Ever-increasing impacts from climate change, including more frequent and intense bouts of extreme weather, are among the greatest challenges faced by mankind.
A changing climate is a daunting prospect. But, says the Met Office’s Theo McCaie, there is one ally standing increasingly firm in our corner: technology; in particular, the benefits from machine learning and artificial intelligence.
We are in the midst of an artificial intelligence revolution where the world’s fastest-growing deep technology has the potential to rewrite the rules of entire industries, fundamentally changing the way we work and live.
Data science advances – including machine learning and artificial intelligence – mean computers can now analyze, and learn from, vast volumes of information at high levels of accuracy and speed, providing exciting new opportunities.
To maximize the advantage of these technological breakthroughs, many scientific disciplines – including weather and climate science and prediction – are revising their operating plans. For example, the Met Office recently published its Data Science Framework outlining how we will “harness the power of data science to push the frontiers of weather and climate science and services”.
There are already promising signs for embracing machine learning in weather and climate. Building on reanalysis data (a fusion of observation data and numerical weather models) several tech companies have produced exciting research indicating similar accuracy to traditional weather forecasting techniques as a fraction of the compute cost at run time.
Theo McCaie – part of the team spearheading the Met Office’s data science capability – credit: Simon Hammett
Embracing change
Professor Kirstine Dale, principal fellow for data science at the Met Office, said, “These results are encouraging and show the benefits that can be gained using machine learning to build upon a rich observation platform, physics-based modeling and data assimilation.
“Critically, these developments pave the way for a hybrid approach bringing together the strengths of both data-driven and physics-based approaches to weather forecasting.
“At the Met Office we are developing systems which will harness the benefits of both physics-based and AI approaches. For example, by using physical models to produce expensive, but high-quality, data that can be used to train fast AI-based emulators,” she said.
A fundamental principle of most AI is that it needs to be trained. The Met Office has rich data sets from a wide spectrum of spatial and time scales which provide a unique training resource. Added to this rich data set is a wealth of knowledge and models based on the physical laws that define how the earth works.
Dale added, “Combining these assets, using cutting-edge AI research in a trustworthy and reliable way is at the core of what we are doing at the Met Office.”
Working with partners
Such a large and important challenge cannot be tackled alone. The Data Science Framework highlights ‘partnership’ as a core pillar to success. One particularly exciting project is fusing machine learning and meteorological expertise from across the Met Office and our partners. Using cloud-scale computing and big data, our experts are leading research into AI-based systems for forecasting UK weather. Progress in this area is also timely. Ever-increasing weather extremes and growing climate change impacts means we can draw together our combined skills and experience to tackle challenges identified in IPCC reports.
Professor Simon Vosper, the Met Office’s director of science, concluded, “Machine learning and artificial intelligence are among the fastest growth areas of science. We are excited to incorporate the valuable benefits of these technologies within our weather forecasts.
“In an era of ever-increasing weather extremes and growing climate change impacts, we believe the most promising developments will come from fusing the benefits of all these technologies rather than simply relying on one or the other.”
To view the original post and read more from the Met Office blog, click here.
