Deep Learning Weather Prediction and Earth-System Modeling

Prof. Dale Durran

Professor of Atmospheric Sciences

Adjunct Professor of Applied Mathematics

University of Washington

Monday September 12, 2022, 2 PM ET

Abstract:

We examine cases in which a machine learning model trained using re-analysis data with convolutional neural networks (CNNs) learns atmospheric dynamics in a nontraditional framework.

We compare the performance of an ensemble-weather-prediction system based on a global deep-learning weather-prediction (DLWP) model with reanalysis data and forecasts from the European Center for Medium Range Weather Forecasts (ECMWF) ensemble for sub-seasonal weather prediction.

The model is trained on a cubed-sphere grid using a loss function that minimizes forecast error over a single 24-hour period.  The model predicts seven 2D shells of atmospheric data on roughly 150×150 km grids with a quasi-uniform global coverage. Notably, our model can be iterated forward indefinitely to produce forecasts at 6-hour temporal resolution for any lead time.  We present case studies showing the extent to which the model is able to learn “model physics” to forecast two-meter temperature and diagnose precipitation.  Sources of ensemble spread and the performance of the ensemble are discussed relative to the ECMWF S2S ensemble forecasts.

We discuss the avenues along which the current DLWP model can be expanded into an earth-system model.  We conclude by describing the new doors for scientific investigation that would be opened by a reliable, well performing DLWP model.

Biosketch:

Dale Durran is a Professor and past Chair of Department of Atmospheric Sciences at the University of Washington and an Adjunct Professor of Applied Mathematics.  His research spans several areas in atmospheric science including predictability, mountain and mesoscale meteorology, atmospheric waves, and numerical methods for the simulation of atmospheric flows. Most recently he has been exploring how deep learning can change our current paradigm for numerical weather prediction, sub-seasonal, and seasonal forecasting.  He is a fellow of the American Meteorological Society (AMS) and has written over 120 scientific publications, the graduate-level textbook “Numerical methods for Fluid Dynamics with Applications to Geophysics (Springer 2010 2nd ed.)  and “perspective” articles about climate change for the Washington Post. Professor Durran will receive the AMS’s Jule Charney Award in January 2023.

Webinar:

Webinar: https://go.umd.edu/durranwebinar

Event site: https://go.umd.edu/durran

Meeting ID: 929 8858 8552
Webinar password: essic

To join the audio conference only:
US Toll: +13017158592
Global call-in numbers: https://umd.zoom.us/u/aMElEpvNu

For IT assistance:
Cazzy Medley: cazzy@umd.edu

Resources:

Seminar schedule & archive: https://go.umd.edu/essicseminar

Seminar Google calendar: https://go.umd.edu/essicseminarcalendar

Seminar recordings on Youtube: https://www.youtube.com/user/ESSICUMD