Deep learning sheds light on past and future atmospheric blocking events

Atmospheric blocking events are persistent, high-impact weather conditions that occur when large-scale high-pressure systems become stationary and divert jet stream and storm tracks for days or even weeks, and can be associated with to floods or record heat waves, such as in Europe in 2023.

In a recent study published in Earth and Environment CommunicationsChristina Karamperidou, an atmospheric scientist at the University of Hawaii at Mānoa, used a deep learning model to infer the frequency of blocking events over the past 1,000 years and shed light on future impact of climate change on these important phenomena.

“This study aimed to extract a paleoweather signal from paleoclimate records using a deep learning model that infers the atmospheric blocking frequency from surface temperature,” Karamperidou said. “This is a unique study and the first attempt to reconstruct a long record of blocking frequencies based on their relationship to surface temperature, which is complex and unknown. Machine learning methods can be very powerful for such tasks.”

Train the deep learning model

Karamperidou developed a specialized deep learning model, which she trained using historical data and large sets of climate model simulations. The model was then able to infer the frequency of blocking events from anomalies in seasonal temperature reconstructions over the last millennium. These reconstructions of past temperatures are relatively well constrained by large networks of temperature-sensitive tree ring records during the growing season.

“This approach demonstrates that deep learning models are powerful tools for overcoming the long-standing problem of paleoclimate extraction,” Karamperidou said. “This approach can also be used for the instrumental period of climate history, which began in the 18th century, when routine weather measurements were made, since we only have reliable data to identify blockages from the years 1940, or perhaps only since the satellite era (after the satellite era 1979).”

Frequency of future blocking events

There is no scientific consensus yet on how climate change will change the frequency of blocking events. These strong and persistent mid-latitude high pressure systems can have significant impacts in Hawaii, where flooding has accompanied persistent North Pacific blocks, and also around the world, for example in the Pacific Northwest and Europe, where Summer lockdowns can lead to extreme heat. waves.

It is therefore very important for Hawaii to understand changes in the frequency of these events, especially as they relate to other major climate players, such as El Niño and long-term trends in sea surface temperatures in the Tropical Pacific. This study allowed Karamperidou to relate the blocking frequencies in the mid and high latitudes to the tropical climate variability of the Pacific in the long context of the last millennium, which is essential for the validation of the climate model and to reduce uncertainties in the blocking future climate projections.

Open research and transparency

Karamperidou worked with two UH Mānoa students to create a unique web interface for exploring the deep learning model and the resulting reconstructions. She emphasized that sharing results and methods in this way is important for best practices and open research transparency, especially as the application of machine learning and artificial intelligence grows rapidly in many aspects of daily life.

In the future, Karamperidou plans to explore a range of features and architectural enhancements to the deep learning model to expand its applications to climate phenomena and variables directly linked to high socio-economic impacts.