Using machine learning for nowcasting to improve the safety and efficiency of aviation
With an increasing number of aircraft passing through our skies every day, any hazardous weather is extremely disruptive. This work, by Senior Met Office Scientist, Claire Bartholomew as part of her PhD at the University of Leeds, looks at how we can improve current Met Office data with machine learning to look at short-term forecasting and improve the safety and efficiency of aviation.
Turbulence, increased risk of hail strike, severe downdraughts and electrical disturbances are all hazardous risks to the aviation industry linked to convection (or thunderstorms). When there is little warning of convection, it can lead to various issues including reactive flight rerouting which increases fuel usage, reduced airspace capacity and delays in airport arrival and departure times. As Claire explains: “If a thunderstorm isn’t forecast, a pilot would have to react to this real–time weather event on their radar and take a routing decision to avoid it, but by predicting this in advance, the flight route would be changed to avoid the hazardous weather conditions, which could save fuel and avoid additional workload for air traffic control to communicate and co-ordinate aircraft deviating from their planned flight route”.
Forecasting this convection however, even just a few hours ahead, is an ongoing challenge within meteorology. With an increasing number of aircraft passing through our skies every day (when the aviation industry is working at normal capacity around 3000 flights arrive and depart every day in the world’s biggest airspace sector – the London Terminal Manoeuvring Area (LTMA) (Source: NATS, 2018)) any hazardous weather is extremely disruptive.
There are an ever-growing number of real-time weather observations available across the UK, from satellite data, surface data, automated weather stations on the ground or from aircraft, plus radar data. Currently all these observations are used to create a weather forecast using the Met Office’s Numerical Weather Prediction (NWP) model. The data is put into a global 10km grid or a UK 1.5km grid and then the NWP combines this to estimate what the weather is doing now. The model then uses the laws of physics to create a forecast.
With machine learning we are able to integrate these data sources on a more frequent basis to improve convective short-term forecasts (commonly referred to as nowcasting) using computation models. One advantage of machine learning is that it can go into much more granular detail than the 1.5km squares the NWP model uses. Machine learning also has the potential to add value in not only the predicted movement of convective features but also their development. This all means that this way of working could add value to the more traditional extrapolation nowcasts currently used by the Met Office.
Claire Bartholomew, Senior scientist at the Met Office says: “Machine learning also has the advantage over other physically-based methods in that it is not constrained by our incomplete knowledge of the physics and mathematics of the relationships between all the chaotic structures in a given weather system.” But she adds that “…our physical understanding will still be able to inform the solution, for instance in the selection of datasets which we expect to influence the solution.”
Accurate forecasts of convection and increased confidence in the forecasts will help to significantly reduce disruption, allowing for airspace to be managed more safely and efficiently, such as by air traffic control pre-emptively applying flow regulations in certain sectors of affected airspace. This will reduce the need for pilots to take more reactive avoidance actions and give airlines more time to find suitable alternative routes in pre-tactical planning.
The ultimate goal of this work is to help to further improve the safety and efficiency of aviation, in an age of ever-increasing air travel. This work will help air travel be more efficient, be safer and reduce fuel usage and therefore also reduce its environmental impact. It could also have wider implications for societal benefit by minimising the other impacts of hazardous weather.
For more information on this work please contact Claire Bartholomew (sccsb@leeds.ac.uk).