Using machine learning to help reconstruct changes in extreme rainfall over British & Irish Isles

Flooding often occurs due to an intense rainfall event, with severe impacts on human society and biodiversity, and climate change is increasing the intensity of such extreme rainfall. As well as mitigating climate change by reducing greenhouse gas emissions, there is a need to adapt to climate change by ensuring society becomes more resilient to changing weather conditions. Understanding how much the climate is changing is crucial information in making decisions about, e.g. building new reservoirs, or planning flood defences.

To better understand climate trends and their consequences, and to create appropriate policy responses, we need to build longer and more detailed reconstructions of how the weather has varied in the past, including extreme rainfall. However, detailed UK reconstructions of daily rainfall are currently only considered very reliable since around 1960. The main aim of this project is to extend backwards this reliable period to around 1870, giving an additional 90 years of detailed information about trends and variations in daily extreme rainfall. This period includes 1872 – known to be the wettest year on record for the UK – and October 1903 – known to be the wettest month on record for the UK.

Vast quantities of rainfall (and other weather) observations have been made over decades to centuries in the UK & Ireland, but the vast majority are still only available on paper in their original hand-written or published form. As one example, the UK National Meteorological Archive has recently scanned 350,000 sheets of paper which contain approximately 100 million individual daily rainfall observations taken across Great Britain since the 1860s. Virtually none of these data are available for scientists to use or analyse as they still need to be transcribed from the images into digital data.

This project will use AI/ML methods such as Optical Character Recognition (OCR) and Large Language Models (LLMs) to transform images of historical weather observations into invaluable data. There are numerous sources of weather observations to consider, some hand-written and some typeset. A high priority is digitising the 350,000 images of rainfall observations, which would enable a detailed understanding of how extreme rainfall is changing and inform policy decisions about how to adapt to this changing climate. Other options are also possible, depending on the student’s interests and ongoing progress with developing ML tools.

A partnership with the Met Office will allow any rescued observations to be used to update the official UK climate reconstructions. Similar efforts to recover historical daily weather observations are being undertaken in other countries such as Ireland and there are collaborative opportunities to link approaches and datasets across jurisdictions with collaboration with different national meteorological services.

Eligibility requirements:

Please note: This funding is only available to UK/Republic of Ireland candidates

Applicants should have a good bachelor’s degree (minimum of a UK Upper Second (2:1) or equivalent)/master’s degree in physical or mathematical sciences or a strongly related discipline.

How to Apply:

Please go to the Apply for a Programme website and create your account, and use the link sent by email to start the application process. During the application process please select the PhD in Atmosphere, Oceans and Climate programme.  Please quote the reference ‘DRC25-014’ in the ‘Scholarships applied for’ box which appears within the Funding Section of your on-line application.