Introduction

Following on from the success of the Deep Learning for Industry workshop in January 2023, the Maths4DL team organised a workshop for industry in the style of a hackathon. This took place at the University of Bath from 14-15 September.

This meeting was attended by the Met Office. A team of scientists brought data and problems and it was an exciting opportunity to engage with a challenging environmental problem involving data and machine learning with a great team from the Met Office. 

               

Agenda

Thursday 14 September	Activity	Location
10.15 – 10.30	Tea and coffee	8W 2.6
10.30 – 11.00	Introduction, plus short presentation about the challenge; formation of teams	8W 2.5
11.00 – 12.30	Hack	8W 2.5
12.30 – 13.30	Lunch and chat	8W 2.6
13.30 – 15.00	Hack	8W 2.5
15.00 – 16.00	Coffee break and fresh air	8W 2.6 and around campus
16.00 – 17.30	Hack	8W 2.5
18.30	Dinner	Bathwick Boatman
Friday 15 September	Activity	Location
09.30 – 11. 00	Hack	8W 2.5
11.00 – 11.30	Coffee and prepare for final push	8W 2.6
11.30 – 12.30	Final push	8W 2.5
12.30 – 13.00	Wrap up	8W 2.5
13.00	Lunch and chat	8W 2.6
14.00	Finish

Useful information

The atmosphere is a fluid, flowing over the Earth’s surface. As you get near to the ground, the wind speed decelerates and in certain idealised conditions does so in a well understood and mathematically easy to describe manner. More generally however, the wind speed’s variation with height is more complex and depends on many factors. These can include variations in surface roughness, temperature and moisture and typically will vary depending on the time of day, weather, location and season.  

Weather forecast and climate models use empirical functions fitted to data from a small number of locations and apply this globally. An accurate representation of near-surface wind impacts the weather forecast near the surface and interacts with the transport of constituents away from the Earth’s surface.

In this challenge, we will use a large dataset comprising thousands of hours of observations from around a hundred observations sites around the world, to try to machine learn a more general correction. As well as developing machine learning models, there will be tasks focussing on data exploration, data re-balancing, scouring the literature and producing informative visual content.

In order to optimise your participation in this hackathon, it is advised that:

• you have eduroam set up on your laptop
• you check are able to run python when away from your home institute
• you have some file space available
• you setup an environment with your favourite/most used python libraries, numpy, matplotlib and either keras/tensorflow and/or pytorch

 

How to register

If you have any questions about this meeting, please get in touch: maths4dl@bath.ac.uk.