Values of meteorological parameters in Turkey had been recorded manually from the late 1920s to the beginning of the 21st century. After the year 2003, starting from the western regions, existing meteorological stations were replaced by automatic ones (automatic weather observing systems, AWOSs), and uncultivated land was also covered with AWOS stations by the support obtained from large projects. These projects can be explained in the four parts as follows:

AWOS 206: excessive rainfalls on 21–25 May 1998, which also triggered landslides, resulted in many flash floods over the western Black Sea Region of Turkey. In order to eliminate damage that originates from floods, the TEFER project (Turkey Emergency Flood and Earthquake Recovery) was introduced, which was financially supported by the International Public Works and Development Bank (World Bank) with a fund of USD 369 million, to strengthen the emergency early warning systems in the west of Turkey. As a part of this project, a total of 206 AWOS stations were started and became operational during 2003 and 2004 (red points in Fig. 1a). With this project, 120 classical meteorology stations were replaced by the new automated ones, and an additional 86 AWOS stations were installed into new areas.

The first meteorology radar over Turkey was installed in Ankara for nowcasting purposes during the year of 2000 (Fig. 2). Afterwards, Istanbul, Zonguldak, and Balikesir radars were installed during 2003 by the TEFER project. Later, to detect EPEs that can be effective over the Mediterranean and Black seas, another six C-band radars were setup in the İzmir, Muǧla, Antalya, Hatay, Samsun, and Trabzon cities during 2007. Due to the forecast difficulties of convective precipitation by the numerical weather prediction models, another four C band radars (Bursa, Afyon, Karaman, and Gaziantep) were located in the inner parts of the country during 2013. From the network of these 14 radar stations, we used Istanbul, Zonguldak, Ankara, and Balikesir radar 8 min PPI (plan position indicator) and Max products, which are provided by the Turkish State Meteorological Service (TSMS) to check the characteristics of EPEs (cyclonic, convective, or sea effect originated) in combination with LWT classification (e.g., Hellström, 2003; Burt and Ferranti, 2012; Moore et al., 2015). Additionally, data of the other radars, İzmir and Muǧla, were analyzed in the study from the beginning of 2007, and data of Bursa and Afyon from the year of 2013.

The subjective version of Lamb's work (Lamb, 1972) was first developed as an objective version by Jenkinson and Collison (1977) and refined by Jones et al. (1993) to indicate the circulation types (CTs) influencing the British Isles. According to the objective methodology, vorticity and directions of the geostrophic flows are calculated using sea level pressure (SLP) fields over a predetermined central point. As a consequence of the six parameters and certain thresholds for the defined region, a total of 27 different CT types were defined (16 hybrid, 8 directional, 1 cyclonic, 1 anticyclonic, and 1 unclassified). In this study, we used daily mean SLP values on 16 grid points (between 5∘ W–55∘ E and 30–60∘ N, Fig. 1a), centered over the Marmara Region and separated by 5∘ from each other. The six parameters, namely the westerly flow (WF), southerly flow (SF), resultant flow (FF), westerly shear vorticity (WSV), southerly shear vorticity (SSV), and total shear vorticity (Z) are computed as follows: WF=12p12+p13-12p4+p5,SF=1.30514p5+2⋅p9+p13-14p4+2⋅p8+p12,FF=WF2+SF20.5, WSV=1.12⋅12p15+p16-12p8+p9-0.91⋅12p8+p9-12p1+p2, SSV=0.85⋅14p6+2⋅p10+p14-14p5+2⋅p9+p13-14p4+2⋅p8+p12+14p3+2⋅p7+p11, Z=WSV+SSV, where pi is the daily mean SLP at grid point i (Fig. 1a). Finally, classification of CTs is done according to the following criteria:

Directional type (N, NE, E, SE, S, SW, W, NW) is found by tan⁡-1 (WF/SF) and adding 180∘ to the final value if WF is positive. A value of 45∘ is allocated for each sector.

An extreme precipitation event is generally defined as a daily amount exceeding a certain threshold (e.g., Brooks and Stensrud, 2000; Ralph and Dettinger, 2012; Hitchens et al., 2012, 2013). For example, Karl et al. (1996) used 50.8 mm to define extreme precipitation events for the United States. For our country, this and other threshold limits were not suitable because of the large topographical difference and irregularity of the precipitation distribution. For this reason, similar to the previous studies (e.g., Jones, 2000; Zhang et al., 2001; Piccarreta et al., 2013; Krichak et al., 2014), we chose a methodology that defines the threshold levels of each station according to its own precipitation characteristics. Thus, this relative technique is based on considering the largest 10 % of the daily precipitation amounts for each station separately as its own extreme. Then, the annual contribution of EP for each station is determined by a standardized total that is the division of the cumulative totals of EP for each station by 10.

We defined EPE types by using the LWT technique and the radar outputs. With the LWT technique, basic airflow coming from the sea and generating extreme precipitation is defined as sea-effect precipitation. The cyclones, which generate severe precipitation over a defined region, are also characterized with cyclonic EPEs. We identified the convective EPEs as the precipitation bands coming from the terrestrial areas. As a result of these definitions, EPEs having northerly (N) and northeasterly (NE) CTs for the Marmara Region were considered to have a Black Sea effect. For the sea-effect EPEs over the Aegean, westerly (W) and southwesterly (SW) CTs were selected. For the convective EPEs, easterly (E), southeasterly (SE), and southerly (S) CTs for Marmara and E and SE types for the Aegean were chosen. In terms of cyclonic (C) EPEs, a low-pressure center over the Marmara and Aegean regions was selected as a cyclonic CT in accordance to LWT methodology.

The physical mechanisms behind the EP were investigated by using NCEP/NCAR Reanalysis products (Kalnay et al., 1996). For this purpose, sea level pressure and temperature data of 850 hPa were examined on a 2.5∘ × 2.5∘ grid resolution of the reanalysis data. For the sea surface temperature (SST) distribution over the neighboring sea areas around Turkey, NOAA High Resolution SST data provided by the NOAA, OAR, and ESRL PSD, Boulder, CO, USA, from their website at http://www.esrl.noaa.gov/psd (last access: 1 August 2017) (Reynolds et al., 2007) were used in the study.

For the first time, different daily precipitation threshold limits of 97 stations were constructed from a 10-year dataset (Fig. 3a). According to the results, the highest daily precipitation rates exceeding 100 mm are observed in the southern Aegean Region where it can be classified as “rich” in terms of extreme amounts of precipitation. This suggests that if the daily precipitation amount of a station located in the South Aegean exceeds this limit, that day is recorded as an EPE for that station. A daily precipitation threshold ranging from 60 to 100 mm is shown to be mainly located on the coastal regions in the west of Turkey. When one moves towards interior continental areas, the daily EP threshold decreases from 60 to 20 mm. The lowest limits are observed in the semi-arid continental areas of the Aegean and Marmara regions, having threshold values lower than 40 mm, as illustrated in blue in Fig. 3a and can be classified as “poor” in terms of extreme amounts of precipitation.

The annual contribution of EP for each station (cumulative totals of EP for each station divided by 10) is shown in Fig. 3b. We observe that the largest normalized annual amounts of EPEs are mainly located on the southwest of the Aegean and the mid-south and northeast of the Marmara Region, with values larger than 60 mm. It is interesting to see that the interior continental areas of the Aegean and Marmara regions that were characterized as poor in terms of extreme amounts of precipitation (Fig. 3a) now exhibit a better picture in their normalized value, generally having a better value between 40 and 60 mm. The reason for this can be the convective precipitation, generating intensified rain that can accumulate higher amounts of precipitation during a single rainstorm. On the other hand, western regions of Marmara that exhibited considerably larger threshold values with precipitation totals larger than 60 mm (Fig. 3a) show a worse image, with the normalized precipitation values being between 40 and 60 mm.

As an example, on 28 October 2010 intense daily rainfalls and many associated flash floods occurred in the western parts of Turkey. As a result of this extraordinary event, daily precipitation amounts exceeded 70 mm in the Bandirma province (south seaside station of Marmara Sea, in Fig. 4). During this day, daily precipitation totals exceeding 50 mm are shown in yellow in Fig. 4, extending from the coastal Aegean Region towards Marmara as an enlarging region that reaches up to Black Sea and passes over the Gulf of İzmit and Silivri. This squall line affected majority of the Marmara Region and to a lesser extent the Aegean Region. However, many stations located outside of this critical yellow region also had rainfall totals above their extreme daily precipitation limits.

The seasonal distribution of the EPE frequencies can provide important information to understand the physical mechanisms forcing these extreme events. For this reason, we analyzed the total counts of EPEs for four seasons, and the results are depicted in Fig. 5. It can be stated from Fig. 5 that winter (DJF) and autumn (SON) are more significant than the other seasons (Fig. 5a and d). During winter, two areas over the Aegean resulted in more than six extreme precipitation days (Fig. 5a). Spring is mainly characterized as having EPEs between 2 and 4 days on the eastern portions of the Aegean Region (Fig. 5b). During summer, the highest count of the EPEs, with 3 days, is shown to be located over the areas of the Marmara Region that were affected by the Black Sea (Fig. 5c). Seasonally, the second highest frequency of EPEs can be found in the autumn. In this season, an area extending from northeast to south of Marmara receives a frequency considerably higher than 6 days (Fig. 5d). From this point of view, a detailed analysis of the atmospheric systems generating EPEs and effecting the Aegean Region mainly during winter and the Marmara Region during autumn becomes important. The next section focuses on this aim.

In this section, we carried out frequency analysis of the seasonal EP events for the Marmara Region and documented the results in Table 1. The Marmara basin seems to be the most sensitive to these occasional precipitation events during the autumn season. It is clear in Table 1 that 53 % of the EPEs (sea effect, cyclonic, and convective in origin) occurred a total of 43 days in this season, followed by 27, 15, and 22 days in the summer, spring, and winter months, respectively. During autumn, convective, cyclonic, and the Black Sea effect EPEs are most influential over Marmara with the percentages of 21 %, 17 %, and 15 %, respectively.

At the Aegean Region, which is represented by a total of 46 stations, EP is more frequent during the winter months and a total of 35 different winter days ended up with cyclonic EPEs during the 10-year period (Table 2). It can be seen from the table that cyclonic EPEs represent 61 % of the wintertime extreme precipitation events belonging to the region. The second highest frequency belongs to the autumn, with a value of 43 % corresponding to 28 events.

The annual distribution of the total counts of EP days together with their precipitation characteristics was also analyzed for the Marmara and Aegean regions during autumn and winter months and depicted in Fig. 9. In terms of Marmara, cyclonic CTs were more active in the years of 2009, 2011 and 2013 (Fig. 9a). On the other hand, the highest counts of Black Sea effect EPEs happened during the years of 2008 and 2015. It is known that 2010 was a wet year for Marmara and dense daily rainfall amounts were generated by convective activity. The annual distribution of the cyclonic EPEs in winter indicates that the Aegean Region had been under the influence of midlatitude cyclones during the years 2009 and 2010 (Fig. 9b). As previously indicated by Türkeş and Erlat (2003), negative relationships were found between NAO indices and Turkish precipitation series. Statistically significant changes in the precipitation amounts during the extreme NAO phases are more apparent in the west and central Turkey. Hence, one reason for this cyclonic EP could be the negative phase of NAO pattern, where warm and moist air over the Mediterranean Sea can be transferred to the Aegean Region by strong westerly or southwesterly flows.

Peak times of extreme precipitation in a day can give us important information for understanding the possible causes triggering this event. For this reason, we investigated hourly behavior of the mean EP for the Marmara and Aegean regions belonging to the autumn and winter months (Fig. 10). In Marmara, the highest daily mean extreme precipitation is shown to occur under the convective types, followed by Black Sea effect and cyclonic CTs. During convective activity, we showed a peak during afternoon hours of the day. The main reason of this event can be the diurnal heating, and this is further investigated in the next section. For the Black Sea effect EPEs, we observe an hourly peak of the precipitation close to noontime, and this suggests that when maximum solar radiation reaches the sea surface, a significant amount of moisture and heat are transferred by northerly flows to Marmara, generating a considerable amount of precipitation (Baltacı et al., 2015, 2017). During the cyclonic CT, the region takes dense hourly precipitation at the mid-afternoon of the day. In regard to the Aegean in winter, cyclones generally release dense precipitation potentials from nighttime to noontime.

From the previous studies, it can be said that the link between precipitation intensity and temperature was explained by the Clausius–Clapeyron (C–C) relation. The C–C relation presents the moisture-holding capacity of the atmosphere to temperature, hinting a roughly 7 % increase in atmospheric moisture storage per degree Celcius. Pall et al. (2007) found a high agreement between the C–C relation and the changes in the rainfall extremes at midlatitudes. Lenderink and van Meijgaard (2008) found for the Netherlands that changes in hourly and daily precipitation intensity generally increased at the 7 % ∘C-1 rate anticipated by the C–C relation at temperatures below 10 ∘C, but that hourly precipitation exhibited a “super C–C” relation (increase greater than 7 % ∘C-1). Later, Lenderink and van Meijgaard (2009) considered that stronger updrafts due to greater latent heat release are the main physical mechanism in the formation of the super C–C relationships. Haerter and Berg (2009) suggested that super C–C scaling may be prevalent in regions that have a relatively balanced coexistence of both convective and large-scale rainfall events.

In this part, to examine the C–C relation for the convective EPEs, we extracted 10 daily mean temperature records and extreme hourly precipitation records of the extreme precipitation days for the selected south (Bursa) and east (Kocaeli) stations of Marmara in autumn (Table 3). It was shown that hourly extreme precipitation is more linked to daily mean temperature (r=0.69, statistically significant at 95 % confidence level) in the south of Marmara under the proper synoptic conditions and daily mean temperature changes from 12.2 to 18.4 ∘C.