# 2025 / March TRIC Seminar

## Digital Twinning, AI Infrastructure and Software at the National Centre for Atmospheric Science

:::info
**Speaker**: Laurents Marker
**Date**: March 12th 2025, 11:00 - 12:30
**Location**: The Alan Turing Institute, Margaret Hamilton Meeting Room
:::

In this talk Laurents Marker will cover three areas of ongoing work at the National Centre for Atmospheric Science (NCAS) that pose fertile ground for collaboration and the sharing of knowledge and tools with the Alan Turing Institute:

Jointly funded by the UKRI and the UK Met Office, the VISION (Virtual Integration of Satellite and In-situ Observation Networks) project aims to help close the gap between observed (satellite and airborne) and modelled data in weather and climate research. The VISION team have produced a python toolkit which provides sophisticated interpolation routines to link observations and model data and have applied this to a digital twin of the NCAS airborne laboratory for use in research campaign decision-making.

The user interface for this digital twin is based upon a generic browser-based mapping framework written on top of OpenLayers, React and Redux. Laurents will explore the design principles behind this mapping software, its current usage within NCAS, such as a web app for nowcasting over Africa, and the potential for developers to use it in their own applications.

Alongside his own work, Laurents will discuss JASMIN, the UK's data analysis facility for the environmental sciences and its recent upgrade: the ORCHID GPU cluster. It's co-location with the vast archive of atmospheric, oceanographic and earth observation data hosted at the Centre for Environmental Data Analysis (CEDA) holds significant potential for machine learning and artificial intelligence applications with which the JASMIN and NCAS communities are keen to explore and collaborate. Laurents has worked as a software developer at NCAS for two and a half years. He was brought onboard to develop software services and architecture in operational forecasting for observational research campaigns in weather, climate and air quality. He is based between the University of Leeds, the FAAM Airborne Laboratory at Cranfield University and South East London.

## Links
- [VISION project](https://gtr.ukri.org/projects?ref=NE%2FZ503393%2F1#/tabOverview)
- [UKRI TWINE programme](https://www.ukri.org/news/digital-twin-projects-to-transform-environmental-science/)
- [Orchid GPU cluster](https://help.jasmin.ac.uk/docs/batch-computing/orchid-gpu-cluster/)

## Questions
- [name=person]
- Aim is to fly more efficiently, what is the metric you use to decide?
- the main decision is between flight vs no flight (on the next day)
- wind weak? algo bloom present?
- multi-objective decision problem then?
- our system won't make decisions, it will provide data in a more helpful way.

- do you provide uncertainty quantification for the real-time decision?
- vision works with cf standard
- cnoversation with user to communicate possibility of interolation error
- whole suite of tools for aircraft plannign in general

- Is there enough overlap in flight trajectories from different campaigns?
- it depends how much data for a flight exists
- but optimally post-campaign you'd want to archive your data, e.g. on CEDA
- comes down to who is running the campaign and what they used for decision making. If they follow the standard it should be compatible
- What would be possible to do wtih this app if we had AI (cheap forecasting methods)
- ensemble forecasting would give you better likelihoods, also for atmospheric phenomenons that cannot be modelled (eg turbulence)
- also, injecting data in very last minute (no long load-up times) and longer ongoing forecasts
-