The eastern Black Sea region is well covered by automated meteorological stations. In addition to the eight long-term stations in the region, 41 new automated meteorological stations have been added since 2013. To present the high spatial resolution and to retrieve a homogeneous data set, hourly and daily precipitation data of 49 stations operated by the Turkish State Meteorological Service (TSMS) were used in the study (Fig. 1). The main characteristics of the stations are described in Table 1.

The synoptic context of the extreme event of 24 August 2015 as well as atmospheric conditions from the previous day were analyzed with NCEP/NCAR 2.5∘×2.5∘ latitude/longitude reanalysis data. To track the intense rainfall episodes, radar PPI (plan position indicator) images, which provided by TSMS, were used. Rainstorm development stages associated with the flash flood were evaluated with Meteosat 10 images.

Operationally, one global and two regional NWP models are run regularly twice a day (00:00 and 12:00 UTC) for the precipitation forecast by TSMS. In terms of the global NWP, the horizontal grid resolution of ECMWF (European Centre for Medium-Range Weather Forecasts) the IFS (Integrated Forecast System) covers almost 16 km and uses 91 vertical levels. For the regional weather forecasts, the Alaro meteorological model has been designed to be run at convection-permitting resolutions. The key concept is in the precipitation and cloud scheme called Modular Multiscale Microphysics and Transport (3MT) developed by Gerard and Geleyn (2005), Gerard (2007), and Gerard et al. (2009). In the usage of the Alaro by TSMS where the outer domain has grid spacing of 10 km, the inner domain has almost 5 km of grid spacing as well as 60 vertical levels.

The mesoscale NWP system of a nonhydrostatic mesoscale model (NMM) core of the Weather Research and Forecasting (WRFg) is developed by the National Oceanic and Atmospheric Administration (NOAA)/National Centers for Environment Prediction (NCEP), WRF-NMM is a fully compressible, non-hydrostatic mesoscale model with a hydrostatic option (Janjic, 2003). The model uses a terrain-following hybrid sigma-pressure vertical coordinate. The grid staggering is the Arakawa E-grid. The model uses a forward–backward scheme for horizontally propagating fast waves, an implicit scheme for vertically propagating sound waves, the Adams–Bashforth scheme for horizontal advection, and the Crank–Nicholson scheme for vertical advection. The dynamics conserve a number of first- and second-order quantities including energy and enstrophy. In the study, model has a horizontal grid spacing of 30 km in its outer computational domain, and the inner domain has a grid spacing of 10 km together with 46 vertical levels. To compare precipitation forecasts of these models with the observation results, daily precipitation forecasts of the models belonging on the last runtime for 24 August 2015, at 00:00 UTC outputs were assessed.

The coastal part of the region is restricted by the EBS Mountain chain in the south and the BS in the north (Fig. 1). This mountain chain extends parallel to the Black Sea and has an average altitude of 2000 m. It rises to 3973 m at its highest point (Eris et al., 2012). Apart from the basic synoptic scale circulations such as continental polar and tropical air masses, the region is also affected by orographic precipitation. Colder air masses are prevented by the Caucasus Mountains (the highest point of Georgia) from the north; therefore, more dry climates are seen in the southern part of the region. The rain shadow effect on the lee side of the mountainous area causes a more continental climate in the southern parts of the EBS (Biyik et al., 2010). When compared with the other regions, highest winter and summer precipitation totals are observed in this part of Turkey due to the interactions of synoptic weather patterns and orographic lifting. (Unal et al., 2012). To better visualize the seasonal precipitation variability in the EBS, long-term precipitation data from 1960 to 2014 were extracted from the available eight meteorology stations (stations marked by stars in Table 1 were used for the climatological approach in Fig. 3). Five stations are located in the north of the region. According to the results, mean annual precipitation (MAP) varies from 438 mm in the south (Bayburt) to 2243 mm in the north (Hopa). This high spatial precipitation variability generates different land cover terrain. Interestingly, the highest seasonal precipitation amounts in the coastal areas were observed in the fall (SON) months. This can be explained by the significance of MCS, flow directions and SST variations over EBS. In the second wettest season (DJF), highest precipitation records were observed at Hopa, Rize, and Pazar stations with the values 606, 636, and 550 mm, respectively.

In Fig. 4a, spatial distribution of daily precipitation totals for 24 August 2015 was extracted from 49 meteorological stations. It can be seen that three main cores of precipitation are measured at the Arhavi, Hopa, and Borçka stations with the values of 135, 144, and 149 mm, respectively. In Hopa, 27 % of the long-term mean of summer rainfalls was recorded on this day. As a consequence of the intense daily rainfall episodes, these three surrounding areas of the Artvin district were the most influenced by flash floods and landslides (i.e., Hopa, Arhavi, and Borçka). Among these stations, Hopa (33 m altitude, no. 1 in Fig. 1b) is at the lowest altitude and is located in the northern coastal part of Artvin city. Borçka station is shown with an altitude of 190 m (the second lowest altitude in Artvin, no. 7 in Fig. 1b). Another coastal station, Arhavi (290 m altitude, no. 6 in Fig. 1b), is located in the northwest and has the third lowest altitude among all Artvin stations. Temporal precipitation distribution of these selected stations was extracted as shown in Fig. 4b. Rainstorms started in the evening (22:00 UTC) of 23 August 2015 and ended at midday on the following day. Hourly observations at the three stations showed the torrential rains increased from 27 to 32 mm between 22:00 and 24:00 UTC on 23 August. Thereafter they suddenly dropped from 4 to 2 mm between 01:00 and 05:00 UTC on 24 August. Later, uninterrupted extreme rainstorms hit the north and coasts of the Artvin district. According to the hourly rainfall observations, the highest precipitation amounts were recorded at Hopa over the 8 h of the flash-flood day (Fig. 4b). The maximum daily precipitation value was observed with 144.3 mm in 6 h (starting at 05:00 UTC and ending at 11:00 UTC) in Hopa, and maximum hourly rainfall measured 51.5 mm at 09:00 UTC. In Arhavi, daily total precipitation was 135.5 mm and reached a maximum value at 00:00 UTC with 32.4 mm. In Borçka, while daily precipitation amounts were higher (148.9 mm) than at Hopa and Arhavi, peak values of hourly precipitation intensities were lower. According to the results from these three stations, hourly precipitation reached a maximum value at 09:00 UTC in the low altitudes of the region; this implies that the precipitation was much lower in the upper sectors of the mountainous area.

This section summarizes the atmospheric circulation and associated physical mechanisms that were responsible for the flash flood in the region. In order to better evaluate the phenomenology of the event, pre-existing synoptic conditions starting from 23 August were investigated. At 00:00 UTC on 23 August, the summer Asian monsoon low extends to the eastern Black Sea (Fig. 5a). During the summer months, in consequence of the excessive surface heating over the arid regions of the Middle East, the monsoon low expands westward and generates the Persian trough (Alpert et al., 2004; Saaroni et al., 2010), which extends to Turkey, forming a thermal low over the eastern Mediterranean (Tyrlis et al., 2015). Besides the surface synoptic conditions, low-level moisture convergence, specific humidity content, and geopotential height values of 850 hPa were extracted. It is known that low-level moisture convergence is a good indicator for large-scale precipitation (e.g., Fragoso et al., 2012), and eastern Turkey (and Azerbaijan) has good synoptic precipitation conditions. In the upper levels, interaction between a weak ridge over northern Africa and a trough over the Aegean Sea (because of the upper-level cold low over central Europe) concludes with southwesterly winds over the Artvin district (Fig. 5c).

On 24 August at 00:00 UTC, a high-pressure center (HPC) over northern Russia moved to the south, located around 30∘ E, 60∘ N. While the cyclone remained almost stationary, a new cutoff cyclone occurred over the EBS (Fig. 5d). Thus, high northeasterly winds brought moisture from the Black Sea to the eastern coasts of Turkey (Fig. 5e). As a result, deep precipitation areas were observed over these regions according to the low-level moisture convergence results. In the upper-level chart (500 hPa), shifting cold core of upper-level high to the south cause the moving of midlatitude low to the west, and thus southwesterly winds turn into the westerly together with a decrease in temperature from -7.5 to -10 ∘C (Fig. 5f).

At the start of the rainstorm (24 August, 06:00 UTC), similar surface and upper-level large-scale circulations appeared and were compared with the midnight synoptic conditions (Fig. 6a and c). Strong moisture convergence zones were detected over the flash-flood region (Fig. 6b). For this reason, thermodynamic analysis was needed to better understand the evaluation of physical mechanisms that developed severe precipitation. Hence, as a consequence of analyzing the nearest radiosonde measurements from Samsun station (41.34∘ N, 36.25∘ E), instability indices such as CAPE (convective available potential energy) and LI (lifted index) showed that there was no strong convective activity before and during the rainstorm (not shown).

In order to follow the distribution convective cells and cloud droplets in a large area, it was necessary to use satellite and radar image data.

Repeated temporal resolution data are an excellent tool for understanding the spatial distribution of the convective cells. Therefore, SEVIRE (Spinning Enhanced Visible and Infrared Imager) MSG (Meteosat Second Generation) outputs were used to examine the atmospheric conditions on 24 August at 06:00 UTC. It is known that “convective storms RGB” product visualizes the particle size features of high-level cloud tops with good contrast (Kerkmann et al., 2006). Whereas yellowish cloud tops indicate opaque ice clouds with small particles, high-level opaque ice clouds with large particles are shown as reddish. The RGB product in Fig. 7a was produced by assigning the brightness temperature difference (BTD) 6.2–7.3 values as the red component, the BTD 3.9–10.8 as the green component, 1.6–0.6 as the blue component. In Fig. 7a, numerous convective storms with large ice particles are shown over the EBS. On the other hand, more intense storms were observed over the land areas such as Georgia and this implies the storm intensity. Separately, SYNOP observations indicate that southerly winds over the coast of the EBS stations met with humid northerly flows throughout the seaside area. If the land (21 ∘C) and sea surface temperatures (SSTs) were sufficiently different, the convective instability and storm severity could have increased with time. As seen in Fig. 7b, high PPI (plan position indicator) reflectivity values from the radar image showed that two cores of the extreme precipitation were over Hopa and Çayeli sub-basins.

The influence of SSTs on precipitation over Turkey was investigated in detail by Bozkurt and Sen (2011). They found that increased SSTs led to increased precipitation over the peninsula, especially downwind of the sea. Later, Kömüşçü and Çelik (2013) explained that the warm Aegean SST is one of the significant causes of the development of rainstorms. In this study, exploring the role of Black Sea surface temperatures on storm development, long-term (1982–2015) means of August SSTs were extracted for the BS using NOAA high-resolution SST data (provided by NOAA/OAR/ESRL PSD, Reynolds et al., 2007). As seen in Fig. 8a, cold SSTs of the BS were north of the latitude of 44∘ N. The warmest pool of the BS was in the eastern BS, and SSTs exceeded 27 ∘C in this month. During the day of the extreme event, spatial distribution of the SSTs indicates negative anomaly values in north of 44∘ N latitudes (Fig. 8b). The EBS region has the highest SST anomalies and 1.5 ∘C higher SST variations compared with the August means for the EBS.

According to the ECMWF daily precipitation product, spatial coverage of the maximum daily precipitation values (over 160 mm) is shown in the northern Rize and northwestern Artvin cities (Fig. 9a). Compared with the model output (Fig. 4a), station observations were clearly underestimated in northern Rize. On the other hand, model predictions for the Arhavi and Borçka settlements, except Hopa, were good. With regard to the Alaro model results, the highest daily precipitation totals were only well predicted for the Hopa district at 150 mm (Fig. 9b). Although precipitation forecasts of this limited-area model described Hopa well, the other two flood regions were not well predicted. Optimum spatial coverage of the daily precipitation forecasts is shown in the mesoscale WRF outputs (Fig. 9c). The problem with this model is the underestimated forecasts compared with the observation data. In TSMS, meteorologists merge the outputs of these models (the so-called “poor man ensemble”) with their own experience and provide quantitative precipitation forecasts for the alert subregions in predefined time windows. As a consequence of this subjective prediction, TSMS and its regional weather forecast offices issued alert messages related to natural hazards including severe precipitation events. These organizations also carry the responsibility for nowcasting and monitoring rainfall events. According to the main alert on 23 August 2015 at 09:00 UTC prepared by TSMS Weather Forecast Centre, very intense precipitation between 51 and 100 mm was predicted at the Rize, Artvin, and Trabzon districts within 12 h of 24 August. The authorities and the public were alerted to the risk of flash-flood, lightning, and landslide events.