Uncovering Processes Responsible for Winter Storm Track Anomalies over North America Using a Causal Inference Framework

Since storm tracks produce most of the severe weather over the mid-latitudes in North America, especially during the cold seasons, reliable dynamical predictions of severe weather can only be achieved by accurate storm track simulations. While climate models have improved over generations, biases in storm track simulations still exist, indicating the need for identifying bias-related processes to further refine the models.

To identify these processes, we systematically examine the winter weekly interactions between the storm tracks and large-scale circulation variability over North America and its neighboring oceans by introducing a novel causal inference framework. This framework aggregates the causal discovery results from bootstrapped samples, enhancing high-dimensional variable selection and providing uncertainty estimates for causal effects.

Our causal inference framework suggests potential pathways influencing the winter storm tracks over North America. One pathway diagnosed in the ERA5 reanalysis indicates that the storm track activity over the eastern and central North America is influenced by the atmospheric circulation over the North Pacific. This effect is mediated by a mid-tropospheric wave train across North America. This wave train changes the vertical wind shear and thus the baroclinicity over the West Coast, affecting the downstream synoptic wave propagation. Another pathway originates from the North Atlantic and influences storm track activity over much of the eastern North America via the lower-tropospheric wind over the East Coast.

These pathways revealed in the ERA5 reanalysis are then compared with their counterparts in the CESM2 and GFDL SPEAR large ensembles. While examination of the discrepancies in interactions provides quantitative measures of model departures from the reanalysis, our framework produces the uncertainty estimates to quantify the robustness of these discrepancies. In addition, investigating the large ensembles provides information on internal variability to determine whether the discrepancies in the interactions are likely to arise from genuine model biases or internal variability.

By evaluating the discrepancies between the reanalysis and models in the pathways influencing the North American storm tracks, our study will help inform further model development and improve severe weather predictions. Additionally, the bootstrapping and mapping methods of our framework, which can be integrated into other causal inference methods, have the potential for extensive research applications.