**1. Introduction**

Cold fronts are significant components of the weather and climate systems, and can be closely associated with extreme events. The passage of a cold front is indicated by, and associated with, substantial variations of temperature, humidity and wind. The identification of cold fronts has attracted more scientific interest than their warm counterparts because of their more discrete character and their connection with severe weather phenomena [1–6].

Despite the availability of numerical prediction and analysis models, the manual identification of fronts on weather charts is a time-consuming process that introduces a high degree of subjectivity, even for an experienced operational meteorologist [7,8]. The complexity of the task dictates that the compilation of frontal climatologies by manual methods is not feasible. Hence, there is practical and scientific interest in developing automated schemes to create such climatologies from observed data, reanalyses, and climate model outputs [8]. The advantage of automated detection methods is that they are objective, reproducible, and fast.

The majority of such objective and automated front identification methods in the literature use thermal criteria [6,9–13]. Most of these studies focus on large-scale fronts that develop and move across vast areas of oceans and continents. Thus, their identification is facilitated by the large scale of fronts and the homogeneity of surfaces. However, verification of these algorithms has shown that the recognition of the frontal surfaces, taking into account only temperature gradients, is inadequate

in many cases for complex features [14,15]. Some of these methods are used routinely in weather forecasting [16], whereas others focus on extreme events, such as widespread fires [17]. A number of automated algorithms have been applied in order to generate frontal climatologies for southwest Western Australia [18], for the globe [7], and for the Southern Hemisphere [19,20].

Since the Mediterranean Sea is a closed basin surrounded by complex topography, its fronts tend to exhibit small spatial and temporal scales, as well as complicated kinematic and thermodynamic features during their lifetime [21]. Climatological studies focusing on the Mediterranean fronts are relatively few, and the early studies were based on subjective approaches utilising synoptic surface maps [22]. The investigation of a nine-year period (1971–1979) of daily charts [22], demonstrated that fronts appear very frequently in the Mediterranean throughout the year with maximum frequency one every seven days in winter. While identification schemes have been applied to diagnose the climatologies of cyclonic [23–25] and anticyclonic centers [26] in the Mediterranean, there is no corresponding application for the analysis of cold fronts.

The objective of this study was to develop and evaluate a scheme for the identification of cold frontal systems in the Mediterranean basin which is based on the Frontal Tracking Scheme (FTS) [19]. In Section 2, the scheme and the modifications performed are presented in brief, while in Section 3, typical results of the sensitivity tests are given for specific high impact cases connected with cold fronts passages over the Mediterranean. In Section 4, a statistical validation of the scheme for a decade is given and, finally, in Section 5, the main conclusions are summarized.

#### **2. Description of the Identification Scheme**

FTS was developed at The University of Melbourne, Australia [19], and has been used for the climatological study of Southern Hemisphere cold fronts. Unlike other similar schemes which use thermal criteria [7,10,16], FTS uses only wind-related criteria to identify fronts and has proved to work well compared to the other schemes. More specifically, thermal based methods are known to have difficulties identifying fronts in the areas of high temperature contrasts, such as coastal areas and regions with elevated topography [15]. Hence, the Mediterranean region would be a particularly difficult site for frontal identification using thermal variables. Furthermore, thermal based methods may not be able to reliably distinguish between cold/warm fronts [27].

FTS is based on Eulerian changes of the 10 m meridional wind component (*v*) which is valuable in diagnosing various aspects of frontal behavior [8,28]. The criteria for identification are [19]: (a) at a time *t*, grid points are flagged where the wind changes from the southwestern quadrant (westerly zonal wind *u* > 0, southerly meridional wind *v* > 0) to the northwestern quadrant (westerly zonal wind *u* > 0, northerly meridional wind *v* < 0) between subsequent time points *t* and (*t* + 6 h), (b) the change of the meridional wind *dv* exceeds a specific threshold value *dvcrit* during the same 6 h period.

The grid points which satisfy the above-mentioned criteria are flagged and a component labelling technique is applied [29]. Then, each flagged pixel is related and connected to its nearest eight neighbors, giving clusters of grid points. The location of the front is determined by the eastern edge of each cluster. As this approach is applied to all of the eastward edge points, the output is a set of latitude and longitude points which mark the location of the front. Thus, the location of the front at the time *t* + 6 h is given by a single series of longitude values. These values will have a stepwise character, since they represent discrete grid points. For this reason, the longitude values are treated as a simple series and smoothed by a resistant smooth method [30] appropriate for equally-spaced data. This robust statistical technique employs a set of short-window running median and running mean filters, which are successively applied multiple times to the series, to achieve adequate smoothing. Then, the obtained smoothed eastern edge determines a "mobile front". This method is particularly suited for the detection of strongly elongated, meridionally oriented fronts, which typically extend far from a cyclone center [19].

Since FTS was developed to identify cold fronts in the oceans while the topography of the Mediterranean affects the formation and characteristics of cold fronts [31], in this study FTS is modified to better identify the position, scale and tilt of cold fronts in the Mediterranean. From the records of the Hellenic National Meteorological Service, twenty cases of cold fronts are selected that entered Mediterranean from di fferent regions (e.g., Atlantic, North Africa, northern Europe) or formed in di fferent parts of the Mediterranean (Western, Central and Eastern) during di fferent months throughout the year, having caused intense precipitation.

The initial criterion used in MedFTS is that the zonal component *u* is westerly both at *t* and *t* + 6 h and the meridional wind changes sign from positive to negative. Then, sensitivity tests were performed on the other wind related criteria. First, sensitivity tests are made on the criterion of meridional wind change within the 6 h time step (*dv*), in order to find the optimum threshold value of the meridional wind magnitude change *dvcrit*. Second, instead of using the change of the meridional wind component (*dv*), the shift of the vector wind direction is employed duringa6h period (*d*ϕ), where ϕ = arctan (*v*/*u*) and a specific minimum threshold value *d*ϕ*crit* is also investigated. This criterion is examined to better identify zonally elongated cold fronts and at the same time, to filter out erroneously identified front segments. Third, an additional criterion of the magnitude of vector wind |*U*| exceeding a specific critical threshold |*U*|*crit* in each cluster of grid points is added to optimise the scheme, considering the operational experience of forecasters that the intensity of the northwesterly wind is significant behind the cold front. This criterion allows the discarding of shallow fronts.
