With extreme weather becoming more frequent due to global warming, accurate forecasting is becoming more and more important for all of us, from farmers to city dwellers to businesses around the world. To date, climate models have not been able to accurately predict precipitation intensity, especially extreme precipitation intensity. In nature, precipitation is highly variable and there are many extremes of precipitation, but climate models are biased toward light precipitation and predict smaller variations in precipitation.
The Missing Piece of Current Algorithms: Cloud Organizations
While researchers have worked to develop algorithms to improve prediction accuracy, climate scientists at Columbia Engineering report that traditional climate model parameterizations include very fine-scale cloud structures and Information was missing, a way to describe the organization. It is not captured on the computational grid being used. These tissue measurements influence predictions of both precipitation intensity and its stochasticity, that is, the variability of random fluctuations in precipitation intensity. Until now, there has been no effective and accurate way to measure cloud structure and quantify its impact.
A new study by a team led by Pierre Jeantine, Director of the Artificial Intelligence and Physics (LEAP) Center, uses global storm resolution simulations and machine learning to develop an algorithm that can independently address two different scales of clouds. is created. Organization: what is solved by climate models and what is too small to be solved. This new approach addresses missing information in traditional climate model parameterizations and provides a way to more accurately predict precipitation intensity and variability.
“Our findings are particularly exciting because the scientific community has been debating whether cloud texture should be included in climate models for many years,” said Genting and Maurice Ewing, professors of geophysics in the School of Global Environmental Engineering. said Professor J. Lamar Worzel. He majored in environmental science and is a member of the Data Science Institute. “Our study answers the debate, provides a new solution for including tissue, and shows that the inclusion of this information can significantly improve predictions of precipitation intensity and variability.”
Design neural network algorithms with AI
Sarah Shamekh, a PhD student working with Gentine, has developed a neural network algorithm that learns relevant information about the role of fine-scale cloud texture (unresolved scale) in precipitation. Since Shamekh did not define any metrics or formulas beforehand, she implicitly learns how to measure cloud clustering, an organizational metric, and uses this metric to improve precipitation forecasts. . Shamekh trained an algorithm on high-resolution water fields to encode the degree of small-scale organization.
“We find that our institutional measure accounts for precipitation variability almost completely and can replace stochastic parameterization in climate models,” said the study’s first author. said Shamek. PNAS. “The inclusion of this information has significantly improved precipitation projections at scales relevant to climate models, allowing them to accurately predict extreme precipitation and spatial variability.”
Machine learning algorithms improve future predictions
Researchers are now using machine learning approaches in climate models that implicitly learn subgrid cloud organization metrics. This will greatly improve predictions of precipitation intensity and variability, including extreme precipitation events, and enable scientists to better predict future changes in the hydrological cycle and extreme weather patterns in a warming climate.
How it will be done in the future
The study also opens up new avenues for investigation, such as exploring the possibility that precipitation can create memories, in which the atmosphere retains information about recent weather conditions and influences subsequent atmospheric conditions in the climate system. increase. This new approach has potential for a wide range of applications beyond precipitation modeling, such as better modeling of ice sheets and sea levels.
