1 Introduction

Sea effect snow / Lake effect snow (SES/LES) is a complex meteorological process involving the interaction between warm water bodies and cold atmospheric conditions, which leads to the formation of snow clouds and heavy snowfall downwind of the water body (Wiggin 1950). The term “LES” refers to the occurrence of convective boundary layer and cloud growth resulting from the input of heat and moisture from the Great Lakes, as described by previous works such as Lenschow (1973) and Kristovich (1993). Large bodies of relatively warm water, such as lakes, can induce convective instability in cold air masses such as Arctic and Continental Airmasses (Niziol et al. 1995). Large amounts of heat and moisture are exchanged between the water surface and the air, creating turbulent fluxes. This leads to the development of shallow convection, which induces intense and localized convective precipitation. This precipitation can take the form of snowbands that are carried towards the coast by the prevailing winds (Mazon et al. 2015a, b). It was discovered that the clouds tended to occur when the 850 hectopascal (hPa) temperature was at least 13 °C colder than the lake surface temperature, and were often organized into parallel bands that resembled cloud streets but with larger dimensions (Holroyd 1971). Also, the location and movement of these bands may not directly be controlled by surface conditions, but rather by winds aloft. This phenomenon is influenced by various factors, such as wind direction and temperature gradients (Peace et al. 1966). The importance of steering winds in determining the location of the axis of heaviest snowfall was demonstrated in a study conducted by Jiusto and Kaplan (1972), with a particularly strong role played by the gradient-level wind direction, revealing that the path and intensity of snowstorms can be significantly influenced by winds at different altitudes. Climate change is likely to have a significant impact on lake-effect snow patterns, with the frequency of heavy snow events expected to decrease in some areas. However, the impacts of climate change can be greatly influenced by subtle regional differences, and the exact effects may vary depending on the specific location and climate conditions (Kunkel et al. 2000). For example, evidenced by warmer Great Lakes surface waters and decreased ice cover, may be contributing to an increasing trend in lake-effect snowfall in areas downwind of the Laurentian Great Lakes (Burnett et al. 2003).

In recent years, the significance of the Black Sea snow-effect has been increasingly acknowledged as a severe winter weather phenomenon. This is due to the unexpected occurrence of massive amounts of snow over highly populated areas along the southwestern coast of the Black Sea during Sea snow-effect events (Kindap 2010). Its effects go beyond hazardous driving conditions, leading to transportation disruptions, school closures, and power outages. Furthermore, the accumulation of thick snow can result in property damage, roof collapses, and tree limbs breaking due to the weight of the snow. In 17–19 February Istanbul Sea-Effect Snow (SES) event, intermittent snowfall with lightning (see Appendix A) began at the Ataturk International Airport (AHL), which intensified on February 18th, causing visibility to drop to 100 m according to METAR reports. The snowfall, which lasted briefly on February 19th, ended completely at the airport at 07:20 UTC. The heavy rainfall that preceded the snow caused severe disruptions to traffic in the sprawling metropolis, bringing it to a near standstill (Fig. 1). The Istanbul Metropolitan Municipality (İBB) released a statement regarding the ongoing snowfall that has been affecting the city. The statement indicated that the heavy snowfall, which continued throughout the night in blizzard form, resulted in the temporary closure of the TEM highway between Büyükçekmece and Mahmutbey due to trucks without chains getting stuck on the road and over the past 48 hours, numerous districts experienced 189 trees falling as a result of the blizzard and the snow's weight (Sözcü, 2015a).

Fig. 1
figure 1

Images of the damage caused by snowfall (Sözcü, 2015b)

Full size image

Frequent lake-effect snow (LES) events during the winter season are known to be experienced in the Great Lakes region in North America. Carpenter (1993) conducted a study examining the features of LES in the Great Salt Lake region in Utah, USA. Additionally, Steenburgh et al. (2000) highlighted shared atmospheric variables in their investigation of the climatology of lake-effect snowstorms originating from the Great Salt Lake. Andersson and Gustafsson (1994) highlighted the significance of terrain height, topography, and surface roughness in influencing land-sea interactions during the formation and structure of convective snowbands.

Baltic Sea and North Sea region in Europe, SES and LES occurs and studies on this region were conducted by Bergeron (1949) and Pike and Webb (2020). In The Hokuriku Region of Japan, a study was conducted by Veals et al. (2019) to investigate the spatial and temporal variability of SES, and it was found that the precipitation patterns are influenced by the topography and proximity to the Sea of Japan. An separate investigation, Veals et al. (2020) examined the variations in distribution and intensification of lake-effect precipitation (LPE) during a SES spanning from February 2nd to 7th, 2010 in the Hokuriku region.

The variability of SES in The Caspian Sea was investigated by Ghafarian et al. (2018), and it was found that the precipitation patterns are influenced by the sea surface temperature and wind patterns. Bednorz et al. (2022) examined SES events along the southern Baltic Sea coast, highlighting the presence of a bipolar sea-level pressure pattern and the transportation of cold air masses over the comparatively warmer sea, creating conducive conditions for convection and snowfall. In addition, Olsson et al. (2023) shown the climatology of convective SES in Finland over a 48-year period, identifying the spatial and temporal occurrence of snowbands. Demirtaş (2023a) has brought attention to the fundamental, identifiable, and predictable dynamic indicators, regional topography, and thermodynamic components that led up to the early January 2017 lake-effect snowstorm occurrences in the eastern Mediterranean. Specific circulation types (NW, N, NE) blowing from the Black Sea, along with a warm sea surface temperature and a cold temperature anomaly at the near ground level of the atmosphere, were favorable for the development of intense snowstorms, and there was a positive relationship between snow cover depth and land-sea temperature contrast in the Black Sea (Baltaci et al. 2021). Over the years, numerous studies have been conducted on SES in the Black Sea Region. These studies have focused on a variety of topics, including the atmospheric conditions that lead to SES (Özdemir & Yetemen 2019; Yavuz et al. 2021b; Demirtaş, 2023b) and Climatological analysis (Baltaci et al. 2021; Yavuz et al. 2021a). The impact of these events has become a cause for concern, as they can lead to a range of societal and economic challenges, such as transportation disruption, power outages, and other issues that could potentially put the lives and livelihoods of local residents at risk. The need for a better understanding of the Sea snow-effect in the Black Sea region has therefore become increasingly urgent, and efforts to investigate and model this phenomenon using advanced techniques such as WRF parametrization have become an essential component of research in this field.

Accurately simulating SES in WRF requires the careful selection of parameterizations since SES events involve complex interactions between various atmospheric and surface processes. McMillen and Steenburgh (2015) investigated how the microphysics parameterization (MP) choice affects the simulation of the Great Salt Lake-effect snowstorm of 27 October 2010, using the WRF Model. Results indicate the Thompson scheme (THOM) is the most precise while the Goddard, Morrison, and WRF double-moment 6-class microphysics schemes (WDM6) produce more precipitation, and WDM6 overpredicts the most. Hydrometeor mass tendency analysis reveal that WDM6 generates more graupel, less snow, and more total precipitation compared to other schemes. In another study, Different microphysics and planetary boundary layer schemes were evaluated in the simulation of a lake-effect snow event in the Caspian Sea, revealing that the best configurations for precipitation and snow estimation were Morrison- MYJ and Goddard-MYJ, respectively, and that precipitation intensity and distribution were better simulated by the Mellor–Yamada–Janjic Scheme (MYJ) scheme compared to the Yonsei University Scheme (YSU) scheme (Ghafarian et al. 2021). Through the use of numerical sensitivity tests, Hjelmfelt (1990) discovered that surface fluxes and upwind atmospheric stability play the most significant roles in shaping the development of lake-effect boundary layers and mesoscale circulations. The inversion layer creates a stable layer of air that prevents the mixing of air masses, resulting in different temperature and moisture profiles above and below the inversion layer. The profiles below the inversion layer determine the location of the upwind cloud edge, which in turn affects the distribution of snowfall and precipitation over the surrounding regions (Kristovich & Laird 1998).

Advanced modelling techniques are crucial for comprehending the intricate dynamics of SES and enhancing the prediction accuracy of its intensity, duration, and location. The study aims to identify the most effective boundary layer parameterization schemes for modeling the SES event that occurred on 17–18 February 2015. To accomplish this, the state-of-the-art numerical weather prediction model, WRF, is run to conduct sensitivity experiments. By examining the various schemes' impact on precipitation amounts and latent fluxes over the sea, and cloud formation the study hopes to provide insights into which PBL parameterization scheme would produce the most accurate forecast for such weather events.

2 Study area, data and method

2.1 Study area

The Black Sea, a large and important body of water, is situated at the boundary between southeastern Europe and Asia Minor. It is surrounded by six different countries: Romania, Bulgaria, Turkey, Ukraine, Russia, and Georgia (Fig. 2). With an area of approximately 436,400 square kilometers and a coastline of around 2,500 km, it is one of the world's largest inland seas. The sea's maximum depth is 2,212 m, located at the Yalta Abyssal Plain off the coast of Ukraine, and its average depth is approximately 1,225 m.

Fig. 2
figure 2

Study area and types of stations

Full size image

2.2 Data

2.2.1 Model data

ERA5 reanalysis data

The ECMWF Reanalysis (ERA) data is a meteorological dataset created by the European Centre for Medium Range Weather Forecasts (C3S, 2017) that provides a comprehensive description of the Earth's atmosphere, including variables such as temperature, wind, humidity, and pressure. It is a long-term, consistent, and quality-controlled data product that is generated by assimilating observations from weather balloons, satellites, and weather stations into a numerical weather model (Hersbach et al. 2020).

Observation data

This study incorporates diverse observational data, encompassing meteorological observations on both land and sea from the Turkish State Meteorological Service (TSMS), mole (breakwater), upper-level data obtained through radiosondes, and aviation data, including METAR codes. Throughout Turkey, TSMS operates numerous meteorological stations, crucial for industries such as aviation, providing essential data on weather conditions, temperature, humidity, precipitation, and wind dynamics (TSMS 2023). Sea surface temperature data is collected from mole stations, which are specifically designed to measure this variable. METAR codes offer concise, standardized information on current airport weather conditions, supporting informed decision-making by pilots and air traffic controllers, with special reports (SPECI) issued in cases of significant weather changes (TSMS 2022; Bureau of Meteorology 2023.). Radiosondes record atmospheric parameters during ascent, contributing valuable data, while GPS tracking enables measurements of upper atmosphere wind speed and direction (NOAA 2021). The research also incorporates TSMS radar data, focusing on the Maximum Display product. This product is generated from volumetric scanning data, illustrating the highest values detected by the radar within two specified horizontal levels. The MAX product offers insights into levels characterized by increased activity and the dense cores of air masses (TSMS 2021). Figure 2 visually presents the diverse array of observational data types employed in this investigation.

2.3 Method

2.3.1 WRF model

The WRF model, extensively detailed by Skamarock et al. (2008), is a robust numerical weather prediction system renowned for its two computational cores, data assimilation system, and flexible software architecture enabling parallel computation and system extensibility. It accurately simulates various atmospheric conditions, incorporating observational and idealized data, making it widely applicable in meteorological studies. The open-source WRF model is accessible at https://www.mmm.ucar.edu/models/wrf.

The WRF model incorporates multiple parameterization schemes, categorized into various types such as microphysics, cumulus, radiation, planetary boundary layer (PBL), land surface, shallow convection, surface layer, urban, and ocean-related schemes. The aim of PBL parametrizations is to accurately simulate the transport of heat, moisture, and momentum within the boundary layer. The parameterization options used in the WRF model are provided in the table below (Table 1).

Table 1 WRF parametrizations
Full size table

2.3.2 PBL schemes

Local schemes focus solely on vertical levels that are directly adjacent within the model, while non-local schemes have the capability to encompass deeper layers spanning multiple levels, especially when depicting the influence of vertical mixing within the Planetary Boundary Layer (Yoon et al. 2021). For the PBL schemes, we selected three non-local and three local closure schemes. The non-local closure schemes are YSU, TEMF and NCEP GFS schemes, The local closure schemes are MYJ, MYNN2, and University of Washington (UW) scheme.

Yonsei university scheme (YSU)

The YSU (Yonsei University) scheme integrates a sophisticated non-local closure approach, enhancing the MRF formulation by introducing explicit entrainment through an asymptotic term that scales with surface flux, derived from extensive large-eddy simulations (Hong et al. 2006). Moreover, it offers an exhaustive treatment of subgrid-scale turbulent mixing phenomena, ensuring a comprehensive representation of atmospheric dynamics at finer scales.

Mellor–yamada–janjic scheme (MYJ)

MYJ combines the Mellor-Yamada turbulence closure scheme with modifications by Janjic to improve its performance, especially in simulating boundary layer processes in complex terrain and coastal regions (Janjić 1994). It includes parameterizations for both turbulent kinetic energy and its dissipation rate, which allows for a more detailed representation of turbulent mixing within the boundary layer.

Mellor–yamada nakanishi niino (MYNN) Level 2.5

MYNN Level 2.5 is an advanced turbulence closure scheme that incorporates improvements by Nakanishi and Niino, enhancing its representation of boundary layer processes and convective phenomena (Nakanishi and Niino 2006). It features a multi-level approach to parameterizing turbulent mixing, allowing for a more accurate representation of vertical profiles of turbulent kinetic energy and other key variables.

NCEP global forecast system scheme

The MRF (Medium Range Forecast) PBL scheme incorporates a counter-gradient term into the local mixing formulation for both potential temperature (θ) and specific humidity (q), thereby introducing the influence of nonlocal mixing from larger eddies. The Planetary Boundary Layer Height (PBLH) is iteratively determined by assessing the bulk stability between the surface layer and an upper level (Hong and Pan 1996).

University of washington (TKE) boundary layer scheme

The UW (University of Washington) PBL scheme originated from the GBM scheme, with a notable distinction being that Turbulent Kinetic Energy (TKE) is diagnosed rather than prognosticated (Bretherton and Park 2009). This scheme conducts calculations across multiple layers identified through thermodynamic stability diagnosis, utilizing the moist gradient Richardson number as a criterion.

TEMF scheme

The TEMF scheme is an advanced 1.5-order non-local closure scheme renowned for its incorporation of the subgrid-scale total energy prognostic variable and its implementation of mass-flux type shallow convection (Angevine et al. 2010). In addition to these features, it utilizes the Rib method to compute the Planetary Boundary Layer (PBL) height, employing a threshold value set at 0. The Rib method involves calculating the bulk Richardson number, a dimensionless quantity that describes the ratio of buoyancy to mechanical shear in the atmosphere. This is done by integrating variables such as virtual potential temperature and horizontal wind components from the surface upward until the Richardson number reaches zero, indicating the transition point between turbulent and non-turbulent layers. This approach ensures robust and accurate representation of boundary layer dynamics within atmospheric simulations.

2.3.3 Model configuration

The selection of specific schemes is determined by the modeling objectives and application key components of setting up a WRF simulation is configuring the model domain and its associated boundary conditions. These include defining the study area and desired spatial resolution, choosing the appropriate map projection, selecting boundary conditions, specifying sea surface temperature dataset, determining land use and land cover, setting up model grids, and specifying physics options. All configuration and parametrization schemes can be found below (Table 2). The WRF model was run for one week including one day spin-up period, starting on 14 February 2015, at 00:00 UTC and ending on 20 February, 2015, at 00:00 UTC. The decision to select a domain with an outermost resolution of 9 km was based on the fact that a polar air mass came from the 60-degree latitudes on 15th February 2015 and the model needed to accurately capture the progression of this air mass. The second domain, with a resolution of 3 km, was chosen to cover the Western Black Sea region and resolve the dynamics that could potentially result in SES. The third domain, which has a resolution of 1 km, encompasses the study area of Istanbul. Point analyses were conducted using data obtained from the third domain (Fig. 3).

Table 2 Configuration of WRF model
Full size table
Fig. 3
figure 3

Domain configuration of WRF model

Full size image

SST affects atmospheric models, and it plays a critical role in determining the formation of sea fog, sea breezes, and tropical cyclones. Heat from underlying warmer waters can significantly modify the characteristics of an air mass over short distances, resulting in phenomena such as narrow lake-effect snow (LES) or SES bands (Kavak et al. 2016). To ensure dynamic acquisition of Sea Surface Temperature (SST) input, the model is configured to receive updates every 6 h from the ECMWF. These updates are facilitated through the incorporation of auxinput4 into the model, while setting sst_update to "1."

3 Result and discussion

3.1 Synoptic situation

Synoptic maps are available from the Met Office through the Wetter3 website, and were evaluated in this study. On February 15th, 2015 at 00:00 UTC, a continental arctic cold front was located at coordinates 15°E – 45°E and 60°N, spanning a large area. This front started to accelerate southward along 30°E, moving through the Western Black Sea and Istanbul before eventually reaching the Sea of Marmara. As the cold front passed over the Western Black Sea, it caused a trough connected to the frontal system to move through, creating conditions that resulted in SES. This phenomenon occurred because the cold air behind the front caused the warm sea to heat the air, leading to increased instability and isobar flows of the Siberian high pressure moving towards the Marmara Region via the western and central Black Sea at an angle of approximately 40–45 degrees.

At 00:00 UTC on February 17th, 2015, the continental arctic cold front had made significant progress in its southward movement and had reached the Marmara Sea, having advanced rapidly along its longitude. Interestingly, even before the cold front had completely passed over the Western Black Sea, the conditions that led to SES had already begun to take place. Based on the analysis of six different runs of the WRF model, it is evident that the synoptic conditions during the passage of the continental arctic cold front and current pressure gradients on February 17th, 2015 at 00:00 UTC were consistent with the observed weather chart (Fig. 4). At atmosphere near the ground, the wind velocity was measured to be in the range of 20–25 knots, with a predominant northeast flow over the water. Pressure gradients within Istanbul exhibited variations in the range of 1026 to 1028 hPa. Additionally, all simulations consistently depicted low-pressure gradients situated in the eastern Black Sea. However, the 2nd and 6th runs of the WRF model revealed an interesting difference in temperature. Specifically, the 2 m temperature in these two runs had higher values, ranging from 0.5–1 °C, compared to the other runs in terms of spatial distribution of 2-m temperature.

Fig. 4
figure 4

Analysis weather chart and weather charts gathered from WRF on 17 February 2015 00:00 UTC

Full size image

At 00:00 on February 17, 2015, there was an increase in the occurrence of pressure gradients in the southeastern area of the Marmara Sea. The sudden decrease in temperature at a 2-m height and the convergence of these gradients suggest the occurrence of a cold front event. When the cold front enhances the SES, it adds an extra layer of instability to the already unstable atmosphere above the sea. The colder air from the front, when mixed with the relatively warmer air above the sea, creates an even greater contrast in temperature and increases the instability of the air mass. The rising air currents then become even stronger, leading to more intense and long-lasting snowfall.

3.2 1000–500 hPa Thicknesses

In this study, the incorporation of 1000–500 hPa thicknesses into meteorological analyses, which are represented in decameters (dam) and meters (m), offers invaluable insights into the depth of atmospheric instability, particularly in scenarios where cold airstreams interact with warm oceanic surfaces (Pike & Webb 2020). The models consistently produced nearly identical results for February 17, 2015, at 12:00 UTC, revealing a 1000–500 hPa thickness range of 500–510 decameters in the southern region of Crimea, coinciding with the central expanse of the Black Sea (Fig. 5). This observation highlights an exceptionally profound cold atmospheric column, evidenced by the sub-510 decametre coverage over Crimea and the northern Black Sea. Simultaneously, the geopotential height at 500 hPa registered 5300 m for this same geographical area. On the subsequent day at 12:00 UTC, all models depicted a widening of the cold atmospheric column's expanse, gradually enveloping the western Black Sea before ultimately positioning itself directly over the Bosphorus by 1200 UTC (Fig. 6). This shift was accompanied by changes in thickness above the Bosphorus ranging between 500–520 decametres. Furthermore, the geopotential height at 500 hPa showed a slight increase to 5320 m. Additionally, charts revealed a prominent upper-level low/vortex situated just east of the Bosphorus, with its expansive circulation encompassing a significant portion of southeast Europe. Notably, areas exhibiting intense vorticity within this system indicate the presence of potent short-wave troughs, further emphasizing the intricate and dynamic nature of the atmospheric conditions under scrutiny.

Fig. 5.
figure 5

1000–500 hPa Thickness and 500 hPa geopotential height gathered from WRF on 17 February 2015 12:00 UTC

Full size image
Fig. 6.
figure 6

1000–500 hPa Thickness and 500 hPa geopotential height gathered from WRF on 18 February 2015 12:00 UTC

Full size image

3.3 Sensible and latent heat flux

In order to unravel the factors contributing to the spatial distribution of precipitation throughout our research timeframe, our initial focus involves scrutinizing the attributes and driving forces behind SES convection. The transfer of heat energy between the sea surface and the atmosphere through latent heat flux is a key factor in the formation and intensity of SES. The transfer of heat from the sea surface to the atmosphere through latent heat flux can create a temperature gradient between the water and the air, which can lead to the formation of clouds and precipitation. Areas of the map with higher latent heat flux (i.e., those colored red) indicate locations where there is a greater transfer of energy from the sea surface to the atmosphere (Fig. 7). A substantial sensible and latent heat flux, reaching up to 1000 W/m2, is observed in the southwest and southeast regions of Crimea. This elevated value can be attributed to the presence of a synoptic-scale continental cold front. Conversely, in the central part of the Black Sea, the values range between 400 to 500 W/m2 in open-sea, with a noticeable decrease in magnitude moving southward. Simulations indicate an average total heat flux ranging from 250 to 400 W/m2 over Istanbul. Particularly on the European side of Istanbul, values peak between 300 to 400 W/m2, surpassing those on the Anatolian side where they range from 200 to 250 W/m2. Our findings align with a related study (Veals et al. 2020) on SES in the Hokuriku region, which attributes the phenomenon to a cold winter monsoonal flow. In this study, total heat flux over open water reached 850 W/m2, with comparatively lower values observed over Toyama Bay ranging from 400 to 500 W/m2. In a separate investigation on snowstorms in Greece (Patlakas et al. 2024), it was observed that the total heat flux varied between 100 and 300 W/m2 across mainland Greece. Meanwhile, over the Aegean and Ionian Seas, the total heat flux exhibited a broader range, spanning from 100 to 900 W/m2 during daytime hours. In this study, a cold front associated with a synoptic system further enhances this effect, leading to the formation of a very large gradient. Winter conditions, especially those associated with SES, are characterized by extreme temperature contrasts. The warm sea surface serves as a significant source of heat and moisture, which are transferred to the cold air above through turbulent mixing. This process is further enhanced by the presence of synoptic systems, leading to the observed high turbulent fluxes as in (Patlakas et al. 2024). In this study, the presence of a cold front associated with a synoptic weather system further amplifies the temperature gradient. The cold front introduces frigid air masses, which interact with the warmer sea surface, thereby enhancing turbulent fluxes.

Fig. 7
figure 7

Mean Total Heat Flux and temporally averaged 10 m wind vectors on 17–18 February 2015

Full size image

According to Bowen ratio (ß) (see Appendix B), over the sea, both maps predominantly show blue shades, indicating low values (0 to 1), suggesting that latent heat flux, typical of evaporation from the sea surface, is dominant. Over the land, there is more variability in the Bowen Ratio. On 18.02.2015 at 12:00 UTC, some land regions exhibit higher Bowen Ratios (yellow to red shades), indicating dominant sensible heat flux, especially in coastal areas and further inland regions due to convectional heating. By 19.02.2015 at 00:00 UTC, there is a general shift towards lower Bowen Ratios (blue to light blue shades) over land, suggesting increased latent heat flux during this period. This noticeable change in the Bowen Ratio between the two times suggests an overall reduction in Bowen Ratio values from 12:00 UTC on the 18th to 00:00 UTC on the 19th, indicating a shift towards increased latent heat flux or reduced sensible heat flux across both sea and land areas. This change can be attributed to the lack of direct solar heating at night and the temperature gradient caused by the passage of a cold front and north—northeasterly atmospheric flow.

This temperature difference can enhance the rate of evaporation and energy transfer from the sea surface to the atmosphere, which can in turn contribute to the formation of moist air masses that are then cooled over land to produce snow. The observed values of latent heat flux during the event days of February 17–18, 2015, provide important insights into the dynamics of the atmospheric processes in the region (Fig. 8 and Fig. 9). The range of average latent heat flux values of 350–500 W/m2 on indicate the intensity of the heat exchange between the atmosphere and the underlying surface in the southern and southwestern regions of the Crimean Peninsula. The study reveals notably elevated mean latent heat flux values, reaching 175–200 W/m2 in the northern regions and 100–150 W/m2 around the Bosphorus, immediately following the frontal transition at the synoptic scale. This transition amplifies the temperature differential between the air and sea surface, consequently resulting in the observed heightened Heat Flux. Figures illustrates a noteworthy escalation in the 12-h mean latent heat flux, surging to 250–300 W/m2 in the northern regions and 175–250 W/m2 around the Bosphorus during after this transitional phase. Additionally, the image demonstrates that the European side exhibits greater latent heat values compared to the Asian side, with a range of 25–50 W/m2. In literature, surface heat fluxes directed from the lake to the overlying air typically exceed 150W/m2 for the western Great Lakes. On the contrary, the over-lake surface fluxes found in the eastern lakes region, including Erie and Ontario, range from 50-150W/m2 (Lang et al. 2018). In another study, Controlled WRF simulation revealed that the mean maximum latent heat flux, which reached 200 W/m2, was observed over the southern half of the Caspian Sea. (Ghafarian et al. 2021).

Fig. 8.
figure 8

12-Hourly mean latent heat flux just after front passage (17.02.2015 00:00 UTC—17.02.2015 12:00 UTC)

Full size image
Fig. 9.
figure 9

12-Hourly mean peak latent heat flux (18.02.2015 12:00 UTC – 19.02.2015 00:00 UTC)

Full size image

Specifically, the TEMF scheme exhibits higher mean total heat flux and latent heat flux compared to alternative schemes. Within the WRF model, the TEMF scheme employs a turbulence kinetic energy (TKE)-based approach to parameterize turbulent fluxes, particularly for moist shallow cumulus convection (Angevine et al. 2010). This methodology facilitates a more dynamic representation of turbulence, particularly pronounced in conditions characterized by strong turbulent mixing. Furthermore, the scheme explicitly incorporates moisture effects, including latent heat release, thereby intensifying convective mixing in moist environments. Consequently, this dual consideration contributes to greater sensible and latent heat fluxes by altering the buoyancy and stability of the boundary layer.

3.4 Temperature gradient between 850 hPa and sea surface temperature (\(\Delta \text{SST}-850\text{ hPa}\))

The temperature gradients between the sea water temperature and 850 hPa level are influenced by various factors such as air-sea heat exchange, ocean currents, atmospheric circulation, and solar radiation. In the study area, these gradients extend in the northeast-southwest direction, and their magnitude varies similar to latent heat fluxes. On 17–18 February 2015, the sea water temperature SST– 850 hPa difference was found to be 17.3 ± 0.6 °C, 18.2 ± 0.5 °C, 16.8 ± 0.4 °C, 17.1 ± 0.5 °C, 18.1 ± 0.4 °C, 17.6 ± 0.4 °C in the study area respectively (Fig. 10). It is seen that the temperature difference in the MYJ and TKE is higher than other runs spatially.

Fig. 10
figure 10

Mean SST-850 hPa temperature difference of six WRF runs (18.02.2015 00:00 UTC – 18.02.2015 12:00 UTC)

Full size image

The temperature at 850 hPa during the event dates was recorded as an average of -10.15 °C based on the Radiosonde measurements conducted at the Istanbul Regional Station. In addition, the Mole stations located in the Western Black Sea recorded a sea surface temperature of 6.9 °C. This implies that the measurements yielded an SST-850 hPa value of 17.1. However, this value varies throughout the duration of the event and reaches its maximum value of 19.2 °C on February 18th, 2015 at 12:00 UTC. In all model runs, the criteria for the occurrence of SES have been fulfilled, which necessitates a temperature difference of more than 13 degrees between the sea surface temperature and the 850 hPa level in literature (Niziol et al. 1995; Veals & Steenburgh 2015). Specifically, Baltaci et al., (2021) states that the temperature difference between the sea surface and that at 850-hPa changes from 17.2 to 18.2 °C in all SES events occurred in Istanbul during the 1971–2006 winter months. The temperature difference produced by the WRF model simulations ranged from 17.1 °C to 18.8 °C. Although this range is not significantly different from the temperature difference reported in the literature, it confirms the value within the expected range.

3.5 Temperature gradient between 700 and 850 hPa (\(\Delta 850\mathbf{h}\mathbf{P}\mathbf{a}-700\mathbf{h}\mathbf{P}\mathbf{a}\))

The 700 hPa pressure level, located approximately 3 km (10,000 feet) above sea level, plays a crucial role in weather forecasting. It serves as a significant reference point due to its positioning within the middle or upper portion of intense LES/SES events in the atmosphere. These events often involve critical atmospheric processes, making the 700 hPa level a valuable indicator for understanding and predicting weather phenomena.

According to the Radiosonde data, the average air temperature at 700 hPa was found to be -21.2, and there was a temperature difference of 11.1C between 850 hpa and 700 hPa. The temperature difference between 700 and 850 hPa varies spatially, ranging from 5.1 (on land) to 11.3 (on open-sea) on average. Also, it was determined that there is a temperature difference of 28.15 °C between 700 hPa and SST according to radiosonde measurements. In WRF model, this difference found to be 28.2 ± 0.4 °C, 28.6 ± 0.3 °C, 28.3 ± 0.2 °C, 28.4 ± 0.4 °C, 28.9 ± 0.5 °C, 28.7 ± 0.3 °C respectively (Fig. 11). However, in the northwest regions of Marmara, the temperature gradient between 700–850 hPa decreases, resulting in a temperature difference of around 5 °C. Notably, this differs from the 850-SST temperature gradient. The SST-700 hPa temperature differences predicted by the WRF model are significantly higher than the temperature thresholds described in the literature, 16 °C (Steenburgh et al. 2000) and 17 °C for Great Salt Lake (Carpenter 1993).

Fig. 11
figure 11

Mean 850 hPa—700 hPa temperature difference of six WRF runs (18.02.2015 00:00 UTC – 18.02.2015 12:00 UTC)

Full size image

3.6 Fetch distance and shoreline geometry

The quantity of fetch has a significant impact on snowfall intensity. This distance plays a crucial role in determining the duration for the air to absorb heat and moisture from the sea surface, making it a crucial factor. Laird et al. (2003) utilized the ratio of wind speed (V) to fetch over the open water (L) to determine the lake-effect for idealized scenarios. Their study revealed that bands are formed when the value of V/L falls between 0.02 and 0.09 m/s/km.

$$0.02 <V/L<0.09$$

In order to strengthen the development of the snow band, a longer fetch is necessary, which consequently requires larger fetch distances for stronger winds. When examining the distance from the Crimean Peninsula to Istanbul, it becomes apparent that the flow over sea is relatively consistent in terms of both direction and speed. In all WRF runs, the average wind speed was found to be between 20–25 knots within this region. Considering the wind direction, the fetch distance was calculated to be roughly 600 km. V/L value ranging 0.030 – 0.041 falls well within the thresholds mentioned above, thus making the identification of SES cloud bands a relatively straightforward process.

Orographic lifting is a phenomenon observed in Marmara region where the hilly terrain forces air to ascend. As the air rises, it undergoes adiabatic cooling and condensation, resulting in intensified precipitation rates on the windward side of the terrain (Suriano & Leathers 2017a). This effect is particularly notable during SES events in Marmara region, where cold air flows over the relatively warmer waters of the Black Sea, leading to significant snow accumulation on the windward slopes of the city's hills. Moreover, Marmara Region’s topographic features, including its hills and coastal areas, play a crucial role in channeling and concentrating air flow. This leads to the formation of enhanced convergence zones downstream, contributing to the development of localized bands of intense snowfall (Steenburgh & Campbell 2017). Additionally, coastal formations influence the direction of air flow, thereby influencing the spatial distribution of SES events in Istanbul.

3.7 Radar and cloud band analysis

The Maximum Display (MAX) product uses volumetric scan data to show the maximum values measured by radar between two defined horizontal levels, providing information on the more active event layers and dense cores of air masses, and includes three different displays simultaneously (TSMS 2024). On February 17th at 10:00, as the passage of the cold front marked the onset of the first day of the SES event, observations revealed intriguing similarities between MAX radar product imagery and Aqua MODIS corrected reflectance imagery (Fig. 12). Notably, the cloud bands exhibited a distinct northeast-southwest orientation with a pronounced band structure, mirroring simulations from the WRF model. The thickness of these bands, particularly evident in their dBZ representation, underscored the model's success in capturing complex cloud formations. As stated by Fujisaki-Manome et al., (2022) Ensuring adequate accuracy in forecasting LES or SES poses a significant challenge, particularly in predicting the timing, location, and intensity of snowfall with enough precision for local communities to prepare in real-time. Our study specifically encountered challenges related to pinpointing the location of snowfall. It's worth noting a discrepancy: the models consistently placed the cloud bands further west geographically than observed. This deviation stems from discrepancies in ERA5 renalysis initial condition values and computational inaccuracies within the model framework. Clearly, The ERA5 reanalysis data was checked, revealing deviations in wind direction from the actual values across different altitude levels.

Fig. 12
figure 12

MAX Radar product, Aqua MODIS Satellite Imagery, and Visualized MAX dBZ from WRF runs on 17 February 2015 10:00 UTC

Full size image

While radar measurements typically indicated dBZ values ranging between 28–37, the WRF models depicted a slightly higher range of 34–44, with localized peaks reaching 44–50 in certain areas. On February 18th, at 09:00 UTC, the intensity of the SES event peaked, marked by larger and denser cloud band structures compared to the preceding day. Similar to earlier observations, discrepancies persist in the geographical positioning of these cloud bands within WRF simulations, with the model consistently placing them slightly further west than observed in radar and satellite imagery (Fig. 13). Furthermore, there's a recurring trend of dBZ overestimation within the WRF simulations. While radar measurements typically registered dBZ values ranging between 28–37, the WRF simulations occasionally depicted higher values, reaching 44–50 in some instances. An a comparable study (Ghafarian 2021), 30–35 dBZ values exhibit notable clarity, whereas in WRF, across various simulations, 40–50 dBZ values appear fragmented. Regarding boundary layer comparison, the MYJ parameterization demonstrates greater consistency with radar observations. This shows that the model overestimates the maximum dBZ as in this study.

Fig. 13
figure 13

MAX Radar product, Aqua MODIS Satellite Imagery, and Visualized MAX dBZ from WRF runs on 18 February 2015 09:00 UTC

Full size image

When conducting comparisons among model simulations, YSU, MYJ, NCEP GFS, MYNN schemes emerged as the most visually congruent with observed data. This determination was reached through a comprehensive assessment encompassing various factors including dBZ values, the convective structure of clouds, visual similarities in cloud formations, and the directional extension of clouds. Despite the remarkable similarity observed among these four parameterization options, subtle discrepancies do exist. These variations, though minor in magnitude, are nonetheless discernible upon closer examination. They manifest in nuanced differences in cloud morphology, spatial distribution, or the intensity of convective activity within simulated cloud systems.

Upon comparative analysis, the TEMF parameterization emerges as the least effective, yet it still stands out for its distinct portrayal of cloud structures across the study area when compared to alternative options. Notably, the TEMF scheme manifests a greater dispersion of cloud formations within the study region, accompanied by an increase in the occurrence of clouds exhibiting high dBZ values. This observed phenomenon aligns seamlessly with the depiction of elevated latent heat values showcased in the preceding Fig. 7. The heightened presence of clouds with robust convective activity, as indicated by elevated dBZ values, correlates directly with increased latent heat release associated with vigorous atmospheric processes. In a separate investigation, it was discovered that the three planetary boundary layer (PBL) schemes—QNSE, MYNN2, and TEMF—employing distinct surface-layer schemes, exhibit the highest simulated surface heat fluxes. Additionally, it was noted that the significant deviation in model-simulated surface heat fluxes using the TEMF scheme in Anavyssos, Greece, remains unidentified (Banks et al. 2016).

3.8 Snow depth

The choice of PBL scheme can affect the simulated snowfall distribution, as it can influence the vertical profile of temperature, humidity, and wind speed. In regions with complex terrain, the PBL scheme can affect the simulated temperature and moisture profiles, which can impact the formation and distribution of snowfall. Snow measurements collected by TSMS at 06:00 UTC are depicted by red dots on the station map. To ensure consistency, WRF model simulations were assessed concurrently, at 06:00 UTC on February 21, 2015.

Firstly, it's evident that snowfall occurs in distinct bands within each simulation, indicating that the WRF model adeptly captures the snowfall patterns corresponding to cloud bands. However, a crucial observation emerges: the cloud snowfall bands delineated in the preceding section predominantly manifest in the western region. As illustrated in Fig. 14, this results in a notable accumulation of snow thickness, ranging from 25 to 35 cm. Notably, the NCEP GFS model exhibits the poorest performance, yielding markedly lower snow thickness values with the highest error rates among the simulations evaluated. Conversely, the TEMF parameterization tends to overestimate values spatially due to previously mentioned high heat fluxes. Among the parameterizations, MYNN emerges as the most successful, consistently demonstrating superior predictive capabilities, albeit occasionally falling short compared to other simulations. Following closely is the MYJ model, which ranks as the second-best performing simulation. Additional research validates that MYJ and MYNN is the most accurate parameter in representing snowfall. the optimal configuration of the WRF for the research region involved utilizing the Thompson microphysical parameterization and Mellor-Yamada-Janjic scheme, as it produced the most accurate results in terms of spatial distrubition (Fernández-González et al. 2015). Also, in terms of predicting precipitation and snow water equivalent, the Morrison-MYJ schemes demonstrate the best performance in Caspian Sea (Ghafarian 2021). For MYNN, Conrick & Reeves (2014) examined the impact of different boundary layer parameterization schemes on forecasts of lake-effect snow, using the WRF-ARW model to test six schemes in a case study of Lake Erie in December 2009. The results demonstrate that parameterization schemes have a significant impact on precipitation amounts, with higher heat and moisture fluxes over the lake resulting in more precipitation. The Mellor-Yamada-Nakanishi-Niino scheme provides the most accurate precipitation forecast in their study. But, in our case, due to the presence of cloud bands and their misrepresented distribution towards the west, snowfall and consequently snow depth appear elevated in western regions, deviating from actual conditions. Both the MYJ and MYNN models share a common characteristic in projecting snow depths of up to 30–35 cm in these areas. However, MYNN outperforms MYJ, particularly evident on the Anatolian side of Istanbul. Notably, MYNN models accurately capture the accumulation of snow on Aydos and Kayışdağ hills, which stand at altitudes of 400–500 m on the Anatolian side. Put differently, parametrizations utilizing two local closure schemes, as opposed to TEMF, proved effective in simulating snow depth compared to non-local schemes.

Fig. 14
figure 14

Observed and simulated snow depth from the WRF simulations during the SES event

Full size image

3.9 Taylor diagrams of radiosonde measurements

During the occurrence of the SES event, comparisons between temperature variable data from the model at various pressure levels and data collected from radiosondes at 6-hour intervals were conducted using Taylor diagrams (Fig. 15). The simulations demonstrated remarkable accuracy particularly in atmospheric temperature levels close to where snow cover is simulated. MYJ emerged as the most successful parameterization, boasting a correlation coefficient of 0.94, indicative of its superior performance. Conversely, the TEMF model lagged behind with a correlation coefficient of 0.84, signifying its lower level of success. Furthermore, some parameterizations yielded favorable outcomes for the temperature at the 850 hPa level. Notably, MYJ exhibited a high degree of consistency in modeling the 850 hPa temperature, boasting a correlation coefficient of 0.95. It was closely followed by MYNN with a correlation of 0.92. On the contrary, TEMF performed the worst in simulating the 850 hPa pressure level temperature. At the 700 hPa level, all models displayed remarkable consistency, with correlation values ranging from 0.94 to 0.97, except for TKE, which exhibited a lower correlation of 0.88 compared to the other models. Moving to the 500 hPa level, consistent modeling results were observed across all parameters, likely due to the greater uniformity of the temperature gradient and atmospheric flow in the upper atmosphere. Correlation values ranged from 0.88 to 0.92 among all models, with the MYJ parameterization yielding the highest correlation of 0.92 for the 500 hPa level.

Fig. 15
figure 15

Taylor diagrams of WRF simulations for temperature at different pressure levels

Full size image

3.10 Wind shear

The direction and strength of wind shear, as indicated by changes in wind direction (veering or backing) and magnitude, between the lower and upper atmospheric levels (specifically, the 850 and 700 hPa levels), are crucial factors that influence the formation, structure, and stability of SES bands. To achieve optimal and robust organization of these bands, it is important to ensure that the wind direction does not deviate more than 30° between the surface and the 850 hPa level (Suriano & Leathers 2017b) and 60° between the surface and the 700 hPa level (Niziol 1987). Utilizing radiosonde data retrieved from the station identified by the ICAO code "KRIB" (Fig. 2). This specific radiosonde station, played a pivotal role in acquiring atmospheric values at both the 850 hPa and 700 hPa levels, which were subsequently employed in grid-based calculations. Throughout the meteorological event, the wind direction difference between surface level and 850 hPa, as determined from both observation and model data, consistently fell within the range of 0 to 30 degrees, with the exception of the initial stages of the event (Fig. 16). This indicates that the wind shear was within the acceptable range for the formation and maintenance of SES bands for all models. The models showed that the wind direction difference between surface level and 700 hPa occasionally exceeded 60 degrees. However, radiosonde data collected every 12 h indicated that the range of 0–60 degrees was generally observed. In the model data, this range was mostly met, although some deviations were noted at certain time periods.

Fig. 16
figure 16

a Wind shear of Surface level and 850 hPa b Windshear of Surface level and 700 hPa

Full size image

According to the radiosonde and model data, it has been noted that the tropopause level experienced a decrease from 250 hPa during the Cold front transition on February 17th, 2015 at 00:00 UTC, to around 525 hPa by February 18th, 2015 at 12:00 UTC (Fig. 17 and Fig. 18). After a transition of cold front, there is a sharp change in temperature and pressure across the front, which can cause the tropopause level to decrease. On February 18th, 2015 at 1200 UTC, the temperature at ground level was -0.9 °C, at the 850 hPa level it was -10.7 °C, at the 700 hPa -22.7 °C, and at the 500 hPa level it was -38.1 °C (Fig. 18). The presence of an upper-level inversion layer is an important atmospheric condition for the development of SES bands (Yavuz et al. 2022). The inversion layer plays a role in the development of clouds and precipitation by curbing the disturbance arising from the transfer of heat and moisture fluxes in the lower atmosphere (Yavuz et al. 2021a). In literature inversion layers located upper atmosphere have been found to enhance the intensity of the SES bands (Niziol 1987). As of 17.02.2015 at 00:00 UTC, it appears that there is an inversion layer present within the atmospheric column ranging from approximately 700 hPa to 725 hPa (Fig. 17). This region can be identified by the red marking in the corresponding figure. Also it should be noted that by 12:00 on the same day (17.02.2015), the inversion layer has shifted to a higher altitude range, extending from 591 to 511 hPa (not depicted in the given figure). All the models runs were able to successfully model the inversion layers that were present. The identification of tropopause and inversion levels has been facilitated through the utilization of radiosonde data (Table 3). The presence of colder air below the 500 hPa level induces the occurrence of thunderstorm events within a layer located below 6 km. This is attributed to the limited ascent of moisture flow caused by atmospheric instability. As a result, the simultaneous manifestation of thunderstorms and snowfall is observed.

Fig. 17
figure 17

Skew-T Log P diagrams gathered from radiosonde and WRF runs at 17 Feb 2015 00:00 UTC

Full size image
Fig. 18
figure 18

Skew-T Log P diagrams gathered from radiosonde and WRF runs at 18 Feb 2015 12:00 UTC

Full size image
Table 3 Inversion levels observed according to radiosonde data
Full size table

4 Conclusion

SES/LES is a complex meteorological process resulting from the interaction between warm water bodies and cold atmospheric conditions. It leads to the formation of snow clouds and heavy snowfall downwind of the water body. The Istanbul SES event that occurred from February 17–19, 2023, resulted in significant disruptions to the city's transportation, with heavy snowfall and blizzard conditions causing visibility to drop to 100 m at Ataturk International Airport. It has been revealed that the SES was intensified by a continental arctic cold front, which moved southward along 30°E and caused a trough to move through, creating conditions that led to SES. The examination of six distinct runs of the WRF model confirms the alignment of synoptic conditions during the Arctic cold front's passage on February 17th, 2015, with the observed weather chart. The intensification of sea effect snow is driven by the substantial temperature difference between the sea surface and the atmosphere. In this study, a cold front associated with a synoptic system amplifies this effect, creating a significant temperature gradient. The warm sea surface transfers heat and moisture to the cold air above, enhanced by synoptic systems, leading to high turbulent fluxes. The cold front introduces frigid air masses that interact with the warmer sea surface, further increasing turbulent fluxes and contributing to the large energy values observed. Following the transition of the Arctic cold front, the range of 12-h average latent heat flux values of 175–250 W/m2 around the Bosphorus across all runs denotes a heightened level of heat exchange intensity during the event, potentially accentuating the formation and intensity of SES. The models consistently showed a 1000–500 hPa thickness range of 500–510 decametres over Crimea's southern region on February 17, 2015, at 12:00 UTC, indicating a deep cold atmospheric column extending to the northern Black Sea. By the following day at the same time, the cold atmospheric column expanded, covering the western Black Sea and positioning directly over the Bosphorus, accompanied by changes in 1000–500 hPa thickness.

The temperature difference between the sea surface and 700 hPa/850 hPa level during the snow event was observed to be within the range reported in the literature, and all model runs fulfilled the criteria for the occurrence of SES. Also, the wind direction difference between 1000 hPa and 850 hPa during the meteorological event fell within the acceptable range for the formation and maintenance of SES bands for all models. During the event's peak snowfall period, the MAX radar product closely resembled the WRF models, albeit with some slight overestimation. Notably, thick cloud bands were visibly apparent, stretching from the Black Sea to the southernmost point of Istanbul. However, a notable inconsistency emerged: the models consistently positioned the cloud bands further west than observed, attributed to differences in ERA5 reanalysis initial conditions. MYNN scheme was found to be the most accurate in simulating the 2-day cumulative new snowfall data, with the Morrison-2 moment microphysical parameterization as the optimal configuration for the research region.

In the context of this research, the evaluation reveals that among the local closure schemes, the Mellor-Yamada-Nakanishi-Niino (MYNN) scheme stands out as the most effective, closely followed by the Mellor-Yamada-Janjić (MYJ) scheme in terms of performance. Local closure Planetary Boundary Layer (PBL) schemes such as MYJ and MYNN operate under the assumption that turbulence primarily responds to the immediate atmospheric conditions, encompassing factors like local temperature, moisture gradients, and wind speed. This localized approach proves highly effective in replicating the intricate, small-scale mechanisms that underpin SES events. Given the localized and heavily influenced nature of SES phenomena within the lower atmosphere, these schemes excel by honing in on local gradients and processes. Notably, they can accurately model the crucial roles played by elements such as sea surface temperature and the moisture content of the overlying air. Furthermore, SES dynamics entail intricate feedback mechanisms between boundary layer processes and snow generation. For example, the cooling effect induced by falling snow can intensify thermal stratification, thereby impacting turbulence and mixing within the boundary layer. Leveraging their meticulous treatment of local processes, local closure schemes effectively capture these feedback loops, resulting in a more nuanced and faithful representation of SES dynamics.