Extracting Paleoweather from Paleoclimate: Using Deep Learning to Unravel the Role of the Tropical Pacific in Atmospheric Blocking over the Last Millennium (Invited Presentation)

Projected changes in atmospheric blocking and associated extreme weather events are marked by significant uncertainties due to climate model biases, the complex nature of the phenomenon, and its strong natural variability. Paleoclimate proxy records, suggesting shifts in blocking activity, extend observational records and may help reduce these uncertainties. However, extracting synoptic-scale signals from low-temporal-resolution paleoclimate records is challenging. In this talk I will present a Deep Learning (DL) model that infers summertime blocking frequencies from tree-ring-based gridded reconstructions of Northern Hemisphere surface temperatures over the Last Millennium. The model is shown to be constrained implicitly by the paleoclimate proxies, even without directly incorporating them or having any knowledge of their specific locations. The DL blocking reconstructions highlight the significant influence of the tropical Pacific in driving strong blocking variability from interannual to centennial scales. A weakened tropical Pacific zonal temperature gradient during the Little Ice Age correlates with a hemispherically reduced -yet more variable at interannual scales- blocking frequency and altered regional patterns. Interannual variability in N. Hemisphere blocking is enhanced post-1600, coinciding with both the reduction in tropical Pacific zonal gradient and an apparent increase in NINO3.4 variance. Further, I will show that changes in ENSO diversity, and not just NINO3.4 variance, are important to consider in order to explain multiscale variability in N. Hemisphere blocking, as Eastern vs. Central Pacific ENSO events can have opposite-sign impacts on the main centers of blocking activity. These results underscore that any model biases or erroneous projections in tropical Pacific mean state and variability change can be carried over to their projections of blocking with implications for impact projections.

This study uses paleoclimate evidence to support an affirmative answer to a pivotal question that has been central to Mark Cane's career: "Do the Tropics Rule?" (Cane & Evans, 2000; Science). It is also the latest in a series of studies originating from work during my PhD where I used the Lorenz-84 model forced by a heuristic version of the Cane-Zebiak model to explore the impact of ENSO on extratropical atmospheric dynamics. This "toy-model" led to my first encounter with Mark Cane as a PhD student, marking the beginning of his profound influence on my career and life through his mentorship and support.