Search for: All records

Abstract The quasi-biennial oscillation (QBO) is the dominant mode of variability in the equatorial stratosphere. It is characterized by alternating descending easterly and westerly jets over a period of approximately 28 months. It has long been known that the QBO interactions with the annual cycle, e.g., through variation in tropical upwelling, lead to variations in the descent rate of the jets and, resultantly, the QBO period. Understanding these interactions, however, has been hindered by the fact that conventional measures of the QBO convolve these interactions. Koopman formalism, derived from dynamical systems, allows one to decompose spatiotemporal datasets (or nonlinear systems) into spatial modes that evolve coherently with distinct frequencies. We use a data-driven approximation of the Koopman operator on zonal-mean zonal wind to find modes that correspond to the annual cycle, the QBO, and the nonlinear interactions between the two. From these modes, we establish a data-driven index for a “pure” QBO that is independent of the annual cycle and investigate how the annual cycle modulates the QBO. We begin with what is already known, quantifying the Holton–Tan effect, a nonlinear interaction between the QBO and the annual cycle of the polar stratospheric vortex. We then use the pure QBO to do something new, quantifying how the annual cycle changes the descent rate of the QBO, revealing annual variations with amplitudes comparable to the 30 m day⁻¹mean descent rate. We compare these results to the annual variation in tropical upwelling and interpret them with a simple model. Significance StatementThe quasi-biennial oscillation (QBO) is a periodic cycle of winds in tropical atmosphere with a period of 28 months. The phase of QBO is known to influence other aspects of the atmosphere, including the polar vortex, but the magnitude of its effects and how it behaves are known to depend on the season. In this study, we use a data-driven method (called a Koopman decomposition) to quantify annual changes in the QBO and investigate their causes. We show that seasonal variations in the stratospheric upwelling play an important but incomplete role. 

Abstract We train random and boosted forests, two machine learning architectures based on regression trees, to emulate a physics‐based parameterization of atmospheric gravity wave momentum transport. We compare the forests to a neural network benchmark, evaluating both offline errors and online performance when coupled to an atmospheric model under the present day climate and in 800 and 1,200 ppm CO₂global warming scenarios. Offline, the boosted forest exhibits similar skill to the neural network, while the random forest scores significantly lower. Both forest models couple stably to the atmospheric model, and control climate integrations with the boosted forest exhibit lower biases than those with the neural network. Integrations with all three data‐driven emulators successfully capture the Quasi‐Biennial Oscillation (QBO) and sudden stratospheric warmings, key modes of stratospheric variability, with the boosted forest more accurate than the random forest in replicating their statistics across our range of carbon dioxide perturbations. The boosted forest and neural network capture the sign of the QBO period response to increased CO₂, though both struggle with the magnitude of this response under the more extreme 1,200 ppm scenario. To investigate the connection between performance in the control climate and the ability to generalize, we use techniques from interpretable machine learning to understand how the data‐driven methods use physical information. We leverage this understanding to develop a retraining procedure that improves the coupled performance of the boosted forest in the control climate and under the 800 ppm CO₂scenario. 

Abstract Blocking events are an important cause of extreme weather, especially long‐lasting blocking events that trap weather systems in place. The duration of blocking events is, however, underestimated in climate models. Explainable Artificial Intelligence are a class of data analysis methods that can help identify physical causes of prolonged blocking events and diagnose model deficiencies. We demonstrate this approach on an idealized quasigeostrophic (QG) model developed by Marshall and Molteni (1993),https://doi.org/10.1175/1520‐0469(1993)050<1792:taduop>2.0.co;2. We train a convolutional neural network (CNN), and subsequently, build a sparse predictive model for the persistence of Atlantic blocking, conditioned on an initial high‐pressure anomaly. Shapley Additive ExPlanation (SHAP) analysis reveals that high‐pressure anomalies in the American Southeast and North Atlantic, separated by a trough over Atlantic Canada, contribute significantly to prediction of sustained blocking events in the Atlantic region. This agrees with previous work that identified precursors in the same regions via wave train analysis. When we apply the same CNN to blockings in the ERA5 atmospheric reanalysis, there is insufficient data to accurately predict persistent blocks. We partially overcome this limitation by pre‐training the CNN on the plentiful data of the Marshall‐Molteni model, and then using Transfer learning (TL) to achieve better predictions than direct training. SHAP analysis before and after TL allows a comparison between the predictive features in the reanalysis and the QG model, quantifying dynamical biases in the idealized model. This work demonstrates the potential for machine learning methods to extract meaningful precursors of extreme weather events and achieve better prediction using limited observational data. 

Abstract Atmospheric blocking is characterized by persistent anticyclones that “block” the midlatitude jet stream, causing temperature and precipitation extremes. The traffic jam theory posits that blocking events occur when the Local Wave Activity flux, a measure of storm activity, exceeds the carrying capacity of the jet stream, leading to a pile up. The theory's efficacy for prediction is tested with atmospheric reanalysis by defining “exceedance events”, the time and location where wave activity first exceeds flow capacity. The theory captures the Northern Hemisphere winter blocking climatology, with strong spatial correlation between exceedance and blocking events. Both events are favored not only by low carrying capacity (narrow roads), but also a downstream reduction in capacity (lane closures causing a bottleneck). The theory fails, however, to accurately predict blocking events in time. Exceedance events are not a useful predictor of an imminent block, suggesting that confounding factors explain their shared climatological structure. 

Abstract Neural networks (NNs) are increasingly used for data‐driven subgrid‐scale parameterizations in weather and climate models. While NNs are powerful tools for learning complex non‐linear relationships from data, there are several challenges in using them for parameterizations. Three of these challenges are (a) data imbalance related to learning rare, often large‐amplitude, samples; (b) uncertainty quantification (UQ) of the predictions to provide an accuracy indicator; and (c) generalization to other climates, for example, those with different radiative forcings. Here, we examine the performance of methods for addressing these challenges using NN‐based emulators of the Whole Atmosphere Community Climate Model (WACCM) physics‐based gravity wave (GW) parameterizations as a test case. WACCM has complex, state‐of‐the‐art parameterizations for orography‐, convection‐, and front‐driven GWs. Convection‐ and orography‐driven GWs have significant data imbalance due to the absence of convection or orography in most grid points. We address data imbalance using resampling and/or weighted loss functions, enabling the successful emulation of parameterizations for all three sources. We demonstrate that three UQ methods (Bayesian NNs, variational auto‐encoders, and dropouts) provide ensemble spreads that correspond to accuracy during testing, offering criteria for identifying when an NN gives inaccurate predictions. Finally, we show that the accuracy of these NNs decreases for a warmer climate (4 × CO₂). However, their performance is significantly improved by applying transfer learning, for example, re‐training only one layer using ∼1% new data from the warmer climate. The findings of this study offer insights for developing reliable and generalizable data‐driven parameterizations for various processes, including (but not limited to) GWs. 

Abstract While a poleward shift of the near-surface jet and storm track in response to increased greenhouse gases appears to be robust, the magnitude of this change is uncertain and differs across models, and the mechanisms for this change are poorly constrained. An intermediate complexity GCM is used in this study to explore the factors governing the magnitude of the poleward shift and the mechanisms involved. The degree to which parameterized subgrid-scale convection is inhibited has a leading-order effect on the poleward shift, with a simulation with more convection (and less large-scale precipitation) simulating a significantly weaker shift, and eventually no shift at all if convection is strongly preferred over large-scale precipitation. Many of the physical processes proposed to drive the poleward shift are equally active in all simulations (even those with no poleward shift). Hence, we can conclude that these mechanisms are not of leading-order significance for the poleward shift in any of the simulations. The thermodynamic budget, however, provides useful insight into differences in the jet and storm track response among the simulations. It helps identify midlatitude moisture and latent heat release as a crucial differentiator. These results have implications for intermodel spread in the jet, hydrological cycle, and storm track response to increased greenhouse gases in intermodel comparison projects.