‘Artificial Intelligence and Machine Learning to improve surface weather forecasts’ – webinar – November 9th, 2021

  1. Home
  2. news
  3. ‘Artificial Intelligence and Machine Learning to improve surface weather forecasts’ – webinar – November 9th, 2021
On November 9th from 3.30 pm to 5 pm (CET), IFAB, in collaboration with ECMWF, will present the research project “Empirical modeling of 2m temperature and 10m wind”.

PROGRAM:

INSTITUTIONAL GREETINGS
Francesco Ubertini, IFAB President

INTRODUCTION
Sanzio Bassini, IFAB Technical-Scientific Coordinator
Florence Rabier, ECMWF Director-General
Tiziana Paccagnella, Member of the Steering Committee of Italiameteo

PRESENTATION OF THE PROJECT AND ITS RESULTS
Peter Dueben, ECMWF Coordinator for Machine Learning and AI Activities
Fenwick Cooper, ECMWF Visiting scientist

Q&A

Live stream event. The event will be held in English.

Rewatch the event:

About the project:

The research project, realised by ECMWF (European Centre for Medium-Range Weather Forecasts) and supported by IFAB (International Foundation Big Data and Artificial Intelligence for Human Development) aims to reduce errors in the surface weather forecasts. Three machine learning algorithms are applied to model errors in the ECMWF operational 0-48 hour forecasts of temperature at 2m above the surface and wind at 10m above the surface. A reduction in global root-mean-squared error of between 5 and 15%, depending upon season, is observed. 1 year of training data and 3 months of verification data between December 2019 and May 2021 is used for each of four seasons. The three machine learning algorithms used are linear regression (of cubic polynomial functions), the random forest and a 3-layer neural network. Each algorithm returns similar error reduction perhaps indicating that they are all approximating the same physics. Examining errors at individual stations indicates large scale regions where error reduction is robust. Large regions where error reduction was non-existent or increased and individual stations where the error substantially increased also exist. Atmospheric variability, excessive model extrapolation and reduced station data quality are identified as three possible explanations.

Previous Post
Second IFAB Lecture: ‘The energetic transition between growth and balance’ with Riccardo Valentini and Alberto Piatti – October 12, 2021
Next Post
Francesco Ubertini is the new IFAB President
Menu