New method predicts extreme weather events more accurately
This story was originally posted by Columbia Engineering.
With the increase in extreme weather events, which are becoming more frequent in our warming climate, accurate forecasts are becoming more and more essential for all of us, from farmers to city dwellers to businesses around the world. To date, climate models have failed to accurately predict rainfall intensity, especially extremes. While in nature precipitation can be quite varied, with many extremes of precipitation, climate models predict a smaller variance in precipitation with a bias towards light rain.
The missing piece in current algorithms: the organization of the cloud
Researchers have been working to develop algorithms that will improve the accuracy of predictions but, as Columbia Engineering report climatologists there is information missing from traditional climate model parameterizations – a way of describing the structure and organization of clouds that is at such a fine scale that it is not captured on the computing grid used.
These organizational measures affect predictions of precipitation intensity and stochasticity – the variability of random fluctuations in precipitation intensity. Until now, there was no efficient and accurate way to measure cloud structure and quantify its impact.
A new study of a team led by Pierre Gentindirector of the Earth Learning Center with Artificial Intelligence and Physics (LEAP), used global storm resolution simulations and machine learning to create an algorithm capable of separately processing two different scales of cloud organization: those resolved by a climate model and those that cannot be resolved because they are too small. This new approach addresses the missing piece of information in the parameterizations of traditional climate models and provides a way to more accurately predict precipitation intensity and variability.
“Our findings are particularly exciting because for many years the scientific community has debated whether to include cloud organization in climate models,” said Gentine, Maurice Ewing and J. Lamar Worzel, professor of geophysics at the departments of earth and environmental engineering and environmental sciences of the Earth and member of the Data Science Institute. “Our work provides an answer to the debate and a new solution for the inclusion of the organization, showing that the inclusion of this information can significantly improve our prediction of rainfall intensity and variability.”
Using AI to design a neural network algorithm
Sarah Shamekh, a PhD student working with Gentine, has developed a neural network algorithm that learns relevant information about the role of fine-scale (unresolved scales) cloud organization on precipitation. Because Shamekh didn’t define a metric or formula up front, the model implicitly learns – on its own – how to measure cloud clustering, an organization metric, and then uses that metric to improve prediction of precipitation. Shamekh trained the algorithm on a high-resolution moisture field, encoding the degree of fine-scale organization.
“We found that our organization metric almost entirely explains rainfall variability and could replace a stochastic parameterization in climate models,” said Shamekh, lead author of the study, published May 8, 2023, by PNAS. “The inclusion of this information has significantly improved precipitation forecasting at the scale relevant to climate models, accurately predicting extreme precipitation and spatial variability.”
The researchers are now using their machine learning approach, which implicitly learns the organization metric of the subgrid cloud, in climate models. This should significantly improve the prediction of precipitation intensity and variability, including extreme precipitation events, and enable scientists to better predict future water cycle changes and extreme weather patterns in a climate that heats.
This research also opens up new avenues of investigation, such as exploring the possibility that precipitation creates memory, where the atmosphere retains information about recent weather patterns, which in turn influences later atmospheric patterns, in the climate system. This new approach could have many applications beyond simple precipitation modeling, including better modeling of the ice sheet and ocean surface.