Free

Entering a new era – climate science in the time of big data and machine le...

Event Information

Share this event

Date and Time

Location

Location

Imperial College London

Data Science Institute

William Penney Lab

London

SW7 2AF

United Kingdom

Event description

Description

Speaker Biography

Peer Nowack is a computational physicist who uses state-of-the-art numerical models and machine learning to address key challenges in the atmospheric sciences. His central research interest is to improve our understanding of the physics of the Earth system and its relation to changes in atmospheric composition. Since August 2017, he has been an Imperial College Research Fellow at the Department of Physics and the Grantham Institute for Climate Change. Previously, he worked as a Postdoc and PhD student at the University of Cambridge after completing his undergraduate degree at ETH Zurich in Switzerland.

Talk Abstract

Over the last two decades, Earth Science has rapidly evolved into a big data research field. This development is mainly the result of ever increasing computational power, which has led to a swift expansion of a number of computer modelling projects, and parallel advances in measurement systems such as satellites. For example, climate model simulations that inform the Intergovernmental Panel on Climate Change (IPCC) produced 1GB of data in 1996, an amount that has increased to an estimated 100PB this year (via 35TB in 2005). NASA’s Earth observing data archive alone now contains around 18PB of data, up from 9PB in 2014.

Here, I highlight why typical data science methods, from machine learning to data visualisation, have the potential to revolutionise scientific disciplines that rely on these increasingly large datasets. In this context, I emphasise two central challenges:

1) the ongoing quest of reducing uncertainty in climate change projections, especially on the regional (county-city) scale. I show how statistical models of reduced complexity can help us to understand drivers of uncertainty in climate change projections. In addition, I explain how self-learning algorithms could be used to build better mathematical representations of physical processes that cannot currently be resolved in climate models. Ultimately, I argue that the combination of the now abundant data and machine learning could provide a regime change in our ability to test scientific ideas in this area and to improve model projections.

2) the ongoing deterioration of air quality, a separate but related concern which is known to have serious health impacts in cities like London and Beijing. I touch on how machine learning and more generally advanced statistical methods could help us find effective ways to improve air quality in cities.


Share with friends

Date and Time

Location

Imperial College London

Data Science Institute

William Penney Lab

London

SW7 2AF

United Kingdom

Save This Event

Event Saved