Analysis of tropical-like cyclones over the Mediterranean Sea through a combined modeling and satellite approach



[1] Several Mediterranean vortices with characteristics similar to tropical cyclones are analyzed by means of numerical simulations, satellite products and lightning data. Numerical analysis suggests that the broad tropical-like cyclone category includes in reality a set of different cyclones, ranging from very small and weak vortices to larger and stronger cyclones. One case displays a much longer persistence of tropical features than the other events. The analysis of the tracks identifies two preferred areas of occurrence: the Ionian sea and the Balearic Islands. The satellite analysis of cloud top height and retrieved rainfall indicates that the stage characterized by the most intense convective activity and rainfall anticipates the mature phase, when the cyclone is more intense and characterized by tropical features, during which convection is shallower and rainfall weaker. This result is confirmed by a preliminary analysis of the lightning activity.


[2] Since satellite images became available in the early 1960s it has been possible to identify vortices in the Mediterranean basin whose characteristics are similar to those of tropical cyclones, such as spiral-like cloud bands and the presence of an “eye.” The evidence of a warm core [Lagouvardos et al., 1999] and of a weak vertical wind shear [Reale and Atlas, 2001] made the similarity of these vortices to tropical cyclones more apparent from a dynamic perspective too. Tropical-like cyclones (TLC) have mostly been studied on a case study basis. Although several studies focused on different aspects of their structure and evolution [e.g., Pytharoulis et al., 2000; Homar et al., 2003; Emanuel, 2005], a general perspective on their properties, lifetime, and evolution is still missing. A preliminary effort toward a more systematic approach was provided by the Meteorology Group of the University of the Balearic Islands (UIB), which has identified 43 different cyclones showing tropical characteristics from 1982 to 2008 [website:] based on the visual identification of tropical-like features from satellite images.

[3] More recently, Tous and Romero [2013] implemented an algorithm for the identification of TLCs from satellite images. Unfortunately, they could not discriminate among tropical and extratropical features without using very restrictive criteria, which however subsampled the number of TLCs. Following a dynamical approach, Cavicchia [2013] analyzed the presence of TLCs in a climatological 60 year model run, by applying the Hart [2003] phase-space criteria. This approach introduces the first climatology of this kind of cyclones.

[4] The present study aims at a better understanding of the characteristics of different TLCs by combining a modeling and satellite approach with a view toward a future automatic detection algorithm for the discrimination of TLCs from baroclinic cyclones. The approach is based on the numerical simulation of a set of events considered as potential candidates for being TLCs. The data set includes only cases after May 1998, when the National Oceanic and Atmospheric Administration (NOAA)-15 satellite was launched with the Advanced Microwave Sounding Unit-B (AMSU-B) radiometer that provides the data for the 183-WSL precipitation retrieval algorithm used in the study.

Model Simulations

[5] Large-scale field analyses (e.g., those of the European Centre for Medium-range Weather Forecasts, ECMWF) cannot be directly used to identify the properties of TLCs, as their availability every 6 h and with a horizontal resolution of some tens of kilometers is comparable with the lifetime and the dimension of TLCs [Miglietta et al., 2011; Chaboureau et al., 2012], thus a downscaling becomes necessary. Model simulations are forced here with 6-hourly ERA-INTERIM reanalysis data, which have a spectral resolution T255 (approximately 79 km). Compared with the ECMWF analysis, the ERA-INTERIM fields suffer from a coarser resolution, but they are homogeneous, as they are characterized by the same methodology of assimilation over the whole period.

[6] Numerical simulations are performed here using the Weather Research and Forecasting model (WRF-ARW) [Wang et al., 2010], version 3.1. The model is implemented with the following parameterization schemes: WRF Single Model-5 class microphysics, Rapid Radiative Transfer Model for longwave radiation, Dudhia shortwave radiation, five-thermal diffusion scheme for land surface, Yonsei University planetary boundary layer, and Kain-Fritsch cumulus convection. Simulations are undertaken using two grids, with a horizontal resolution of 22.5 and 7.5 km, in a two-way nesting configuration, and 40 vertical levels. The results are analyzed hereafter only within the inner domain. The use of grids covering the whole Mediterranean would be computationally demanding and thus four different sets of grids are selected covering the track of the cyclones. The different domains have a similar number of grid points (103 × 151 for the outer grids; 199 × 277 for the inner grids, apart from the eastern domain, which has 199 × 307 points) and differ mainly in their center point. The extension of the inner grids is shown in Figure 1.

Figure 1.

Tracks of the cyclones in the four inner domains considered for the simulations: west (top left), south (top right), central (bottom left), east (bottom right). The tracks are constructed using mslp minima every 3 h; the day/hour corresponding to the first and last point in the track are shown. Thicker lines indicate the part of the track with tropical features.

[7] Twenty-eight cases of cyclones are selected as candidates having tropical features. They are extracted from the UIB website, from the cases analyzed in literature [Fita et al., 2007; Tous and Romero, 2013; Claud et al., 2010; Moscatello et al., 2008b], and from those emerging in the last few years from a visual analysis of satellite images performed by the authors, mainly the RGB images of the Moderate Resolution Imaging Spectroradiometer (MODIS). All these cases are simulated with the WRF model; the simulations cover the lifetime of the cyclones, and in particular their most intense phase, with a duration of 3–5 days.

[8] For each case, a validation is first performed. As these cyclones spend most of their lifetime over the sea, surface measurements are not available for comparison. Thus, large-scale analyses are considered as a reference even if they generally underestimate the depth of the cyclones [Cavicchia and von Storch, 2012]. As an additional check, WRF model simulations are compared with the available Meteosat images and the NOAA “Blended Sea Winds” scatterometer data product (6 h time resolution and 0.25° horizontal resolution) to verify the coherence of the model fields. Because the accuracy of scatterometer-retrieved winds is small for wind speeds higher than 20 m s–1 [Ebuchi et al., 2002], and the horizontal resolution is not adequate to represent the detailed circulation of some TLCs (e.g., the case of 26 September 2006), such data are used only for a qualitative comparison. In most cases the model is able to reproduce a track consistent with that emerging from the large-scale analysis and the scatterometer data, as the pressure minima locations generally differ by no more than 2°. The agreement is very good for the most intense cyclones; as expected, the simulations generally predict a minimum significantly deeper than the analysis. On the contrary, for four cases, all affecting the southern part of the Mediterranean sea (28 September 2003, 27 October 2005, 16 October 2006, 16 October 2007), the tracks are far apart and are not considered in the following.

[9] Over the remaining 24 cases, the task of identifying tropical characteristics is pursued by using the Hart [2003] three-dimensional phase-space diagram. This tool provides an objective classification of the simulated cyclone characteristics and allows for the analysis of the cyclone structural evolution. Three parameters are considered: the storm-motion relative 900–600 hPa thickness asymmetry across the cyclone, and two parameters representing the thermal-wind magnitude in the upper and the lower troposphere.

[10] To apply the Hart [2003] analysis, the meteorological fields need to be analyzed within an appropriate circle centered at the mean sea level pressure (mslp) minimum. The radius of 500 km is appropriate for tropical cyclones but too large for TLCs. After several trials and errors, here the radius is chosen considering the extension of the warm core anomaly at 600 hPa, as implicitly suggested by Hart [2003, Figure 3a].

[11] Ten cases affecting respectively the Ionian sea (19 March 1999, 15 September 2005, 22 October 2005), the Libyan sea (17 September 2003, 1 February 2006, 9 March 2012), the Tyrrhenian sea (22 March 2007), and the surroundings of the Balearic Islands (16 October 2007, 19 October 2007, 10 September 2008) are discarded as they do not satisfy at least one of the conditions of the Hart diagram. Thus, only 14 of 24 cyclones show symmetric, warm-core features throughout the troposphere for at least one instant (the model output is saved every 3 h) and are considered hereafter. Such cases and their main properties are reported in Table 1. The cyclones occur mainly in the fall and their tracks, shown in Figure 1, are mostly concentrated in two areas: around the Balearic Islands and over the Ionian Sea (in agreement with Tous and Romero [2013] and Cavicchia [2013]). In some cases the tracks change suddenly their direction, drawing loops that suggest a possible role of the topography. In other cases the TLCs follow a more straight direction, following the jet stream, and strongly deepen when they are positioned on the left side of the jet exit. Among the different regions, the cyclones in the western Mediterranean basin appear, on average, to be wider, of longer duration and more intense.

Table 1. Case Studies Showing Tropical Features and Relative Properties: Affected Area, Date, Radius, Duration of the Phase With Tropical Characteristics, Pressure Minimum, Maximum Wind Speed at 900 hPa, Sea Surface Temperature and Anomaly, Nearby Presence of the Jet Stream (300 hPa Wind Speed > 30 m s–1)
DomainAreaDate (Day/Month/Year)Radius (km)TLC Phase (h)Min. Pressure (hPa)Max. Wind at 900 hPa (m s–1)SST and Anomaly (°C)Jet Stream
  1. All values refer to the phase when the cyclones show tropical characteristics. Inner domain fields are analyzed at 3 h interval. Data of daily SST and of its anomaly with respect to the monthly average are evaluated along the tracks of the TLC from the data of Istituto Nazionale di Geofisica e Vulcanologia, available from January 2001; for the earlier periods, the ECMWF analysis absolute values, without anomaly, are considered.
3SE Italy26/09/2006100129804520–26 (+1)yes
4Ionian04/12/200810039904216–18 (+1)yes
5S Italy13-14/04/201250–150219823914–17(+1)in the initial phase
7Sicily Strait03-06/11/200450–503-31000–100530–2422–25(+3)yes
8Aegean Sea09/10/200050310002419–23yes
10West Sardinia27/05/200350–1002410003018–20 (−1)yes
11Balearic Islands17-18/10/2003100–150189904219–23(+1.5)yes
12Balearic Islands17/10/200750–1001210112421–23(+2)no
13S France26/10/200710069963817–20 (0)no
14Balearic Islands and S France06-08/11/2011150–200639904617–22(+2.5)in the initial phase

[12] The radius of the Mediterranean TLCs ranges from about 50 to 200 km [see also Tous and Romero, 2013], some of them presenting a different extension in different phases of their lifetime. The 900 hPa maximum wind speed ranges from 18 to 46 m s–1, and the intensity of the cyclones does not appear generally related to their extension. In some simulations the maximum 10 m wind speed is close to, but never exceeds, 33 m s–1, the threshold for a Category 1 hurricane in the Saffir-Simpson scale. Considering that the simulated wind speed would be larger—and more realistic—by reducing the resolution to 1 km or less [Rotunno et al., 2009], it is reasonable to suppose that the most intense cyclones belong to Category 1.

[13] Tropical features (a symmetric, deep warm-core structure) are detected for 3–6 h in 6 cases, and for 12 h to 1 day in other 7 events. No correlation was identified between the duration of this phase and the total cyclone lifetime. The case of November 2011 displays a peculiar behavior, as tropical features persist for two and a half days: this long persistence is probably favored by the slow movement of the cyclone over the gulf of Lyon waters (Figure 1), in the area where the sea surface temperature (SST) anomaly is higher. The presence of a high SST anomaly is detected in most cases, while SST covers a wide set of values, only rarely above the threshold for tropical cyclone development (SST > 26°C): this is a consequence of the key role played by cold cut-off lows, which are absent in the Tropics [Palmén, 1948].

Satellite Data

[14] In the literature, TLCs are discriminated from satellite images based on the typical characteristics of a tropical cyclone, e.g., a symmetric cloud structure rotating around a clearly defined eye [Tous and Romero, 2013]. However, among the TLCs considered here and for the time when the Hart [2003] criteria are satisfied, it comes out that: the eye may not be well defined, e.g., for the case of December 2005 the eye is not clearly visible because of high cloud cover; the shape may be nonsymmetric, although the eye of the storm is clearly detectable, as at 0900 UTC, 4 December 2008. Thus, the way to define an algorithm for the automatic detection of TLCs appears not straightforward and a further investigation of these cyclones is necessary. Here an analysis is made using microwave satellite data; in particular, the water vapor absorption band at 183.31 GHz is used through the precipitation retrieval method Water vapor Strong Lines 183 GHz (183-WSL) developed by Laviola and Levizzani [2011], which allows to analyze the precipitation at the ground and the cloud type in terms of its precipitation characteristics.

[15] The method makes use of the AMSU-B on board the NOAA satellites. The AMSU-B is a cross-track scanning instrument with a nominal field of view of 1.1° (about 15 km at nadir) and five spectral channels, two centered within the atmospheric windows at 89 and 150 GHz, and the other three selected within the water vapor absorption band at 183.31 GHz. The radiation at 89 GHz is strongly absorbed by cloud liquid water whereas the signal at 150 GHz is markedly affected by ice particles. A scattering index is used to estimate the probability associated with the surface rain intensities (convective and stratiform, in a number of categories) dwelling on the value of the brightness temperature difference between 89 and 150 GHz.

[16] The occurrence of TLCs is anticipated by a phase characterized by high precipitation intensity and deep convection, while, during the mature stage, rainfall is less intense and convection relatively shallow. For example, considering the very well documented case of September 2006 [Moscatello et al., 2008a; Davolio et al., 2009; Conte et al., 2011; Laviola et al., 2011], the maximum rainrate (about 20 mm h–1) occurs a few hours earlier than the mature tropical-like phase of the cyclone, and is due to deep convective clouds (Figures 2a and 2b). However, when the cyclone shows tropical features, from the late morning of 26 September, convective cells are confined to the first 6 km (Figures 2c and 2d) of the atmosphere, and the rain rate is limited to about 10 mm h–1.

Figure 2.

Cloud type identification [for midlatitudes, stratiform type: ST1 (cloud top at: 1–3 km), ST2 (3–5 km), and ST3 (5–6 km); convection type: CO1 (6–7 km), CO2 (7–9 km), and CO3 (>9 km); left] and rain rate (right) from the 183-WSL algorithm at 0139 UTC (top), 1431 UTC (bottom), 26 September 2006.

[17] Similar results are obtained for most of the cases analyzed here and have also been previously documented for some TLCs [Claud et al., 2010]: heavy convective clouds and intense rainfall are observed in the initial phase, before the appearance of the eye, while the phase of maximum intensity of the cyclone is characterized by a lower rainfall intensity, mainly distributed in rainbands. During the mature stage, the clouds are usually located in the middle and lower troposphere, even for the most intense cases, with characteristics of stratiform and convective type 1 (6–7 km), according to the classification of the 183-WSL algorithm. The relatively low height of the cloud tops is consistent with the presence of a slanted eyewall as a consequence of the secondary circulation along surfaces of constant angular momentum [Emanuel, 1986].

Lightning Data

[18] The space-time distribution of the electrical activity is important to penetrate the TLC structure. Among the available lightning data systems, the World Wide Lightning Location Network ( was the most convenient for the present analysis. The observed lightning frequencies over the Mediterranean basin can be compared with the simulated mslp minimum and the maximum surface wind. Figures 3a and 3b show the results for the TLC of November 2011. The initial stage of the storm is mainly characterized by high lightning frequency; however, when the cyclone starts to show tropical features (t = 4; 1200 UTC, 06 November), there is an abrupt decrease of the electric activity and a corresponding deepening of the cyclone (Figure 3a), with a remarkable increase of surface wind speed (Figure 3b) until 08 November, 1200 UTC (t = 19). The MSG infrared images show that in the initial phase the number of strokes is closely distributed around the cyclone center (Figure 3c), while during the mature stage (Figure 3d) the electric activity is spotted mainly in correspondence of the comma cloud crossing the coastline. The lightning activity peaks 48 h before the maximum wind speed (>160 km h–1), when the majority of clouds shows convective characteristics; during the mature stage, convection is shallower and the number of strokes drastically decreases. These conclusions agree well with the results by Price et al. [2009], who analyzed 56 hurricanes, showing that in most cases the maximum frequency of lightning activity anticipates the deepening of the cyclones.

Figure 3.

(top) Lightning strokes cumulated in 3 h, from 0300 UTC, 06/11/2011 to 1500 UTC, 10/11/2011 (solid line), counted in a moving window of 5° (latitude) × 6° (longitude) centered at mslp minimum location, versus mslp minimum (left) and maximum surface wind (right). (bottom) MSG images and number of lightning strokes at 1800 UTC, 6 November (left) and 0000 UTC, 8 November (right); one time step corresponds to 3 h.


[19] Several tropical-like cyclones are analyzed using the Hart [2003] parameter space applied to WRF model simulations. Even on a case study basis, peculiar aspects of this type of cyclones are identified. The analysis suggests that the broad TLC category includes in reality cyclones with different extension and intensity. Two preferred areas of occurrence are identified, the Ionian Sea and the Balearic Islands. Among the fourteen cases analyzed, one shows a much longer persistence of tropical features and of high wind speeds than the other cases.

[20] The passive microwave satellite analysis indicates that the phase characterized by the most intense convective activity anticipates the mature phase of the TLC, when the cyclone is more intense and characterized by tropical features. This result is suggested by the analysis of cloud top height and rainfall amount, and is also confirmed by an analysis of the lightning activity in some events. The latter results are currently investigated for other Mediterranean TLCs and will be the subject of a forthcoming paper.


[21] Authors S. L. and V. L. were funded by EUMETSAT's H-SAF and EU 7FP GLOWASIS project. The authors wish to thank the World Wide Lightning Location Network for the lightning data, Richard Rotunno (NCAR) for proof-reading the manuscript, an anonymous reviewer and Kostas Lagouvardos (NOA) for their very helpful comments and suggestions.

[22] The Editor thanks Konstantinos Lagouvardos and an anonymous reviewer for their assistance in evaluating this paper.