Machine learning could improve extreme weather warnings

In June 2021, the U.S. Pacific Northwest and southwestern Canada experienced a significant heatwave, with temperatures reaching 43°C (109°F) in Portland, Oregon. New weather forecasting approaches based on machine learning can more accurately predict such phenomena and improve lead times. Credit: Contains modified Copernicus Sentinel data (2021) processed by the European Space Agency. CC BY-SA 3.0 IGO
Traditional weather forecasts are only issued about 10 days in advance because small changes in atmospheric or surface conditions can have large and difficult to predict effects on future weather. Longer lead times can make communities more vulnerable to future events, especially the record-breaking heat wave in the U.S. Pacific Northwest in June 2021 that melted railroad power lines, destroyed crops, and killed hundreds of people. We may be able to better prepare for extreme events such as
Meteorologists commonly use adjoint models to determine how sensitive a forecast is to inaccuracies in initial conditions. These models can help determine, for example, how small changes in temperature or water vapor in the atmosphere affect the accuracy of predicting conditions several days in the future.
Understanding the relationship between initial conditions and the amount of error in a prediction allows scientists to make changes until they find the set of initial conditions that produces the most accurate prediction.
However, running the adjoint model requires significant financial and computing resources, and the model can only measure these sensitivities up to 5 days in advance. The researchers tested whether a deep learning approach could provide an easier and more accurate way to determine the optimal set of initial conditions for a 10-day forecast.
The findings will be published in the journal Geophysical Research Letters.
The researchers used two different models, the GraphCast model developed by Google DeepMind and the Pangu-Weather model developed by Huawei Cloud, to create a prediction for the Pacific Northwest heat wave in June 2021.
They compared the results to see if the models behaved similarly, and then compared their predictions to what actually happened during the heat wave. (The dataset used to train the predictive model did not include heatwave data to avoid affecting the results.)
The researchers found that using deep learning techniques to identify optimal initial conditions reduced the GraphCast model’s 10-day forecast error by approximately 94%. This approach produced similar error reductions when used with the Pangu-Weather model. The researchers noted that the new approach improved predictions up to 23 days in advance.
More information: P. Trent Vonich et al., Prediction limits for the 2021 Pacific Northwest heat wave using deep learning sensitivity analysis, Geophysical Research Letters (2024). DOI: 10.1029/2024GL110651
This article is republished courtesy of Eos, sponsored by the American Geophysical Union. Read the original story here.
Citation: Machine learning could improve extreme weather warnings (October 11, 2024) from https://phys.org/news/2024-10-machine-extreme-weather.html October 11, 2024 get to date
This document is subject to copyright. No part may be reproduced without written permission, except in fair dealing for personal study or research purposes. Content is provided for informational purposes only.