People in the UK love talking about the weather. Regularly checking the forecast has become as much of a daily habit as checking Facebook and email for many, but how do meteorologists "predict" what the weather will be like weeks in advance, as in long-range weather forecasts?
Modern weather forecasting
When the UK’s national weather service, the Met Office, was established in 1854, forecasters predicted the weather using hand-drawn charts and human observations. Read more: How to predict the weather using nature
Now, forecasters are using sophisticated supercomputers performing trillions of calculations every second. These machines process hundreds of thousands of atmospheric global observations, which are converted into numerical data and fed into the supercomputers to solve equations. With increased computing power, calculations take minutes.
Consequently, today’s four-day forecasts are as accurate as the next-day forecasts were 30 years ago, and the exciting thing is; this increasing forecast accuracy is expected to continue into the future as technology advances.
"These supercomputers use hundreds of thousands of global weather observations as a starting point for running an atmospheric model containing over a million lines of code," Met Office meteorologist, Emma Sharples, told WIRED. "We use these models to produce weather forecasts and long-term predictions about Earth’s future climate."
The Met Office supercomputer is just the tip of the meteorological iceberg, Sharples continued.
"Weather forecasting involves separate but linked processes. The Met Office gathers and processes data on existing weather before processing data to set up the initial conditions in a numerical model. It then computes the evolving weather conditions with elapsed time and repeats the process to refine the forecast."
When it comes to predicting rain, snow and other precipitation, regular weather forecasting is able to gather local and satellite measurements and assimilate them into the heavy models of atmosphere, which then return the forecasted precipitation amount over each region. The weather forecasting programs include the codes for the changes in clouds that cause precipitation. Here’s how it works:
The Met Office observation network receives around 500,000 daily observations, all recording global atmospheric conditions. In the UK, the Met Office maintains 200 unmanned ground-based weather stations reporting on the state of the atmosphere. These stations are spaced around 25 miles (40km) apart to record weather associated with the low pressure and frontal systems that cross the UK. Read more: The truth behind the 'toxic haze' caused by the UK heatwave
"Hundreds of other stations gather more data, including measurements of rainfall and data from ships and buoys," explains Sharples. "These observations are supported by weather balloons that measure atmospheric parameters, and aircraft that collect observations of temperature and wind.
"Data from weather aircraft and radiosondes – a type of high-altitude balloon – is compared with data from ground stations and satellites to check accuracy and improve atmospheric models."
Land-based weather stations provide essential observations, too, but because they are spaced fairly far apart, some relatively small-scale isolated features, such as thunderstorms, can evade the network. That’s when radar and weather satellites come into play.
"The UK weather radar network emits short pulses of radio waves which, when intercepting precipitation, have part of the energy reflected or scattered back to the station," Sharples explains. "These readings are analysed to determine the extent, type and location of any precipitation. Doppler radar provides extra detail on wind speed and direction of a storm."
Satellites have also revolutionised forecasting, making it possible for meteorologists to identify weather patterns days in advance of ground weather stations. They provide accurate temperature and humidity data to enhance the accuracy of data from ground weather stations.
There are two types of weather satellites: polar-orbiting satellites, which orbit the Earth from north to south, passing over specific locations twice every day; and geostationary satellites, which orbit the Earth above a fixed point on the equator providing constant data measuring the hemisphere below. Polar-orbiting satellites orbit nearer to the surface of the Earth at an altitude of 530 miles (850km). This means they provide higher resolution than geostationary satellites, which orbit at a height of 22,000 miles (35,400km). Both satellite types provide the coverage needed for 3D grid-based weather modelling.
The next step is to combine and convert this data into computer code to numerically represent atmospheric conditions, a process known as atmospheric data assimilation. Thanks to increased computing power and improved understanding of atmosphere dynamics, sophisticated assimilation schemes can represent the current atmosphere more accurately than ever before. However, in order to provide the models with unbiased initial data, observation and modelling errors need to be taken into account.
A numerical weather prediction model is used to calculate how weather conditions will evolve. Results from these billions of mathematical calculations provide guidance to forecasters and form the foundation of detailed forecasts.
Meteorologists divide Earth’s surface and atmosphere into a 3D grid and map all weather data onto this grid. Each grid box represents the atmospheric processes in that region. Numerical equations are applied to each box of the grid, and researchers evaluate the forecast results. Numerical weather prediction models run globally as local weather is influenced by what’s happening in other parts of the world.
Depending on the type of weather forecast, models are run at different grid resolutions. At low resolution, models can accurately produce large-scale weather patterns such as wide areas of rain or regions of shower activity. Higher-resolution models and increased computing power are needed to produce small-scale features such as individual showers, valley fog and heavy rainfall in mountainous regions.
Many weather providers are using the power of artificial intelligence to take weather forecasting to the next level, providing real-time weather updates to allow for personalised experiences in very specific locations. Read more: Forget Google, Uber, Amazon and Spotify. In Russia, Yandex rules
Yandex, Russia’s largest internet company, uses a technology called ‘Nowcasting’ in its Yandex.Weather product, which is used in many apps and services across the country. The firm’s Head of Yandex.Weather, Dima Solomentsev, explains that it “takes precipitation prediction to the next level” by using data from weather radars as well as a state-of-the-art tool to gather the precipitation map 124 miles (200km) around the radar location.
“Our Nowcasting technology uses radar imagery to forecast the changes in weather using a deep learning neural network called Yandex.Meteum, which performs the transformations, transitions and other changes with the latest radar shot,” Solomentsev explains.
“Essentially, it allows us to do the forecasting on a minute-to-minute basis for the next two hours with a very high degree of accuracy, calculating the weather forecasts not only for the pre-defined grid points but also for the exact coordinates requested by the user.”
Yandex.Meteum also provides hyper-local forecasts down to the exact block in the city where the user is located, as opposed to most services that offer city-based forecasts.
“Weather can actually be really diverse within one city,” adds Solomentsev. “For example, it may rain heavily in Central Park while the sun is shining on Wall Street.
“When a Yandex.Meteum user checks the forecast, it prompts a request to the Yandex.Weather API to deliver that person's location. We gather the pre-computed values for the requested regions, blend them with the data that is gathered operationally from the users and local data feeds.”
After the calculation, Yandex sends the precise weather information back to the Yandex.Weather user, completing the whole process in one-tenth of a second. This results in different products, such as a future precipitation map, seen in the figure below, as well as the smart push notifications, alerting a user to the exact moment when precipitation will stop.
Although data assimilation, numerical models and weather-focused machine learning have progressed in recent years, a single numerical weather prediction forecast can still be wrong. Weather forecasters use ensemble forecasting to cope with such uncertainty.
Rather than one numerical weather prediction forecast, the model is run several times using slightly different starting conditions, based on observations. Analysing this collection of forecasts makes it possible to assess the chance of a particular sort of weather.
As a result, meteorologists’ knowledge and experience is crucial in comparing numerical weather predictions against observations, and are needed to correct for known biases or weaknesses in computer model predictions. They determine the emphasis in weather forecasts and tailor forecast information for all audiences.
This article was originally published by WIRED UK
