Advancing Weather and Climate Prediction with Data Driven Methods
Will Chapman ()
Colloquium
Tuesday, February 20, 2024, 3:30 pm
Abstract
The increasing frequency of climate and weather-related disasters globally underscores the urgent need for accurate and timely meteorological and climatic prediction. While traditional forecasting methods have laid a solid foundation, the advent of innovative, data-driven approaches marks a significant rapid divergence in atmospheric sciences. Among these advancements, Machine Learning (ML) emerges as a particularly transformative technology, helping to redefine the capabilities of weather and climate prediction. Additionally, this new attention to machine learning in the atmospheric science community offers a way to strengthen convergent actions in the computer and atmospheric science fields, leading to novel frameworks and theoretical advancement in both fields. This presentation delves into ML's growing role in atmospheric sciences, highlighting its pivotal contributions in four broad themes. Firstly, model post-processing, where ML improves the utility and accuracy of weather model outputs, making forecasts more reliable and user-friendly. Secondly, it fosters scientific discovery, which will be highlighted in an example of enhancing our understanding of key atmospheric teleconnections. Thirdly, hybrid modeling bridges the gap between empirical knowledge and computational efficiency by integrating ML's predictive power with the detailed physical representations of traditional climate models during online model runs, capturing complex climate dynamics more effectively. Lastly, ML facilitates model emulation for full model replacement, offering a computationally efficient alternative that maintains, or in some instances exceeds, the predictive accuracy of comprehensive climate models. By enhancing the precision and reliability of weather and climate forecasts, ML not only complements but, in certain cases, surpasses the accuracy of traditional methods, underscoring its significant impact and critical role in advancing atmospheric science through improved forecasting and scientific discovery.
Bio
Will Chapman is a project scientist with the Climate and Global Dynamics Group at the National Science Foundation's National Center for Atmospheric Research (NSF-NCAR). Before assuming his current role, he was a postdoctoral fellow in NCAR's Advanced Studies Program. Will is an alumnus of the Scripps Institution of Oceanography, where he earned his Ph.D., Stanford University, where he completed his M.S., and the University of California San Diego, where he obtained a B.S. in Engineering. His current research focus is hybrid machine learning (combining physics based and machine learning methods in online runs) within climate models and advancing data-driven numerical weather prediction. Committed to the democratization of science, Will actively develops open and user focused software to make ML climate and weather prediction technologies accessible to the wider atmospheric science community.