New method more accurately predicts extreme weather events
This story was originally published on Columbia Engineering.
As the climate warms and extreme weather events become more frequent, accurate forecasting becomes increasingly important to all of us, from farmers to city dwellers to businesses around the world. To date, climate models have failed to accurately predict precipitation intensity, especially in extreme cases. While in nature, precipitation is highly variable, with many extremes, climate models predict less variability in precipitation, favoring light rain.
Credit: “Heavy Rain Colorado Springs Colorado” go through Broken Taco/Flickr Licensed under CC BY 2.0
The Missing Piece in the Current Algorithm: Cloud Organization
Researchers have been working on developing algorithms that can improve the accuracy of predictions, however, Columbia Engineering Climate scientists report that one piece of information is missing in the parameterization of traditional climate models — a way to describe cloud structure and organization that is so granular that it cannot be captured in the computational grids in use.
These tissue measurements affect predictions of precipitation intensity and its stochasticity (the variability in random fluctuations in precipitation intensity). So far, there has not been an efficient and accurate way to measure cloud structure and quantify its impact.
a new study from freedom Pierre Jeandindirector Understanding Earth with the Center for Artificial Intelligence and Physics (LEAP), using global storm-solving simulations and machine learning to create an algorithm that can separately handle two different sizes of cloud organizations: those solved by climate models, and those that are too small to resolve. This new approach addresses missing information in the parameterization of traditional climate models and provides a way to more accurately predict precipitation intensity and variability.
“Our findings are particularly exciting because for years, the scientific community has debated whether to include cloud organization in climate models,” said LettingMaurice Ewing and J. Lamar Worzel Professor, Department of Geophysics Earth and Environmental Engineering and members of Earth Environmental Sciences and Data Science Institute“Our work provides answers to the debate and a novel solution for including organizations, showing that including this information can significantly improve our predictions of precipitation intensity and variability.”
Designing Neural Network Algorithms Using Artificial Intelligence
Sarah Shamekh, a doctoral student working with Gentine, developed a neural network algorithm that learns about the contribution of fine-scale cloud organization (unresolved scales) to precipitation. Since Shamekh did not define a metric or formula in advance, the model implicitly learned on its own how to measure cloud clustering, an organizational metric, and then used that metric to improve precipitation forecasts. Shamekh trained the algorithm on high-resolution moisture fields, encoding the extent of small-scale organization.
“We found that our organizational metric can almost completely explain precipitation variability and can replace stochastic parameterization in climate models,” said Shamekh, lead author of the study, published May 8, 2023 in Member of the National Academy of Sciences“Including this information could significantly improve precipitation forecasts associated with climate models, accurately predicting precipitation extremes and spatial variability.”
future forecast
The researchers are now using their machine learning approach, which implicitly learns subgrid cloud organization metrics in climate models. This should significantly improve predictions of precipitation intensity and variability, including extreme precipitation events, and allow scientists to better predict future changes in the water cycle and extreme weather patterns in a warming climate.
The study also opens up new avenues of investigation, such as exploring the possibility that precipitation creates memory, the atmosphere retaining information about recent weather conditions that in turn influences later atmospheric conditions in the climate system. This new approach could have wide-ranging applications beyond modeling precipitation, including better modeling of ice sheets and ocean surfaces.
