ClimateAi researchers realize artificial intelligence


SAN FRANCISCO, March 24, 2022 (GLOBE NEWSWIRE) — ClimateAi, a pioneer in applying artificial intelligence to climate risk modelling, today announced that its team has solved a critical challenge in weather forecasting. Leveraging advances in AI to improve weather and climate predictions, ClimateAi researcher Dr Stephan Rasp and Price from the University of Oxford have created an innovative machine learning approach using generative adversarial networks (GANs) trained on global weather forecasts to correct for biases that exist in current weather models. After extensive peer review, the report “Increasing the Accuracy and Resolution of Precipitation Forecasts Using Deep Generative Models” will be presented at 25th International Conference on Artificial Intelligence and Statistics (AISTATS) later this month.

“Going beyond the recent flurry of activity to improve nowcasts, our new model achieves the same accuracy for horizon forecasts from several hours to several days,” noted Dr. data for ClimateAi. “By taking a two-pronged approach, we can account for systematic errors in global models and increase forecast resolution so that regional extremes are accurately captured.”

While a warmer and wetter world due to climate change makes extreme weather events more frequent and intense, it is notoriously difficult to develop accurate regional forecasts due to the complex physics behind heavy rainfall. and extreme weather events. Global forecasts can take advantage of the availability of a large number of weather data and models, but lack precision and are prone to error, as any small detail not taken into account can lead to large-scale divergence. Regional forecasts, on the other hand, require expensive and time-consuming supercomputers with trained local practitioners, which limits access to rich countries.

The new model scales down the global forecast to be as accurate as a local forecast, without requiring the vast amounts of computing, financial and human resources previously required for such a small scale. Offering accurate local forecasts for precipitation and extreme weather without the traditional (and expensive) constraints of current forecasting systems, these results could provide a new paradigm for forecasting extremes in low-income countries that cannot afford high-resolution local forecast technology.

By training GANs – a subset of machine learning in which two neural networks fight and train each other until they reach a conclusion – to first look at the coarse global weather forecast and fix the errors, then to reduce the forecasts to a high resolution for the local/regional scale, the researchers produced accurate local forecasts (beyond the immediate window of other recent breakthroughs) with the same high resolution and the same quality than regional forecasts from expensive supercomputers. GANs generate several different images (or potential realizations) that show the different potential scenarios, on time scales of several days, all with equal probability.

For example, rather than simply confirming a “40% chance of rain this week” for an entire region, the new model would allow users to easily answer more useful questions such as: How likely is it to rain or that it won’t rain tomorrow? Where exactly is it going to rain? If it rains, will it drizzle everywhere, pour in one place, or pour in several places but drizzle in others? While coarse, large-scale predictions hide all that important information, this new machine learning approach effectively brings it to light.

“Current global and regional forecasts lack precision and are prone to error,” Dr Rasp added. “Advances in artificial intelligence and machine learning are changing weather forecasting, and resource-intensive regional weather models may soon be completely replaced by machine learning approaches. Actionable predictions will help businesses and governments looking to protect their initiatives and operations from climate change. »

Based on this research, low-income countries – often also those most affected by the impacts of climate change – may soon have access to accurate, high-resolution forecasts that will provide a climate adaptation tool for agriculture. , infrastructure planning and more. The ClimateAi researchers also expect this method to work for longer-term forecasts (weeks, months, years, decades), where the need for increased resolution is even greater.

The authors will present the full results on March 28 at the 25th International Conference on Artificial Intelligence and Statistics (AISTATS). Read the arXiv pre-print server report now.

About ClimateAi
ClimateAi helps businesses better manage climate risk with actionable intelligence to design and protect global supply chains. Leveraging proprietary AI modeling, ClimateAi provides unparalleled weather and climate forecasting and business impact conversion that sets a new industry standard for accuracy, precision and scale. Its first-of-its-kind enterprise climate planning platform quantifies climate impacts, provides company-specific insights, reduces risk and creates new value opportunities for its customers. Based in San Francisco, ClimateAi is committed to working with industry leaders to standardize climate risk assessments across supply chains to help businesses adapt and bring climate resilience to our economy. world. Learn more about and follow ClimateAi on Twitter and LinkedIn.

Contact media relations: [email protected]


About Author

Comments are closed.