Format: Hybrid

Abstract: Traditional data assimilation techniques in weather forecasting focus on matching model parameters to observed state trajectories, such as daily temperature series. However, for climate modeling, capturing the long-term distributions of state variables—like the frequency and range of temperatures—is crucial. This talk introduces a new probabilistic framework designed to address this need by emphasizing probability density functions (PDFs) of state variables, using the Lorenz’96 system as a toy model. We have developed a PDF emulator utilizing conditional normalizing flow models, which are adept at transforming simple initial distributions into complex data distributions through a series of invertible mappings. This emulator efficiently replicates the state distributions of climate variables without simulating every possible state trajectory, offering a significant computational advantage. Building on this foundation, our framework includes a sophisticated, distribution-driven approach to model parameter inference. This method aligns the emulated PDFs with observed data distributions and incorporates uncertainty quantification, enhancing model reliability and applicability in various climatic conditions.

The framework’s potential extends to applications such as extreme value estimation and joint quantile analysis, which are pivotal for understanding and predicting climate extremes. By focusing on the tails and joint distributions of state variables, our approach promises improvements over traditional data assimilation methods, presenting a new tool for advanced climate modeling and analysis.

Bio: Shawn Li is currently a Postdoctoral Scientist Fellow at the Data Science Institute of Columbia University, jointly advised by Professors Pierre Gentine, Tian Zheng, and Upmanu Lall. His research focuses on developing process-based mathematical/physical models and data-driven machine-learning techniques to study the earth and environmental processes.

Time: 12:00 pm EST

Free!

Registration