3.1. Hydrometeorological Analysis
The flood event in Volos city was associated with a severe storm caused by a very slowly moving barometric low over the Aegean Sea, supported by an upper air trough [44
]. The synergy between the prevailed synoptic conditions and the orographic convergence of unstable air over Pelion mountain developed a persistent storm, characterised by torrential rainfall.
In more detail, on 8 October at 06:00 UTC, a cold upper air trough (500-hPa level) moved from central Europe southward to the Mediterranean Sea, supporting the formation of a barometric low over Greece centred over the Aegean Sea (Figure 5
A). The upper air temperature in the core of the trough over southern Italy was about −18 °C, quite lower than the surrounding areas. Six hours later, at 12:00 UTC, the southeastward movement of the trough triggered deepening of the barometric low, which reached a minimum MSL pressure of 1008 hPa (Figure 5
B). The combination of low atmospheric pressures over Greece, with an anticyclonic (high atmospheric pressure) system over central Europe, provoked isobar thickening over the central Balkan Peninsula and, subsequently, enhanced near-surface easterly winds over this area (Figure 5
A,B). At 18:00 UTC, the trough, continuing the slow southeastward movement, was isolated from the western atmospheric circulation and it was finally cut off (Figure 5
C). The cutoff of the trough was critical for the evolution of the barometric low and, subsequently, for the severe storm causing the flash flood. The isolation of middle and upper air from the main western circulation, which can cause the formation of cutoff troughs, is often associated with extreme weather events characterised by heavy rainfall [13
]. The studied cutoff trough preserved its characteristics, like positive vorticity (not shown) and low temperature, offering favourable conditions to the barometric low staying in place for many hours.
On 9 October, during the period from 00:00 UTC to 12:00 UTC, the cutoff trough slowly moved to the Ionian Sea and regional Greece, supporting the displacement of the barometric low to the southern Aegean Sea and the formation of a secondary barometric low over the southern Ionian Sea (Figure 5
D–F). During the same period, the high-pressure barometric system over central Europe was intensified by a relatively warmer ridge (500 hPa), which originated from the western Mediterranean Sea, and as a result, MSL pressure reached 1026 hPa at its centre. Under these atmospheric conditions, high pressures were favoured to expand to the Balkan Peninsula, while low pressures remained dominant over southern Greece. The balance of these adverse conditions pushed the thick-isobar area, characterised by high-pressure gradients and intense easterly winds, to move from the central Balkans to the Aegean Sea. The easterly winds over the Aegean Sea transferred moist and unstable air to the wider area of Pelion mountain over Volos city.
A,B presents the advection of air characterised by specific humidity of 12–14 g kg−1
and convective available potential energy (CAPE) of 800–1200 J kg−1
towards the windward side of Pelion mountain on 9 October at 09:00 UTC. This upwind airflow orographically lifted over the mountain, as shown by the positive vertical velocities reaching 5.2 m s−1
C). The moist and unstable updrafts over Pelion and Volos, as shown by the vertical cross-section illustration of Figure 6
D–F, generated water-rich clouds with specific humidity up to 16.7 g kg−1
. These conditions led us to conclude that the buildup of moisture and CAPE in the updrafts over the precise area for several hours drove to the evolution of the persistent storm. The combination of the orographic convergence of humid–unstable air originated from the Aegean Sea, with the approaching cutoff trough accompanied by positive vorticity and reduced geopotential heights, favoured the enhancement of atmospheric instability, which caused the severe storm.
It is noteworthy that during the period from 06:00 to 18:00 UTC on 9 October, the model estimated precipitation of 213 mm and 446 mm in Volos city and at Pelion mountain peaks, respectively. This amount of precipitation in only 12 h is considered as extreme because it approximately corresponds to the annual accumulated precipitation of Volos city, which varies from 400 to 770 mm, based on data retrieved from the meteorological station of Volos city in the network of the National Observatory of Athens (NOA) with operation period from February 2007 to July 2017. Indeed, the local news talked about a “nonstop” rainfall that lasted from 10:00 local time until the late evening of the same day, creating chaos in the main city of Volos and the surroundings. The authorities characterised the rainfall event as the highest in duration and intensity within the last twelve years and issued the prefecture of Magnesia in an emergency state. The model results indicate an overestimation in comparison with measurements at Fytoko meteorological station (Source: Institute of Industrial Plants and Livestock—Department of Plant Protection, Volos) [10
]. The measured precipitation from 06:00 to 18:00 UTC at Fytoko station was 210 mm [10
]; however, the model estimated about 278 mm in the same area. On the other hand, the model estimated precipitation of 213 mm in Volos city during the same period, presenting a small underestimation in comparison with the measured precipitation of 232 mm [44
]. These differences between spatially averaged model-generated (over a grid area) and the measured point-values are the results of sub-grid scale atmospheric processes during such severe phenomena and the various errors introduced in measurements and simulations.
The results of hydrometeorological simulations of the flash flood event in Volos city are presented in this section. Discharge simulated by the WRF-Hydro model at 01:00, 08:00, 09:00, and 18:00 UTC of 9 October is shown in Figure 7
A–D, superimposed by the corresponding 1-h accumulated precipitation (mm) simulated by the WRF-ARW model. Figure 7
presents different characteristics in the spatial pattern and intensities for the four different times. Figure 7
A presents the beginning of the event at the Pelion mountain slopes, while Figure 7
B,C show the peak of the flash flooding reaching precipitation rates of 84 mm h−1
and discharge of 715 m3
. Moreover, Figure 7
D demonstrates the second peak of the storm, re-triggering the flash flooding during the evening, with precipitation rates exceeding 40 mm h−1
over Xerias drainage basin, while discharge reaching 350 m3
. Afterwards, discharge presents a higher peak of 497 m3
at 20:00 UTC. The differences in maps are attributed to the different mature stages of orographic lifting of warm and humid air masses originating from the Aegean Sea, following the easterly atmospheric flow which was induced by the almost stagnant barometric low. It is noteworthy that the spatiotemporal distribution of precipitation affects the available surface and soil moisture determining surface- and channel-water runoff and, therefore, the streams’ routing. Figure 8
presents the hydrologic response of Xerias watershed for the extreme storm event of 9 October 2006 at the locations (A) R42 (Upper Xerias sub-watershed), (B) R32 (Seskouliotis and Kakaviotis sub-watershed), (C) R21, and (D) OUT (Xerias Watershed outlet). These flood hydrographs were used as an input boundary condition in the 2D HEC-RAS hydraulic–hydrodynamic model for river flood modelling.
demonstrates the maximum discharge values of all examined flood hydrographs used as input to the hydraulic–hydrodynamic model at junctions R42, R32, and R21. In comparison to the other two flood hydrographs (i.e., Clark IUH and Design Hydrograph), WRF-based hydrograph yielded a lower maximum discharge value at R32 and a higher one at R42. This is attributed to the capability of the WRF-ARW weather simulation to physically represent the inhomogeneous pattern of precipitation across the watershed. However, the other two approaches distribute equally the precipitation over the study area. Hence, the spatial differences in precipitation affect the results of the WRF-hydro hydrological simulation, leading to higher discharge values at R42 due to heavy rainfall at Pelion mountain and lower discharge values at R32 which was partly covered by the storm.
3.2. Hydraulic Simulation Results
presents the maximum water depths and the flood extent simulated by the 2D flood inundation model using three different boundary conditions for each junction point (R42, R32, R21) of Xerias river: (A) WRF hydrograph, (B) Clark IUH, and (C) design hydrograph. Τhe spatial distribution of the simulated flood extent reveals that the results derived from the WRF hydrograph coincide with the spatial pattern obtained using the Clark IUH. Both simulations depict major flooding in the areas expanding west of Volos Port along Xerias torrent, indicating particularly high local values within Agii Anargiri, Neapoli, and Pedion Areos areas (framed areas in Figure 9
A,B). However, this is not the case for the design hydrograph, which fails to capture localised flooding, especially in Agii Anargiri area (Figure 9
C). Design hydrograph was not generated for the specific event but for a designed flood of 100 years and, as a result, the simulated maximum discharge value in junction R42 is lower than the maximum flow peaks of the other two hydrographs (i.e., WRF hydrograph and Clark IUH). The aggregated scores derived from the validation process for the three flood hydrographs are detailed in Table 4
. In general, all of the study simulations manage to provide acceptable scores based on the percentage of validation points that fall within the simulated flood extent. WRF hydrograph produces the highest success rate score with 84% of the validation points to be observed within the simulated flood extent. As expected, the WRF physically-based approach provides the best correspondence with the observed flooded area and, therefore, its results are considered as more “accurate” than the ones of the other methods. These results will be further discussed and compared with the qualitative data of Section 2.2
. Nevertheless, it is important to point out that despite the difference between the WRF hydrograph and the Clark IUH validation scores, Clark IUH achieves high success rate scores as well (79%).
The cumulative distribution functions of the simulated water depth calculated for the three flooded areas, separately, show that WRF and Clark IUH simulations converge to maximum water depths of about 2.2 m (Figure 10
). Indeed, qualitative analysis of the local press indicated cases in which floodwaters reached large values, around or even higher than 2 m, creating major catastrophes to the infrastructure of Agii Anargiri and Neapoli areas. Characteristically, mass-media and press reports mentioned that residents and business owners in Agii Anargiri and Neapoli were desperate, trying to clean their property from mud and floodwaters, which sometimes exceeded 1.5 m until the next evening on 10 October. Based on the WRF-based simulations, the median values of the maximum water depths were estimated as 0.43 m, 0.67 m, and 0.63 m in Agii Anargiri, Neapoli, and Pedion Areos, respectively (Figure 10
). U.S. Federal Emergency Management Agency (FEMA) estimates that 0.3 m of flash flood water could float many vehicles, whereas 0.6 m of rushing water could carry away most vehicles, including SUVs and pickups. In Neapoli and Pedion Areos, the fire services undertook multiple emergency calls from 21 recorded flooded roads. Although flood velocity is the primary determinant for incidents to motorists, the simulated floodplain gives a strong indication of the vulnerable roads, which concentrated a mass of validation points. In Figure 11
, it is apparent that the incidents reported by the Fire Department shape the dense road network of the commercial area (green points in Figure 11
), and they are indicative of the high exposure of citizens during the active commercial hours of a weekday that the event occurred. In addition to that, flooding caused the roads of Volos to get overloaded with private and public vehicles, as well as hundreds of taxis that attempted to drive people to their relatives and properties. In times of crisis, people consider their place as a safe destination where “nothing bad can happen” or as the reference point where the family gathers for a collective response in case of emergency [76
]. There were many cases in the past in which individuals lost their lives in their effort to reach a destination like home or while attempting to retrieve their belongings and pets from flooded buildings [77
In the flood event of 9 October, the Traffic Police of Volos struggled to deal with dozens of stranded cars and to coordinate the traffic jams across the city. Unfortunately, the lack of detailed flood records on the validation points does not allow for a thorough comparison with the simulated water depths on the road network. A rough estimation of the flood water depth was possible only to certain locations which were detailed in the press or to which relevant pictures could be assigned based on the knowledge of the authors. From the stranded cars pictured on the green dots in Figure 11
, the authors estimate water depth around 0.5 m in Larisis Street, which crosses Neapoli and Agii Anargiri areas. This is to be compared with the 0.8 m extracted as the median value of the simulated maximum depths in the validation polygon of Larisis street (Figure 11
). Traffic also increased as municipality services started to transfer students from flooded schools to their homes late in the morning on Monday 9 October. Many schools in central Volos City and the Prefecture of Magnesia were evacuated and remained closed the next day. The authors managed to associate a picture collected in the validation data with the location of a flooded school in Neapoli (4th high school of Volos city) to get an indicative value of the observed water depth. WRF simulations present a median value of 0.78 m that is slightly higher than the 0.4–0.5 m assessed from the corresponding image (see Figure 11
). Slight overestimations could be attributed to the fact that the observed water depth data do not correspond to the maximum levels reached in the current event because of the differences between the timing of the simulated water depth and the actual timing that the photo was taken. It should be noted that flash floods are characterised by a sharp rise followed by a relatively rapid decline, causing high flow velocities. In flash floods, discharge quickly reaches a maximum level and diminishes almost as rapidly [79
The validation points also include a bunch of emergency calls (more than three hundred) from residents who asked for help in flooded homes and stores from 07:00 UTC (10:00 local time) and after on 9 October, and especially after noon in Neapoli, following the hydrologic peaks presented in the previous section. However, there are no witnesses about the depth of floodwaters in these locations to assess the average situation in the flooded areas. In the commercial area that extends north of Neapoli to Pedion Areos, the media refer to floodwaters reaching 1.5 m and immobilising owners inside their stores. In accordance with this statement, the WRF simulations give a median of 1.21 m in the validation polygon that corresponds to the main commercial area affected by the floodwaters. Notably, 80% of the companies that claimed compensation for flood damages to the Greek government are located in this polygon (see validation polygons in Figure 11
). One of the largest printing companies, which was highly affected by the flood in this area, serves as an additional validation point (Figure 11
). The press highlighted that the building was completely destroyed, with floodwaters exceeding 1.5 m. WRF-based simulations provide a good approximation in this spot, with a median water depth of 1.68 m in the validation polygon that surrounds the structure (see Printing company in Figure 11
). The model gives particularly high values in some locations in Upper Volos. In that area, the iron railway bridge at Xerias intersection collapsed at 13:00 UTC (16:00 local time) due to the significant volume of floodwater from Xerias torrent and the intense debris flow hitting its foundations (bridge shown in Figure 11
). WRF-simulated water depths reached maximum values as high as 4 m at the flood peak around the bridge. Although interviewed locals mentioned that floodwaters overcame the bridge, the local press reported about 2 m of water, and it is probably a more reliable source of information. An overestimation of this value could be related to the exceptional conditions generated by the debris flow (i.e., carried cobbles, boulders, olive trees, and mud) at the preceding bridge, leading to lower water depths with higher flow velocities at the location of the railway bridge. Despite the unavoidable uncertainties on the exact values, it appears that the simulations provide valuable indications about the harshness of the situation. As a result of the catastrophic event, the national railway was interrupted, disconnecting Volos from the rest of the country for many weeks after the event. Given that the event occurred during the morning rush hours, it is fortunate that, after the intervention of local residents, authorities discontinued the train schedule between Larisa and Volos just before the bridge collapsed. This measure prevented possible accidents and losses of human lives.