Euro-Atlantic weather Regimes in the PRIMAVERA coupled climate simulations: impact of resolution and mean state biases on model performance
Climate Dynamics Springer Science and Business Media LLC 54:11-12 (2020) 5031-5048
Euro-Atlantic weather Regimes in the PRIMAVERA coupled climate simulations: impact of resolution and mean state biases on model performance
Climate Dynamics Springer Nature 54:11-12 (2020) 5031-5048
Abstract:
Recently, much attention has been devoted to better understand the internal modes of variability of the climate system. This is particularly important in mid-latitude regions like the North-Atlantic, which is characterized by a large natural variability and is intrinsically difficult to predict. A suitable framework for studying the modes of variability of the atmospheric circulation is to look for recurrent patterns, commonly referred to as Weather Regimes. Each regime is characterized by a specific large-scale atmospheric circulation pattern, thus influencing regional weather and extremes over Europe. The focus of the present paper is the study of the Euro-Atlantic wintertime Weather Regimes in the climate models participating to the PRIMAVERA project. We analyse here the set of coupled historical simulations (hist-1950), which have been performed both at standard and increased resolution, following the HighresMIP protocol. The models’ performance in reproducing the observed Weather Regimes is assessed in terms of different metrics, focussing on systematic biases and on the impact of resolution. We also analyse the connection of the Weather Regimes with the Jet Stream latitude and blocking frequency over the North-Atlantic sector. We find that—for most models—the regime patterns are better represented in the higher resolution version, for all regimes but the NAO-. On the other side, no clear impact of resolution is seen on the regime frequency of occurrence and persistence. Also, for most models, the regimes tend to be more tightly clustered in the increased resolution simulations, more closely resembling the observed ones. However, the horizontal resolution is not the only factor determining the model performance, and we find some evidence that biases in the SSTs and mean geopotential field might also play a role.Through a Jet Speed Darkly: The Emergence of Robust Euro-Atlantic Regimes in the Absence of Jet Speed Variability
ArXiv 2003.04871 (2020)
Abstract:
Euro-Atlantic regimes are typically identified using either the latitude of the eddy-driven jet, or clustering algorithms in the phase space of 500hPa geopotential height (Z500). However, while robust trimodality is visibly apparent in jet latitude indices, Z500 clusters require highly sensitive significance tests to distinguish them from autocorrelated noise. As a result, even small shifts in the time-period considered can notably alter the diagnosed regimes. Fixing the optimal regime number is also hard to justify. We argue that the jet speed, a near-Gaussian distribution projecting strongly onto the Z500 field, is the source of this lack of robustness. Once its influence is removed, the Z500 phase space becomes visibly non-Gaussian, and clustering algorithms easily recover three extremely stable regimes, corresponding to the jet latitude regimes. Further analysis supports the existence of two additional regimes, corresponding to a tilted and split jet. This framework therefore naturally unifies the two regime perspectives.Machine Learning for Stochastic Parameterization: Generative Adversarial Networks in the Lorenz’96 Model
Journal of Advances in Modeling Earth Systems American Geophysical Union 12:3 (2020) e2019MS001896
Abstract:
Stochastic parameterizations account for uncertainty in the representation of unresolved subgrid processes by sampling from the distribution of possible subgrid forcings. Some existing stochastic parameterizations utilize data‐driven approaches to characterize uncertainty, but these approaches require significant structural assumptions that can limit their scalability. Machine learning models, including neural networks, are able to represent a wide range of distributions and build optimized mappings between a large number of inputs and subgrid forcings. Recent research on machine learning parameterizations has focused only on deterministic parameterizations. In this study, we develop a stochastic parameterization using the generative adversarial network (GAN) machine learning framework. The GAN stochastic parameterization is trained and evaluated on output from the Lorenz '96 model, which is a common baseline model for evaluating both parameterization and data assimilation techniques. We evaluate different ways of characterizing the input noise for the model and perform model runs with the GAN parameterization at weather and climate time scales. Some of the GAN configurations perform better than a baseline bespoke parameterization at both time scales, and the networks closely reproduce the spatiotemporal correlations and regimes of the Lorenz '96 system. We also find that, in general, those models which produce skillful forecasts are also associated with the best climate simulations.
AMOC and summer sea ice as key drivers of the spread in mid-Holocene winter temperature patterns over Europe in PMIP3 models
Global and Planetary Change 184 (2020) 103055
Abstract:
The mid-Holocene (6000 years before present) was a warmer period than today in summer in most of the Northern Hemisphere. In winter, over Europe, pollen-based reconstructions show a dipole of temperature anomalies as compared to present-day, with warmer conditions in the north and colder in the south. It has been proposed that this pattern of temperature anomaly could be explained by a persisting positive phase of the North Atlantic Oscillation during this period, which was, however, not reproduced in general by climate models. Indeed, PMIP3 models show a large spread in their response to the mid-Holocene insolation changes, the physical origins of which are not understood. To improve the understanding of the reconstructed temperature changes and of the PMIP3 model spread, we analyze the dynamical response of these model simulations in the North Atlantic for mid-Holocene conditions as compared to pre-industrial. We focus on the European pattern of temperature in winter and compare the simulations with a pollen-based reconstruction. We find that some of the model simulations yield a similar pattern to the reconstructed one, but with far lower amplitude, although it remains within the reconstruction uncertainty. We attribute the northern warm part of the latitudinal dipole of temperature anomaly in winter to a lower sea-ice cover in the Nordic Seas. The decrease of sea ice in winter indeed reduces the local sea-ice insulation effect, allowing the released ocean heat to reach continental northern Europe. This decrease in winter sea-ice cover is related to an increase in the Atlantic meridional overturning circulation (AMOC) and its associated ocean heat transport, as well as the effect of insolation changes on sea ice in summer, which persists until winter. We only find a slight cooling signal over southern Europe, compared to reconstructions, mainly related to the insolation-induced cooling in winter over Africa. We show that the models that failed to reproduce any AMOC increase under mid-Holocene conditions are also the ones that do not reproduce the temperature pattern over Europe. The change in sea level pressure is not sufficient to explain the spread among the models. The ocean-sea ice mechanisms that we proposed constitute an alternative explanation to the pattern of changes in winter temperatures over Europe in the mid-Holocene, which is in better agreement with available model simulations of this period. Finally, we evaluate if reconstructions of the AMOC for the mid-Holocene can provide interesting emerging constraints on key changes in European climate, and indirectly on AMOC response to on-going and future radiative changes. Although there is a significant link between the response of the mid-Holocene and projections, it remains limited. The proposed mechanism does not appear to be sufficient to explain the large discrepancies between models and reconstruction data for the summertime period.