The roughness and temperature differences between land and water surfaces often lead to local sub-climates such as mesoscale circulation systems or stable and unstable conditions. For instance, during autumn and winter when the water surface temperature is still warmer than the average air temperature an unstable stratification develops in low levels since the ice-free water surface appears as a source of moisture and heat to the overlying air mass. As a result, mesoscale convective precipitation events like convective snow bands may develop.

Convective snow bands are also known and studied as lake effect snow (e.g. at the Great Lakes), cloud streets, horizontal convective rolls or vortices with solid precipitation. At the Swedish east coast, they are often referred to as a snow canon (snökanon). Convective snow bands develop commonly over the open water surface of lakes or seas when cold air approaches from the continent. Enhanced heat and moisture fluxes from the comparatively warm water body trigger shallow convection as the colder air mass travels across the sea. An unstable atmospheric boundary layer builds up and the formation of shallow convective clouds is favoured. Relatively strong winds can organize this convection into wind-parallel quasi-stationary cloud bands with moving individual cells. Depending on various factors such as the strength of the horizontal wind, the vertical wind shear or the shape of the coast (Andersson and Gustafsson, 1993), different snow band structures can form (Niziol et al., 1995). When the prevailing atmospheric conditions imply a strong development of convective snow bands, intense precipitation occurs locally where the snow bands hit the coast. The highest precipitation has, however, been shown to occur over the sea close to the coast (Andersson and Nilsson, 1990). The topographic changes from sea to land involve additional convergence and orographic lifting, which intensify the snowfall further. Since the circulation is organized in steady bands along the wind direction, usually only a limited area is affected by the snowfall. Thus, a reliable wind field forecast is crucial for a prediction of the hazard area.

Convective snow bands lead repeatedly to severe precipitation events in the cold season of the high mid-latitudes around the world. This phenomenon occurs frequently at several regions situated in high latitudes, including the Great Lakes (Kelly, 1986). However, even the Swedish east coast and the coastal regions along the Gulf of Finland frequently experience convective snow band events. The large amounts of snow along with strong wind speeds can cause serious problems for traffic, infrastructure, and other important establishments of society. Due to global warming it is in general expected that larger areas of lakes and inland seas will stay longer ice-free (IPCC, 2015). As a result, the occurrence of those convective snow band events may become more frequent as well as more intense.

We here investigate the ability of the numerical regional climate model RCA4 (Rossby Centre regional atmospheric model) to simulate snow bands, with the purpose of identifying the sensitivity of the results to the model resolution and surface forcing conditions and to derive climate statistics of the occurrence. Convective snow bands in the Baltic Sea area have been studied widely using a variety of methods for the Gulf of Finland as well as other parts of the Baltic Sea (e.g. by Andersson and Gustafsson, 1993; Andersson and Nilsson, 1990; Vihma and Brümmer, 2002; Mazon et al., 2015; Savijärvi, 2012, 2015). The focus of our study is on the performance of a regional climate model to represent convective snow bands affecting the Swedish east coast. This emphasis allows the determination of snow band conditions based on very specific atmospheric properties associated with lake effect snow for the region.

The large-scale synoptic situation leading to convective snow band development over the Baltic Sea can be very different. However, a strong pressure gradient over the Baltic Sea is required to guide cold air masses from the north-northeast over the warm water surface. Strong prevailing northeast winds with small vertical wind shear are unusual for this latitude, but they can result from a deep low-pressure system southeast of the Baltic Sea and/or indirectly by a local high-pressure development over the cold north of Scandinavia. Therefore, convective snow bands can occur in the Baltic Sea area, and furthermore along the north- and northeast-facing coasts of Estonia, Latvia, Poland, and even Germany.

Studies by Evans and Wagenmaker (2000), Niziol et al. (1995), Andersson and Gustafsson (1993), and Niziol (1987) provided evidence that convective snow bands occur under specific conditions. The most important element for the formation of snow bands is the thermal difference between the water surface and the overlaying air (Mazon et al., 2015), which determines the extent of the essential heat and moisture fluxes. Therefore, it is necessary that a large part of the water surface is ice-free in order to ensure sufficient sensible heat release and evaporation from the sea. A partially frozen sea changes the coast line and affects the conditions for snow band developments considerably. Another important condition is the presence of instability to trigger convection. For the Great Lakes it has been observed that the minimum temperature difference between the water surface and the 850 hPa level must be 13 ∘C to initiate convective snow bands without additional synoptic-scale forcing (Holroyd, 1971). This vertical temperature gradient matches approximately the dry adiabatic lapse rate. A larger value for the atmospheric lapse rate implies the presence of an absolute unstable layer within the lowest 850 hPa. A big thermal difference may furthermore enhance the moisture flux towards the air mass and support the formation of clouds and precipitation. The conditions for an intense development of convective snow bands are more favourable with increasing instability. However, the convection is vertically restricted. A capping subsidence inversion usually determines the height of the unstable boundary layer. This convective layer should extend at least 1 km above the surface in order to allow adequate convective cloud growth. Nevertheless, Niziol et al. (1995) indicate that large heat and moisture fluxes from the water surface can significantly lift and even erode the inversion layer.

The wind field throughout the boundary layer plays an essential role for the evolution of convective snow bands. The most common and severe snow bands are aligned parallel to a strong prevailing wind (type I snow bands defined by Niziol et al., 1995) larger than 10 m s-1 (Andersson and Nilsson, 1990). It should be noted that lower wind speeds of less than 5 m s-1 may also lead to the development of shoreline-parallel cloud bands, initiated by a thermally driven land-breeze circulation (type IV snow bands defined by Niziol et al., 1995). A combination of wind speed and wind direction, and thus the distance and path that the air travels across the water surface, determines how much time an air mass of certain properties has to absorb heat and moisture from the water. A longer fetch allows a stronger development of the snow band. Laird et al. (2003) found for idealized cases that cloud bands form when the ratio between the wind speed and the fetch distance over the open water is between 0.02 and 0.09 m s-1 km-1. Accordingly, for a wind speed of 10 m s-1 the fetch distance has to be between 110 and 500 km. Therefore, stronger winds require larger fetch distances.

The directional wind shear within the convective boundary layer is observed to be small for the time period of the snow band event. Niziol (1987) established a criterion for the likelihood of lake effect snow depending on the wind shear within the steering layer of the snow band (which determines approximately the first 50 hPa above the ground up to 700 hPa). Thus, convective snow bands are likely to occur for a directional wind shear of less than 30∘. At a directional wind shear between 30 and 60∘ snow bands are possible to develop; however, beyond 60∘ the band-like structure will break down.

Even the shape and the topography of the coast surrounding the water body can be essential for the snow band evolution. Andersson and Gustafsson (1993) investigated that the genesis areas for convective cloud bands are often bays at the “coast of departure”. A convergence zone develops as two opposed land breezes meet in the centre of the sea. The secondary circulation system forces convection, which continues downwind as bands of convective rolls. When reaching the “coast of arrival”, the convection can be enhanced by another land breeze raising the capping inversion. Snow bands tend to organize themselves parallel to a concave shaped shoreline. Islands that are located along the fetch disturb the heat and moisture flux from the sea to the air mass locally and can cause multiple bands to reorganize and merge or split up.

Case studies with high-resolution numerical weather prediction (NWP) models have been carried out successfully to represent convective snow bands. The present study, however, evaluates the potential of applying a regional climate model with relatively coarse resolution to conduct climatological studies for this mesoscale phenomenon. Performing simulations at a high resolution is computationally expensive and time consuming, and therefore it is less feasible to use a high-resolution NWP model at climatological timescales. An appropriate model for climatological studies of snow bands must balance computational expense and accuracy of the simulated physical processes. Case studies of different model configurations are therefore essential to evaluate the model's performance in order to be aware of potential weaknesses and to give a justified interpretation of the climatological results. In a regional climate model the information on the large-scale forcing is given at the lateral boundaries, and regional response to local conditions (like surface information) can be studied. We also use additional models coupled to the regional atmospheric model (RCA4) to study the importance of more accurate sea surface temperature (by using the ocean model NEMO) or impact of surface waves (by using the wave model WAM).

In order to give a good representation of the Baltic Sea, regional models with high resolution are essential to reproduce topographical features and substantial processes. In connection with the investigation of the mesoscale processes determining convective snow band events, this study was carried out in two parts: the statistical analysis of snow band events based on an 11-year RCA4 dataset and the evaluation of the use of different regional climate model systems.

First, the atmospheric regional climate model RCA4 was used in a high resolution to simulate the atmosphere over the 11-year time period from 2000 to 2010 with a spin-up of 2 months. The horizontal RCA4 resolution was set to 0.16∘, which corresponds to approximately 18 km grid spacing. In order to keep the model numerically stable, the time step is 10 min. Based on the criteria listed in Sect. 1, days of convective snow band conditions were selected and statistically analysed with respect to the season and the strength of the snowfall. The applied criteria are summarized in Table 1 and distinguish with different threshold values between moderate and favourable conditions for snow band development reflecting the local as well as the large scale for the occurrence of precipitation related to convective snow bands. The investigation is focused on convective snow bands which develop over the Baltic Sea and lead to snowfall at the Swedish coast. The criteria were therefore applied to either the Baltic Sea area as seen in Fig. 2a or the specified precipitation sector as seen in Fig. 2b.

In the second part of this study, two case studies are presented, comparing the atmospheric properties simulated by five different model setups in order to give an assessment of the specific model performance concerning snow bands. An overview over the five model systems is given in Table 2. The regional atmospheric climate model RCA4 has been used with resolution of 40 vertical model layers and 0.22∘ (about 25 km) horizontal grid spacing with a spin-up time of approximately 2 months ahead of the snow band event. A coupled simulation of the RCA4 with the ocean model NEMO was carried out to investigate the impact of the SST. In order to provide enough time for adjustment between the models, a spin-up of almost 2 years was used. The two components in the coupled system of RCA4 and WAM are identical with regard to horizontal resolutions and time steps, while WAM provides the RCA model with a sea surface roughness corresponding to the atmospheric wind field above. The final two experiments were performed using RCA4 simulations with increasing resolutions either in horizontal spacing or in vertical direction. The horizontal resolution was refined from 0.22 to 0.11∘ (about 12.5 km) and the 40 model layers of the original RCA4 set-up were increased in a different run to 62 layers. A spin-up time similar to the RCA simulation was used. The higher resolution, however, requires a lower time step, which was therefore decreased from 15 to 5 min.

The simulations by RCA and RCA-WAM resulted in a rather weak development of the convective snow bands and the comparison with observational data showed that the atmospheric conditions were often underestimated by these two models. The coupling of the atmospheric RCA model with the wave model component WAM employed a different roughness length computation at the sea. Hence, the largest impact was observed on the wind field. The magnitude of the maximum 10 m wind speed reaches similar values; however, the RCA-WAM model possesses a time shift due to the wave feedback on the roughness length and wind speed. With regard to all other investigated parameters, RCA and RCA-WAM show very similar results and the different roughness length calculation due to the coupling of RCA and WAM has a negligible impact on the atmospheric conditions describing convective snow band events.

The coupled atmosphere–ocean system RCA-NEMO benefited clearly from the high resolution SST of the NEMO-Nordic component. The reanalysis ERA-40 data which otherwise provided the uncoupled RCA model with the SST had a coarser resolution than the original RCA itself. For water bodies of the size of the Baltic Sea, this resolution is insufficient and the quality of the simulation is impaired as local extremes cannot be represented. The interpolated ERA-40 SST shows a negative bias towards the NEMO SST as well as independent datasets such as OISST. The difference is significant with up to 5 ∘C between ERA-40 and NEMO. While the NEMO simulated SST was shown in some locations to be slightly higher than the OISST data, it provided a better representation of the sea surface properties than ERA-40. The higher SST of the RCA-NEMO model caused furthermore larger heat fluxes and it increased the instability through a larger temperature difference within the lower layers resulting in an enhanced convection and a higher boundary layer height and finally, larger local precipitation rates.

A high resolution is of great importance when it comes to the regional modelling of mesoscale high-impact events. Increasing the resolution of the atmospheric RCA model resulted in a great improvement for the model performance. Both high-resolution simulations indicate larger values for the local maximum of the 10 m wind speeds over the Gulf of Bothnia. The best agreement with observational data, however, was obtained by the RCA model of increased horizontal resolution. Although the temperature and heat fluxes did not show any impact from an increased resolution in any direction, the high horizontal resolution RCA was able to represent the mesoscale atmospheric circulation process associated with convective snow bands in more detail and resolved the local precipitation rate and area more precisely when compared to the other models.

Finding a method to select convective snow band events is not straightforward. Even though the atmospheric conditions are typical and can be described by certain criteria, the thresholds applied imposed a larger impact on the number of days which are selected. Although the criteria themselves were chosen based on references describing snow bands developing in other regions of the world, they are generally valid and the specific values for the selection were adjusted for the Baltic Sea region. The determination of two different categories distinguishing between weak and moderate conditions hints at a need to be more careful with the pool of days fulfilling only weak conditions. Since convective snow bands arise from cloud bands that may initially not give any precipitation, it is not always clear how to define the transition and where to set the threshold for heavy precipitation due to snow bands. The average precipitation did not appear very intense, as some days have been selected with rather low snowfall rates of snow bands which could not have developed as strongly. According to the results of the case studies, the RCA model often underestimates the exact amount of precipitation. The increased resolution improves the results on the representation of the snow bands; however, for a long dataset of 11 years it is very computationally intensive to run the simulations in such a high horizontal resolution as was done in the case studies. Also, the use of the ERA-40 data as sea surface input has shown its drawbacks, and caused the RCA model to simulate snow bands in a less intense evolution. The exact amount of the snowfall and its return period should therefore be interpreted with caution. However, the result for the distribution and the relative amount of the snowfall can be understood in a qualitative rather than a quantitative sense. The Gävle region possessed the largest average snowfall rates and the shortest return periods for comparatively high hourly snowfall rates as a result of snow bands developing over the Gulf of Bothnia. It seems natural that convective snow bands which require a cold air outbreak over the warmer sea will develop more frequently in the most northern part of the Baltic Sea. The shape of the Gulf appears furthermore conducive to the generation of convergence zones for the initial formation and the bay-shaped coast in the Gävle region enhances the precipitation due to orographic forcing once again. The snow precipitation occasionally reaches 100 km inland; however, the most intense lake effect snow falls approximately within a radius of 50 km.

Northeast winds also often lead convective snow bands to the Västervik region; however, the average snowfall here is not as intense nor as frequent. Convective snow bands occur often in multiple band structures and it is common that various regions along the coast are affected by the lake effect snow when the atmospheric conditions are favourable in a large area. Hence, the Gdansk region was also perceived to experience convective snow bands on the same days as the Swedish coast, when all criteria were fulfilled as for convective snow bands along the Swedish coast (see Table 1 applied to the regions represented in Fig. 2).

Since convective snow bands occur in different strengths around the year, a statistical analysis as performed in the first part of this study must define the range of snow band criteria depending on their intensity. The present study includes snow bands that caused moderate snowfall. However, in order to investigate stronger snow band events with hazardous consequences a longer period of time should be studied with stricter criteria for the precipitation since their occurrence it not as frequent.

The investigation of an 11-year RCA4 climate model dataset indicated the heaviest and most frequent lake effect snow due to convective snow bands in the Baltic Sea area affecting the Gävle region. The Västervik region at the Swedish coast and even the Gdansk area at the Polish coast also experienced enhanced snowfall on days of favourable atmospheric conditions for snow bands. When including days of moderate conditions for convective snow band development, a total of 11 days per year on average results. Most convective snow band events occur in the months of November and December, when the sea surface is still warm from the summer and the cold air approaches frequently from the cold Finnish land.

The application of RCA4 in different model setups has indicated for the two case studies that any of the investigated RCA4 model configurations simulate the atmospheric conditions for convective snow bands and fulfil the criteria established in previous research. Nevertheless, significant differences have been observed between the different model systems.

The RCA and RCA-NEMO model varied largely due to the different SST input provided by coarse reanalysis data or the high-resolution ocean model respectively. The coupled RCA-NEMO model provided a superior representation of the sea surface with significantly higher SST values when comparing with the ERA-40 data. The direct impact of the higher NEMO SST on the heat fluxes and convective development manifests itself through a more intense convective snow band development and higher local precipitation rates. Even if on a larger scale all models agree well on the overall precipitation area, the exact location on a smaller scale as well as the amount and time of the snowfall remain a challenge. The models differed to a great extent in the amount of accumulated precipitation. The largest precipitation rates were given by the two high-resolution models as well as the atmosphere–ocean model.

Since the atmosphere–ocean interaction is of great importance for the regional climate modelling of events like convective snow bands, the coupling with the high-resolution ocean model NEMO is advantageous compared with the use of the coarse reanalysis data used in the original RCA model. Moreover, the increased resolution of the atmospheric RCA model had a positive impact on the model results. The high horizontal resolution led to an especially significant improvement in the representation of the cloud bands, the precipitation area, as well as the wind speed. Based on the investigation of the two cases the use of a coupled atmosphere–ocean system in connection with a high horizontal resolution of the atmospheric component is suggested for a more accurate representation of convective snow bands in regional climate models.