University of Hertfordshire and Goonhilly hail deep learning-led UK mapping data breakthrough

The University of Hertfordshire says its technology tie-up with the Goonhilly Earth Station in Cornwall will open up a raft of business opportunities for retailers, insurance firms and the agricultural industry by producing better-quality mapping data about the UK.

The university claims that tapping into Goonhilly’s artificial intelligence (AI) and deep learning platforms has enabled it to create a service that will provide access to satellite mapping data of the UK that is free from cloud cover for the first time.

It relies on the use of satellite radar imaging, which can pass through clouds, to produce detailed images of the Earth’s surface regularly.

The work is being led by astrophysicist professor James Geach and his PhD student Mike Smith, whose team have used Goonhilly’s NVIDIA GPU supercomputer to train its ClearSky algorithm on a huge set of satellite images of the Earth’s surface.

“Goonhilly’s deep learning platform has allowed us to massively accelerate time to market,” said Geach.

“The platform’s phenomenal processing speed has made it possible for us to significantly scale up our models and increase the scope of our analysis. It is rewarding to see how techniques developed for astrophysics can be applied to Earth observation data to deliver real-world impact.”

Geach and his team already have a number of potential use cases in mind for the data, including using it to monitor coastal erosion, track the impact of climate change on crop growth patterns, and predict where flooding and wildfires might occur. These, in turn, could open up business opportunities for retailers, insurance firms, commodity traders and supermarkets, they said. 

They have also forged ties with agricultural technology company Agrimetrics to provide capabilities for the firm to monitor the health and growth rate of 2.8 million fields across the UK on a weekly basis.

The team is looking to spin out its own commercial venture, known as DeepEO, later this year on the back of this work by creating a continuously updated database of Earth observation data that organisations can use to inform their decision-making around extreme weather events.

As previously reported by Computer Weekly, the team at Goonhilly has set its sights on becoming a major incubation hub for UK enterprises and academic organisations looking to create machine learning, deep learning and AI-based services.

The site officially went live in July 2019, and is already home to a renewably-powered datacentre, which is where a lot of the computational work carried out by organisations such as the University of Hertfordshire is carried out.

As an extension to this, the Goonhilly team is now turning its attention to building a testbed of platforms that will allow the site’s users to tap into edge computing capabilities, and conduct near real-time analysis of live satellite data streams.

Chris Roberts, head of datacentre and cloud at Goonhilly, said it is hoped this work will lead to more research organisations taking steps to create commercial ventures, as the University of Hertfordshire team plans to do.

“Our wraparound service nurtures startups like DeepEO with the resources they need on their journey from a research project to commercial growth and profitability,” said Roberts.

“And our datacentre’s green credentials are an increasingly important factor, ensuring that valuable environmental efforts to mitigate the effects of climate change are not themselves creating volumes of CO₂.”