Informal Seminar: Disentangling regional impacts of teleconnections using causal representation learning

Large-scale modes of variability in the climate system and their teleconnections are key sources of predictability at subseasonal to seasonal timescales. However, disentangling the joint and causal influences of multiple teleconnections on surface-level extremes remains a challenge.  In this work, we introduce a deep learning-based approach that jointly learns low-dimensional and physically interpretable representations of climate processes and the causal relationships among interacting teleconnections. We apply this approach to identify predictable and informative dynamical drivers of extreme precipitation over Morocco in the form of targeted circulation regimes. Furthermore, we demonstrate how the method can be extended to analyse multiple interacting teleconnections influencing the short rains over the Greater Horn of Africa at seasonal timescales.