The 1987 storm can be seen as something of a ‘game changer’ in UK meteorology. It led to changes in the way in which weather forecasters dealt with episodes of extreme weather and acted as the catalyst for numerous advances in the science of meteorology and numerical weather prediction (NWP).
The National Severe Weather Warning Service
A major improvement made in the wake of the 1987 storm was the creation of the National Severe Weather Warning Service (NSWWS). At that time, during the run-up to severe weather, warnings were issued to emergency services, emergency responders and even the armed forces. However, the only way to warn the public, over and above the routine forecasts on radio, television and in newspapers, was via a FLASH warning. These warnings were issued during the weather event itself, usually in response to observations of severe weather, and covered the 18 largest urban areas in the UK; they were based on certain thresholds (e.g. 15mm of rain falling in 3 hours). This system was thought to be adequate at the time, but the 1987 storm demonstrated that, although it could be effective by day, it was far less so at night as the main dissemination method was via radio and television (Met Office, 1988). Accordingly, the FLASH system was upgraded to become the NSWWS. Among other things, the improvements included the facility to issue warnings nationwide and to provide warnings with some advance notice.
The NSWWS was given a major overhaul in 2011 following a review by the Public Weather Service Customer Group (Goldstraw, 2012). The current system allows warnings for rain, wind, fog, snow and ice to be issued across the UK, and it is split into two categories depending on the lead time: alerts may be issued between one and five days ahead, and warnings within 24 hours, of an expected severe weather event. The alerts and warnings are then further sub-divided into yellow, amber and red based on an assessment of the likelihood and projected impacts of the event, rather than just the simple thresholds which were used when issuing the FLASH warnings. For example a yellow warning may represent the high likelihood of a low impact event (towards the top left in Figure 1) or the low likelihood of severe impacts (towards the bottom right).
Updates are normally issued during the late morning to give responders and media outlets time to get the message out, but warnings can still be updated at any time if the weather warrants it.
How the modern NSWWS worked during two Scottish windstorms
The latest NSWWS was put to good use during a period of very disturbed weather in December 2011 and January 2012, when two particularly violent storms moved across Scotland. As in the storm of 1987, pressure fell below 960mbar in the centre of the depressions and gusts at low-level sites exceeded 80kn, although the storm tracks were much further north this time. The two Scottish storms also differed greatly in terms of the forecast confidence, which in turn had an impact on the NSWWS warnings issued as they approached the British Isles.
The first storm, which crossed northernmost Scotland on 8 December (Figure 2), was consistently forecast several days in advance by the Met Office to bring a period of high winds to much of Scotland. A yellow alert was issued on 5 December and an updated yellow alert a day later. A combination of consistent model output, convergence between different models and agreement amongst forecasters that there was high confidence of a high-impact event for parts of Scotland led to an upgrading directly to a red warning on the 7th, the first red warning under the new impact-based NSWWS (Figure 3).
In the event the wind speeds and areas most affected were in accordance with the general warning envelope: there were gusts of 60–70kn across much of Scotland, and much higher gusts in some upland and exposed coastal locations. The highest gust from a site not on a mountain top was 91kn at Tulloch Bridge (Highland), the highest on record there, whilst Tiree and Fair Isle reported 79kn and 80kn respectively. At the mountain station of Cairngorm (altitude 1237m) a gust of 143kn was recorded. The strongest winds are thought to have been associated with a sting jet (see inset). Gusts of 50–60kn were recorded over north Wales, northern England and Northern Ireland.
The storm left an estimated 150 000 homes without power across Scotland and caused major travel disruption. However, the collaborative approach between the Met Office and Scottish authorities in assessing the potential impacts in advance resulted in some key mitigating actions being taken. The benefits cannot be easily quantified but there is anecdotal evidence from Traffic Scotland that there was around 20% less traffic on the roads as a result of the warning being issued, which allowed them to react more readily to the incidents that did occur than might otherwise have been the case. Also, hundreds of schools were closed in response to the early issue of the red warning, thus reducing the risk to life. This was brought into sharp focus as a school bus was blown over in Dalry (Ayrshire) in the morning: no pupils were on board owing to the school closures and the driver escaped without injury.
The storm of 3 January 2012 was a different ‘animal’ entirely with respect to its forecast evolution and confidence, and to some degree the level of impact. By 31 December a significant wind event was signalled to affect England and Wales on 3 January, and with some model consensus the Met Office issued a yellow alert. At this stage Scotland was not expected to be affected by the high winds. However during 1 January model solutions diverged, with an increasing number suggesting that northern, rather than southern, parts of the UK may be at most risk. Accordingly, the yellow warning was extended to also include Scotland and Northern Ireland, but confidence remained relatively low in both the track of the low and the wind detail. On 2 January within 24 hours of the expected arrival of the storm, a further shifting of model consensus led to growing confidence that the main core of high winds would primarily affect Scotland and Northern Ireland, rather than the original alert area of England and Wales, and with a lead time of around 17 hours an amber warning was issued to cover southern and central Scotland and the north coast of Northern Ireland.
On the night of 2 January into the early morning of the 3rd there were still -significant doubts as to the detail of both the depth and track of the depression centre, and hence the location and magnitude of the strongest gusts. This can be seen from the content of the Met Office Operations Centre forecast guidance, a product produced by the on-duty Chief Forecaster and issued to all forecast offices. The guidance is designed to give a unified approach across all forecast output and provides forecasters with information on the general weather evolution, including any uncertainties and the reasons for these. The forecast guidance at 2110 utc on the 2nd stated that maximum gusts of 70kn were to be expected around Islay and Malin Head by 0600 utc on the 3rd, with inland gusts of 50–60kn, possibly reaching 65kn on some of the hills in the southern half of Scotland.
The routine update of the Chief Forecaster's guidance at 0300 utc on the 3rd stated that the low was now explosively deepening in line with earlier expectations, but the track was shifted slightly to the south. All-in-all, the forecast wind speeds were expected to be similar to those stated earlier but there was still considerable uncertainty, even just a few hours prior to the storm's onset. At around 0400 utc, as the storm approached Northern Ireland, another forecast guidance product was produced, the Synoptic Evolution. This stated that the Met Office North Atlantic and European (NAE) model was failing to deepen the low enough when compared with observations, hence winds in its forecast were not strong enough. On the other hand the Met Office Global Model was too low with the surface pressure and therefore the forecast winds from this product were thought to be too strong, although this scenario was given as a 25% chance of occurrence. At 0615 utc the guidance was updated again in light of the first observations in Northern Ireland, which showed that the stronger wind scenario could be happening. Forecast gusts across the central belt of Scotland were increased to approximately 70kn, with a 30% chance of gusts reaching around 78kn. The issuance of a ‘red’ warning was also mooted at this point.
As the storm unleashed itself across Scotland (Figure 4), Islay reported a mean wind speed of 65kn and gust of 84kn at 0700 utc, just touching hurricane force 12, whilst by 0800 gusts of 79kn and 80kn affected Glasgow Bishopton and Glasgow Airport respectively. At 0815 the Met Office issued a red warning (Figure 5) for this high-impact event across Central Scotland, and a maximum gust of 89kn (102 mph) was recorded soon afterwards in Edinburgh city, at Blackford Hill (altitude 134m).
There was evidence from satellite imagery that a sting jet feature may have enhanced the gust speeds in the westsouthwestern sector of the low centre, over the isles and coasts and possibly further inland across western Scotland. However the isobaric gradient was so tight that this alone is sufficient to explain the extreme gusts in the 70–90kn range. An aircraft inbound to Glasgow airport at 0850 utc reported a westerly mean speed of 73kn at an altitude of 610m: this was around an hour after the peak strength had passed and corresponds well with surface gusts at the time.
The 3 January storm was the most significant across central and southern Scotland since the Boxing Day 1998 storm, and in some locations winds even exceeded those of 1998, with a greater impact. There was widespread damage across central and southern Scotland, with roofs ripped from buildings, many chimney stacks collapsing and thousands of trees brought down. Strathclyde Fire and Rescue Service said it attended 488 incidents by 1300 utc on the 3rd. The entire rail system out of central Scotland was halted with Network Rail having to remove 856 fallen trees from the lines. Air, sea and road transport was also severely disrupted, with critical motorway bridges such as the Forth, Tay, Erskine and Kingston closed.
A sting jet is a phenomenon associated with some rapidly developing cyclones, whereby exceptionally strong winds can be experienced at the surface. It is thought to be a result of snow evaporating rapidly, which in turn quickly cools the surrounding air. This cooler, denser air then descends, transferring momentum to the ground. The strongest winds typically occur just ahead of the ‘cloud head’ (a hooked area of very cold-topped clouds near the centre of the developing circulation) and can lead to a great deal of damage (Browning, 2004). However, as Baker (2009) commented, and as is noted here in the storm of 3 January, a sting jet need not be the cause of the strongest low-level winds in an extreme windstorm and further work is needed to determine the relationship between low-level winds and strong gusts. Figure 6 indicates the location of the possible sting jet in this storm.
The modern NSWWS system is a very effective way of notifying the public, emergency responders and other interested parties of potentially hazardous weather days in advance, as was illustrated during the 8 December storm. However, as shown by the events surrounding the storm on 3 January, there are still occasions when forecast uncertainties in the run-up to a period of severe weather can lead to the highest level warnings not being issued until very late, in a similar fashion to the Great Storm of 1987.
Another significant change since 1987, though, is that there have been more thoughtful attempts to convey to the public the uncertainty in the forecast, which was the subject of some criticism levelled at the Met Office in the wake of the 1987 storm. A number of letters in newspapers then included comments implying that readers were unhappy with the forecasters’ constant use of the word ‘will’, rather than ‘could’ or ‘may’. It was felt that if the uncertainty in the forecast had been better communicated, people would have been more prepared for what actually happened. For example, a letter in The Independent shortly after the storm stated:
The public are not fools and always will respect the genuine, perhaps reasoned, uncertainty of an expert; indeed a clear indication of the considered degree of uncertainty of a forecast would be of great help in ordering one's affairs.
A good example of the progress that has been made in communication methods since that time was demonstrated during the events of the 15 and 16 December 2011, which utilised not only the NSWWS system, but also brand new forms of interaction with the public, including social media and mobile phone apps.
In December 2011, much as in October 1987, the potential for an explosively--deepening storm moving up from the southwest to affect parts of the UK was indicated by numerical models several days in advance. However, there was considerable uncertainty about the track and depth of the depression, with model output from a wide range of sources showing differing solutions.
Figure 7 shows the probabilities of 1km winds (indicative of surface gust strength) ex-ceeding 60kn at 0000 utc on 16 -Decem-ber, taken from the European Centre for Medium-Range Weather Forecasting (ECMWF) ensemble initiated at 0000 utc on 11 December. An ensemble -system such as this takes a deterministic model (a model that produces only one forecast) and runs it several times with slightly different starting conditions (Richardson, 2011). These different starting conditions (the state of the atmosphere at the time the model starts) allow for differing solutions to be produced by the same model. This then enables a forecaster to compare the deterministic output with the various ensemble solutions to see whether the deterministic output is, in itself, a potentially extreme solution and to gauge the likelihood of different scenarios occurring. Model output such as that shown in Figure 7 led to a low-probability, high-impact yellow alert being issued early on 12 December for strong winds across the entire UK (except Shetland) on 16 December.
Over the next couple of days, confidence remained relatively low with regard to where the strongest winds would be felt. The Met Office Global Model was reasonably consistent in indicating a track across Northern Ireland and northern England (which would keep the strongest winds to the south of this), but other deterministic models (including ECMWF, the American GFS and the Canadian and Japanese models) moved towards a much more southerly track. The ensembles continued to produce a wide range of solutions but still highlighted the potential for exceptionally windy conditions, particularly in southern England and the English Channel. By 13 December, it was becoming accepted that the most likely track of the low centre would be somewhere across southern Britain, but with higher-than-normal levels of uncertainty and still the distinct possibility of a deeper and more northerly track.
As the expected focus of the strongest winds shifted southwards, the warning area in the NSWWS was reduced so that the update issued during the morning of 14 December covered only southern coastal counties of England. However, given the continuing uncertainties in the final track of the low and consequent wind speeds, confidence in the evolution remained low enough that the warning had to stay yellow. In the context of the above, it was felt that extra effort should be put into conveying these uncertainties to the public, over and above the NSWWS. Consequently, the Met Office produced a pair of graphics on the 14 December (Figure 8), which were widely communicated through the BBC and various social media outlets. These showed both the most likely track and the low probability alternative, with a clear indication of what each would mean in terms of gust strengths in the affected areas (Met Office, 2011a).
Along with the strong winds, there was also always the potential for significant rain and snow associated with this system, and by 15 December it was felt that rain would be more disruptive than wind along the south coast. Therefore the wind warning was replaced by a rain warning, which nevertheless still mentioned the risk of strong winds within its text. The graphics showing the differing tracks were also reissued (Figure 9) with the original ‘most likely’ option (that of 70 mph gusts in Kent and Sussex) now becoming the ‘low risk’ alternative. A new ‘most likely’ scenario showed no particularly unusual winds across the UK but continued to emphasise the rain and snow in the parts of the country expected to be affected (Met Office, 2011b).
In the event, a marked but not exceptionally deep frontal depression tracked along the English Channel, before deepening -further over the Low Countries. As a result, the strong winds associated with it largely missed the UK, but gusts of over 50kn were recorded along the French coast and there was a significant amount of disruption to transport and infrastructure on the near continent, particularly in Brittany. However, it is encouraging that the communication of uncertainty has evolved to the extent that forecasters can issue ‘alternative’ scenarios, such as those shown in the diagrams, and expect the majority of the public, media, emergency responders and the Government to take away the intended message. An email received by the Met Office on 15 December included the words:
I was very pleased to see in the weather forecast after the BBC 10 o'clock news last night the uncertainty in the next storm and its impact explained well in terms of two possible trajectories for its centre!
This is very positive feedback. However, it must be balanced by the sentiment expressed in another email to the Met Office on 16 December, regarding the same event:
Your organisation has not apparently had any degree of certainty itself about the expected weather for even the next day since last week. Why is that? I would really like to understand why this period of weather has been so problematical to you to the extent that you have not been able to give clear information and yet have found it necessary to post various warnings which merely concern people with apparently no basis. Are you using seaweed too?
Clearly there is still work to be done in communicating the reasons behind the uncertainty, but the fact that people are asking questions such as these is surely a step in the right direction.
Advances in Numerical Weather Prediction
The improvements made in warning the public of impending severe weather have been enabled by a combination of experience, scientific advances and improvements in Numerical Weather Prediction (NWP). Following the storm of 1987 a substantial study into the numerical model forecast performance was undertaken (Shutts, 1990). This concluded that the model performance prior to the storm, particularly in the preceding 36 hours, was fairly poor and that improvements in the numerical models available to duty forecasters should be pushed forward as a priority – particularly the introduction of higher-resolution models.
At the time of the 1987 Storm, Met Office forecasters in the Central Forecasting Office (CFO) based in Bracknell routinely had access to five models, two of which were produced by the Met Office. The other three models were the ECMWF model and two models based on the United States Global Model (Met Office, 1988).
Improvements in NWP have moved apace since then, thanks in large measure to the huge increase in available computing power. For example, the supercomputer that was used by the Met Office to run the various models in 1987, the CDC Cyber 205, was able to undertake 200 × 106 floating point operations (calculations) per second, also known as FLOPS. The current, IBM Power 6, supercomputer can undertake around 140 × 1012 FLOPS, making it around a million times more powerful.
This increase in computing power has allowed Met Office forecasters, now largely based in the Operations Centre in Exeter, to access a far larger array of models, most of which with a far superior resolution than in 1987 and that are available at more frequent intervals throughout the day. The greater number of models available to Met Office forecasters allows them to evaluate whether there is a consensus in the forecast evolution, or whether there is a wide range of solutions. This suite of models includes four deterministic models from the Met Office supercomputer alone:
- The Global model (GM), which aims to model the weather across the world.
- The NAE, which covers the North Atlantic and Europe.
- The UK4 and UKV which are designed to cover the UK in high detail.
In addition to these deterministic models, there is also a greater number of models available from other forecasting centres. In 1987 only the ECMWF model, the two American models and the French model (which was running, but was not actively used then), were available. In 2012 forecasters can access not only the ECMWF, USA and French models, but also the German, Japanese and Canadian output.
Another major improvement since the 1987 storm has been the introduction and proliferation of Ensemble Prediction Systems (EPS). These systems give forecasters a larger array of potential outcomes, which in turn should cover the vast majority of solutions, owing to their different starting conditions. Met Office forecasters have access not only to the Met Office EPS, known as MOGREPS (Met Office Global and Regional Ensemble Prediction System), but also to the ECMWF and the American versions.
An EPS could have helped forecasters in 1987. In fact it was shown shortly after the storm of 1987 that the Fine-Mesh model would have done a far better job at forecasting the intensity and position of the storm had it had better starting conditions afforded to it by the inclusion of several late observations (Lorenc et al., 1988). The initial perturbations in an EPS should, therefore, have been able to replicate the impact of these missing observations. Indeed a reforecast of the 1987 storm using the ECMWF EPS showed that, had it been available at the time, a reliable warning could have been issued up to 96 hours ahead (Jung et al., 2005).
The access to increased computing power has not only allowed a larger range of model output to be produced, but it is also available at shorter intervals and in a faster time. In 1987 the UK Fine-Mesh model was available twice daily. It was run at 0000 utc and 1200 utc and was available around three hours later for use in the CFO. The Global model was run at the same time, but took appreciably longer to complete its calculations, taking around five hours to become available.
On the other hand, the present day models mentioned above, with the exception of some of the non-Met Office models, can be run four times a day. The NAE and Global run at 0000, 0600, 1200 and 1800 utc and are available around two or three hours later respectively, whereas the UK4 and UKV are run at 0300, 0900, 1500 and 2100 utc and are available around three to four hours later. This increased model production allows forecasters to react more quickly to changes in model outcome and potential forecasts than was possible in 1987. It also means that there is a reduced wait time for new model output, which is useful to see if a change in a model outcome is consistent, in which case more confidence can be placed in the forecast.
Another major improvement since 1987 is in the resolution of the models themselves. Numerical models work by representing the atmosphere at different points both in the vertical and horizontal, making up a three-dimensional grid. The resolution is the spacing between these points: the finer the grid, the smaller the feature that can be modelled. The modelling of very small-scale features is very important for gauging the exceptional wind speeds associated with strong wind storm events, including those arising from sting jets.
In 1987 the highest resolution model was the Fine-Mesh model (Gadd, 1985). This had a horizontal resolution of approximately 75km. However, for a model to accurately resolve a feature within the atmosphere it must be at least two grid squares across, so in the case of the Fine-Mesh model it would have to be 150km across. The highest--resolution model currently available to Met Office forecasters is the UKV, which has a horizontal resolution of 1.5km, so can resolve a feature that is just 3km across, 50 times smaller than the Fine-Mesh model.
The vertical resolution has also improved markedly. In 1987 both the Global and Fine-Mesh models had 15 model levels in the vertical. The lowest level was approximately 20m off the ground, with the highest level at around 25km. The levels were not evenly distributed though, with a higher density near the surface and in the upper troposphere in order to model the boundary layer and the jet stream more accurately. However the modern GM, NAE, UK4 and UKV models all have 70 vertical levels. This allows the same number of model levels which were present in the entire Fine-Mesh model, i.e. 15, to be given over to modelling just the bottom 5500 ft in the GM and NAE. However, the vertical resolution in the layers closest to the Earth's surface is increased in the UK4 and UKV, which means that the first 15 levels model just 2750ft or so of the lowest atmosphere.
However, even with all these advances, modern NWP models cannot always capture the full intensity of the strongest windstorms. For example during the 3 January storm that struck Scotland, the MOGREPS output indicated less than a 1% chance of gusts exceeding 80kn through the central belt between Edinburgh and Glasgow, even in the model runs just a few hours before the onset of the strongest winds, although it did signal a greater than 80% chance for the coastal areas close to Islay. The UK4 also fared poorly, placing the core of the strongest winds around 100km further north than they occurred in reality, and also underestimating the large spatial extent over which the extreme winds occurred. This storm illustrates that, despite all the massive gains in NWP since 1987, there are still some rapidly-deepening storms for which the models do not adequately capture the observed wind speeds with any great prescience. Local knowledge of topographical features, funnelling, and an alert eye on actual observations are still vital aids to forecast accuracy.
The Great Storm of 1987 changed the way in which forecasters deal with major windstorm events. It is now possible to warn the public of impending periods of strong winds up to five days ahead. It is also more likely that the intensity of the storm, and hence the strength of the winds, will be correctly forecast. Forecasters will still face challenges similar to those of 25 years ago, with differing NWP scenarios and explosively-deepening lows. However, the experience gained since 1987, in conjunction with improvements in communicating the forecasts to the public, should, in future, help to counteract these issues.
Thanks to Nick Grahame from the Met Office for his constant support and guidance throughout the writing of this article. We also greatly appreciate the input and advice from Ewen McCallum, Pat Boyle, Robert Neal and Charles Powell, also from the Met Office. We would also like to acknowledge the help given by Tim Hewson, jointly of the Met Office and ECMWF, particularly in reference to the 3 January Storm.