontal directions, respectively. The coarsest domain was centered at 44° N, 12° E, while only the finest domain is shown in the results (Section 4). There were 75 vertical levels with finer resolution at lower altitudes. The set of parameterizations used includes the Mellor-Yamada-Janjic scheme [Janjic, 2002] for turbulence, the Lin et al. scheme [Chen and Sun, 2002] for microphysics and the Grell-Devenyi scheme [Grell and Devenyi, 2002] for cumulus clouds, which was used only for the coarsest domain.
3. Atmospheric Analyses
3.1. Ground Observations
[13] Exceptional sea level oscillations (around 4 m) with a period of less than 10 min coupled with a lack of any extreme weather conditions (apart from the thunderstorm cloud passing above the area shortly before the inundation, according to eyewitnesses reports), as well as the similarity to the meteotsunamis sporadically observed in the Adriatic Sea, led us to the assumption that the oscillations hitting Široka Bay around 1530 UTC on 22 August 2007 were caused by a propagating air pressure disturbance. To test this hypothesis, we examined the air pressure time series for 22 August 2007 (Figure 2), in which we noticed a distinctive disturbance propagating through the atmosphere shortly before the Široka Bay inundation. This pressure disturbance was propagating over the investigated area between approximately 1430 and 1515 UTC and was observed at the Mali Lošinj, Rab, Senj, Zavižan and Zadar meteorological stations. The disturbance was completely absent from the meteorological stations located to the northwest of the Mali Lošinj–Senj line, indicating that the disturbance developed either somewhere near that line or that it came from a more southerly direction. A pronounced and sudden pressure drop (up to 4 hPa over 15 min) has the potential to cause meteotsunami generation. As was shown by Hibiya and Kajiura [1982] and later confirmed with numerical simulations [Vilibić, 2005], the sea response to the air pressure forcing is mainly proportional to the tendency of the forcing. Consequently, the most pronounced air pressure oscillations cause the most extreme cases of meteotsunamis. Another pressure disturbance with a large tendency (up to 3 hPa over 10 min) was propagating through the atmosphere between 1830 and 1900 UTC. However, no extreme sea level oscillations were associated with this disturbance. A closer look at the time series reveals that the second disturbance was losing its shape much faster than the first one and was almost completely absent at Zadar. That makes this disturbance a less likely candidate for the Proudman resonance since a long-lasting shape and longevity above the sea of uniform depth are prerequisites for the successful resonant transfer of energy between the atmosphere and the sea.
[14] We determined the speed and direction of propagation of the air pressure disturbance using both isochronal and cross-correlation function analysis methods. The isochronal method yielded a speed of v = 24 ± 2 m s−1 and a direction of α = 86 ± 2°, while the cross-correlation method gave a speed of v = 21 ± 1 m s−1 and a direction of α = 86 ± 5°. In the latter case, the speed and direction represent the mean values calculated in different frequency intervals. Values for specific intervals, as well as the number of triangle sets of stations that satisfied quality criteria, are given in Table 1 (for quality criteria see Appendix A). The similarity of the calculated speeds and directions of propagation (both for different methods and for different frequency intervals), the small deviation of these results around their mean values, and the fact that for most frequency intervals all triangles satisfy quality criteria are remarkable results. They indicate that the disturbance passing through the atmosphere around 1500 UTC was highly nondispersive. Analysis by Šepić et al. [2008] of pressure disturbances passing over the northern Adriatic and having pronounced tendencies showed, however, that even pressure oscillations that are not as well preserved spatially (i.e., disturbances with changing shape) can cause considerable sea level oscillations (wave heights up to 80 cm in Bakar Bay), provided that they have propagation speeds comparable to the speed of open sea waves. The estimated propagation speed of the air pressure disturbance preceding the Široka Bay inundation on 22 August 2007 (21–24 m s−1) was close to (or within the limits of) the speed of open sea waves (22–26 m s−1), most likely propagating over the 200 km wide and 50–70 m deep Adriatic shelf, in a region to the west and southwest of the island of Ist. The essential condition for the Proudman resonance was presumably satisfied, a conclusion that could be confirmed by a targeted ocean modeling study. That was not the case for the pressure disturbance that propagated through the atmosphere later the same day at around 1900 UTC: its speed was between 13 and 17 m s−1. Thus, the second disturbance was not only spatially nonhomogenous, but also propagated through the atmosphere too slowly for the full resonance to occur. Namely, its Froude number (Fr, the ratio between the speed of atmospheric and oceanic long waves) was around 0.6, meaning the resonant effects were an order of magnitude lower than in fully resonant conditions (Fr = 1.0 [Vilibić, 2008]).
| Frequency Band (h−1) | v (m s−1) | mv (m s−1) | α (°) | mα (°) | n |
|---|---|---|---|---|---|
| 0.5–1.0 | 21 | 1 | 94 | 5 | 5 |
| 1.0–1.5 | 20 | 1 | 88 | 5 | 5 |
| 1.5–2.0 | 21 | 1 | 85 | 3 | 5 |
| 2.0–2.5 | 20 | - | 80 | - | 1 |
| 2.5–3.0 | 22 | 2 | 82 | 2 | 5 |
[16] Next, we estimated the time-running coherence spectra between several stations at which pressure disturbances were observed (Figure 4) using the same procedure and parameters as in the power spectra computations. Additionally, records from the Mali Lošinj, Rab, Senj and Zavižan meteorological stations were shifted in time on the basis of the estimated speed and direction of propagation to match the arrival time of the pressure disturbance at the Zadar meteorological station. Cross-spectral analysis showed noticeable coherence (above 0.9) between all the stations around 1500 UTC, once more displaying the nondispersive nature of the pressure disturbance. To summarize, results of the air pressure time series analysis legitimize our assumption that the sea level oscillations hitting Široka Bay fall into the category of meteotsunamis.
3.2. Synoptic Conditions and Vertical Structure
[17] We will now try to assess the prevailing meteorological conditions during 22 August 2007. The synoptic situation over Europe was characterized by an intense low centered over Northern France and the Benelux countries (Figure 5). A deep southwesterly flow to the southeast of the low advected very warm and dry air originating from northern Africa across southern Italy and into southeastern Europe (Figure 5). Meanwhile, the Atlantic maritime air mass, recognizable by its high humidity, spread into the western Mediterranean over France (Figure 5). Because of the interaction of the two air masses, the frontal boundary developed in the early hours of 22 August 2007 above Sardinia and central Italy (Figure 5). The boundary was visible in the 850 hPa temperature and 700 hPa relative humidity fields at 1200 UTC and evident in the form of the cloud line in satellite imagery taken earlier that day. Furthermore, the sequence of satellite photos revealed that the convection over the Apennines was well developed in the morning and midafternoon hours, apparently triggered during the displacement of the frontal zone from Sardinia and central Italy toward the northeast (not shown). The result was the growth of several convective clouds, some of which attained further strength while crossing the Adriatic Sea. In fact, the passage of a well developed convective cloud above the northern Adriatic closely matched the arrival time of the air pressure disturbance observed around 1500 UTC at different meteorological stations, leading us to believe that this disturbance was actually related to the cloud. The satellite images clearly showed the development of this cloud upon passage of the cold front over Sardinia (around 1100 UTC), its strengthening over the Apennines and the Adriatic Sea (between 1100 and 1400 UTC), and its attainment of full potential upon arriving above the coastal northeastern Adriatic (around 1500 UTC; not shown). The satellite images indicated that the cloud propagated in the northeastward direction, which is somewhat different from the previously calculated eastward direction of the air pressure disturbance. Assuming that the air pressure disturbance was indeed related to the cloud, this discrepancy may be explained as follows. The meteorological stations used in this study fall along a more or less east–west line. Because of the cloud's significant elongation in the north–south direction, its northeastward propagation, and the assumption of both methods that the direction of propagation of the disturbance is perpendicular to the lines of constant phase, it is highly plausible that at these stations, mainly the east–west component of speed was seen. Linking the pressure disturbance with the cloud satisfactorily explains why the pressure oscillations were observed at only some of the meteorological stations in the area. Because of the cloud's line of propagation and the limited spatial dimension, the stations to the east of the Mali Lošinj–Senj line caught at best only the flanks of the cloud. Furthermore, analysis of the wind and pressure data for the Rab and Zadar meteorological stations showed that the disturbance-related air pressure rise was closely followed by an increase in wind speed in the direction of propagation of the air pressure disturbance (not shown). A pattern of this kind could have been caused by the passage of a gust-front [e.g., Kingsmill, 1995], which normally precedes convective clouds, and is often associated with pressure jumps and wind shifts. Thus, the air pressure oscillation might have been caused by the propagating convective cloud.
[19] Atmospheric soundings from the Zadar station show a height profile with similar characteristics to those discussed above. Apparently, the moist Mediterranean air spread from the surface to 1500 m (850 hPa) altitude where a pronounced temperature inversion persisted, marking the inflow of the warm African air (Figure 6). Furthermore, the moist Brunt-Väisälä frequency and Richardson number profiles indicated that a statically unstable layer, recognizable by negative values of both parameters, existed at around 4.5 km (Figure 7). It should be noted that the Brunt-Väisälä frequency was calculated as the moist Brunt-Väisälä frequency [Durran and Klemp, 1982] on levels where relative humidity was above 90% and as the dry frequency otherwise. However, the wind speed in the unstable layer around 4.5 km was somewhat smaller than the speed of propagation of surface air pressure oscillations, revealing imperfect conditions for the ducting of a gravity wave over large distances. Still, it is possible that during the 3.5 hours between the atmospheric sounding and passage of the meteotsunami-generating air pressure disturbance the atmospheric height profile changed enough to allow for the ducting of atmospheric gravity waves. To test this hypothesis, as well as to evaluate the previous assumption of a thunderstorm-related pressure disturbance, we numerically modeled the event. The results of our simulations are presented in the next section.
4. Numerical Simulations
[20] First, the WRF model results are validated through comparison of the atmospheric vertical profiles derived from the model with those obtained from the soundings at Zadar. Temperature and dew point profiles for Zadar at 1200 UTC on 22 August 2007 from both the model and the measurements are given in Figure 6. Both modeled and measured profiles show the presence of colder moist air below 1500 m (850 hPa). Above this level is a pronounced temperature inversion, somewhat smoothed in the model results, which represents the inflow of warmer and drier African air. This layer of warmer and drier air reaches an altitude of about 4.5 km. Its upper boundary is defined by a saturated layer, indicated by the convergence of dew point and temperature lines. Measurements show that the saturated layer was also statically unstable, as the Richardson number obtained negative values in this region (Figure 7). Although the model results do not reveal the presence of the statically unstable layer, a layer with somewhat lower Richardson number is modeled between 4 and 4.5 km. Aside from the absence of the statically unstable layer in the model simulation, the measurements and the simulation are quite similar (in terms of temperature, dew point, wind, Brunt-Väisälä frequency and Richardson number vertical profiles (Figures 6 and 7)). We thus use the model results to explain the meteotsunami source and its dynamics.
[22] Model results further reveal that the gravity wave, presented as a disturbance in the potential temperature field in Figure 10, propagated across the northern Adriatic from the Apennines in the west to Velebit Mountain in the east, arriving at the eastern coast at approximately 1530 UTC on 22 August 2007, and closely matching the arrival time of the measured pressure disturbance at the Zadar meteorological station. The potential temperature disturbance and the strong vertical wind updraft were approximately 90° out of phase (one quarter of the horizontal wavelength), with the wind updraft preceding the potential temperature disturbance (Figure 10). The pressure disturbance, on the other hand, was 180° out of phase with the potential temperature disturbance (not shown) and consequently also 90° out of phase with the vertical velocity. These kinds of phase shifts are characteristic for propagating gravity waves and have previously been detected in their dynamics [e.g., Belušić et al., 2007; Zhang et al., 2003]. The amplitude of the gravity wave was significantly diminished at heights between 4 and 5 km (Figure 10), i.e., in the critical and unstable layer (Figure 9), as expected for ducted gravity waves. The small-amplitude potential temperature disturbance is also visible above the critical layer (Figure 10), but this disturbance is apparently due to the convective warming inside the high cloud. Presumably, this modeled high-level convective cloud represented the one that was observed in the satellite images and which we also suspect to be the cause of the surface air pressure oscillations.
[23] Both the cloud and the pressure disturbance can also be seen in Figure 11, where the mean sea level pressure and column-integrated cloud water are depicted at 1512 UTC on 22 August 2007. The cloud (clouds are recognizable from Figure 11 as the areas where column-integrated cloud water is above 0.1 mm) seems to be located directly above the pressure disturbance. However, the entire system (i.e., the pressure disturbance and the cloud) is displaced about 40 km to the southeast in the model compared to the measurements. It is therefore reproduced only at the Zadar station (Figure 12) and is missing in the Mali Lošinj, Rab, Senj and Zavižan modeled air pressure series. Furthermore, it should be noted that the model failed to reproduce the two air pressure disturbances that preceded the disturbance that generated the meteotsunami. However, comparison of the air pressure time series at Zadar reveals a very good match in the arrival time of the meteotsunami-generating pressure disturbance and satisfactory reproduction of the pronounced pressure oscillation. We thus believe that these results may be used to explain the origin of the meteotsunami. Moreover, the exact reproduction of such highly variable events is still not possible with present state-of-the-art mesoscale atmospheric models, and the results of numerical modeling often fail to reproduce all the elements of the atmospheric component of meteotsunami events. For example, although the MM5 model simulation successfully reproduced the bulk characteristics of the wave-CISK system responsible for the summer 2003 meteotsunami in the eastern Adriatic Sea [Belušić et al., 2007], it placed the event 2 hours too early and missed the detailed features.
5. Discussion and Conclusions
[24] Our main intention was to trace the source of the inundation hitting Široka Bay on the island of Ist around 1530 UTC on 22 August 2007, injuring one person and causing substantial material damage. In order to do so, we have approached the problem from two different angles: data analysis and numerical modeling.
[25] Data analysis revealed that the inundation of Široka Bay on the island of Ist was caused by a pronounced air pressure disturbance (negative pressure tendency up to 4 hPa in 15 min) that passed above the island and the surrounding area shortly before the event. The air pressure disturbance propagated above the northern Adriatic with a speed of about 21–24 m s−1 and was capable of exciting barotropic ocean waves over a nearby shelf (50–70 m deep and 200 km wide) via the Proudman resonance mechanism. Upon hitting the coastal areas, these open sea waves presumably provoked normal modes of the bay, which eventually reached the extreme amplitudes observed at the top of the bay.
[26] Our assumption of the occurrence of the double resonance process can be justified only by comparison of this event to other previously observed meteotsunamis, as no ocean data were available in the region. (This also implies that the double resonance should be reproduced by process-oriented ocean numerical modeling.) Nonetheless, the Ist event differs somewhat from other observed meteotsunamis. Namely, Široka Bay opens to the southeast, and is thus protected from the direct influence of open ocean waves propagating from the west or southwest. This means that during the meteotsunami of 22 August 2007, the air pressure disturbance and resonantly generated open sea waves were moving perpendicular to rather than along the main axis of the bay, in contrast to the direction of wave propagation in all other studied meteotsunami events. As shown by Vilibić et al. [2004, 2008], the direction of propagation is an important parameter in meteotsunami generation, resulting in high energy content for air pressure disturbances propagating (within some span of angles) toward the open end of the bay and in a significant drop of energy content otherwise. Consequently, an additional process is needed to explain the case of the Široka Bay meteotsunami. Possibilities include wave reflection on the nearby island of Molat (Figure 1) or evolution of circular waves around Ist, making this event a unique starting point for exploring and numerically modeling such events in the ocean, a research direction that we intend to pursue.
[27] We investigated the source of the observed air pressure disturbance that caused the Široka Bay meteotsunami. The data analysis left us with two possible explanations, namely that the pressure oscillation was (1) a result of a convective cloud propagating above the area at approximately the same time as the pressure oscillation or (2) a surface manifestation of a ducted gravity wave. To choose the more probable answer, we simulated the event using the WRF numerical model.
[28] The model results suggested that the air pressure disturbance was in fact a ducted gravity wave. This wave developed over the Apennines and propagated across the northern Adriatic to its eastern coast. Shortly before the gravity wave crossed the Adriatic, an unstable layer characterized by Richardson number values less than 0.25 became well developed at altitudes between 4 and 6 km throughout the gravity wave's route of propagation. Furthermore, this layer contained a region where wind speeds reached the calculated speeds of the surface pressure disturbance (around 22 m s−1). The layer therefore satisfied the essential condition for a wave duct, namely the existence of an unstable layer with an imbedded critical level above a stable layer adjacent to the ground [Lindzen and Tung, 1976]. The presence of a gravity wave was further confirmed by analysis of the phase relationships between the potential temperature, pressure disturbance, and vertical wind speeds, which closely matched the known phase relationships for propagating waves. The amplitude of the gravity wave rapidly diminished above the critical layer, confirming the wave duct. The numerical model placed the pressure disturbance somewhat to the southeast of its measured location. Despite this discrepancy, we believe that the model reproduced the air pressure disturbance sufficiently well. However, we cannot completely discard the influence of cloud dynamics acting in concert with the ducted wave, especially as eyewitnesses claim that the thunderstorm cloud passed above the area shortly before the flood, and as a convective cloud was observed in the satellite images. The convective cloud and the related pressure disturbance are also present in the model results, where they are found propagating in and above the critical layer and directly above the gravity wave. This coexistence of the gravity wave and the convective cloud somewhat resembles the wave-CISK mechanism, indicating that perhaps both wave-CISK and wave duct mechanisms were active during the propagation of the meteotsunami-generating air pressure disturbance over the northern Adriatic.
[29] The above finding stresses the importance of numerical modeling when attempting to explain observed mesoscale atmospheric events. Available meteorological measurements are often sparse in both space and time and are therefore not sufficient for the reliable reproduction of an event, but are certainly needed for the verification of numerical model results. Thus, in order to make a trustworthy case study of mesoscale atmospheric events, both measurements and numerical modeling should be utilized.
[30] Finally, we would like to stress that the gravity wave duct was suggested by the meteorological data as one of the possible explanations of the observed pressure disturbance primarily because of the similarity between the atmospheric vertical profile observed above the Adriatic Sea on 22 August 2007 and the profiles usually observed over the Balearic Islands during the “rissaga” events. This similarity was presumably a consequence of the similarity of general synoptic conditions, including cyclonic circulation over western Europe, resulting in advection of colder Atlantic air to the Mediterranean from the west and warmer African air from the south. Such synoptic similarities point to an interesting possibility that there are atmospheric conditions generally favorable for meteotsunami generation in the Mediterranean area.
Acknowledgments
[33] Meteorological data collected at stations in the northern Adriatic were provided by the Meteorological and Hydrological Service of the Republic of Croatia. Satellite imagery was obtained from EUMETSAT. The work was supported by the Ministry of Science, Education and Sports of the Republic of Croatia (grants 001-0013077-1122 and 119-1193086-1323).
Appendix A
A1. Isochronal Analysis Method
denote the measured difference between arrival times of pressure disturbances at various stations, while n is the number of station pairs. Speed v and direction γ can then be obtained by minimizing the following function using the least squares approach:
A2. Cross-Correlation Function Analysis Method
