Machine Learning for Environmental Sciences

21 July, 2019

Rachel Furner is a PhD student at British Antarctic Survey, which has recently opened up its new AI Lab, that aims to foster the application of various machine learning (and adjacent) techniques to the rapidly growing, complex, and heterogeneous body of data found in atmospheric, oceanic, and Earth sciences. This blog is a reflection of the two-day workshop at BAS on 17-18 June 2019.

Understanding high-impact environmental risks around the world, and their potential effects on industry and human health, is crucial for future economic growth and supporting a strong society. In the “Big Data” era, the abundance of highly-relevant data offers great potential for reducing uncertainties in the prediction of such events. To continue cutting-edge science discovery and innovation, there is an increasing need to adopt modern technologies, such as Artificial Intelligence (AI) which has transformed bioinformatics and genomics, amongst other scientific disciplines, and has the potential to revolutionise environmental risk assessment.

To do this it is essential that we bring together the various research disciplines (machine learning, environment, health, structural engineering etc) with decision makers from business and policy groups. To this end, the British Antarctic Survey and the University of Cambridge coordinated a workshop  – Machine Learning for Environmental Sciences – on the 17th/18th June 2019 to discuss the application of data science techniques to environmental challenges. The workshop was hugely successful, with 80 attendees from around the UK and beyond, and up to 30 remote participants watching online.

Keynote presentations were given by Dr Emily Shuckburgh from the University of Cambridge, and Associate Professor Claire Monteleoni from the University of Colorado. These were joined by a number of short talks from delegates, as well as discussion sessions, and a lively poster session. Presentations and posters covered a variety of topics, including examples of applications in atmospheric, oceanographic, and biological research.

Recordings

For those who could not join us, we have uploaded some of the presentations from the workshop to the BAS YouTube channel:

• Environmental Data Science – Emily Shuckburgh (Univ. Cambridge)
• Constructing a Digital Environment: A NERC-led initiative – Stephen Hallett (Univ. Cranfield)
• How to use deep learning in weather and climate models – Peter Dueben (ECMWF)
• Applying machine learning to improve simulations of a chaotic dynamical system using empirical error correction – Peter Watson (Univ. Oxford)
• Deep Gaussian Processes for Multi-fidelity Modelling – Mark Pullin (Amazon)
• The model is simple until proven otherwise – Anita Faul (Univ. Cambridge)
• Detecting anthropogenic cloud perturbations with deep learning – Duncan Watson-Parris (Univ. Oxford)
• Unsupervised classification: identifying dynamical and biogeochemical regimes in oceanographic data – Dan Jones (BAS)
• Unsupervised Classification of Convective Organisation with Deep Learning – Leif Denby (Univ. Leeds)
• Using machine learning to decrease uncertainties in tropospheric ozone loss – Tomás Sherwen (Univ. York) (paper)

The workshop was followed by a hands-on data challenge, with a number of environmental datasets made available for participants to experiment with. Many participants had no previous experience with machine learning tools and by the end they could see ways in which they could apply them to their own research challenges.

The workshop and data challenge was organised jointly by the BAS AI Lab (https://www.bas.ac.uk/ai) and the Cambridge Environmental Data Science Group (Encompassing research from across the University, BAS and industry partners) and made possible by support from the British Antarctic Survey, the Alan Turing Institute, the CCIMI, Cambridge Big Data, Cambridge Spark and the University of Cambridge.

Mailing list

To continue to grow the Machine Learning/Environmental Science network we have setup a mailing list which anyone can subscribe (https://lists.cam.ac.uk/mailman/listinfo/ucam-cedsg-news). Further meetings and related events will be announced through here.

Centre for Doctoral Training

Going forward, BAS are now preparing to welcome new future leaders in AI environmental science. With its first cohort of students starting in October 2019, Cambridge and BAS will lead the UK in this area with the recently funded UKRI Centre for Doctoral Training (CDT) in the “Application of Artificial Intelligence to the study of Environmental Risks” (AI4ER). Over a four-year research programme, MRes and PhD students will receive high-quality training in research, professional, technical and transferable skills through a focused core programme with an emphasis on development of data science skills through hackathons and team challenges.

Embedded in the outstanding research environments of the University of Cambridge and the British Antarctic Survey (BAS) – and with over 30 industrial partners, including Google DeepMind, Microsoft, Mott MacDonald, and the Environment Agency – the AI4ER CDT will specifically address problems that are relevant to building resilience to environmental hazards and managing environmental change.

For more information on the CDT please check out the website (https://ai4er-cdt.esc.cam.ac.uk), or the Twitter account (https://twitter.com/AI4ER_CDT).