Session A10: Rare Events, Tipping Points, and Abrupt Changes in the Climate System

The Path to the next IPCC Assessment: Known Knowns and Known Unknowns

We summarize the key findings of the Sixth Assessment Report (AR6) by the Working Group I of the Intergovernmental Panel on Climate Change regarding the physical science basis for climate change.  We summarize the empirical evidence of accelerating human-induced changes in the atmosphere, ocean, cryosphere, and land surface as evaluated in the AR6.  The interpretation of this evidence has been bolstered by significant advances since the last Assessment in our theory, computational models, and observations of the forcings, responses, and feedbacks driving the evolution of the climate system.  We note the impacts of these advances on the levels of evidence and confidence underlying some of the core findings by Working Group I. Despite this progress, many areas of climate science present ongoing research opportunities for physicists where discoveries would materially improve our understanding of past, present, and future climates. We conclude by discussing several of the most important directions for future research.     

This Climate is Changing

Using observational data and an elementary rigorous statistical fact it is easily shown that the distribution of Earth's climate is non-stationary. Examination of records of hundreds of local Industrial Era temperature histories in the Northern Hemisphere were used to show this fact. Statistically, the mean of the ensemble has been rising during the Industrial Era. All of this confirms what climate scientists already know. The issue of predictions under uncertainties was tackled as well: a simple balance model was tuned to track an ensemble of climate records. Stochastic parametrizations were created to capture natural and anthropogenic CO2 forcings. The resulting stochastic model was then tested against historical data and then used to make future predictions. This exercise confirmed as well climate science attribution to significant global warming during the Industrial Era to anthropogenic activities. The variability of the model due to uncertainties is simply not large enough to obfuscate a clear rise in the mean temperature in the Industrial Era. Further, even if the variance of the natural CO2 contribution is greatly increased artificially (in the model), the fluctuations cannot account for the current change in the historical mean.

These outcomes weaken the factual validity of the US administration, 2016-2020, claims that there are too many uncertainties in climate and climate science to make climate predictions, and further that contemporary reports of floods, extreme weather, even a rising global mean temperature are simply manifestations of a climate that always fluctuates within a nature-derived statistical distribution.   

Lévy-noise versus Gaussian-noise-induced Transitions in the Ghil-Sellers Energy Balance Model

We study the impact of applying stochastic forcing to the Ghil-Sellers energy balance climate model in the form of a fluctuating solar irradiance. Through numerical simulations, we explore the noise-induced transitions between the competing warm and snowball climate states. We consider multiplicative stochastic forcing driven by Gaussian and α-stable Lévy - α∈(0,2) - noise laws, and examine the statistics of transition times and most probable transition paths. While the Gaussian noise case - used here as a reference - has been extensively studied in a plethora of studies on metastable systems, much less is known about the Lévy case, both in terms of mathematical theory and heuristics, especially in the case of high- and infinite-dimensional systems. In the weak noise limit, the expected residence time in each metastable state scales in a fundamentally different way in the Gaussian vs. Lévy noise case with respect to the intensity of the noise. In the former case, the classical Kramers-like exponential law is recovered. In the latter case, power laws are found, with the exponent equal to −α, in apparent agreement with rigorous results obtained for additive noise in a related - yet different - reaction-diffusion equation as well as in simpler models. The transition paths are studied in a projection of the state space and remarkable differences are observed between the two different types of noise. The snowball-to-warm and the warm-to-snowball most probable transition path cross at the single unstable edge state on the basin boundary. In the case of Lévy noise, the most probable transition paths in the two directions are wholly separated, as transitions apparently take place via the closest basin boundary region to the outgoing attractor.   

Atmospheric dynamics across scales: Jet streams and gravity waves

On planetary scales, the midlatitude jet streams and storm tracks set the weather patterns experienced by a large fraction of humanity. Understanding and preparing for their variability on daily to decadal timescales is a critical challenge. Events that bring about abrupt transitions in their behavior, such as the breakdown of the polar vortex, have the potential to suddenly alter weather patterns and have been associated with extremes such as cold snaps and floods. I will present work investigating these transitions with application to the behavior of the jet stream in the Atlantic basin, which exhibits three preferred positions, and show results suggesting that the northernmost of these is a consequence of the presence of Greenland. Additionally, our recent work demonstrates that jet responses in the Atlantic basin may be described as regime transitions, resulting in a shifting of the probability distribution under forcing by polar vortex events, rather than a straightforward shift of the position of the jet stream. On smaller scales, atmospheric gravity waves (GWs) are ubiquitously excited on the Earth and are critical drivers of the atmospheric circulation, however, they present a challenge to climate prediction. I will describe collaborative efforts aimed at developing an observationally constrained, physically meaningful representation of the effects of GWs on the resolved flow for use in global climate models. We have leveraged high-resolution data from tens of thousands of balloon flights with high-resolution measurements of position, pressure, and temperature from which we have inferred statistics of gravity wave motions in the lower stratosphere. We have also developed a machine learning GW parameterization, coupled it to a global climate model, and showed that it is stable and accurate when run online, and that it reproduces features of the climate that depend critically on GWs.   

Interpretable transfer learning: Applications to climate change modeling

Recent studies have found promising results using machine learning (ML) techniques such as convolutional neural networks (CNNs) to improve climate models, e.g., by better representing subgrid-scale processes. However, applying NN-enhanced models to a different climate system, for example with a different radiative forcing, can lead to inaccurate and even unstable simulations. This is because NNs and similar techniques cannot be expected to work accurately outside their training manifold, i.e, they often do not extrapolate. Transfer learning (TL), which involves re-training some layers with a small amount of new data, offers a solution to this and a few recent studies have found promising results in simple test cases. However, the general understanding of TL, mainly from applications involving static images, does not apply to climate modeling. Here, we present a framework to guide TL for applications involving climate modeling. The framework is based on the spectral analysis of the new data and inputs, weights, and outputs of the re-trained layers in TL. Using 2D turbulence as the test case, we show how this framework connects with the physics of the flow and some of the recent advances in the ML community on the training of NNs.           

Learning a weather dictionary of atmospheric patterns using Latent Dirichlet Allocation to study climate change & extreme events

Mid-latitude circulation dynamics is often described in terms of weather regimes. Each pattern is given by a given combination of several synoptic objects (cyclones and anticyclones). Such intrication makes it arduous to  quantify recurrence and intensity of climate extremes. Here we apply Latent Dirichlet Allocation (LDA), used for topic modeling in linguistic, to build a weather dictionary: we define daily maps of a gridded target observable as documents, and the grid-points composing the map as words. LDA provides a representation of documents in terms of a combination of spatial patterns named motifs, which are latent patterns inferred from the set of snapshots. For atmospheric data, we find that motifs correspond to pure synoptic objects (cyclones and anticyclones), that can be seen as building blocks of weather regimes. We show that LDA weights provide a natural way to characterize the climate change for the recurrence of regimes associated with extreme events.   

Short weather forecasts inform long-term climatology of sudden stratospheric warming

The sub-seasonal-to-seasonal (S2S) time horizon is a frontier of weather forecasting, exemplified by sudden stratospheric warming (SSW): a breakdown of the winter polar vortex, altering surface weather for months. SSW events are complex and diverse, unpredictable beyond  2 weeks, and often analyzed case by case. The historical scarcity of observations, and an unusually SSW-rich 2000's decade, lead to uncertain SSW climatology: when do they occur, how often, and how predictably? Long, expensive model runs could answer these statistical questions, but with a tradeoff between cost and bias. We instead utilize weather forecast ensembles that are high-resolution, but short (subseasonal) in duration. A simple coarse-graining procedure chains them together to estimate key climate statistics, such as annual frequencies and timing distributions of SSW events, as formulated in Transition Path Theory. We use S2S forecasts between 1996-2018, but find that the SSW statistics match well with 20th-century reanalysis. Our method extrapolates the climatology well beyond what is possible with the short observational dataset that initialized the forecasts, yielding accurate estimates of 1 in a century events. This suggests exciting new uses for ensemble forecasts in rare event analysis.    

Understanding extremes in a warming climate: On acknowledging uncertainty, embracing complexity, and asking societally relevant questions

Human-caused global warming is now an observable physical reality—but the most important consequences of climate change do not come from the incremental increase in global mean temperature itself. Instead, many human and natural systems are facing increased frequency of historically unprecedented extreme weather events. Indeed, there is now strong evidence that the frequency and/or intensity of certain types of extremes—particularly those most directly related to changes in atmospheric temperature and moisture—are already increasing in a statistically robust manner. Despite this, the complex spatiotemporal dynamics surrounding far-from-mean state conditions in the Earth system, combined with the noisy statistical signal inherent to infrequently occurring extreme events within a still-short observational record, continue to complicate research efforts and often lead to confusion in community and public interpretation of results. In this talk, I will explore recent advances in the burgeoning sub-field of extreme event attribution—and offer thoughts on how embracing “Earth system complexity” and focusing on societally-relevant physical science questions can help move the field forward in the climate change era.