Creating forecasts for hours ahead

At short time ranges a higher level of detail can be forecast with more reliability. The forecasting of the weather in the 0- to 6-hour time frame is often referred to as nowcasting

Traditionally, numerical computer models have been poor at forecasting thunderstorms and other small-scale details. Therefore, the human forecaster has had an advantage over computer NWP models when it comes to forecasting small- (meso-) scale features. The forecaster is able to compare a model field against actual observations and respond quickly and amend a forecast, should the situation warrant it. Rainfall radar observations are very useful in this time frame, and post-processing is used to make very short-range predictions.

More than six hours ahead, numerical model forecasts gain an advantage over other forecasting techniques. Ongoing research at the Met Office, to develop the next generation high-resolution numerical weather prediction system over the UK, should eventually allow numerical model forecasts to become the dominant nowcasting tool as the model's ability to forecast thunderstorms and small-scale features dramatically improves.

National Severe Weather Warning Service

The Met Office has a responsibility to provide warnings of severe weather under the National Severe Weather Warning Service. Nowcasting is used to decide when a Flash warning, which indicates a high confidence of severe weather occurring in the next few hours, needs to be issued. Many forecasts are provided to our customers in this time range.

Mesoscale modelling

Forecasting with high-resolution models

A system called numerical weather prediction (NWP) forms the basis of modern weather forecasting. The system uses a mathematical model of the atmosphere which has been derived from the laws of physics. This model provides a set of equations to solve in order to predict the future weather. These equations are solved by averaging over 'chunks' of the atmosphere (grid boxes) and short periods of time (time steps) to then give us numerical equations which are put into the supercomputer.

Model resolution

The size of a chunk is called the 'resolution' of the model, similar to the resolution of an image from a digital camera. More than one grid box is needed to represent weather features in the much the same way as a digital image needs more than one pixel to represent something like a face. The number of equations which have to be solved depends on the total number of grid boxes. As the equations have to be solved long before the weather happens, the size of grid boxes (given the forecast area) is limited by how quickly they can be solved.

The current global forecast model has a horizontal resolution of about 40 km over the UK, meaning 160 million equations have to be solved just to step the atmosphere 15 minutes in time. This resolution is very good for information about the general weather conditions over the UK, and the Met Office's current computer power means a 5-day forecast can be produced in a few hours. At shorter range (1-2 days) a higher-resolution model (about 12.5 km) is used, because it provides more regional detail.

Convective clouds

Much of the UK's most damaging weather involves clouds which 'bubble up' from near the surface when a layer of cool air lies above a layer of relatively hot, moist air. These are called convective clouds, and the most energetic are often thunderstorms which may produce torrential rain, snow, damaging hail, flash flooding and strong winds, including tornados.

The resolution of current models is unable to represent individual convective clouds. A large thunderstorm may be about 10 km across, with a very strong core less than 1 km across, so the current models completely average over even the largest thunderstorm. While the forecast might state that a particular region is at risk from thunderstorms and torrential rain, it is currently impossible to know where individual thunderstorms might form.

Convective-scale NWP

As computer power increases it will be possible to reduce the grid box size in the models. In preparation, the Met Office is already experimenting with models which make a major jump in resolution. Instead of completely averaging over thunderstorms, models are being developed which actually allow thunderstorms to form. So far there have been good results for large storms from a resolution of around 1 km. This is known as 'convective-scale' NWP. This resolution is also very useful for other aspects of weather, such as major areas of fog and detailed features of the wind around ranges of hills.

The animation on the right is an example forecast from an NWP model running at 1.5 km resolution, showing a representation of cloud (grey) and rainfall (colours).

Convective-scale NWP in action

We have tested our experimental system using a wide variety of recent events over the UK. For example, the major flash flood in High Wycombe on 3 August 2004.

Nowcasting

Nowcasting is a technique for very short-range forecasting that maps the current weather, then uses an estimate of its speed and direction of movement to forecast the weather a short period ahead — assuming the weather will move without significant changes. Since it takes time to gather and map the weather observations, a short forecast is needed to even know what the weather is 'now'.

How nowcasting works

Rainfall and associated severe weather, such as hail and lightning, are the most widespread and most advanced applications of nowcasting. In the UK, rainfall nowcasts can be useful up to three or four hours ahead in widespread rain bands in winter, but only one to two hours ahead for summer thunderstorms.

To extend the period of predictability nowcasts can be combined with output from numerical weather prediction models.

The Met Office uses nowcasting for many weather variables including wind, temperature, snow and fog. Because it is a forecasting technique that can be applied quickly, either by human forecasters or by modest-sized computers, it is possible to update the forecasts frequently — every time there are new observations available. In the Met Office most nowcasts are updated every hour.

As computer models improve, the lead times will become shorter and, ultimately, these simple techniques may be used for instant forecasting, such as the immediate path of a tornado.

Short Term Ensemble Prediction System (STEPS)

Fig 1. Historic example of radar

The Met Office has STEPS (Short Term Ensemble Prediction System), a state-of-the-art rainfall nowcasting system, developed in collaboration with the Australian Bureau of Meteorology.

In STEPS, the rainfall distribution is separated into different sizes of rainfall feature, so the large rainfall events (which are the more predictable) can be nowcast for longer, while the small events are only nowcast for a very short time.

Beyond this predictability limit, information is used from the NWP model for larger rainfall features, and the smaller features are filled in realistically using a random statistical method.

The 'ensemble' in the title refers to the fact that many forecasts are produced, with the rainfall areas moving at slightly different speeds, and with the small rainfall features represented by slightly different random statistics. Using this approach enables a realistic range of uncertainty to be estimated for flood forecasting.

History of nowcasting

Nowcasting is a very old technique. When Admiral FitzRoy first produced forecasts at the Met Office in the 1860s, he did it by collecting reports of storms from around the coast, and then sharing these reports with coastal ports that may be downwind, so that they knew there was bad weather coming. This was a simple form of nowcasting.

The term 'nowcasting' was actually coined in the 1980s by Met Office scientist Professor Keith Browning, to describe the process of extrapolating a sequence of radar images to produce a very short-range rainfall forecast.

Creating forecasts for days ahead

To produce weather forecasts for one to 15 days ahead, we use a variety of techniques. Our North Atlantic and European numerical forecast model provides more detailed information over this region in forecasts out to two days ahead. In addition, our short-range ensemble prediction system, MOGREPS, provides information on the degree of uncertainty associated with the forecasts.

Forecasting the oceans

For those working at sea or living near the coast, forecasts of wave height, ocean currents or storm surges up to days ahead are just as vital as forecasts of the weather.

Forecasting days ahead

When forecasting more than two days ahead we need to use our global forecast model, because the weather happening many thousands of miles away today will affect the weather over the UK in a few days time. In medium-range forecasts, two to 15 days ahead, the use of an ensemble is essential as the uncertainty in the large-scale weather patterns becomes greater.

Seasonal forecasting

What is a seasonal forecast?

Seasonal forecasts provide information on how weather, averaged over the next few months, is expected to vary from normal, e.g. "Are UK rainfall totals this winter likely to be above or below the long-term average?". The UK/Europe forecasts relate to the conventional seasons — winter, spring, summer and autumn. For other parts of the world the period of the forecast may vary, e.g. the

North Atlantic tropical storms forecast refers to the June to November season. Seasonal forecasts are indications of an overall picture, as it is impossible to forecast individual events so far ahead; the short-range forecasts are where the details begin to appear.

Because of uncertainty in forecasting at long range, seasonal forecasts are generally expressed in terms of probabilities. For example, our forecast for mean UK temperatures for the winter of 2007/8 gave probabilities for a relatively warm winter, an average winter and a relatively cold winter, as 50%, 30% and 20% respectively.

Long-range forecasts are part of the advice provided to the public on prospects over a range of time scales, and can help government agencies and companies with their long-term strategic planning. The forecasts have global coverage and are used in areas like Africa to help plan for year-to-year variability in rainy seasons.

How are seasonal forecasts possible?

Slowly varying aspects of the Earth's climate, in particular fluctuations in the surface temperature of the global oceans, can influence patterns in the weather. These influences are not easily noticed in day-to-day weather events but become evident in long-term weather averages.

The slow fluctuations of sea-surface temperature (SST) can be predicted, to some extent, at least up to six months ahead. The links between SST and weather can be represented in computer models of the atmosphere and ocean. Computer models developed at the Met Office, like those used in making both daily forecasts and long-term climate change predictions, form the basis of our seasonal prediction systems.

The strongest links between SST patterns and seasonal weather conditions are found in tropical regions, and it is here that seasonal forecasting is most successful. The best known links are those associated with sustained large-scale warming (or cooling) of SST in the tropical Pacific known as El Niño (or La Niña) events. These events can disrupt the normal pattern of weather around the globe, bringing, for example, large changes in seasonal rainfall that lead to droughts in some regions and floods in others.

Although the strongest links between SST and seasonal weather are found in the tropics, there is good evidence that similar, if weaker, links are present in other parts of the globe. The computer model forecasts can thus provide the best available guidance on likely seasonal conditions in many parts of the world, including Europe.

Because the link between weather and SST is best detected in long-term weather averages, and because the uncertainty in forecasts generally rises as the forecast range increases, seasonal forecasts look rather different in format compared to the familiar daily forecasts. The two key differences are:

• forecasts are for conditions averaged over three-month periods
• forecasts are stated in terms of probability

How are the forecasts produced?

The same computer models of the atmosphere that are used to make the daily weather forecasts are used, with some differences:

• they are run forward in time up to many months ahead, rather than just for a few days
• active oceanic, as well as atmospheric, components are included
• they are run many times, with slight variations to represent uncertainties in the forecast process

We occasionally use statistical forecasting methods on the seasonal timescale — in winter and summer for UK and Europe. This is done where physical relationships between weather and the state of the oceans have been found, but where models do not yet show sufficient skill to pick up these particular relationships. This gives rise to a mixed statistical and physical model forecast process.

We also use this mixture of methods for forecasting the mean global mean surface temperature for a year ahead. However, on even longer time scales, such as a century ahead, only physical models are used, as no more skilful statistical approach has been found.

Forecasts for other regions

We make seasonal forecasts for all parts of the world, focusing on specific regions where there is particular vulnerability to seasonal anomalies such as droughts, and where relatively good predictability has been identified. These include forecasts for rainfall in the north-east corner of Brazil during their wet season (February-May), in East Africa during the October-December wet season and in tropical West Africa, including the Sahel region, for July to September.

Creating forecasts for months ahead

From forecasting whether the coming season will be warmer or drier than normal to predicting what the world will be like in 100 years, our scientists use the same process used to produce weather forecasts hours or days ahead.

These long-range forecasts incorporate more of a global view to look out a month or more into the future. They factor in details such as the current average state of the atmosphere and the ocean at distances often thousands of miles away from the specific location of interest. Long-range forecasts estimate only the average weather, not specific weather events. Therefore, in seasonal forecasts, language such as 'warmer than normal' and 'wetter than normal' is common.

Seasonal forecasting

Seasonal forecasts provide information on how weather, averaged over the next few months, is expected to vary from normal.