Abstract

Accurate numerical weather prediction of intense snowfall events requires the correct representation of dynamical and physical processes on various scales. In this study, a specific event of high-impact wet snowfall is examined that occurred in the northwestern part of Germany in November 2005. First, the synoptic evolution is presented, together with observations of precipitation type and vertical temperature profiles, which reveal the existence of a so-called potential melting layer during the early period of wet snowfall. During the main part, the performance of the operational forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) is investigated. It is shown that only the short-term predictions captured the snowfall event, whereas earlier forecasts were in error concerning the phase and/or amount of precipitation. However, even the short-term forecasts produced the onset of surface snowfall too late (i.e., during the dry snowfall period). Reasons for the misforecasts are errors on various scales. For the early forecasts, they include an inaccurate representation of the upper-level trough and a misplacement of the surface cyclone. For the later forecasts, a slight overestimation of the depth of the potential melting layer and a potentially too fast snow melting process in the model lead to the erroneous prediction of surface rainfall during the wet snowfall period. Hindcast experiments with the high-resolution Consortium for Small-Scale Modeling (COSMO) model also point to the necessity of improving its snow melting parameterization in order to provide useful predictions of potentially high-impact wet snowfall events.

1. Introduction

Winter storms in midlatitudes associated with heavy snowfall or freezing rain often produce hazardous surface conditions that can paralyze community life by causing communication problems and traffic hazards. As a recent example, the so-called Great 2008 Chinese Ice Storm, consisting of four waves of freezing rain during the 4 weeks after 10 January, led to economic losses of more than $20 billion U.S. in the south-central region of the country (Zhou et al. 2011). Due to these potentially dramatic socioeconomic consequences, the atmospheric conditions and meteorological processes associated with snowfall and ice storm events have long been of particular interest. A better understanding of these events is also a prerequisite for improving their prediction.

In terms of the timing, amount, and phase of precipitation, events of heavy snowfall are particularly demanding for current numerical weather prediction (NWP) models. Seemingly small temperature errors, for instance, can cause an erroneous prediction of a moderate rain event instead of intense snowfall. Heavy snowfall events often occur with ambient air temperatures close to 0°C, that is, in a regime where microphysical parameterizations are particularly sensitive to temperature uncertainties. However, accurate forecasts of surface snowfall pose challenges on all scales involved. Uncertainties in the simulated flow on the planetary and synoptic scales can lead to a misrepresentation of the weather system that is causing the precipitation event (typically an extratropical cyclone). On the mesoscale, shortcomings in the prediction of the frontal structures can cause significant errors in the vertical temperature and humidity structure. Eventually on the microscale, the treatment of the microphysical processes in particular with the formation, sublimation, and melting of snow can lead to erroneous forecasts of the phase and amount of precipitation. It appears evident that an accurate numerical prediction of a midlatitude snowstorm is only possible if the forecast system captures the essential processes and their interactions on all scales.

An early study on the prediction of snow versus rain by Penn (1957) has previously highlighted the importance of accurately forecasting the atmospheric temperature profile and emphasized the relationship between this profile and the melting process. This relation has been intensively investigated during the last decades (e.g., Lumb 1961; Bocchieri 1980; Lumb 1983; Czys et al. 1996; Zerr 1997; Bourgouin 2000; Rauber et al. 2001). Stewart (1985) and Thériault et al. (2006) introduced classes of temperature profiles that are associated with distinct precipitation types. Following these studies, a simple classification of temperature profiles with three categories will be used in the present study (Fig. 1):

  1. profiles without a potential melting layer,1 that is, profiles where T(Z) < 0°C for all z;

  2. profiles with a potential melting layer [T(Z) > 0°C; z1 > z > z2 ≥ 0] and a potential refreezing layer below the melting layer [T(z) < 0°C; z < z2]; and

  3. profiles with a potential melting layer [T(z) > 0°C; z1 > z ≥ 0] and without a potential refreezing layer below.

Snow is typically observed with profiles of category 1 but might also occur with profiles in categories 2 and 3 because of the finite time it takes for the snow to melt. The depths of the potential melting and refreezing layers as well as their maximum or minimum temperature, respectively, play an important role for the resulting precipitation type at the surface.

Fig. 1.

Three categories of atmospheric temperature profiles, which are typically associated with different surface precipitation types (dmelt denotes the depth of the potential melting layer): 1) profile without a potential melting layer typically associated with snow, 2) profile with potential melting and refreezing layers typically associated with ice pellets/freezing rain, and 3) profile with a near-surface potential melting layer, typically associated with rain or wet snow.

Fig. 1.

Three categories of atmospheric temperature profiles, which are typically associated with different surface precipitation types (dmelt denotes the depth of the potential melting layer): 1) profile without a potential melting layer typically associated with snow, 2) profile with potential melting and refreezing layers typically associated with ice pellets/freezing rain, and 3) profile with a near-surface potential melting layer, typically associated with rain or wet snow.

In addition to examining the linkage between the surface precipitation type and the vertical structure of the atmosphere, the snow melting process itself has been an active area of research (e.g., Knight 1979; Matsuo and Sasyo 1981a,b; Fujiyoshi 1986; Mitra et al. 1990). Mitra et al. (1990) conducted experiments of melting snowflakes under free-fall conditions in a vertical wind tunnel and developed a detailed theoretical description of the melting process for snowflakes based upon their experimental results. They also investigated the water generated by melting within a single snowflake. Knight (1979) studied the different forms of snowflakes in the various stages of melting and concluded that the generated meltwater does not uniformly cover stellar snow crystals, but flows from the periphery to the linkages of its branches. Additionally, Matsuo and Sasyo (1981b) emphasized the role of atmospheric humidity in that the melting process is slower under drier conditions, as later confirmed by Mitra at al. (1990). Makkonen (1989) presented a critical curve that separates dry and wet snow depending on air temperature and humidity.

Finally, the previously mentioned large-scale atmospheric conditions play an important role for forecasts of snowfall events, as investigated for instance by Langland et al. (2002). They found a strong dependency of forecast quality on errors on the synoptic scale and highlighted the strong sensitivity to the initial conditions. Also Zhang et al. (2002) identified a strong sensitivity of the mesoscale precipitation distribution for a U.S. east coast snowstorm to small changes in the initial conditions. Several other studies investigated the evolution of the meteorological conditions associated with intense snowfall events (e.g., Halcomb and Market 2003; Evans 2006; Mass 2008). The analysis of the synoptic conditions leading to extreme lake-effect snow over the state of New York by Lackmann (2001) revealed that for the examined region these events are connected with an upper-level trough or a closed low at 500 hPa. Bednorz (2008, 2011) focused on heavy snowfalls in the Polish–German lowlands and found that these events can be associated with negative anomalies in sea level pressure and also in the 500-hPa geopotential height field over central Europe. The presence of a typically short-wavelength upper-level trough and a typically small-scale surface cyclone are also frequent features of other intense snowfall events in North America (e.g., Ferber et al. 1993; Marwitz and Toth 1993), eastern Europe (Georgescu et al. 2009), and China (Wang et al. 2011). In contrast, Evans (2006) studied heavy, banded snowstorms that are particularly difficult to predict because of the lack of synoptic-scale forcing. The examination showed that a correct humidity field was essential for an exact prediction of the snowband position. The investigations of, for example, Halcomb and Market (2003) focused on even smaller-scale convectively driven events of so-called thundersnow, which can lead to banded areas of intense snowfall.

It is striking that most of the previous case studies focused on heavy snowfall events in North America. Therefore, the present study is one of the first dealing with a detailed analysis of the dynamics and predictability of a specific high-impact event of heavy snowfall in central Europe, more precisely in North Rhine-Westphalia in northwestern Germany in November 2005. The key objectives of this study are

  1. to investigate the synoptic-scale meteorological evolution and the vertical temperature structure associated with the event, based upon observations and operational analyses from the European Centre for Medium-Range Weather Forecasts (ECMWF) (section 2),

  2. to assess the forecast performance of the deterministic and ensemble-based ECMWF forecasting system for precipitation amount and type at various lead times up to 6 days (section 3), and

  3. to quantify the relationship between the depth of the potential melting layer and the surface precipitation type in deterministic ECMWF forecasts and hindcast experiments with the mesoscale limited-area model from the Consortium for Small-Scale Modeling (COSMO; sections 3 and 4).

The main results of this study will be summarized in section 5, together with a brief outlook on potential implications for further research.

2. Observational analysis of the event

On 25/26 November 2005, heavy snowfall occurred in the northwestern part of Germany, especially in North Rhine-Westphalia (NRW), the most populous state of Germany. In some regions of NRW, a disaster alert was declared. Near-surface temperatures were slightly above the freezing point when snowfall started and therefore the snow was very wet. The wet snow load and strong winds caused severe damage to 82 power line towers, several of which were snapped (Makkonen and Wichura 2010). As a consequence, approximately 250 000 residents were without electricity supply for several hours, some even for several days. The airport at Duesseldorf was closed for several hours, trains were delayed, and roads were blocked by fallen trees. Thousands of people spent the night of 25/26 November in their cars. Makkonen and Wichura (2010) investigated the wet snow load and the snow deposition diameter of the affected power lines in NRW due to wet snow accretion (Makkonen 1989). They found the diameter of the wet snow cylinders surrounding the cables to be up to approximately 15 cm, indicating a maximum wet snow load up to 50 N m−1. Using climatological data, they identified a return period of around 50 yr for such a hazardous wet snowfall event in northwest Germany.

In the Münsterland, snow depths up to 30 cm were reported on 26 November 2005. The measurements at the weather stations in Essen (51.24°N, 6.58°E) and Münster/Osnabrück (52.08°N, 7.42°E) provide insight into the evolution and intensity of the precipitation event (Fig. 2). (The location of each station is indicated by an asterisk in Fig. 6a.) The measurements reveal that snowfall in Essen started at 0300 UTC 25 November, 6 h earlier than in Münster/Osnabrück, and lasted until the end of 27 November (colored dots at the top of the panels represent the observed precipitation type). During the previous day, on late 24 November, surface rainfall occurred, which gradually changed into wet snowfall and later dry snowfall. The time evolution of the snow depth (yellow lines) indicates approximate snowfall rates. The period of maximum snow accumulation at Münster/Osnabrück occurred between 1200 UTC 25 November and 0000 UTC 26 November when snow height increased from about 5 to 28 cm (Fig. 2b). At Essen there was a first phase of wet snowfall on early 25 November and a second, intense dry snowfall period between 1800 UTC 25 November and 0000 UTC 27 November, leading to a maximum snow depth of 23 cm (Fig. 2a). The weather stations at Diepholz (52.35°N, 8.21°E) and Bad Salzuflen (52.06°N, 8.45°E) reported similar evolutions of snow depth except for lower maximum values (10 and 13 cm, respectively). The black lines in Fig. 2 show 12-hourly accumulated precipitation values (water equivalent of snowfall), which exceed 5 mm (12 h)−1 during two and three 12-hourly intervals at the stations of Essen and Münster/Osnabrück, respectively. Peak values amount to 12–16 mm (12 h)−1, which occur during the transition phase from surface rain to wet snowfall at both stations. The moderate increase in surface snow height during these periods reflects the high density of the wet snow.

Fig. 2.

Observations at the weather stations at (a) Essen (51.24°N, 6.58°E) and (b) Münster/Osnabrück (52.08°N, 7.42°E). The location of each station is indicated by an asterisk in Fig. 6a. Yellow lines represent the snow depth (cm) and black lines the 12-hourly accumulated precipitation [mm (12 h)−1]. The colored dots at the top denote the observed precipitation types.

Fig. 2.

Observations at the weather stations at (a) Essen (51.24°N, 6.58°E) and (b) Münster/Osnabrück (52.08°N, 7.42°E). The location of each station is indicated by an asterisk in Fig. 6a. Yellow lines represent the snow depth (cm) and black lines the 12-hourly accumulated precipitation [mm (12 h)−1]. The colored dots at the top denote the observed precipitation types.

The ECMWF analyses (Figs. 3 and 4 ) serve to illustrate the synoptic-scale evolution of the event. The initial situation at 0000 UTC 24 November was characterized by a large-scale low pressure system north of Scandinavia, which was located underneath a pronounced upper-level trough (Fig. 3a). South of Greenland, a large and very intense high pressure system with a core pressure of almost 1040 hPa blocked the westerly flow and steered polar airmasses toward western Europe. An upper-level cutoff was located over the Alps and the western Mediterranean, as can be seen in the 310-K potential vorticity (PV) field, associated with relatively cold air in the lower troposphere (Fig. 4a). Over the North and Baltic Seas, relatively warm air prevailed in between the colder airmasses farther north and south (Fig. 4a). During the next day, the southward protrusion of the upper-level trough to the east of the quasi-stationary blocking event led to the generation of a secondary surface cyclone (Fig. 3b), which developed first over Norway and later over the North Sea in the area of relatively warm low-tropospheric temperatures (Fig. 4b). The tongue of warmer air narrowed as the cold air advanced and a warm seclusion emerged in the center of the newly formed cyclone (Fig. 4c). Note however that the horizontal temperature contrasts are fairly weak in the warm-frontal zone associated with this cyclone and that frontogenesis (not shown) remains weak during the snowfall event. At 0000 UTC 25 November, the cyclone was located underneath the tip of the upper-level PV streamer and reached the Netherlands coast with a core pressure of less than 985 hPa (Fig. 3c). Snowfall set in along the weak frontal zone over France, the Netherlands, and Germany (Fig. 4c), moving from west to east. Twelve hours later the quasi-stationary cyclone further intensified to a core pressure below 975 hPa, and the PV streamer started to wrap up cyclonically (Fig. 3d). During the following day (i.e., until 1200 UTC 26 November), the meteorological situation in NW Europe remained remarkably stationary (Figs. 3e,f and 4e,f). The intensity of the cyclone and the surface warm anomaly became slightly weaker, but surface snowfall continued to the south of the cyclone center.

Fig. 3.

ECMWF analyses at (a) 0000 UTC 24 Nov, (b) 1200 UTC 24 Nov, (c) 0000 UTC 25 Nov, (d) 1200 UTC 25 Nov, (e) 0000 UTC 26 Nov, and (f) 1200 UTC 26 Nov 2005. Gray shading denotes PV on the 310-K isentrope (in PVU) and black lines represent sea level pressure (contour interval is 5 hPa).

Fig. 3.

ECMWF analyses at (a) 0000 UTC 24 Nov, (b) 1200 UTC 24 Nov, (c) 0000 UTC 25 Nov, (d) 1200 UTC 25 Nov, (e) 0000 UTC 26 Nov, and (f) 1200 UTC 26 Nov 2005. Gray shading denotes PV on the 310-K isentrope (in PVU) and black lines represent sea level pressure (contour interval is 5 hPa).

Fig. 4.

ECMWF analyses at the same times as in Fig. 3. Gray shading denotes equivalent potential temperature (K) at 850 hPa and thin black lines represent sea level pressure (contour interval is 5 hPa). Thick black lines indicate the 12-hourly accumulated short-range ECMWF snowfall prediction [contour value of 2 mm (12 h)−1].

Fig. 4.

ECMWF analyses at the same times as in Fig. 3. Gray shading denotes equivalent potential temperature (K) at 850 hPa and thin black lines represent sea level pressure (contour interval is 5 hPa). Thick black lines indicate the 12-hourly accumulated short-range ECMWF snowfall prediction [contour value of 2 mm (12 h)−1].

The cyclone’s spiral structure, indicated by the upper-level PV and low-level temperature fields, is confirmed by the infrared satellite image at 1107 UTC 25 November (Fig. 5). The cloud band at the end of the spiral and south of the cyclone center is well apparent. The intense snowfall in NRW occurred in the region of this cloud band at the leading edge of the near-surface warm-air cutoff (Fig. 4d). A closer examination of surface observations (Fig. 6a) corroborates the location of the band of precipitation (gray shading) at 1200 UTC 25 November in the northern part of NRW in front of the warm-air cutoff (dashed lines). The observed precipitation types (colored dots) indicate snowfall in this area. At this time, snowfall was observed at Münster/Osnabrück (red asterisk symbol in Fig. 6a). At Essen (black asterisk symbol in Fig. 6a), the wet snowfall period had already stopped and the dry snowfall period started a few hours later (see also Fig. 2a). One day later (Fig. 6b), the measured snow depth exceeds 20 cm in the most affected area, which is marked by a black rectangle.

Fig. 5.

Thermal infrared (10.8–11.3 μm) satellite image of the National Oceanic and Atmospheric Administration polar-orbiting satellite (NOAA-17) at 1107 UTC 25 Nov.

Fig. 5.

Thermal infrared (10.8–11.3 μm) satellite image of the National Oceanic and Atmospheric Administration polar-orbiting satellite (NOAA-17) at 1107 UTC 25 Nov.

Fig. 6.

Weather situations at (a) 1200 UTC 25 Nov and (b) 1200 UTC 26 Nov 2005. Gray shading denotes the amount of precipitation for the time periods (a) 1100–1200 UTC 25 Nov (mm h−1) and (b) 1200 UTC 25 Nov–1200 UTC 26 Nov [mm (24 h)−1] from a gridded observational dataset (Paulat et al. 2008). Note that no data are available outside of Germany. The solid black lines represent sea level pressure (in hPa) and the dashed black lines equivalent potential temperature (K) at 850 hPa from the ECMWF analyses. Colored dots in (a) indicate the observed precipitation type at low-altitude weather stations (below 200 m) between 50° and 55°N in Germany (asterisks for Essen and Münster/Osnabrück stations). Red numbers in (b) represent measured snow depths and the black rectangle (51.15°–52.50°N, 6.35°–8.80°E) indicates roughly the area of heavy snowfall.

Fig. 6.

Weather situations at (a) 1200 UTC 25 Nov and (b) 1200 UTC 26 Nov 2005. Gray shading denotes the amount of precipitation for the time periods (a) 1100–1200 UTC 25 Nov (mm h−1) and (b) 1200 UTC 25 Nov–1200 UTC 26 Nov [mm (24 h)−1] from a gridded observational dataset (Paulat et al. 2008). Note that no data are available outside of Germany. The solid black lines represent sea level pressure (in hPa) and the dashed black lines equivalent potential temperature (K) at 850 hPa from the ECMWF analyses. Colored dots in (a) indicate the observed precipitation type at low-altitude weather stations (below 200 m) between 50° and 55°N in Germany (asterisks for Essen and Münster/Osnabrück stations). Red numbers in (b) represent measured snow depths and the black rectangle (51.15°–52.50°N, 6.35°–8.80°E) indicates roughly the area of heavy snowfall.

The evolution of the vertical temperature structure over NRW between 24 and 26 November can be investigated with the aid of the 12-hourly radiosonde ascents in Essen (Fig. 7). During the first 2 days, all soundings display a potential melting layer (i.e., a layer with temperatures exceeding 0°C). On 24 November, both soundings additionally show a potential refreezing layer with temperatures below 0°C beneath the potential melting layer. The structure of these profiles can be compared with the observed precipitation types in Essen (Fig. 2a). Ice pellets were reported at 1500 UTC 24 November, consistently with the observed refreezing layer below an elevated potential melting layer 3 h earlier at 1200 UTC. Thériault et al. (2006) indicated that this type of vertical profile is typical for the occurrence of ice pellets (Fig. 1). On 25 November wet snow was observed in Essen. The soundings on this day reveal a shallow near-surface potential melting layer, with depths of 23.25 and 16.00 hPa at 0000 and 1200 UTC, respectively. Together with relative humidity values close to saturation in this layer (not shown), this indicates that the snow started to melt before reaching the ground, which explains the observed wet snowfall at the beginning of the event (i.e., shortly after 0000 UTC 25 November). According to the surface observations in Essen (Fig. 2a), there was no precipitation right at the time of the 0000 UTC sounding. This might explain the absence of an isothermal 0°C layer, which is often associated with melting snow. The surface temperature development in Essen shows a decrease by about 1°C from 0200 UTC to temperatures slightly above 0°C at 0600 UTC on 25 November. During this period, at 0300 UTC when surface temperature was 0.4°C, the precipitation type observation in Essen, provided every 3 h, indicated snow for the first time (Fig. 2a). We infer that snowfall started in the presence of a potential melting layer, which possibly was slightly shallower than the 23 hPa observed at 0000 UTC, which was slightly before snowfall started. On 26 November the soundings show no potential melting layer. The surface temperature dropped below 0°C, consistent with the surface observation of dry snow. Note also that the temporal evolution of the starting points of the vertical soundings illustrate the rather spectacular surface pressure decrease in Essen due to the approach of the intensifying cyclone, from more than 1010 hPa at 0000 UTC 24 November to about 965 hPa 36 h later. Clearly, it is not ideal to only have single profile observations at 12-hourly intervals for studying the precipitation evolution in an entire region. However, the qualitative agreement between the profile characteristics and the observed surface precipitation types at Essen indicates that considering the temporal evolution of the temperature profile is meaningful and also relevant when evaluating the performance of the NWP models.

Fig. 7.

Measured temperature profiles at Essen (51.40°N, 6.96°E) at 0000 and 1200 UTC on 24, 25, and 26 Nov 2005, respectively, between the surface and 800 hPa. Gray shading highlights potential melting layers with T > 0°C.

Fig. 7.

Measured temperature profiles at Essen (51.40°N, 6.96°E) at 0000 and 1200 UTC on 24, 25, and 26 Nov 2005, respectively, between the surface and 800 hPa. Gray shading highlights potential melting layers with T > 0°C.

3. ECMWF forecast performance

In this section, various aspects of ensemble and mainly deterministic forecasts from the ECMWF are investigated for this high-impact snowfall event. The focus of the analysis will be on the phase and amount of surface precipitation, and on the processes that are responsible for inaccuracies in the predictions. Due to the presence of a potential melting layer leading to the wet snowfall, the parameterization of snow melting becomes a particularly relevant aspect of the model for predicting this event. The operational ECMWF model in 2005 treated precipitating rain and snow diagnostically (ECMWF 2011), which indicates that horizontal advection of precipitating hydrometeors is neglected on the scale of the model grid resolution. The scheme produces precipitation due to autoconversion from the prognostic total cloud condensate and this precipitation is then removed from a grid-box column within one time step. The instantaneous vertical structures of temperature and humidity are used to calculate potential melting and evaporation during the sedimentation. Melting of snow occurs if the air temperature is higher than 0°C. The model outputs both total precipitation and snow, which allows us to calculate the rain to snow ratio. We just note that since November 2010, a new cloud scheme is operational at ECMWF, which contains prognostic variables for cloud water, cloud ice, rain, and snow (Forbes and Tompkins 2011).

a. Ensemble forecasts of surface snowfall

First, we consider the forecasts of surface snowfall of the ECMWF Ensemble Prediction System (EPS, with a spectral resolution of T255L40, corresponding to a grid spacing of about 80 km) in order to get an impression of the potential of the EPS to predict this intense snowfall event. The snowfall amount in the most affected region, shown by the black rectangle in Fig. 6b, between 0600 and 1800 UTC 25 November (i.e., for the time period of maximum wet snowfall at Münster/Osnabrück), was evaluated for the 50 ensemble members of the forecasts initiated every 12 h from 0000 UTC 19 November to 0000 UTC 25 November. For every EPS forecast, the probability of the snow water equivalent exceeding thresholds between 1 and 5 mm in 12 h was calculated, as shown in Fig. 8. For instance, a probability of 20% indicates that snowfall in 10 out of 50 ensemble members exceeded the considered threshold. Recall that station observations indicated locally larger values than 5 mm during the considered 12-h period.

Fig. 8.

Probabilities for exceeding different threshold values of 12-hourly accumulated snowfall between 0600 and 1800 UTC 25 Nov 2005 in the black rectangle (Fig. 6b) for various start times of the ECMWF EPS between 0000 UTC 19 Nov (1900) and 0000 UTC 25 Nov (2500).

Fig. 8.

Probabilities for exceeding different threshold values of 12-hourly accumulated snowfall between 0600 and 1800 UTC 25 Nov 2005 in the black rectangle (Fig. 6b) for various start times of the ECMWF EPS between 0000 UTC 19 Nov (1900) and 0000 UTC 25 Nov (2500).

For all forecasts started until 1200 UTC 23 November (i.e., for lead times of more than 2 days), the EPS indicated a low probability of about 20% for snowfall of more than 1 mm during the considered 12 h. For larger thresholds the probability of exceedance was close to zero. For the forecasts started at 0000 and 1200 UTC 24 November, the probabilities increase for the lower thresholds, but an appreciable increase in the probability for an intense snowfall event occurs only for the short-range ensemble forecast started at 0000 UTC 25 November. These forecasts indicate a probability larger than 60% for a snow water equivalent of more than 3 mm during the 12 h. We conclude from this brief consideration of the ECMWF ensemble that probabilistic forecasts did not provide useful guidance for predicting the intense snowfall event more than 1 day in advance.

b. Deterministic forecasts

We now investigate deterministic ECMWF forecasts (with a spectral resolution of T511L60, corresponding to a grid spacing of about 40 km) in order to gain insight into the capability of the relatively high-resolution global model to predict this event. The investigated forecasts were started every 12 h between 0000 UTC 19 November and 0000 UTC 25 November. Different comparisons with observations from weather stations, described in section 2, and with ECMWF analysis fields provide insight into the accuracy of the forecasts in terms of surface precipitation, the vertical temperature structure in the snowfall region, and the representation of the synoptic-scale characteristics of the event at the surface and the tropopause level.

1) Surface precipitation phase and amount

In a first step the predicted surface precipitation phase is investigated by calculating the averaged amounts of snow and rain in the area of observed heavy snowfall (i.e., in the black rectangle shown in Fig. 6b). A subjective threshold is required to classify the predicted precipitation type as snow or rain, which then allows a comparison with the observed precipitation categories. We chose the following pragmatic approach: model precipitation in the rectangle is categorized as “snow” if the amount of predicted snow exceeds the amount of predicted rain and vice versa. The comparison with observations (Fig. 9) shows that early forecasts fail completely in predicting surface snowfall (1200 UTC 22 November and 0000 UTC 23 November) or produce snowfall much too late (e.g., 0000 UTC 21 November and 1200 UTC 23 November). Only short-range forecasts (0000 and 1200 UTC 24 November and 0000 UTC 25 November) capture the time evolution of the surface precipitation phase with reasonable accuracy. However, during the wet snowfall period starting in Essen at 0300 UTC 25 November, none of the forecasts predicted snow. Snowfall is only predicted when, according to the observations, surface precipitation changed from wet to dry snow. Here, it is important to note that our analysis is based upon a very simple rain versus snow discrimination and an experienced forecaster [e.g., with the help of a statistical postprocessing approach; Roebber et al. (2003)] might be able to infer an increased likelihood of intense wet snowfall when the forecast indicates a mixture of mainly rain and a smaller contribution of snow during a 6- or 12-hourly time period. Our main intention here is to highlight the rain-dominated surface precipitation forecasts, even of the best short-range ECMWF predictions of this high-impact wet snowfall event.

Fig. 9.

Observed precipitation types at Essen and Münster/Osnabrück (bottom rows), and predicted precipitation types in the black rectangle (see Fig. 6b) derived from deterministic ECMWF forecasts. Every horizontal line is for one forecast labeled along the y axis. Precipitation type has been classified only if the amount of predicted precipitation exceeded 0.5 mm (6 h)−1. Precipitation is categorized as snow if its amount is higher than the amount of rain and vice versa.

Fig. 9.

Observed precipitation types at Essen and Münster/Osnabrück (bottom rows), and predicted precipitation types in the black rectangle (see Fig. 6b) derived from deterministic ECMWF forecasts. Every horizontal line is for one forecast labeled along the y axis. Precipitation type has been classified only if the amount of predicted precipitation exceeded 0.5 mm (6 h)−1. Precipitation is categorized as snow if its amount is higher than the amount of rain and vice versa.

Investigation of the predicted snowfall amount (not shown) reveals that most forecasts underestimate the amount of surface snow during the observed period of maximum snowfall by predicting values less than 1 mm (12 h)−1. Significantly larger values of about 5 mm (12 h)−1 or more are only predicted by the short-range forecasts (1200 UTC 24 November and 0000 UTC 25 November) for the time period from 0600 to 1800 UTC 25 November. This model validation of surface precipitation amount and type indicates a large variability in the quality of the deterministic forecasts, in agreement with the previous inspection of the EPS. In the following, the forecast representation of the synoptic-scale features and of the vertical temperature structure in the area of heavy snowfall are investigated with the aim of potentially identifying processes leading to the surface precipitation misforecasts.

2) Upper-level trough

As shown in the previous section, the genesis and evolution of the cyclone that produced heavy snowfall in NRW was mainly determined by the approach and further development of a prominent and relatively small-scale upper-level trough. In the upper-level PV field this trough took the shape of a PV streamer. Figure 10 shows the dynamical tropopause [i.e., the 2-PVU contour, where 1 PV unit (PVU) = 10−6 m2 s−1 K kg−1] on the 310-K isentrope for selected forecasts (colored) and the analysis (black) at 1200 UTC 25 November, that is, at a time when wet snowfall was about to end in Essen but still continued in Münster/Osnabrück. The diagram shows that all forecasts capture the existence of the upper-level trough, however with significant variability and deviations from the analysis. Forecasts for 1200 UTC 22 November and 1200 UTC 24 November show the best agreement of the upper-level flow structure with the analysis. The PV streamer produced by the forecast for 1200 UTC 19 November extends too far to the south and is too narrow. The PV structure in the forecast for 1200 UTC 23 November corresponds better to the analysis but appears to be erroneous over Denmark and the southernmost part of Norway. It is expected that these structural differences at upper levels significantly influence the evolution of the surface cyclone, the accompanying surface temperature structure, and the resulting precipitation. It is well documented by previous studies that relatively small-scale errors in the upper-level PV field can have a large impact on the evolution of extratropical cyclones (e.g., Fehlmann and Davies 1997; Demirtas and Thorpe 1999) and intense precipitation events in central Europe and the Mediterranean (e.g., Fehlmann and Quadri 2000; Argence et al. 2009).

Fig. 10.

The 2-PVU contours of potential vorticity on the 310-K isentrope for four selected forecasts (colored) and the analysis (black) at 1200 UTC 25 Nov 2005.

Fig. 10.

The 2-PVU contours of potential vorticity on the 310-K isentrope for four selected forecasts (colored) and the analysis (black) at 1200 UTC 25 Nov 2005.

3) Surface cyclone and low-tropospheric temperature

The fields of sea level pressure and 850-hPa equivalent potential temperature at 1200 UTC 25 November (Fig. 11) indicate the large variability of the position and intensity of the cyclone and the associated low-tropospheric temperature distribution. To summarize the variability of the predicted cyclones, the positions of the surface cyclone centers in the forecasts and the analysis at this particular time are plotted in the bottom-right panel of Fig. 11. Only the two short-range forecasts for 1200 UTC 24 November and 0000 UTC 25 November, which predicted intense snowfall, capture the cyclone position well with a displacement error of less than 100 km. The intensity of the cyclone is also captured by these forecasts (with a minimum sea level pressure value below 975 hPa), whereas all earlier forecasts produced a too weak cyclone. In most earlier forecasts the cyclone center is too far north, in particular in the 1200 UTC 23 November forecast, consistent with the erroneous upper-level PV structure over Denmark noted before. Also in agreement with the forecast errors at upper levels, the surface cyclone in the 1200 UTC 19 November forecast is located to the SW of the observed position. This clearly confirms the crucial influence of the upper-level flow evolution on the position of the surface cyclone.

Fig. 11.

Surface cyclones and associated structure of the 850-hPa equivalent potential temperature field (gray shading for 288, 290, and 292 K) for every forecast and the analysis at 1200 UTC 25 Nov 2005. The center of each panel corresponds to the center of the cyclone in the particular forecast (black dot). Thin black lines represent sea level pressure (contours every 5 hPa) and boldface dashed lines denote the 0°C temperature contour on the first model level. (bottom right) Locations of the cyclone center in the forecasts and the analysis (label A).

Fig. 11.

Surface cyclones and associated structure of the 850-hPa equivalent potential temperature field (gray shading for 288, 290, and 292 K) for every forecast and the analysis at 1200 UTC 25 Nov 2005. The center of each panel corresponds to the center of the cyclone in the particular forecast (black dot). Thin black lines represent sea level pressure (contours every 5 hPa) and boldface dashed lines denote the 0°C temperature contour on the first model level. (bottom right) Locations of the cyclone center in the forecasts and the analysis (label A).

The quality of the predicted 850-hPa equivalent potential temperature structure is also highly variable (Fig. 11). Most of the early forecasts missed the formation of the warm band extending from the North Sea toward the Netherlands, which is no surprise given the incorrect location and intensity of the surface low. An interesting exception is the 0000 UTC 20 November forecast, which produced a fairly intense cyclone with a prominent warm band located about 200 km too far to the east (i.e., covering the region of observed heavy snowfall in NRW). As a consequence, this forecast produced liquid precipitation in the considered area (Fig. 9), indicating the critical importance of accurately predicting the location of the key synoptic-scale flow features for predicting the correct phase of surface precipitation. Similarly, the warm band predicted by the 1200 UTC 22 November reaches too far into western Germany, leading also to a misforecast of precipitation type. The latest three forecasts have a reasonable representation of this warm band over the Netherlands and the adjacent region of Germany. So far, we can summarize that our detailed investigation of the forecast performance indicates that accurate predictions of the structure of the upper-level trough, the position and intensity of the surface low pressure center, and of the low-level warm band were essential for predicting the investigated snowfall event. With the exceptions of the 1200 UTC 24 November and 0000 UTC 25 November short-term forecasts, and less so for 0000 UTC 24 November, all earlier predictions failed in accurately predicting these key flow structures in the region of the observed wet snowfall event. The very weak frontogenetic activity in the snowfall area revealed by the analyses indicates that the representation of frontogenesis in the forecasts was of comparatively minor importance.

For the three latest forecasts, the predicted vertical temperature structure at 0000 UTC 25 November (i.e., 3 h before wet snowfall started in Essen) is spatially fairly uniform in the region of interest (not shown). A potential melting layer is present near Essen in all forecasts considered (Fig. 12), indicating that sedimenting snow would start to melt. Figure 12 shows the predicted vertical depth of the potential melting layer (black bars) together with the observed value of approximately 23 hPa from the sounding (cf. Fig. 7, red horizontal line). All forecasts predicted rain at that time (Fig. 9). The differences between the forecasts are fairly large, but all forecasts overestimate the depth of the potential melting layer by typically 10 hPa or more. This finding is also valid for the 1200 UTC 24 November forecast, which produced the highest precipitation intensity compared to the other simulations (not shown). The most accurate depth of the potential melting layer is predicted by the 0000 UTC 24 November forecast, but this forecast also produced mainly surface rain during the hours after 0000 UTC 25 November (see Fig. 9). This indicates that, at least for this event, a fairly well predicted low-tropospheric temperature structure also led to a misforecast of surface precipitation type. In turn, this leads to the hypothesis that the simplified snow melting process in the model might be too fast, that is, that total melting of sedimenting snow might occur already for fairly shallow potential melting layers.

Fig. 12.

Forecasted depths of the potential melting layer at 0000 UTC 25 Nov 2005 from the ECMWF deterministic forecasts at Essen (black), the COSMO hindcasts at Essen (green), and the COSMO hindcasts averaged over the black rectangle (Fig. 6b). The red line indicates the observed depth of the potential melting layer (about 23 hPa, from the radiosounding) and the blue bar the ECMWF analysis at Essen.

Fig. 12.

Forecasted depths of the potential melting layer at 0000 UTC 25 Nov 2005 from the ECMWF deterministic forecasts at Essen (black), the COSMO hindcasts at Essen (green), and the COSMO hindcasts averaged over the black rectangle (Fig. 6b). The red line indicates the observed depth of the potential melting layer (about 23 hPa, from the radiosounding) and the blue bar the ECMWF analysis at Essen.

A statistical approach is used for further investigating this hypothesis. For all 6-hourly time periods of the considered forecasts, and at every grid point in northwest Germany and parts of the Netherlands, the depth of the potential melting layer and the “surface snow fraction” have been evaluated. The latter corresponds to the fraction of the total surface precipitation predicted as snow. Figure 13 shows a scatterplot of the relationship between the two parameters for different categories of the total precipitation amount. Based upon an idealized thought experiment, assuming that snow particles continuously melt as they fall through a potential melting layer until they become completely liquid, one might expect a linear relationship, that is, a linear decrease in the snow fraction with an increased depth of the potential melting layer, and a threshold melting-layer depth beyond which no snow reaches the surface. This idea is comparable to the findings of Mitra et al. (1990), especially their Fig. 4, which illustrates the approximately linear decrease of the melted mass fraction of single snowflakes by the distance below the 0°C level. Qualitatively, such a pattern of behavior can be seen in Fig. 13, however with substantial variability.

Fig. 13.

Scatterplot, based upon ECMWF deterministic forecasts, of the predicted 6-hourly surface snow fraction and the depth of the potential melting layer in a region covering northwest Germany and parts of the Netherlands every 6 h of the forecast. Considered are all grid points where snowfall occurred and a potential melting layer existed.

Fig. 13.

Scatterplot, based upon ECMWF deterministic forecasts, of the predicted 6-hourly surface snow fraction and the depth of the potential melting layer in a region covering northwest Germany and parts of the Netherlands every 6 h of the forecast. Considered are all grid points where snowfall occurred and a potential melting layer existed.

There are at least two potential reasons for the deviation from a compact linear relationship. A first reason for the relatively large scatter is the coarse temporal resolution of this investigation: the snow fraction is calculated from 6-hourly accumulated precipitation values and the depth of the melting layer is evaluated as the average between values deduced from the temperature profiles at the beginning and end of this period. A second reason is the variability of relative humidity in the potential melting layer, which can lead to reduced melting due to evaporative cooling in the case of subsaturation. Nevertheless, the general characteristics of the scatterplot show that the ECMWF microphysical snow melting scheme typically leads to small/moderate surface snow fractions (0%–60%) in situations similar to those observed in Essen at 0000 UTC 25 November with a 23.25-hPa-deep potential melting layer. This indicates that it is also very likely that a “perfect” deterministic short-range forecast in terms of the low-tropospheric temperature structure would have predicted mainly (or even only) liquid precipitation during the period of observed wet snowfall. Considering that most ECMWF forecasts had an overestimated melting-layer depth of about 40 hPa, Fig. 13 makes it plausible that they predicted very small fractions of surface snowfall. It appears that a too shallow potential melting layer would be required to predict a dominant fraction of surface snowfall.

4. Hindcast simulations with the limited-area COSMO model

Here, we consider short-term simulations of the same event with a limited-area model with a more complex cloud microphysical parameterization compared to the global model of the ECMWF. The simulations are mainly evaluated in order to diagnose the relationship between the simulated potential melting-layer depth and surface precipitation in this higher-resolution model. Hindcast simulations have been performed with the nonhydrostatic limited-area COSMO model (Steppeler et al. 2003) with a horizontal resolution of 7 km and 40 vertical levels. The vertical resolution of the COSMO simulations and ECMWF forecasts is comparable in the lower atmosphere with about 10 levels in the lowest 100 hPa. ECMWF analyses were used as the initial and boundary data leading to hindcasts. The model was set up in the operational configuration used by the German Weather Service. A one-moment microphysical scheme is used with the four prognostic hydrometeor species—cloud water, rain, cloud ice, and snow—as described in detail by Doms and Schättler (2004). The cloud parameterization generates meltwater from the snow mixing ratio if the temperature is above 0°C (and the air is saturated). The meltwater is instantaneously converted into rain, leading to a more rapid sedimentation of the melted water compared to the remaining snow.

Figure 14 shows the accumulated snowfall between 0000 and 1200 UTC 25 November for the three hindcast simulations—0000 UTC on 23 and 24 November, and 1200 UTC 24 November—together with the gridded rain gauge observations (Paulat et al. 2008; panel d). Also shown are the observed and simulated zones with 2-m temperatures above and below 0°C, respectively, at 0600 UTC 25 November. The two early forecasts (0000 UTC on 23 and 24 November) capture the warm sector over northwest Germany well but only simulate weak snowfall in this area. The short-range 1200 UTC 24 November forecast also captures the temperature structure but additionally reveals a snowband with adequate intensity but located too far to the west. In qualitative agreement with the ECMWF forecasts, the COSMO hindcast simulations also captured the intense snowfall event only about a day in advance. Figure 12 shows, additionally, that the COSMO hindcasts were relatively good in predicting the depth of the potential melting layer at 0000 UTC 25 November, in particular if taking the averaged values in the target area (black rectangle in Fig. 6b) into consideration. The 0000 UTC 23 November and 1200 UTC 25 November hindcasts slightly overestimated the depth, with 25.51 and 29.51 hPa, and produced surface snow fractions in this area during the next 6 h of 53 and 49%, respectively. In contrast, the 0000 UTC 24 November hindcast slightly underestimated the melting-layer depth (19.45 hPa) and produced a surface snow faction of 66%. Note the increased surface snow fraction that occurs in these simulations with a reduced vertical dimension of the potential melting layer. In absolute terms, the 1200 UTC 24 November hindcast produced the largest values of total surface precipitation and of surface rainfall (see also Fig. 14c). These simulations indicate that similarly to the results from the ECMWF forecasts, the COSMO model produces a large amount of surface precipitation in liquid form (about 50%) even with a fairly accurate low-tropospheric temperature structure.

Fig. 14.

Three COSMO model hindcasts of surface snowfall accumulated between 0000 and 1200 UTC 25 Nov started at (a) 0000 UTC 23 Nov, (b) 0000 UTC 24 Nov, and (c) 1200 UTC 24 Nov, and (d) the observed precipitation distribution. The black lines mark areas with amounts larger than 12 mm (12 h)−1. The green lines represent the predicted 2-m temperature 0°C isolines and the colored dots the observed 2-m temperature with red for above, black for equal to, and blue for below 0°C.

Fig. 14.

Three COSMO model hindcasts of surface snowfall accumulated between 0000 and 1200 UTC 25 Nov started at (a) 0000 UTC 23 Nov, (b) 0000 UTC 24 Nov, and (c) 1200 UTC 24 Nov, and (d) the observed precipitation distribution. The black lines mark areas with amounts larger than 12 mm (12 h)−1. The green lines represent the predicted 2-m temperature 0°C isolines and the colored dots the observed 2-m temperature with red for above, black for equal to, and blue for below 0°C.

As for the ECMWF forecasts, scatterplots are produced for the 1200 UTC 24 November COSMO hindcast to quantify the relationship between the simulated surface snow fraction and the depth of the potential melting layer at grid points in northwest Germany and the Netherlands (Fig. 15). In addition to 6-hourly intervals, the hourly output of the COSMO model also allows us to investigate this relationship for hourly accumulated precipitation, which might elucidate the linkage between the two parameters even more clearly. Results from the 6-hourly fields (Fig. 15a) show a level of scatter similar to that for the ECMWF model (cf. with Fig. 13) and also offer a clear indication for a strong decrease of the simulated surface snow fraction with an increasing depth of the potential melting layer. When doing the same analysis with hourly fields (Fig. 15b), the resulting scatterplot becomes much more compact and reveals a fairly robust pattern of behavior for the snow melting in the model: with a potential melting-layer depth of 20 hPa, typically more than 50% of the surface precipitation is simulated as liquid, in qualitative agreement with the results in the target area discussed above. In most situations when the melting layer depth exceeded 40 hPa, no surface snow is simulated by the COSMO model. The more compact character of the hourly compared to the 6-hourly analyses for the COSMO model, and the qualitative agreement of the 6-hourly analyses between the COSMO and ECMWF models, underline the general relevance of considering this relationship. For the particular case investigated in this study, a balloon sounding at the beginning of the wet snowfall period indicated a potential melting-layer depth of about 23 hPa, that is, a value in the range where the COSMO model already produces intense melting of the falling snow.

Fig. 15.

Scatterplot, based upon the COSMO hindcast simulation started at 1200 UTC 24 Nov, of the predicted surface snow fraction of grid-scale snowfall and the depth of the potential melting layer at all grid points between 47° and 55°N and 4° and 11°E at all model output times, for (a) 6-hourly accumulated model output and (b) 1-hourly model output. Considered are all grid points where snowfall occurred and a potential melting layer existed.

Fig. 15.

Scatterplot, based upon the COSMO hindcast simulation started at 1200 UTC 24 Nov, of the predicted surface snow fraction of grid-scale snowfall and the depth of the potential melting layer at all grid points between 47° and 55°N and 4° and 11°E at all model output times, for (a) 6-hourly accumulated model output and (b) 1-hourly model output. Considered are all grid points where snowfall occurred and a potential melting layer existed.

5. Summary and conclusions

Heavy snowfall over northwestern Germany in late November 2005 was connected to the development of a pronounced upper-level trough from Scandinavia to the north of Germany. The associated surface cyclone followed a similar path and became quasi stationary along the Netherlands coast during the snowfall event. Balloon soundings and surface observations showed that snowfall occurred in an area with near-surface temperatures initially slightly above 0°C. The positive surface temperatures and a potential melting layer with a depth of about 23 hPa at the beginning of the wet snowfall period are prominent features of the event. Snow started to melt partially before reaching the ground, which explains the observed wet, and therefore heavy, surface snow responsible for the infrastructural damages in parts of northwestern Germany. In the later phase of the event, surface temperatures fell below 0°C and the surface precipitation changed from wet to dry snow.

The operational ECMWF forecasts showed a sensitivity of the location of the surface low to the position and structure of the upper-level trough. The predicted position of the cyclone center was wrong by several hundred kilometers in early forecasts and was well captured only by short-range (1 day) forecasts, which also predicted the structure of the upper-level trough well. The predicted vertical temperature structure of the forecasts is influenced by the prediction of the larger-scale structures. Therefore, forecasts with an inaccurately predicted trough or cyclone are typically also inaccurate in terms of low-level temperature and surface precipitation. For the surface precipitation phase the deterministic ECMWF forecasts produced rain instead of snow during the wet snowfall period even for simulations that behaved well on the large and synoptic scales. The short-range predictions captured the dry snowfall event but missed the earlier period of wet snowfall in the sense that they produced predominantly liquid precipitation. The investigation of the ECMWF ensemble predictions corroborates the finding that a high probability for intense snowfall was only predicted with a lead time of about 1 day. Also, high-resolution COSMO hindcast simulations reveal substantial amounts of snow close to the target area during the wet snowfall period only for a short-term simulation (with a substantial location error of the band of intense snow). Overall, it appears that the investigated wet snowfall event had a high societal impact but remarkably low predictability. Note that the present case occurred above fairly flat terrain; previous studies of intense snowfall events in complex topography (e.g., Wesley et al. 1995; Schultz et al. 2002) highlighted the additional complexity in predicting such events in situations where interactions of the large-scale flow with small topographic features become important.

In this case, processes on various atmospheric scales influenced the prediction of surface precipitation. Better upper-level trough and surface cyclone predictions in short-range forecasts led to better temperature forecasts. The strong sensitivity of the predicted surface precipitation type on the characteristics of the low-tropospheric temperature profile are evident by looking at the fairly compact relationship between the depth of the potential melting layer and the associated surface snow fraction. This relationship indicates that even fairly small differences in the predicted potential melting-layer depth can lead to strong misforecasts of surface snowfall. For the event in northwestern Germany in late November 2005, intense surface precipitation in the form of wet snow occurred with a potential melting-layer depth of about 23 hPa at the beginning. Predicting this layer depth correctly, the snow melting schemes of both considered models melt a substantial part of the sedimenting snow, which is likely leading to an erroneous prediction of the surface precipitation type. This is in qualitative agreement with the study by Thériault et al. (2010), who found that small variations in the temperature profiles can have a major impact on the predicted surface precipitation type. Part of the problem is that the melting schemes of the two considered models do not account for an internal mixing of snow and meltwater. They transfer the meltwater instantaneously to the rain category, which is inappropriate in situations involving wet snowfall. With such a model, a postprocessing algorithm would be required to diagnose “wet snow” in situations with a certain ratio between liquid precipitation and snow at the surface accumulated during a short time period. With such an approach, it would be difficult to distinguish an extended period of wet snow from a period with a (rapid) transition from rain to snow.

We conclude that in order to improve future predictions of intense wet snowfall events, it is important to further examine the melting process of snow and its sensitivity to the ambient temperature conditions, and to improve parameterizations of this process. A possible improvement could be the extension of the current schemes by including the possibility of predicting semimelted snow. With such a scheme, the water generated by melting would not be instantaneously converted into rain but would remain part of the falling (and melting) snow. This would slow down the snow to rain transition and therefore probably lead to more realistic snow forecast even under atmospheric conditions as in the presented study. Such an idea has already been realized by Szyrmer and Zawadzki (1999), who obtained good agreement with radar data. The implementation of a similar snow melting parameterization including semimelted snow into the COSMO model is under way and might lead to improved forecasts of heavy snowfall events in central Europe. Another important issue when predicting snowfall in situations with a potential melting layer is the accurate consideration of latent cooling due to melting (e.g., Kain et al. 2000; Lackmann et al. 2002). Clearly, this can be achieved only if the sedimentation and melting of the snow hydrometeors are well represented by the microphysical parameterization. Finally, we mention the additional complexity due to the nonhomogeneous character of snow particles in the real atmosphere. Different types of snow particles might be characterized by a different pattern of melting behavior (e.g., due to varying capacitance and ventilation coefficients). Woods et al. (2007) presented promising results with a bulk microphysical scheme that also allows for snow habit prediction.

Acknowledgments

We thank Axel Seifert for insightful discussions on the parameterization of snow melting and Jörg Trentmann and Astrid Kerkweg for their help with the COSMO simulations. We gratefully acknowledge the constructive comments of three anonymous reviewers and we thank the German and Swiss weather services for granting access to the ECMWF analysis and forecast data.

REFERENCES

REFERENCES
Argence
,
S.
,
D.
Lambert
,
E.
Richard
,
J.-P.
Chaboureau
,
P.
Arbogast
, and
K.
Maynard
,
2009
:
Improving the numerical prediction of a cyclone in the Mediterranean by local potential vorticity modifications
.
Quart. J. Roy. Meteor. Soc.
,
135
,
865
879
.
Bednorz
,
E.
,
2008
:
Synoptic reasons for heavy snowfalls in the Polish–German lowlands
.
Theor. Appl. Climatol.
,
92
,
133
140
.
Bednorz
,
E.
,
2011
:
Synoptic conditions of the occurrence of snow cover in central European lowlands
.
Int. J. Climatol.
,
31
,
1108
1118
.
Bocchieri
,
J. R.
,
1980
:
The objective use of upper air soundings to specify precipitation type
.
Mon. Wea. Rev.
,
108
,
596
603
.
Bourgouin
,
P.
,
2000
:
A method to determine precipitation types
.
Wea. Forecasting
,
15
,
583
592
.
Czys
,
R. R.
,
R. W.
Scott
,
K. C.
Tang
,
R. W.
Przybylinski
, and
M. E.
Sabones
,
1996
:
A physically based, nondimensional parameter for discriminating between locations of freezing rain and sleet
.
Wea. Forecasting
,
11
,
591
598
.
Demirtas
,
M.
, and
A. J.
Thorpe
,
1999
:
Sensitivity of short-range weather forecasts to local potential vorticity modifications
.
Mon. Wea. Rev.
,
127
,
922
939
.
Doms
,
G.
, and
U.
Schättler
,
2004
:
A description of the nonhydrostatic regional model LM. Part II: Physical parameterization. Deutscher Wetterdienst Tech. Rep., 154 pp. [Available online at http://www.cosmo-model.org/content/model/documentation/core/default.htm.]
ECMWF
, cited
2011
:
IFS documentation CY28r1. [Available online at http://www.ecmwf.int/research/ifsdocs/CY28r1/.]
Evans
,
M.
,
2006
:
An analysis of a frontogenetically forced early spring snowstorm
.
Bull. Amer. Meteor. Soc.
,
87
,
27
32
.
Fehlmann
,
R.
, and
H. C.
Davies
,
1997
:
Misforecasts of synoptic systems: Diagnosis via PV retrodiction
.
Mon. Wea. Rev.
,
125
,
2247
2264
.
Fehlmann
,
R.
, and
C.
Quadri
,
2000
:
Predictability issues of heavy Alpine south-side precipitation
.
Meteor. Atmos. Phys.
,
72
,
223
231
.
Ferber
,
G. K.
,
C. F.
Mass
,
G. M.
Lackmann
, and
M. W.
Patnoe
,
1993
:
Snowstorms over the Puget Sound lowlands
.
Wea. Forecasting
,
8
,
481
504
.
Forbes
,
R.
, and
A.
Tompkins
,
2011
:
An improved representation of cloud and precipitation. ECMWF Newsletter, Vol. 129, ECMWF, Reading, United Kingdom, 13–18
.
Fujiyoshi
,
Y.
,
1986
:
Melting snowflakes
.
J. Atmos. Sci.
,
43
,
307
311
.
Georgescu
,
F.
,
S.
Tascu
,
M.
Caian
, and
D.
Banciu
,
2009
:
A severe blizzard event in Romania—A case study
.
Nat. Hazards Earth Syst. Sci.
,
9
,
623
634
.
Halcomb
,
C. E.
, and
P. S.
Market
,
2003
:
Forcing, instability and equivalent potential vorticity in a Midwest USA convective snowstorm
.
Meteor. Appl.
,
10
,
273
280
.
Kain
,
J. S.
,
S. M.
Goss
, and
M. E.
Baldwin
,
2000
:
The melting effect as a factor in precipitation-type forecasting
.
Wea. Forecasting
,
15
,
700
714
.
Knight
,
C. A.
,
1979
:
Observations of the morphology of melting snow
.
J. Atmos. Sci.
,
36
,
1123
1130
.
Lackmann
,
G. M.
,
2001
:
Analysis of a surprise western New York snowstorm
.
Wea. Forecasting
,
16
,
99
116
.
Lackmann
,
G. M.
,
K.
Keeter
,
L. G.
Lee
, and
M. B.
Ek
,
2002
:
Model representation of freezing and melting precipitation: Implications for winter weather forecasting
.
Wea. Forecasting
,
17
,
1016
1033
.
Langland
,
R. H.
,
M. A.
Shapiro
, and
R.
Gelaro
,
2002
:
Initial condition sensitivity and error growth in forecasts of the 25 January 2000 East Coast snowstorm
.
Mon. Wea. Rev.
,
130
,
957
974
.
Lumb
,
F. E.
,
1961
:
The problem of forecasting the downward penetration of snow
.
Meteor. Mag.
,
90
,
310
319
.
Lumb
,
F. E.
,
1983
:
Snow on the hills
.
Weather
,
38
,
114
115
.
Makkonen
,
L.
,
1989
:
Estimation of wet snow accretion on structures
.
Cold Reg. Sci. Technol.
,
17
,
83
88
.
Makkonen
,
L.
, and
B.
Wichura
,
2010
:
Simulating wet snow loads on power line cables by a simple model
.
Cold Reg. Sci. Technol.
,
61
,
73
81
.
Marwitz
,
J. D.
, and
J.
Toth
,
1993
:
A case study of heavy snowfall in Oklahoma
.
Mon. Wea. Rev.
,
121
,
648
660
.
Mass
,
C.
,
2008
:
The Weather of the Pacific Northwest. University of Washington Press, 280 pp
.
Matsuo
,
T.
, and
Y.
Sasyo
,
1981a
:
Empirical formula for the melting rate of snowflakes
.
J. Meteor. Soc. Japan
,
59
,
1
9
.
Matsuo
,
T.
, and
Y.
Sasyo
,
1981b
:
Melting of snowflakes below freezing level in the atmosphere
.
J. Meteor. Soc. Japan
,
59
,
10
25
.
Mitra
,
S. K.
,
O.
Vohl
,
M.
Ahr
, and
H. R.
Pruppacher
,
1990
:
A wind tunnel and theoretical study of the melting behavior of atmospheric ice particles. IV: Experiment and theory for snowflakes
.
J. Atmos. Sci.
,
47
,
584
591
.
Paulat
,
M.
,
C.
Frei
,
M.
Hagen
, and
H.
Wernli
,
2008
:
A gridded dataset of hourly precipitation in Germany: Its construction, climatology and application
.
Meteor. Z.
,
17
,
719
732
.
Penn
,
S.
,
1957
:
The prediction of snow vs rain. Forecasting Guide, No. 2, U.S. Weather Bureau, 29 pp
.
Rauber
,
R. M.
,
L. S.
Olthoff
,
M. K.
Ramamurthy
, and
K. E.
Kunkel
,
2001
:
Further investigation of a physically based, nondimensional parameter for discriminating between locations of freezing rain and ice pellets
.
Wea. Forecasting
,
16
,
185
191
.
Roebber
,
P. J.
,
S. L.
Bruening
,
D. M.
Schultz
, and
J. V.
Cortinas
Jr.
,
2003
:
Improving snowfall forecasting by diagnosing snow density
.
Wea. Forecasting
,
18
,
264
287
.
Schultz
,
D. M.
, and
Coauthors
,
2002
:
Understanding Utah winter storms: The Intermountain Precipitation Experiment
.
Bull. Amer. Meteor. Soc.
,
83
,
189
210
.
Steppeler
,
J.
,
G.
Doms
,
U.
Schättler
,
H. W.
Bitzer
,
A.
Gassmann
,
U.
Damrath
, and
G.
Gregoric
,
2003
:
Meso-gamma scale forecasts using the nonhydrostatic model LM
.
Meteor. Atmos. Phys.
,
82
,
75
96
.
Stewart
,
R. E.
,
1985
:
Precipitation types in winter storms
.
Pure Appl. Geophys.
,
123
,
597
609
.
Szyrmer
,
W.
, and
I.
Zawadzki
,
1999
:
Modeling of the melting layer. Part I: Dynamics and microphysics
.
J. Atmos. Sci.
,
56
,
3573
3592
.
Thériault
,
J. M.
,
R. E.
Steward
,
J. A.
Milbrandt
, and
M. K.
Yau
,
2006
:
On the simulation of winter precipitation types
.
J. Geophys. Res.
,
111
,
D18202
,
doi:10.1029/2005JD006665
.
Thériault
,
J. M.
,
R. E.
Stewart
, and
W.
Henson
,
2010
:
On the dependence of winter precipitation types on temperature, precipitation rate, and associated features
.
J. Appl. Meteor. Climatol.
,
49
,
1429
1442
.
Wang
,
H.
,
Y.
Entao
, and
Y.
Song
,
2011
:
An exceptionally heavy snowfall in northeast China: Large-scale circulation anomalies and hindcast of the NCAR WRF model
.
Meteor. Atmos. Phys.
,
113
,
11
25
.
Wesley
,
D. A.
,
R. M.
Rasmussen
, and
B. C.
Bernstein
,
1995
:
Snowfall associated with a terrain-generated convergence zone during the Winter Icing and Storm Project
.
Mon. Wea. Rev.
,
123
,
2957
2977
.
Woods
,
C. P.
,
M. T.
Stoelinga
, and
J. D.
Locatelli
,
2007
:
The IMPROVE-1 storm of 1–2 February 2001. Part III: Sensitivity of a mesoscale model simulation to the representation of snow particle types and testing of a bulk microphysical scheme with snow habit prediction
.
J. Atmos. Sci.
,
64
,
3927
3948
.
Zerr
,
R. J.
,
1997
:
Freezing rain: An observational and theoretical study
.
J. Appl. Meteor.
,
36
,
1647
1661
.
Zhang
,
F.
,
C.
Snyder
, and
R.
Rotunno
,
2002
:
Mesoscale predictability of the “surprise” snowstorm of 24–25 January 2000
.
Mon. Wea. Rev.
,
130
,
1617
1632
.
Zhou
,
B.
, and
Coauthors
,
2011
:
The Great 2008 Chinese Ice Storm: Its socioeconomic–ecological impact and sustainability lessons learned
.
Bull. Amer. Meteor. Soc.
,
92
,
47
60
.

Footnotes

1

Note that in this study the term “potential melting layer” refers to an atmospheric layer with temperature above the freezing point of water; it does not refer to the actual melting of falling hydrometeors as, for example, is observed by radar.