By: Peter Clark
As I sat in the Kia Oval in Kennington having taken a day off to watch the first One Day International between England and Pakistan, I had plenty of time to appreciate the accuracy and utility of weather forecasts. The afternoon proved to be a microcosm of both the successes of modern weather forecasting and issues surrounding the use of forecasts in more serious applications (though I may well join in with the cries of “there’s nothing more serious than Cricket!”).
First question: to go to the match or not? When we bought the tickets 6 months ahead, we just had climatology to go on. Early May is a risk, but not very different from later in the season. By the time forecasts become available the question is then “is it worth turning up?” By the Friday five days before, there was a very strong consensus amongst computer forecasts that a cyclone would be tracking across England on the day, most likely during the first half of the day. In fact, the Met Office’s ‘deterministic’ forecast proved very accurate, with the continuous heavy rain passing through London by midday. However, behind the surface front close to the cyclone centre, cold air aloft was overrunning warmer air at the surface, which was given an additional boost as it came from the Atlantic and passed over land. Warm (and moist) air beneath colder air leads to the likelihood of dreaded convective showers in the afternoon!
There have been real ‘revolutions’ in forecasting over the last few decades. At the centre lies the combination of vast improvements to computer power, more accurate computer models, vast increases in observations to ‘correct’ the data in the models, and development of much more powerful methods to use (or ‘assimilate’) those observations. An extratropical cyclone, or ‘low-pressure system’, is relatively large and long-lived. In this case, the system was at the small end of the scale and quite intense, roughly the size of England – say 500 km across with a life cycle of at least a day. 30 years ago, our computer models had to represent these systems with a grid of points not much better than 100 km apart (see the Met Office’s history of NWP, for example). Today our forecast models have little problem actually representing a cyclone. In practice, they are often predicted in forecast models even before there’s any clear sign of them in observations. While there will still be uncertainty in track and intensity, on the whole they are astonishingly well forecast several days ahead.
Here lies the problem. Showers are much smaller, say 10 km across with the core less than 1 km, and have a lifetime of an hour or so. These cannot even be directly represented in our global models. The most recent ‘revolution’ in forecasting has been the development of so-called ‘convection-permitting’ models (Clark et al. 2016). Regional models (with a grid spacing around 2 km) at last can represent showers, but not well. Something resembling showers can form and give us some very useful guidance on the probability that we’ll be affected by a shower. Such models are now helping produce more accurate flood forecasts, especially for smaller, faster reacting catchments (Dance et al. 2019). Within the ParaCon project we are working hard to find ways to improve the models.
Figure: Radar estimates of the surface rainfall rate at 17:00, 18:00 and 19:00 BST with inset showing the hail storm that hit the Kia Oval at 17:00 BST. (Courtesy of the Met Office). Showers are triggered along a ‘peninsular convergence’ line extending from Cornwall all the way to London that is present for several hours. Clearly, much depended on whether one was beneath or to one side of this.
The message was the same in the morning before the game. As the rain from the cyclone cleared, a high probability of seeing one or two showers or even thunderstorms during the afternoon – which is precisely what happened. We had a couple of flurries of not very intense rain, which did little to interrupt play, plus two hail storms; pea-sized hail fairly typical of a British summer shower. Each lasted about 5 minutes. The inset in the figure shows the hail storm that hit the oval around 17:00 BST. A mere speck on the scale of England, but locally extremely intense. A perfect forecast! However, a computer model run even a couple of hours before could not predict the precise shower hitting our precise location.
What more could we do? I spent the afternoon trying to look at the Met Office’s weather radar composites on my phone. A new rainfall picture is produced every 5 minutes. On the intermittent occasions when I could access data, the showers were very clearly tracked; interestingly they were forming along a broad ‘peninsular convergence’ line that could be tracked back to Land’s End. Along this line, air coming from either side of the south west peninsula meets and so is forced upwards, triggering showers (Golding et al, 2005). This is shown in the three radar images in the figure. Each is an hour apart, but this convergence line is very persistent. These lines were the topic of the COPE field campaign in 2013 (Leon, et al. 2016). This organisation by topography radically changed the overall predictability of the showers. The sharp-eyed reader might also notice an arc of showers moving east from central England into East Anglia, and it is probably no coincidence that the heaviest storm happened where this met the convergence line. Nevertheless, as we sat on the edge of this line, the best we could hope for several hours ahead was a realistic assessment of the probability of having a shower.
This example illustrates very well that the weather forecast is not the only piece in the jigsaw. First, and foremost, there is the investment in resilience; the Oval ground is very well prepared and drained, but there is a limit to what it can cope with. Similarly, investment in flood defences is often controversial, and the Environment Agency have recently announced that climate change is forcing a ‘new approach to flood and coastal resilience’ that may mean not investing in flood defences in some regions.
Second, there is preparedness. The available forecasts had prepared us well for the likelihood of showers. We equipped ourselves as well as we could. I kept a ‘weather eye’ on the radar, at least as far as technology allowed me. I could see the hail storms coming. In this case, the covers were deployed fast enough to protect the pitch and run-ups. Use of forecasts could enable the deployment of defences that take longer to deploy but ultimately save playing time. Currently, forecasts are used by the authorities to help emergency services prepare for likely (but rarely certain) flooding. How best to educate and prepare users including the public to respond to forecasts is one of the leading questions driving research, for example the World Meteorological Organisation’s ‘HIWeather Project’, which recognises the key importance of “better understanding by social scientists of the challenges to achieving effective use of forecasts and warnings” (HIWeather Impact plan). A key part of this is understanding the inevitability of false alarms. We have to be prepared to see play stopped because a forecast (in this case with a very short lead time) says there is a probability of a heavy shower. The price for not being pre-emptive may be the abandonment of the match. Which happened two and a half hours after the rain and hail stopped.
The modern challenge of forecasting is not just to improve the forecast (which may be an exercise in diminishing returns) but also to find ways to make sure that systems are in place to make full use of them and users are well-prepared to take action and understand the actions of others.