University of Sussex spin-out uses machine learning for flood risk simulations
- Published: Wednesday, 06 May 2020 09:58
- Written by University of Sussex
Success story by the University of Sussex
Ambiental Risk Analytics, a Brighton-based company specialising in flood modelling and flood risk, has teamed up with DataJavelin, a newly established University of Sussex spin-out company, to automate the calibration of flood risk simulations through machine learning and Artificial Intelligence techniques. The Research and Development effort could transform flood forecasting and risk modelling, providing more accurate and cost effective flood predictions, in real time.
Natural disasters across the world led to economic damage costing an estimated $225 billion in 2018 alone. Future flood mapping scenarios recently built by Ambiental for utility companies to see how climate change will affect the UK highlights the scale of the risk. “It came up with some quite staggering numbers – a doubling of the number of properties at risk of flooding in the UK by 2080 under high emissions scenarios,” said Ambiental’s CEO, Dr Justin Butler.
Ambiental’s work has expanded to meet insurers, mortgage lenders, conveyancers, utility companies and governments’ need for flood risk models and digital maps to identify and price risks to land and property from river, sea, surface and ground water flooding. The company’s work now extends far beyond the UK to countries such as Australia, Nigeria and Malaysia, where they are also carrying out real time flood forecasting. This international work is set to expand as Ambiental, which is based at the Sussex Innovation Centre, was recently acquired by Royal HaskoningDHV, a global engineering, design and project management company with 6000 employees worldwide.
Traditionally, flood forecasting systems are set up manually by hydrologists on a catchment by catchment basis. This makes building the systems time consuming, and scaling up simulations is currently not practical. Having seen the potential of Artificial Intelligence for its business, Ambiental has teamed up with DataJavelin to explore how to automate the calibration of flood risk simulations through machine learning and Artificial Intelligence techniques, through Innovate UK funding. With these new AI approaches, Ambiental and DataJavelin aim to be able to create consistent simulations which could run on the scale of continents.
So what is Artificial Intelligence (AI), which nowadays usually refers to a subset of techniques called machine learning? “It's a huge field in which innovation is happening at a tremendous pace,” said Dr Philip Rooney, the CEO of DataJavelin. “At its most basic description, it’s about getting computers to carry out tasks that would have ordinarily required a human to have done them.” You have probably interacted with AI if you have used Amazon’s Alexa to, say, order a cab. While the term was coined in 1956, it needed computing power and fast processing speeds to really take off, which it has done in recent years.
Artificial Intelligence will fuel economic growth
AI could double annual economic growth rates in the next 15 years, increase labour productivity and help people make more efficient use of their time, according to recent research by Accenture on the impact of AI on leading world economies. Set up in 2018, DataJavelin is hoping to build on this momentum using their unique strengths. Both Rooney and Hurley, the Chief technology officer of Data Javelin, completed their PhDs in Astrophysics and carried out postdoctoral research at the University of Sussex where they honed their problem solving skills and methods for analysing large datasets. “DataJavelin is a company where all of its value is within the people working for it and the skills that they have, and those skills were developed here at the University of Sussex,” said Rooney.
“DataJavelin is a company where all of its value is within the people working for it and the skills that they have, and those skills were developed here at the University of Sussex.”
The process of developing machine learning for predicting flood risk is complex as flood modelling involves dynamic processes, such as predicting the total water flow or the height of the river, depending on varying levels of rainfall. “Companies are the experts in their field. We’ll come up with recommendations on what algorithms and methods exist to gain the insights they need to solve business problems,” said Rooney. In this case, DataJavelin is using Ambiental’s real world data to create simulations of, say, the amount of water flowing through a river. They then take a measurement of how their simulation matches the real world data, doing this many times until they have simulations that accurately reflect reality. They’re using open source Python-based tools, called Emukit, developed by Amazon data scientists in the last few years. “It’s cutting edge. That’s why it’s nice to work with companies like Ambiental, who are willing to test things out and take advantage of the latest tools.”
Rooney is quick to add that machine learning tools for flood modelling aren’t meant to replace hydrologists. “With a lot of AI, people are worried that it’s just computers coming to take the jobs of people. And there are areas where that is a concern.” But Rooney believes the new machine learning tools will free up the hydrologists to do more complex, nuanced work. “It provides experts with the ability to carry out experimentation that they just couldn’t do before,” said Rooney. “The idea is to give them a powerful tool so they can do their job even better.”
The collaboration should result in new technology to help Ambiental scale up its business which could help insurers, industry, government agencies and emergency responders have access to more accurate and cost effective flood predictions, in real time. “We’re hoping this technology can be brought to market quickly and boost our revenues and profitability over the next two to three years,” said Butler.
By Suzanne Fisher-Murray, Research Communications Manager, University of Sussex
Published: 6 May 2020