By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Please be advised that we experienced an unexpected issue that occurred on Saturday and Sunday January 20th and 21st that caused the site to be down for an extended period of time and affected the ability of users to access content on Wiley Online Library. This issue has now been fully resolved. We apologize for any inconvenience this may have caused and are working to ensure that we can alert you immediately of any unplanned periods of downtime or disruption in the future.
Weather has a great impact on road traffic, with the risk of accidents especially increasing during wintry weather conditions. Snowfall is often the dominant factor worsening driving conditions through decreased road surface friction and reduced visibility. Andreescu and Frost (1998) found that in Montreal, Canada, the number of accidents increased substantially on days with snowfall, even though people in Montreal are used to driving in snowy conditions during winter. In Sweden, Norrman et al. (2000) developed a method for deriving quantitative relationships between road slipperiness, traffic accident risk and winter road maintenance operations. Their study covered three winters in southern Sweden: it appeared that 50% of all accidents happened during slippery road conditions. They found that the highest accident risk was related to rain or sleet falling on a frozen road surface. The second most hazardous conditions prevailed when snowfall and hoar-frost formation occurred at the same time. During such weather events, accidents happened in spite of full maintenance activity. In order to reduce the amount of crashes, it was, therefore, proposed that the public awareness of such weather-related hazards should be enhanced.
In Finland, Salli et al. (2008) investigated the relationship between accident risk and wintertime road conditions. They found that the risk of accidents resulting in physical damage or injuries was ca four times higher during snowy or icy road conditions compared to dry road conditions. They also pointed out that drivers typically assess the road conditions as less slippery than they actually are.
The situation becomes very dangerous when there is a combination of slippery road conditions and poor visibility, especially if this reduction of visibility occurs suddenly. This can easily bring about the occurrence of severe pile-ups on highways. For example, on 20 March 2008, severe multi-vehicle collisions occurred in the Czech Republic near Prague, with 231 damaged cars, 30 injured persons (three of them seriously) and total costs estimated to be 27.8 million Czech korunas (around 1 million euro). Sudden snowfall, accompanied by reduced visibility and excessive driving speeds, was the main reason for the crashes (Pavek, 2008; Lidove, 2008).
The present paper deals with a similar occurrence of severe pile-ups that happened during the morning rush hours of 17 March 2005 in the Helsinki metropolitan area in southern Finland (Juga and Hippi, 2009). The consequences were appalling, with a total of almost 300 crashed cars, the deaths of 3 people and over 60 people injured. On this occasion, weather conditions deteriorated rapidly as a band of intense snowfall affected the Helsinki metropolitan area when the traffic was at its peak. The main accidents happened within a surprisingly short period of about 10 min at four different locations just before 0600 UTC (0800 local time).
This paper investigates the prevailing weather conditions and other factors, which led to this severe incident. Indeed, some of the media reports highlighted the role of the freezing drizzle as the cause of the accidents. Also some drivers, especially those driving in the western part of the Helsinki metropolitan area, claimed that they had observed freezing drizzle at the time of the crashes. However, the official reports of the Board of Inquiry for Traffic Accidents came to the conclusion that the very poor visibility due to snowfall, reduced road surface friction and excessive driving speeds were the main reasons for the accidents. Thus, the issue of the timing of the snowfall and freezing drizzle is crucial, and requires a detailed analysis to assess and verify the occurrence of the different weather phenomena prior to the accidents.
This analysis is supported by dual-polarization radar observations. The advantage of using dual-polarization radar observations is that they provide three-dimensional depictions of precipitation type, phase and intensity. Even though the analysis of winter precipitation from weather radar measurements is not straightforward, when this is combined with other observations it yields a more complete picture of the event. In this case, it was of decisive importance for reaching firm conclusions on the issues at stake.
The paper first introduces the observations that were available to perform a mesoscale weather analysis and a detailed analysis of imagery from two weather radars (one dual-polarization and one Doppler). The empirical evidence is strengthened by the output data from a numerical model for road weather conditions. Finally, some conclusions and discussions in relation to driving practices and road maintenance operations are presented.
2. Materials and methods
The weather conditions affecting the occurrence of the crashes on 17 March 2005 were analysed. Data included synoptic observations and upper air soundings from southern Finland and the surrounding area, as well as road weather observations from the nearby road weather station to investigate the driving conditions in the vicinity of the accident locations (Figure 1). Helsinki-Vantaa airport (EFHK in Figure 1(b)) is ideally located between the main highways, and, therefore, strong emphasis is laid on present weather and visibility observations from the airfield. The general synoptic situation over southern Finland and surrounding areas during the accident event is depicted in Figure 2.
For this study, measurements from two C-band weather radars were used (locations marked in Figure 1(b)), the University of Helsinki dual-polarization weather radar (KUM) and the Finnish Meteorological Institute (FMI) Doppler weather radar (VAN). The university radar is located on the roof of the Department of Physics building (60°12.26′N, 24°57.78′E) at the Kumpula campus. The radar is positioned 59 m above mean sea level and 30 m above ground level. During the event, the radar was carrying out high time-resolution range-height indicator (RHI) scans. The RHI scans were complemented by plan position indicator (PPI) scans carried out every 20 min. The RHI scans were directed northwards (azimuth angle 11.8°). The scans were carried out in two polarization modes, i.e. the simultaneous transmission and receiving (STAR) mode, and the cross-polarization mode. This gave a cycle time of approximately 2 min for each polarization measurement mode. In this paper, only measurements collected by using the STAR mode are presented. The FMI Vantaa radar (VAN) carried out FMI operational volume scans, repeated every 5 min (Saltikoff et al., 2010), from which the constant altitude PPI (CAPPI) was calculated. The Vantaa radar 500 m altitude CAPPI and the Kumpula radar dual-polarization RHI observations complemented the analysis (Section 3.2).
Both weather radar data and manual visibility observations were combined to analyse the time behaviour of the precipitation intensity. The Helsinki University dual-polarization radar instrument is located ca 12 km south of the airport. Its data were used to investigate the changes in the precipitation mode and especially the timing of the change of snowfall into freezing drizzle. The results were compared to the manual precipitation observations from the airfield. Manual observations have an advantage compared to automatic weather observations: if several weather phenomena are occurring simultaneously they all can be documented. Automatic weather stations may have problems in such a situation, occasionally resulting in false detection of present weather, especially in the case of low intensity precipitation such as drizzle (Merenti-Välimäki et al., 2001).
To obtain a comprehensive picture of the occurrence of the crashes, additional information was used. This included media reports (TV, internet and newspapers) and research reports from the Board of Inquiry for Traffic Accidents. In addition, photographs from the road weather cameras of the Finnish Road Administration (known as the Finnish Transport Agency since 1 January 2010) were used in the study. They disclosed valuable information about the reduction in visibility near the location of the crashes. To assess whether the poor driving conditions were predicted successfully in advance, the weather forecasts and road weather warnings released the previous evening were reviewed and the output from the FMI's road weather model was examined.
In 1999–2000 a road weather model was developed at FMI: since that time it has been used operationally. The model is a useful tool for duty meteorologists forecasting road conditions and issuing road weather warnings to drivers. Furthermore, road maintenance personnel use the model's output when planning and scheduling road maintenance operations. The road weather model is a one-dimensional energy balance model that calculates vertical heat transfer in the ground and at the ground-atmosphere interface, taking into account the special conditions prevailing at the road surface and below it (Kangas et al., 2006). The effect of traffic is also included in the model. At its upper boundary the model is forced with the outputs from a weather forecast model, either directly or as edited by the duty meteorologist. This wider-scale input thus provides the horizontal coupling between the individual points. At the lower boundary, the climatological ground temperature is used as a boundary condition. The most important output of the model is road surface temperature, but several other parameters are also calculated, e.g. the thickness of the water/snow/ice/deposit on the surface, the road surface condition and the traffic condition index, which is an estimate of the prevailing driving conditions (normal, bad, very bad). A short-range model forecast of the driving conditions during the accident day is presented in Figure 3, combined with the radar data and weather observations.
3.1. Synoptic situation
The first half of March 2005 was cold and dry in Finland, so mostly normal winter road conditions prevailed. On 17 March, the weather situation changed, with a low pressure area approaching southern Finland from the west. The night was clear and cold with early morning temperatures about 8 °C below freezing point. The easterly wind increased somewhat and light snowfall reached the Helsinki metropolitan area early in the morning, followed by a band of intense snowfall before 0600 UTC (0800 LST, Figure 2). The visibility significantly decreased due to the snowfall, reaching a minimum value below 1000 m (Figure 3(d)). On the highways, the visibility became very low due to drifting snow. Although the snowfall was intense around 0600 UTC, the precipitation amounts were not large in the Helsinki metropolitan area. Based on gauges and radar information, the total accumulated liquid water equivalent of the precipitation between 0000 and 0600 UTC was 2–4 mm, the change in snow depth at the FMI weather stations in the area being 5 cm or less. Concomitantly, the temperature was slowly rising, but remained below − 5 °C (Figure 3(a)).
Based on the 0600 UTC sounding from Vantaa (Figure 4), the lower troposphere was characterized by a shallow layer of cold air near the ground with somewhat warmer and moister air above it. This type of vertical temperature distribution was related to the approaching warm front. In southern Finland the warmer air did not reach the surface, where temperatures remained below 0 °C the whole day. The intense snowfall was soon replaced by freezing drizzle; manual observations at Helsinki airport showed that the freezing drizzle started there at 0715 UTC. Although the intense snowfall passed, light snowfall (mainly snow grains) was still observed simultaneously with the freezing drizzle (Figure 3(c)). Such a mixture of freezing drizzle and light snow has also been observed in the USA (Bernstein, 2000). In that case the upper clouds probably produced the light snow that fell into the liquid cloud below and became rimed.
Carriere et al. (2000) have investigated the occurrence and formation of surface freezing precipitation in Europe. They found that freezing precipitation typically occurs when the surface temperature is between − 5 and 0 °C. They also pointed out that there are basically two mechanisms of formation of freezing rain and freezing drizzle. Firstly, the supercooled drops can form from the ‘ice process’ originating from ice or snow crystals melting when they fall through a layer with a temperature above 0 °C, and then becoming supercooled when falling through a layer with sub-zero temperatures near the surface. Secondly, supercooled drops (drizzle) can also form from the ‘coalescence process’. In such a case, the drops are directly generated from liquid cloud drop coalescence and diffusional growth in supercooled clouds. The air temperature is below 0 °C throughout from the surface layer to the cloud top. However, the cloud top temperature should be higher than ca − 15 °C, since lower temperatures could favour the ice phase. From the Vantaa sounding of Figure 4 it appears that although there was a relatively warm and moist layer above 500 m, the temperature was clearly below 0 °C. Therefore, in this case, the freezing drizzle was obviously generated by the coalescence process. This is also confirmed by the radar observations discussed in Section 3.2.
Ikeda et al. (2007) found in two case studies that freezing drizzle was observed after the passage of upper cold and humidity fronts, with the freezing drizzle appearing as a weak echo region in the RHI scans during the measurements. The passage of the upper front in the mid troposphere and the consequent drying of the layer above the lower moister layers caused a cessation of the seeder ice-crystals that had previously fallen through the lower layers rimming there and preventing the formation of supercooled precipitation.
In the case studied here, the Vantaa 0600 UTC sounding (Figure 4) displays an equivalent situation at the time of the crashes. It appears that relative humidity was high up into the mid-troposphere, the air being saturated in the height range between 0.3 and 1.3 km, as well as between 3.2 and 3.5 km and around 4 km. This indicates the presence of feeder clouds in the mid-troposphere providing falling snow crystals, which is in line with the intense snowfall reported in the surface observations around 0600 UTC. Unfortunately, further soundings from the Vantaa sounding site were not available during that morning, thus preventing the identification of the possible passage of an upper humidity front. However, later that day the 1200 UTC sounding at Jokioinen (ca 100 km northwest of Helsinki) showed a humid layer below 730 hPa (height ca 2.5 km) with a drier layer above it (relative humidity was 74–82% in the layer between 2.5 and 4 km). This points to the existence of somewhat drier air in the mid-troposphere in that later sounding than in the Vantaa 0600 UTC sounding. However, the simultaneous occurrence of snow grains with freezing drizzle at Helsinki-Vantaa airport might indicate the presence of a feeder cloud in the mid-troposphere.
3.2. Radar-based analysis
The evolution of the meteorological situation during the event as seen from the radar observations is illustrated in Figure 5. The 0505 and 0550 UTC measurements depict the situation that occurred before and at the time of the severe pile-ups, respectively. The 0710 and 0725 UTC observations were selected based on the aviation weather observations at Helsinki-Vantaa (EFHK) airport, which reported changes in the precipitation type at around these times (Figure 3(c)).
For each period, three panels are shown in Figure 5. The left-hand panels show the Vantaa radar CAPPI images of reflectivity measurements. The two right-hand figures show the reflectivity and differential reflectivity RHI scans measured by the Kumpula radar. The differential reflectivity observations are directly related to particle shapes (Bringi and Chandrasekar, 2001). Differential reflectivity values (Zdr) higher than 1 dB indicate the presence of horizontally aligned non-spherical particles, such as dendrites, needles or other types of ice crystal (Straka et al., 2000). Typically, Zdr values not exceeding 0.5–1 dB are associated with low-density aggregates. In the case where supercooled drizzle is the dominating particle type, differential reflectivity values are nearly equal to 0 dB (Straka et al., 2000; Matrosov et al., 2001). It should be noted that in cases where a mixture of different particle types occurs, the interpretation of such radar observations is not easy. Since both reflectivity and differential reflectivity observations are significantly biased towards larger particles, a relatively small number of large aggregates will mask the radar returns from a significantly larger number of small crystals or supercooled drizzle particles (Bader et al., 1987). Therefore, in this work the analysis proposed by Moisseev et al. (2009), which is based on identifying the dominating snow growth mechanisms, such as vapour deposition, aggregation or riming, is followed.
The 0505 and 0550 UTC measurements show a typical snow aggregation pattern (Moisseev et al., 2009). The horizontally aligned non-spherical particles that are present at higher altitudes, as shown by Zdr values larger than 1 dB, are transformed to low-density aggregates represented by lower differential reflectivity values closer to the ground. At the same time, reflectivity is inversely proportional to altitude and increases significantly in the lowest kilometre. Even though observations show that aggregation is at both times the dominating snow growth mechanism, there are still some differences. The differential reflectivity measurements at 0550 UTC exhibit two clearly identifiable high Zdr bands, with values exceeding 2 dB. These bands are located just above the region where the reflectivity growth starts. These bands also correspond to levels where the temperature is between − 10 and − 15 °C. Hogan et al. (2002) have observed similar bands. They concluded from coincident in situ airplane measurements that such bands corresponded to layers where a small amount of supercooled water was present and the growth of dendritic ice crystals by vapour deposition occurred. Such crystals are very oblate and yield high differential reflectivity values. Just below these bands, aggregation is the dominant feature. Typically, aggregation is most efficient at temperatures higher than − 5 °C, but in this case, due to the crystal type, aggregation already takes place at temperatures around − 10 °C (Houze, 1994). It should be noted that even though the high Zdr bands indicate the presence of supercooled water droplets, the liquid water content is rather small, and the droplets are small enough not to precipitate to the ground. Furthermore, the liquid phase is quickly depleted by the Bergeron process. Therefore, it can be concluded that, at the time of the crashes, the precipitation at the surface consisted mainly of aggregates of dendrites, and that no precipitation as supercooled water was present.
At 0710 and 0725 UTC the situation was already rather different. By this time, the high-reflectivity band had moved away from the crash sites towards the northeast, as can be seen in the CAPPI images in Figure 5. The RHI observations of reflectivity exhibit more variability both horizontally and vertically. This is more apparent in the RHI measurements at 0725 UTC between 15 and 25 km. This variability is an indicator of riming and the presence of supercooled water droplets (Moisseev et al., 2009). The most probable cause for the radar reflectivity variability is the occurrence of the vertical air motion that is needed to sustain the growth of liquid droplets, and which is enhanced by latent heat release due to riming. From 0715 UTC onwards, supercooled drizzle was observed at Helsinki-Vantaa airport. The closest RHI measurement point to the airport is shown by the blue arrow in the RHI plots. It can be observed that the corresponding differential reflectivity measurements exhibit values close to 0 dB. These values are characteristic for drizzle and densely rimed particles (Straka et al., 2001; Matrosov et al., 2001). The drizzle-graupel area extends from 12 to 20 km in range and up to 3 km in height. At 0725 UTC this area had moved about 10 km northeast and was centred on one of the crash sites, depicted by the red arrow in Figure 5. At almost the same time, the airport observing station reported the simultaneous occurrence of freezing drizzle and snow grains. The agreement of the airport reports with the dual-polarization radar observations indicates that supercooled drizzle did occur on this day, but that it reached the crash sites more than 1 h after the accidents. Nevertheless, this particular feature was later highlighted in the media.
The radar reflectivity had two maxima, which passed Vantaa at approximately 0600 UTC and 1100 UTC (Figure 3(d)). During the overpass of the first band, the increase in radar reflectivity correlated rather well with the decrease in visibility. However, the precipitation intensity decreased between the bands, but the visibility still remained relatively low. The reduced visibility was caused by drizzle, mist and blowing snow, all of them also being observed at the weather station at Helsinki-Vantaa airport. The road weather stations reported visibility values from 600 to 900 m between 0530 UTC and 0600 UTC, which is consistent with the minimum visibility of 700 m observed at the airport (Figure 3(d)). Very bad visibility was reported on the roads by car drivers, since passing cars increased the effect of the blowing snow. This is not seen so clearly in the measurements, because the visibility sensors of the road weather stations are located at heights of 4–6 m, and so much of the blowing snow did not affect them.
Snow depths and 24 h accumulated precipitation at FMI weather stations were conveniently recorded at 0600 UTC, just in the middle of the event. The snow ratio (i.e. the change in snow depth divided by the liquid water equivalent (Judson and Doesken, 2000)) calculated from eight stations in the area was 10.8 in the early phase, and 6.8 in the later phase of the event. Power et al. (1963) found that unrimed particles will have densities less than 0.1 g cm−3, while rimed particles will have larger densities. Therefore, snow ratios less than 10 would indicate rimed ice particles. This also supports the analysis presented above based on radar measurements.
4. Discussions: actions and consequences
When the light snowfall started early in the morning, around 0300 UTC, the road surface temperature was so low (ca − 8 °C) that, based on maintenance guidelines, the road maintenance contractors were not obliged to carry out salting (although it was optional). There are important reasons for this. Firstly, as temperatures decrease, the amount of salt needed to melt the ice increases significantly. With snowfall in cold conditions, there is a risk that the snow melts only partly due to the salt and thereafter sticks to the road surface and gets packed tightly by traffic, thus forming a slippery layer. Secondly, under very cold conditions the splash water can form an icy layer on cars' windscreens (Jorma Helin, personal communication).
In this case, the two main road maintenance contractors operating in the Helsinki metropolitan area acted somewhat differently. In the western part of the area, salting was performed from 0500 UTC onwards while elsewhere highway junctions were in general gritted (salting was carried out after the occurrence of the main crashes). After 0500 UTC, the snowfall rapidly became heavier, and severe pile-ups occurred at four different locations between 0545 and 0557 UTC. The main highways near Helsinki were totally closed for many hours, and several accidents also occurred on secondary roads. Official investigations into the accidents were started soon after by the Board of Inquiry for Traffic Accidents. The Finnish Meteorological Institute and the Finnish Transport Agency also took part in this inquiry.
The decrease in the visibility (Figure 3(d)) was strongly correlated with the increase in radar reflectivity (precipitation intensity). The minimum visibility values (700 m) were observed at the airport around 0600 UTC and this reduction due to intense snowfall was clearly verified by the weather cameras of the Finnish Transport Agency. These photos also showed that the roadways were partly covered by loose or drifting snow and that the road edges were covered with thicker snow layers. This probably caused the unforeseen increase in braking distances.
One factor that could also have reduced the grip of car tyres on the road is the formation of hoar-frost. This typically occurs when the dewpoint temperature (Td) is higher than the road surface temperature (Tr). The accumulation of hoar-frost on the road increases with increasing mean wind speed, increasing maximum difference between Td and Tr and the increasing duration of conditions with Td > Tr (Karlsson, 2001). Based on observations at the Keimola station (location in Figure 1(b), observations in Figure 3(a)), the road surface temperature had somewhat higher values than the dew-point temperature during the whole morning. This would indicate that a notable continuous formation of hoar-frost probably did not occur. However, the road maintenance contractors had noticed local sublimation of hoar-frost early in the morning. Karlsson (2001) pointed out that on several occasions during her field measurements, the difference between the observed Td and Tr failed to indicate hoar-frost, due to the unsuitable location and poor accuracy of the measuring equipment. The possibility of hoar-frost formation thus cannot be fully ruled out, because the road surface temperature had been lower than the air temperature earlier in the morning (Figure 3(a)), and the relative humidity remained quite high (mostly above 85%). In addition, the prevailing moderate winds might have caused more turbulence and thus a transfer of humidity onto the road surface.
The weather situation and bad driving conditions were forecast quite well beforehand by the FMI. Warnings for poor driving conditions were already issued on radio and TV on the previous day. During the evening TV broadcast the meteorologist specifically pointed out that driving conditions would be troublesome the next morning. The FMI's road weather model predicted that very bad driving conditions would prevail in the Helsinki metropolitan area during the busiest morning traffic hours (Figure 3(b)).
In spite of the quite accurate weather forecasts and warnings for drivers, the average driving speeds in the Helsinki metropolitan area on the accident day were only a few kilometres per hour lower than on the previous day (based on real-time measurements by the Finnish Transport Agency). Moreover, the distances between vehicles were too short. The drivers were probably lulled into a false sense of security by the preceding long period of dry fair winter weather without major problems in driving conditions. Also the sharp leading edge of the snowfall area took the drivers by surprise if they approached Helsinki (and the snowfall) from the northeast. These drivers were forced to encounter simultaneously a rapid decrease in visibility and road surface friction. Based on the investigations of the Board of Inquiry for Traffic Accidents, many of the drivers interviewed pointed out that the visibility became very poor prior to the accidents. Some drivers said that, when braking, the speed of the vehicle decreased too slowly for the collisions to be avoided, indicating low road surface friction.
5. Concluding remarks
It has been shown here that a careful analysis of weather parameters, phenomena and processes using observing networks and especially weather radar data with a high time-resolution can provide a very clear insight into the hazards of sudden changes in weather conditions, and the consequent damage caused by them.
The first reports in the media blamed supercooled drizzle as the main culprit for the pile-ups, because this was the dominant weather phenomenon observed at the time when journalists arrived at the accident sites. However, as the more detailed analysis based on radar data and weather observations confirmed, the accidents took place before the time at which the snowfall turned into supercooled drizzle.
This study emphasizes the value of short intervals between weather observations and the use of new instruments such as dual-polarization radar. This information should be effectively utilized when issuing weather forecasts and warnings, as well as when carrying out investigations and case-studies. The sad fact is that the accidents happened in spite of the warnings for poor driving conditions that were issued the previous evening on radio and TV. Intense snowfall and the related rapid decrease in visibility is a substantial risk for traffic on highways. Lately, similar incidents, for example in the Czech Republic and Austria during March 2008, have occurred.
The provision of real-time weather information and warnings in vehicles and the use of weather-controlled speed limits and displays in road management could be the main ways of preventing massive accidents. New technology is the key here. Traffic safety can be improved with wireless vehicle-to-vehicle and vehicle-to-infrastructure communications. Applications may include the transfer of weather information and warnings as well as traffic information. For example, such an application was developed in the Carlink project of the European Celtic programme (Sukuvaara and Nurmi, 2009). It consists of an intelligent hybrid wireless traffic service platform between cars, supported by roadside wireless base stations.
Last but not least, this case shows that high impact weather does not necessarily have to be extreme: in this case, 5 cm of snow was enough.
The authors wish to express their sincere thanks to Jorma Helin of the Finnish Transport Agency for his insight into driving conditions and road maintenance operations. Also many thanks are due to Sylvain Joffre of the FMI for his review and valuable suggestions for improving this paper.