Atmospheric blocking events are long-lasting and impactful weather occurrences that take place when major high-pressure systems remain in one place, affecting the jet stream and storm paths for extended periods, sometimes lasting from days to weeks. Such events can lead to severe flooding or heat waves, as seen in Europe in 2023. In a recent study, atmospheric scientist Christina Karamperidou from the University of Hawai’i at Manoa utilized a deep learning model to analyze the frequency of blocking events over the past millennium and to understand the potential effects of future climate change on these critical weather patterns.
“This study aimed to extract a paleoweather signal from paleoclimate records by using a deep learning model to determine how often atmospheric blocking occurs based on surface temperature,” stated Karamperidou. “This research is unique, representing the first effort to recreate a long-term record of blocking frequencies by examining their complex relationship with surface temperature. Machine learning techniques can be very effective for this type of analysis.”
Training the deep learning model
Karamperidou created a specialized deep learning model that she trained using historical data along with extensive climate model simulation ensembles. The model can infer how frequently blocking events happened by analyzing seasonal temperature variations from the Last Millennium. These historical temperature reconstructions have been well-documented, supported by many tree-ring records that respond to temperature changes during the growing season.
“This method illustrates the effectiveness of deep learning models in addressing the long-standing challenge of extracting paleoweather information from paleoclimate data,” Karamperidou explained. “This approach is also applicable to the instrumental climate history starting from the 18th century, when regular weather observations began, given we have reliable blocking data only since the 1940s and even more accurate measurements from the satellite era starting in 1979.”
Future frequency of blocking events
Currently, scientists do not have a unified opinion on how climate change might affect the frequency of blocking events. These strong, long-lasting high-pressure systems can have substantial consequences for places like Hawai’i, where severe flooding has been linked to persistent North Pacific blocks, as well as around the globe, such as in the Pacific Northwest and Europe, where summer blocking can lead to intense heat waves.
Understanding how the frequency of these events is changing, particularly in relation to significant climate factors like El Niño and long-term shifts in sea surface temperatures in the tropical Pacific, is crucial for Hawai’i. This study enabled Karamperidou to connect blocking frequencies in mid- and high-latitude regions to climate variability in the tropical Pacific over the last millennium, which is vital for validating climate models and reducing uncertainty in future projections regarding blocking patterns.
Open research and transparency
Karamperidou collaborated with two students from UH Manoa to develop a dedicated web interface for exploring the deep learning model and the reconstructions it produced. She emphasized the significance of sharing research results and methodologies in this way, as it promotes best practices for Open Research and transparency, especially as the use of machine learning and artificial intelligence rapidly expands into numerous areas of everyday life. This web interface is hosted on Jetstream-2, an NSF-supported cloud computing platform that collaborates with regional partners, including the University of Hawai’i Information Technology Services – Cyberinfrastructure and the Hawai’i Data Science Institute.
Looking ahead, Karamperidou intends to investigate different features and architectural improvements for the deep learning model to broaden its application for climate phenomena and variables that directly correlate with significant socioeconomic impacts.
