Climate change has brought about an increase of extreme weather events around the world in recent years, which requires the development of sophisticated and accurate methods of predicting extreme weather events to mitigate their effects. A team of researchers led by Pierre Gentine from Columbia Engineering have developed a machine-learning algorithm that can better predict extreme weather events. Their research, published by PNAS, uses global storm-resolving simulations and machine learning to capture the vital information of cloud organization and quantify its impact on precipitation intensity and its variability.
Traditional climate models have struggled to predict precipitation intensity with accuracy, particularly for extreme events, due to a lack of understanding of the properties of cloud organization. To achieve this, the research team used a neural network algorithm to learn the role of fine-scale cloud organization, which could not be resolved by the computational grid. This enabled them to separate two different scales of cloud organization – those that are resolution and those that are not – and subsequently improve their prediction of precipitation intensity and variability.
The neural network algorithm was trained on a high-resolution moisture field data and encoded the degree of small-scale organization. As the researchers discovered, this metric explained precipitation variability almost entirely and could even replace a stochastic parameterization in climate models – leading to accurate predictions of precipitation extremes and spatial variability.
The success of this algorithm in predicting extreme weather events will enable scientists to project better future changes in the water cycle and extreme weather patterns. It also opens up possibilities for improved modelling of the ice sheet and ocean surface, just to name a few. The research project has been supported by the Learning the Earth with Artificial Intelligence and Physics (LEAP) Center at Columbia.
Pierre Gentine is the Maurice Ewing and J. Lamar Worzel Professor of Geophysics in Departments of Earth and Environmental Engineering and Earth Environmental Sciences. He is also currently a member of the Data Science Institute at Columbia. Sarah Shamekh, a PhD student working with Gentine, developed the neural network algorithm and is the lead author of the study. She has worked tirelessly to develop this important work that could lead to more accurate and reliable predictions of extreme weather events.
