Single cell thunderstorm cloud

EUMETSAT fellow wins Weather4cast 2021 Challenge

Dr Jussi Leinonen developed the competition’s top algorithm for predicting short-term weather

Single cell thunderstorm cloud
Single cell thunderstorm cloud

A custom-designed recurrent neural network was the key to the prize.

Last Updated

02 November 2021

Published on

02 November 2021

On 20 August, the Institute of Advanced Research in Artificial Intelligence announced EUMETSAT fellow Dr Jussi Leinonen as the first place winner of the Weather4cast Stage 1 award.

Leinonen used meteorological satellite data to develop an algorithm that could most accurately predict the short-term weather for distinct regions spanning Moscow to the Azores Islands.

“It was pretty exciting,” said Leinonen, to find out he had won.

Leinonen’s winning algorithm used a recurrent neural network, a trainable system that draws on principles according to which the human nervous system operates. Rather than creating a traditional neural network to predict the weather, which would involve using a static set of rules, a recurrent neural network exploits the “memory” of the system by taking information from prior inputs to influence the current input and output. Recurrent neural networks are likely responsible for the drastic improvement in Google Translate over the years.

Dr Jussi Leinonen
Jussi Leinonen used a recurrent neural network to win the competition’s top prize.

The data contestants used to develop their models came from EUMETSAT’s Meteosat satellites and weather products—information such as temperature, rainfall, and cloud cover for the regions in question—came from the Nowcasting and Very Short Range Weather Forecasting Satellite Application Facility and the Spanish Meteorological Agency.

A EUMETSAT fellow at MeteoSwiss, the Federal Office of Meteorology and Climatology in Locarno, Switzerland since October 2020, Leinonen works on developing models to improve short-term thunderstorm prediction.

Leinonen’s work to improve the accuracy of weather forecasts has significant implications.

“Forecasts may be used to decide, for example, if an outdoor event should be held or cancelled, if flights should be rerouted to avoid bad weather, or if emergency services should prepare to respond to approaching extreme weather,” he said.

“Improving the accuracy of weather forecasts allows us to provide more reliable information, which in turn enables the decision makers to act more confidently.”

For winning both challenges, Leinonen was awarded a total cash prize of €10,000.

He presented his winning solution at the Workshop on Complex Data Challenges in Earth Observation, which was held online on 1 November.

Author:

Sarah Puschmann