Statistical Emulator of Atmospheric Circulation over Extratropical North America using Hidden Markov Models

Abstract:

The identification of recurring, quasi-stationary circulation regimes in the extratropical atmosphere remains a significant challenge, despite substantial research over the past several decades. In particular, it is unclear which method is most suitable to capture the non-linearity and persistence of large-scale regime structures that operate on intra-seasonal timescales, and how these regimes vary across the year. This work focuses on one class of models - Hidden Markov Models (HMMs) – and compares four nested structures to examine how model assumptions influence regime characteristics. The four models, namely (1) HMM, (2) seasonal Non-homogeneous HMM (s-NHMM), (3) Markov Switching Autoregressive model (MSAR), and (4) seasonal Non-homogeneous MSAR (s-NHMSAR), are evaluated in terms of their ability to simulate 38 years of 500 hPa geopotential height (z500) anomaly fields over extratropical North America (30⁰ – 60⁰ N ; 0⁰ – 180⁰ W). Simulation skill is assessed based on the reproduction of a variety of z500 attributes, including marginal distributions, seasonality, and persistence in average and extreme states, as well as the simulated wavelet spectrum of major circulation indices such as the North Atlantic Oscillation (NAO) and Pacific/North American (PNA) teleconnection patterns. The nested structure of the experiment enables a careful assessment of the utility of specific model features, including (1) periodic variability in the latent Markov model and (2) the autoregressive nature of z500 anomalies within regimes. Based on our findings, we comment on the trade-off between model complexity and the accuracy of atmospheric flow simulations. This study advances the development of a statistical emulator of extratropical atmospheric circulation, which can serve as a basis for efficient, stochastic approaches to link weather and climate with application in local weather forecasting, scenario development, and global climate model diagnostics.