A Challenging Tornado Forecast in Slovakia

: An F1 tornado hit the village of Lek á rovce in eastern Slovakia on the afternoon of 3 October 2018. The tornado, which occurred outside the main convective season in Slovakia, was not anticipated by the meteorologists of the Slovak Hydrometeorological Institute. The models available to the forecasters simulated an environment of marginal convective available potential energy (CAPE) and weakening vertical wind shear. This paper addresses forecasting challenges associated with events related to a tornado threat. To investigate conditions before tornado formation, observational datasets, including sounding, and vertical-azimuth display (VAD) data from a radar station and surface stations were used. Hodographs based on observational data and a higher-resolution run of the limited-area model showed stronger lower tropospheric shear than was formerly anticipated over the area of interest. The higher-resolution model was able to better represent the modiﬁcation of the lower tropospheric ﬂow by a mountain chain, which was crucial to maintaining the strong lower tropospheric shear in the early afternoon hours before the tornado’s occurrence. We discuss the importance of using both observational datasets and higher-resolution modeling in the simulation of lower tropospheric wind proﬁles, which a ﬀ ect the lower tropospheric storm relative helicity as one of the key ingredients in mesocyclonic tornadogenesis.


Introduction
Tornado forecasting has its roots in 1948, when Fawbush and Miller issued the first tornado forecast for Tinker Airbase in Oklahoma, United States. Since then, the science of tornado forecasting has advanced considerably, as documented by Brooks et al. (2019) [1]. Currently, tornadoes are forecast using the knowledge of the conditions, or so-called ingredients, required for their formation. Ingredient-based methodology is also used for forecasting of deep moist convection in general. Sufficient lower tropospheric moisture, conditionally unstable stratification, and a lift mechanism are required to form a convective storm [2,3].
In order to develop rotation in a storm (supercell), the presence of strong vertical wind shear in the deep layer of the troposphere is needed. The importance of vertical wind shear for mesocyclogenesis has been understood for decades, and it has been demonstrated both using environmental observations made in proximity to supercells [4,5] and through idealized numerical simulations [6]. A measure of the degree of streamwise vorticity in the inflow to the thunderstorm, storm relative helicity [7], is also used as a supercell predictor [8].
However, the majority of supercells never produce any tornadoes [9]. Scientists have used field experiments to gather more information on the differences between tornadic and nontornadic 1.
Reconstruct the environment of the parent storm using observational data and compare it to the Aire Limitée Adaptation Dynamique Développement InterNational (ALADIN) model made available to forecasters several hours before the event. 2.
Use a higher-resolution run of the ALADIN system to simulate the preconvective environment. 3.
Find out whether observational data or the high-resolution run of the model would help forecasters to recognize favorable conditions for tornadogenesis using the most recent knowledge on the topic.

Methodology
To investigate synoptic-scale weather conditions, we used the European Center for Medium-Range Weather Forecast (ECMWF) operational global hydrostatic numerical weather prediction (NWP) model with a horizontal grid spacing of 0.1 • . For smaller scales, we used local-area NWP models: operational model ALADIN of the Slovenský Hydrometeorologický Ústav (SHMU) (OPM) and experimental local-area model ALADIN/ELAM (ELAM). OPM is a hydrostatic model with convection parametrization, and ELAM is a nonhydrostatic high-resolution model computed as dynamic downscaling from OPM on smaller domains. The 3MT scheme [40] was used for the parameterization of deep convection in the cases of both OPM and ELAM. This scheme proved to work well with partially resolved deep convection in 1-2.5 km grid spacing [41]. We did not experiment with changes in any parameterization schemes when switching between OPM and ELAM. ELAM has higher horizontal and vertical resolution, which results in better topographical descriptions. Both models were utilized with high-performance computers (HPC) at the Slovak Hydrometeorological Institute [42]. The technical settings of the NWP models are summarized in the Table 1, and their domains are shown in Figure 1a. More detailed descriptions can be found in Termonia et al. (2018) [43]. Forecasters did not have the ELAM output available.

Prestorm Environment According to Numerical Models Available to Forecasters
Environmental conditions are discussed on the basis of NWP model data available to the morning shift at the local hydrometeorological service, namely, the 00 UTC run of ECMWF and the 06 UTC run of the local-area NWP OPM.
Slovakia was on the southeastern edge of a low-pressure system ( Figure 2) that moved from the Baltic Sea towards Belarus. In the prevailing northwesterly flow, polar maritime airmass was  In order to reconstruct the hodograph for the time of tornadogenesis, we used observational data from synoptic stations, the radar station from Kojšovská hol'a mountain [45], and the sounding station from Gánovce ( Figure 1b). Local solar time in eastern Slovakia is approximately +1 h compared to Coordinated Universal Time (UTC).

Prestorm Environment According to Numerical Models Available to Forecasters
Environmental conditions are discussed on the basis of NWP model data available to the morning shift at the local hydrometeorological service, namely, the 00 UTC run of ECMWF and the 06 UTC run of the local-area NWP OPM.
Slovakia was on the southeastern edge of a low-pressure system ( Figure 2) that moved from the Baltic Sea towards Belarus. In the prevailing northwesterly flow, polar maritime airmass was advected over central Europe. The airmass was characterized by a lapse rate of 6 to 7 • C·km −1 and surface dew point of 7-10 • C (Figure 3a). This allowed for marginal CAPE values of up to 500 J·kg −1 ( Figure 3b) and a chance for deep moist convection to develop, as the lift was provided by a convergence zone in a surface-pressure trough. The best overlap of conditionally unstable lapse rates and sufficient lower tropospheric moisture, along with the highest CAPE values, was simulated in the afternoon hours over eastern Slovakia. Vertical wind shear decreased over the morning and into the afternoon hours, and 500 hPa surface-layer bulk values (deep-layer shear, DLS) dropped from 40 to 20 m·s −1 , which would still be sufficient for well-organized convection, including supercells [5].
Forecast soundings over eastern Slovakia at 10:00 and 12:00 UTC on Figure 4 showed that the steepest lapse rates were in the lowest 3 km, strengthening at noon with an equilibrium level of 5.5 km, suggesting low-topped convection. While DLS remained strong throughout the day, storm relative helicity in the 0-1 km layer decreased from 60-100 m 2 s −2 in the morning to 10-50 m 2 s −2 in this timeframe. The decrease was due to the veering of surface-wind change from the southern to the western direction ( Figure 5). This would suggest that the environment had become less conducive to tornadogenesis [46]. Meteorologists considered the overall severe convective storm threat to be low, given the marginal values of CAPE and only weak vertical wind shear in the lower troposphere.
( Figure 3b) and a chance for deep moist convection to develop, as the lift was provided by a convergence zone in a surface-pressure trough. The best overlap of conditionally unstable lapse rates and sufficient lower tropospheric moisture, along with the highest CAPE values, was simulated in the afternoon hours over eastern Slovakia. Vertical wind shear decreased over the morning and into the afternoon hours, and 500 hPa surface-layer bulk values (deep-layer shear, DLS) dropped from 40 to 20 m . s −1 , which would still be sufficient for well-organized convection, including supercells [5].   Forecast soundings over eastern Slovakia at 10:00 and 12:00 UTC on Figure 4 showed that the steepest lapse rates were in the lowest 3 km, strengthening at noon with an equilibrium level of 5.5 km, suggesting low-topped convection. While DLS remained strong throughout the day, storm relative helicity in the 0-1 km layer decreased from 60-100 m 2 s -2 in the morning to 10-50 m 2 s -2 in this timeframe. The decrease was due to the veering of surface-wind change from the southern to the western direction ( Figure 5). This would suggest that the environment had become less conducive to tornadogenesis [46]. Meteorologists considered the overall severe convective storm threat to be low, given the marginal values of CAPE and only weak vertical wind shear in the lower troposphere. steepest lapse rates were in the lowest 3 km, strengthening at noon with an equilibrium level of 5.5 km, suggesting low-topped convection. While DLS remained strong throughout the day, storm relative helicity in the 0-1 km layer decreased from 60-100 m 2 s -2 in the morning to 10-50 m 2 s -2 in this timeframe. The decrease was due to the veering of surface-wind change from the southern to the western direction ( Figure 5). This would suggest that the environment had become less conducive to tornadogenesis [46]. Meteorologists considered the overall severe convective storm threat to be low, given the marginal values of CAPE and only weak vertical wind shear in the lower troposphere.

Observational Data
Different observational datasets that could be used to infer the degree of vertical wind shear were available to the forecasters before the time of the tornado. These included radar data, surfacewind observations, and radiosonde data from the Gánovce station.
Before the tornado, a parent storm evolved from a small cluster of showers that formed shortly after 12 UTC. The storm attained supercell characteristics at 13:40 UTC, marked by the presence of a mesocyclone, deviant movement to the right, and the presence of both a bounded weak echo region (BWER) and a hook echo ( Figure 6). Supercell characteristics persisted till 15:00 UTC. The tornado occurred at 14:30 UTC in the Lekárovce village on the Slovak-Ukrainian border and lasted for around 10 min [38]. As the storm strengthened after 13:00 UTC, the present paper focuses on the available observations from around this period.

Observational Data
Different observational datasets that could be used to infer the degree of vertical wind shear were available to the forecasters before the time of the tornado. These included radar data, surface-wind observations, and radiosonde data from the Gánovce station.
Before the tornado, a parent storm evolved from a small cluster of showers that formed shortly after 12 UTC. The storm attained supercell characteristics at 13:40 UTC, marked by the presence of a mesocyclone, deviant movement to the right, and the presence of both a bounded weak echo region (BWER) and a hook echo ( Figure 6). Supercell characteristics persisted till 15:00 UTC. The tornado occurred at 14:30 UTC in the Lekárovce village on the Slovak-Ukrainian border and lasted for around Atmosphere 2020, 11, 821 7 of 17 10 min [38]. As the storm strengthened after 13:00 UTC, the present paper focuses on the available observations from around this period.
Surface observations available from the early afternoon hours showed a discrepancy between model simulation and reality. While the model predicted a westerly wind over southeastern Slovakia and western Ukraine at 12:00 UTC (Figure 5b), surface observations revealed the presence of an easterly flow, reaching up to 5 m·s −1 ahead of a shallow trough (Figures 7 and 8). Furthermore, a southerly wind, instead of a westerly wind as simulated by the models, was observed over the northern part of eastern Slovakia. On the basis of this information, a forecaster could have expected stronger vertical wind shear than what was forecast, given the easterly flow veering to strong westerlies in the mid-to-upper troposphere.
In order to obtain information about the wind above the surface and to recreate a hodograph of the vertical wind profile before the tornado, we also investigated sounding observations from Ganovce at 12:00 UTC (Figure 9), and the vertical-azimuth display (VAD) from the radar station located at Kojšovská hol'a (not shown). Surface observations available from the early afternoon hours showed a discrepancy between model simulation and reality. While the model predicted a westerly wind over southeastern Slovakia and western Ukraine at 12:00 UTC (Figure 5b), surface observations revealed the presence of an easterly flow, reaching up to 5 m . s -1 ahead of a shallow trough (Figures 7 and 8). Furthermore, a southerly wind, instead of a westerly wind as simulated by the models, was observed over the northern part of eastern Slovakia. On the basis of this information, a forecaster could have expected stronger vertical wind shear than what was forecast, given the easterly flow veering to strong westerlies in the mid-to-upper troposphere. Radiosounding, like the model forecast (Figure 4), measured a strong westerly flow, increasing to 50 m·s −1 at 400 hPa. While the wind at the mid-to-upper troposphere may be representative of the environment in which the tornado occurred, about 120 km southeast of the sounding site, its representativeness in the lower altitudes is questionable. First, sounding was launched in the rear side of the trough, with surface flow already having veered to being westerly forecast ( Figure 4). Second, the site is located at a higher altitude (706 m) than the tornado location (108 m). Therefore, a wind profile from the lower levels from sounding was not useful for the investigation of the vertical wind profile in the tornado location. and western Ukraine at 12:00 UTC (Figure 5b), surface observations revealed the presence of an easterly flow, reaching up to 5 m . s -1 ahead of a shallow trough (Figures 7 and 8). Furthermore, a southerly wind, instead of a westerly wind as simulated by the models, was observed over the northern part of eastern Slovakia. On the basis of this information, a forecaster could have expected stronger vertical wind shear than what was forecast, given the easterly flow veering to strong westerlies in the mid-to-upper troposphere.   In order to obtain information about the wind above the surface and to recreate a hodograph of the vertical wind profile before the tornado, we also investigated sounding observations from Ganovce at 12:00 UTC (Figure 9), and the vertical-azimuth display (VAD) from the radar station located at Kojšovská hoľa (not shown).
Radiosounding, like the model forecast (Figure 4), measured a strong westerly flow, increasing to 50 m . s -1 at 400 hPa. While the wind at the mid-to-upper troposphere may be representative of the environment in which the tornado occurred, about 120 km southeast of the sounding site, its representativeness in the lower altitudes is questionable. First, sounding was launched in the rear side of the trough, with surface flow already having veered to being westerly forecast (Figure 4). Second, the site is located at a higher altitude (706 m) than the tornado location (108 m). Therefore, a wind profile from the lower levels from sounding was not useful for the investigation of the vertical  [47]. Using VAD instead of the sounding data in the aforementioned layer yielded a more precise calculation of a storm-motion vector ( Figure 10). VAD measurements were also closer to the tornado location than the sounding. Thus, we considered the VAD to be more representative of the wind profile over the tornado location than sounding. Sounding data were used to infer the vertical wind profile for the altitude above 4 km, where VAD data were not available. Using either sounding or VAD still left us without any knowledge regarding the wind profile in the crucial layer of the bottom 1 km. To fill this gap, we used the measurements from the stations of Michalovce, Vysoká and Uhom, and Zlatá Baňa and Kojšovská hoľa (their locations are shown in Figure 1b), which represented altitudes from 100 to 1200 m. These observations were used to represent the wind field at various heights as if these wind observations were taken over the tornado location. In order to reconstruct the conditions that would be the most conducive to tornadogenesis, we took into consideration the maximal observed wind speeds within 1 h of the change in wind direction associated with the passage of the surface trough. Thus, a 13:27 measurement was taken from Michalovce, 14:14 from Vysoká and Uhom, and 11:22 UTC from Zlatá Baňa. The VAD profile from Kojšovská hoľa was considered only for the time at which a complete wind profile was available to 4 km, which was at 12:50 UTC. While this may be the most accurate way to infer the wind profile on the basis of the limited availability of observational datasets, it is still likely an imperfect representation of the true wind profile near the tornado location. One of the primary limitations is the effect of friction on wind in the surface stations that were used to represent the wind vector that would be hundreds of meters above ground at the tornado location.
Because of the distances between individual weather stations to the tornado site, and the fact that they represent the wind field under the influence of friction, a reconstructed wind profile offers only a rough estimation of reality. Nevertheless, comparing the forecast hodograph using OPM to the reconstructed hodograph clearly showed that the model underestimated the lower tropospheric wind shear ( Figure 10). The forecast wind profile would result in about 6 m . s -1 of bulk shear and 40 m 2 s -2 of storm relative helicity (SRH) in the 0-1 km layer, compared to 12 m . s -1 and 175 m 2 s -2 in the reconstructed profile. Shear in the 0-3 km layer was also substantially stronger in the reconstructed wind profile, with SRH reaching 285 m 2 s -2 , in contrast to 110 m 2 s -2 based on the forecast profile. Such values have been associated with tornadoes both in the U.S. [5,46,48] and Europe [16,17,49]. "Kink" in the lowest 500 m of the reconstructed hodograph also suggested almost purely streamwise vorticity in the inflow to the storm in that layer, a condition that was deemed favorable for VAD measurements were available from the Kojšovská hol'a radar site, starting at an altitude of 1.2 km a.s.l. and about 85 km west-northwest of the tornado location, thus being closer than the sounding measurement. The only complete measurement for the altitude of 1.2 to 4 km was available at 12:50 UTC, with a marked increase in westerly wind, though the increase was not as significant as in the case of the Ganovce sounding station. Data above 4 km were not available from VAD. To decide whether to use sounding or VAD to represent the vertical wind profile between 1.2 and 4 km, we compared the right-moving storm-motion vector calculation using both datasets with the real observed storm motion ( Figure 10). To consider the low-topped nature of the storm on the basis of the forecast ( Figure 4) and observed height (Figure 6c,d and Figure 9) of the equilibrium level, which was close to 5.5 km, we used a 0-4 km mean wind and 0-4 km shear vector when applying the Bunkers et al. (2000) method [47]. Using VAD instead of the sounding data in the aforementioned layer yielded a more precise calculation of a storm-motion vector ( Figure 10). VAD measurements were also closer to the tornado location than the sounding. Thus, we considered the VAD to be more representative of the wind profile over the tornado location than sounding. Sounding data were used to infer the vertical wind profile for the altitude above 4 km, where VAD data were not available.
Using either sounding or VAD still left us without any knowledge regarding the wind profile in the crucial layer of the bottom 1 km. To fill this gap, we used the measurements from the stations of Michalovce, Vysoká and Uhom, and Zlatá Baňa and Kojšovská hol'a (their locations are shown in Figure 1b), which represented altitudes from 100 to 1200 m. These observations were used to represent the wind field at various heights as if these wind observations were taken over the tornado location. In order to reconstruct the conditions that would be the most conducive to tornadogenesis, we took into consideration the maximal observed wind speeds within 1 h of the change in wind direction associated with the passage of the surface trough. Thus, a 13:27 measurement was taken from Michalovce, 14:14 from Vysoká and Uhom, and 11:22 UTC from Zlatá Baňa. The VAD profile from Kojšovská hol'a was considered only for the time at which a complete wind profile was available to 4 km, which was at 12:50 UTC. While this may be the most accurate way to infer the wind profile on the basis of the limited availability of observational datasets, it is still likely an imperfect representation of the true wind profile near the tornado location. One of the primary limitations is the effect of friction on wind in the surface stations that were used to represent the wind vector that would be hundreds of meters above ground at the tornado location.
Atmosphere 2020, 11, x FOR PEER REVIEW 10 of 16 tornadogenesis by Coffer et al. (2017) [20]. Using an observational dataset in lieu of model simulation in this case would allow forecasters to recognize a higher-than-expected tornado threat.

High-Resolution Experimental Local-Area Model (ELAM)
A nonhydrostatic, 1 km grid-spaced run of an ELAM-simulated environment was considerably more conducive to tornadogenesis than the OPM 4.5 km run. The spatial distribution of SRH 0-1 km ( Figure 11) showed local maxima in the vicinity of the town of Michalovce with values up to 230 m 2 s -2 , compared to 10-40 m 2 s -2 from the OPM run and 175 m 2 s -2 from the reconstructed hodograph ( Figure  10). The 0-1 km bulk shear was also much stronger in the ELAM run (16 m . s -1 compared to 6 m . s -1 in the OPM run). However, the simulated SRH maximum was not precisely collocated with the town of Lekárovce, where the ELAM simulated around 80 m 2 s -2 compared to 50 m 2 s -2 in OPM. Differences in the SRH and 0-1 bulk shear between OPM and ELAM runs were due to the treatment of the lower tropospheric wind field. OPM simulated a straight hodograph with a westerly wind at the surface, veering to a northwesterly direction with height. ELAM simulated a curved hodograph with eastsoutheasterly flow at the surface, veering through southerly directions to a west-southwesterly flow in the bottom 1 km.  Because of the distances between individual weather stations to the tornado site, and the fact that they represent the wind field under the influence of friction, a reconstructed wind profile offers only a rough estimation of reality. Nevertheless, comparing the forecast hodograph using OPM to the reconstructed hodograph clearly showed that the model underestimated the lower tropospheric wind shear ( Figure 10). The forecast wind profile would result in about 6 m·s −1 of bulk shear and 40 m 2 s −2 of storm relative helicity (SRH) in the 0-1 km layer, compared to 12 m·s −1 and 175 m 2 s −2 in the reconstructed profile. Shear in the 0-3 km layer was also substantially stronger in the reconstructed wind profile, with SRH reaching 285 m 2 s −2 , in contrast to 110 m 2 s −2 based on the forecast profile. Such values have been associated with tornadoes both in the U.S. [5,46,48] and Europe [16,17,49]. "Kink" in the lowest 500 m of the reconstructed hodograph also suggested almost purely streamwise vorticity in the inflow to the storm in that layer, a condition that was deemed favorable for tornadogenesis by Coffer et al. (2017) [20]. Using an observational dataset in lieu of model simulation in this case would allow forecasters to recognize a higher-than-expected tornado threat.

High-Resolution Experimental Local-Area Model (ELAM)
A nonhydrostatic, 1 km grid-spaced run of an ELAM-simulated environment was considerably more conducive to tornadogenesis than the OPM 4.5 km run. The spatial distribution of SRH 0-1 km ( Figure 11) showed local maxima in the vicinity of the town of Michalovce with values up to 230 m 2 s −2 , compared to 10-40 m 2 s −2 from the OPM run and 175 m 2 s −2 from the reconstructed hodograph ( Figure 10). The 0-1 km bulk shear was also much stronger in the ELAM run (16 m·s −1 compared to 6 m·s −1 in the OPM run). However, the simulated SRH maximum was not precisely collocated with the town of Lekárovce, where the ELAM simulated around 80 m 2 s −2 compared to 50 m 2 s −2 in OPM. Differences in the SRH and 0-1 bulk shear between OPM and ELAM runs were due to the treatment of the lower tropospheric wind field. OPM simulated a straight hodograph with a westerly wind at the surface, veering to a northwesterly direction with height. ELAM simulated a curved hodograph with east-southeasterly flow at the surface, veering through southerly directions to a west-southwesterly flow in the bottom 1 km.
of Lekárovce, where the ELAM simulated around 80 m 2 s -2 compared to 50 m 2 s -2 in OPM. Differences in the SRH and 0-1 bulk shear between OPM and ELAM runs were due to the treatment of the lower tropospheric wind field. OPM simulated a straight hodograph with a westerly wind at the surface, veering to a northwesterly direction with height. ELAM simulated a curved hodograph with eastsoutheasterly flow at the surface, veering through southerly directions to a west-southwesterly flow in the bottom 1 km.  The reconstructed hodograph resembled the one simulated by ELAM in the bottom 500 m, but did not show southerly wind directions between 500 m and 1.2 km ( Figure 10). Instead, the wind immediately turned to the west-southwest. This may have been because there was a large gap in observational data between 585 m and 1.2 km. It is likely that if such data had been available, the reconstructed hodograph would have involved more curvature and a higher SRH. Furthermore, weather-station data were representative of wind 10 m above the ground, strongly affected by surface friction. In reality, altitudes of 585 m and 1.2 km would be hundreds of meters above ground at the tornado location, which was at an altitude of 108 m a.s.l.
Differences in the wind field in the lower troposphere between OPM and ELAM forecasts can be attributed to differences in the simulation of the surface trough, and its shape and movement ( Figure 12). ELAM simulated the slower movement of the surface trough, as it was located more to the west at both 11:00 and 14:00 UTC than in the OPM simulation. At 14:00 UTC, Lekárovce was located north of the small-scale low in the ELAM run, yielding a southeasterly flow, but west of the small-scale low in the OPM run, yielding a westerly flow at the surface. Besides the difference in the trough movement speed, its shape was also distinctive. Smaller-scale troughs and ridges were better identifiable in the ELAM run. These features were also identifiable near the tornado location, with a trough over the lowland and a ridge over the Vihorlat mountain chain. OPM had already forecast a change in the surface-wind direction between 11:00 and 12:00 UTC, while ELAM predicted the change to happen between 13:00 and 14:00 UTC ( Figure 13).
ELAM's predicted change of wind shift between 13:00 and 14:00 UTC was better than that of OPM, but also not completely accurate. Surface observations showed an easterly flow persisting till 15:00 UTC over the tornado area and covering a larger area (Figure 8) than that simulated by the model. Directly over Lekárovce, ELAM simulated a southwesterly surface flow, in contrast to the easterly flow in the surface observations. Thus, it is likely that the SRH bullseye, simulated only to the immediate south of Vihorlat, expanded further to the southeast, covering Lekárovce as well. Nevertheless, the ELAM prediction better represented an environment supportive of tornadoes compared to that of OPM. In the case of operational usage, forecasters could be more aware of the significant lower tropospheric shear over the Slovakia/Ukraine border. small-scale low in the OPM run, yielding a westerly flow at the surface. Besides the difference in the trough movement speed, its shape was also distinctive. Smaller-scale troughs and ridges were better identifiable in the ELAM run. These features were also identifiable near the tornado location, with a trough over the lowland and a ridge over the Vihorlat mountain chain. OPM had already forecast a change in the surface-wind direction between 11:00 and 12:00 UTC, while ELAM predicted the change to happen between 13:00 and 14:00 UTC ( Figure 13).   ELAM's predicted change of wind shift between 13:00 and 14:00 UTC was better than that of OPM, but also not completely accurate. Surface observations showed an easterly flow persisting till 15:00 UTC over the tornado area and covering a larger area (Figure 8) than that simulated by the model. Directly over Lekárovce, ELAM simulated a southwesterly surface flow, in contrast to the easterly flow in the surface observations. Thus, it is likely that the SRH bullseye, simulated only to the immediate south of Vihorlat, expanded further to the southeast, covering Lekárovce as well. Nevertheless, the ELAM prediction better represented an environment supportive of tornadoes compared to that of OPM. In the case of operational usage, forecasters could be more aware of the

Discussion and Conclusions
On 10 October 2018, a tornado hit Lekárovce village in eastern Slovakia, causing F1 damage to numerous structures. While tornadoes are rare in Slovakia, they do occur, as demonstrated by this and some previous cases [36,[50][51][52][53]. Therefore, operational forecasters need to be able to identify environments conducive to tornadogenesis using both NWP models and observational datasets. This paper concentrated on discussing the possibility of identifying such an environment for the Lekárovce case using the NWP models and observational datasets available to forecasters at the time of the tornado, as well as using a higher-resolution run of the NWP model.
The operational NWP ALADIN/SHMU model forecast a collocation of favorable ingredients for deep-moist convection in the afternoon hours over eastern Slovakia at the time, when the degree of lower tropospheric shear was supposed to markedly decrease since the morning hours. However, the reconstructed hodograph from the available observational data showed stronger lower tropospheric shear, approaching magnitudes commonly found with tornadoes in the literature. Thus, it is crucial for the forecasters to utilize these observational datasets in conjunction with NWP models in order to identify tornado threats associated with storms forming over eastern Slovakia.
Experimental NWP ALADIN/ELAM yielded even stronger vertical wind shear than from the reconstructed hodograph in the lower troposphere over the area of interest. The main difference with respect to the OPM was the slower eastward progression of the surface trough through eastern Slovakia. Lekárovce remained on the forward flank of the trough, with southerly-easterly lower tropospheric flow until the early afternoon hours, when the storms formed and utilized the highly sheared environment. Furthermore, ELAM was able to better resolve smaller-scale troughs and ridges. The easterly direction of the surface wind can be explained by a ridge to the north and a trough to the south of the Vihorlat mountain chain that were not present in the OPM run. The easterly wind's changing to southerly and then northwesterly with height yielded a long and curved hodograph with large amounts of streamwise vorticity in the ELAM run, in contrast to the shorter and straight hodograph with pure crosswise vorticity in the OPM run. This could have been caused by orography differences between the 4.5 km OPM and 1 km ELAM ( Figure 14). Using both the higher-resolution model and the observational data would have provided forecasters with enough clues regarding the environment that was prone to tornadoes on the given day. The importance of using observational datasets and high-resolution runs of NWP models for addressing flow modifications and subsequent shear enhancement by local topography was realized in other works, such as Bosart et al.  wind's changing to southerly and then northwesterly with height yielded a long and curved hodograph with large amounts of streamwise vorticity in the ELAM run, in contrast to the shorter and straight hodograph with pure crosswise vorticity in the OPM run. This could have been caused by orography differences between the 4.5 km OPM and 1 km ELAM ( Figure 14). Using both the higher-resolution model and the observational data would have provided forecasters with enough clues regarding the environment that was prone to tornadoes on the given day. The importance of using observational datasets and high-resolution runs of NWP models for addressing flow modifications and subsequent shear enhancement by local topography was realized in other works, such as Bosart et al.  [59] reported overall improvement in the model forecast of tornadic supercells over Poland with both increasing horizontal resolution and the assimilation of surface observations and sounding data. Analysis of more cases of severe convection in the area is necessary to draw general conclusions on the influence of local orography. Nevertheless, this case demonstrated the need for higherresolution modeling of lower tropospheric flow, and for attention to be paid to observational data in situations where environments conducive to severe weather are confined to a rather small area and possibly influenced by topography.
In addition to the operational model discussed in the publication, a convection-allowing model was operationally implemented in the Slovak Hydrometeorological Institute with the same settings as were used for the ELAM (Table 1), but with grid spacing of 2 km. With the purchase of a new supercomputer, it is planned that the model resolution and the size of its forecast domain will be further increased.