Understanding the atmospheric circulation response to climate change (ERC Advanced Grant)


My project aims to better understand the response of the atmospheric circulation to climate change, including how to best characterize the uncertainties associated with that response.  The project WPs represent parallel strands which together achieve the overall aim of the project.

All publications can be found on my main University web page, in the University of Reading’s publicly accessible archive.

WP1: Diagnosing model error

The sensitivity in climate models of key features of atmospheric circulation to poorly constrained aspects of parameterized drag processes — a comparatively unexplored aspect of climate modelling — has been extensively examined. Using free-running climate models, the team has explored the role of both boundary-layer roughness (Polichtchouk & Shepherd 2016) and orographic blocking (Pithan et al. 2016), and their impact on the circulation response to greenhouse warming (van Niekerk et al. 2017). In a collaboration with ECMWF, the model errors associated with both these processes were shown to evolve differently between weather and climate timescales because of dynamical feedbacks (Sandu et al. 2016). A novel nudging technique has been used to suppress these feedbacks to obtain a cleaner understanding of the coupling between the different processes leading to surface drag (van Niekerk et al. 2016). This work is collectively stimulating a renewed emphasis on surface drag processes within weather and climate modelling centres around the world, which should reduce model errors (Sandu et al. 2019). The team has also addressed the effect of gravity-wave drag on model error in the context of stratospheric variability and stratosphere-troposphere coupling (Polichtchouk et al. 2019 JAS, GRL). The group has also made progress on understanding the links between cloud processes and atmospheric circulation — another relatively underexplored area of research (Ceppi & Shepherd 2017).

WP2: Calibrating model projections

The standard approach to using multi-model ensembles of climate projections, e.g. in the IPCC Assessment Reports, is to take the ensemble mean as the central estimate, and the spread as a measure of uncertainty. This is widely admitted to be unfounded. We are seeking better ways of identifying the real information contained within the ensemble. A new approach of optimal seasonal averaging was developed to increase the signal-to-noise ratio of the circulation response (Zappa et al. 2015 JClim), and was used to show that virtually all of the uncertainty associated with the projected cold-season Mediterranean drying — which will have profound socio-economic consequences for Europe — is traceable to the circulation response (Zappa et al. 2015 ERL). Subsequent work (Zappa & Shepherd 2017) has developed a novel, non-probabilistic ‘storyline’ approach to characterizing the uncertainty for climate-change impacts associated with atmospheric circulation. The group has also discovered that the sea-surface temperature mediated circulation response has two distinct timescales, which has profound implications for the difference between transient and equilibrated climate impacts (Ceppi et al. 2018). The impact of Arctic sea-ice loss on midlatitude wintertime circulation has been a major source of uncertainty in model studies; the group used a novel analysis technique to show a consistent response in the North Atlantic sector across the CMIP5 models in late winter (Zappa et al. 2018).

WP3: Understanding natural variability

In order to understand the climate record and to distinguish the response to anthropogenic forcings such as greenhouse gas increases from natural internal variability (climate noise), it is essential to understand the latter. This is challenging because climate noise is highly structured in space and time, with long-memory effects. Two broad approaches are being followed in the project. The first is to understand the implications of non-stationarity, especially interannual variations in the seasonal cycle, on widely used statistical methods of causal inference which focus on anomalies rather than full fields and which generally assume stationarity. This work has already shown the pitfalls of the standard methodology (Byrne et al. 2016), and developed new approaches (Byrne et al. 2017, Byrne & Shepherd 2018). This has led to improved understanding of the signal of stratosphere-troposphere coupling in Southern Hemisphere seasonal prediction (Byrne et al. 2019). The second approach is to develop sound theoretical models for the interaction between different components of the atmospheric circulation, to provide a physical basis for causal inference analysis. PhD student Boljka has done this for extratropical circulation (Boljka and Shepherd 2018; Boljka et al. 2018), and PhD student Gabrielski is doing it for tropical circulation (work in progress).

Novel and/or unconventional methodologies

In order to understand the interactions between different climate processes, the traditional approach has been to use simplified models of climate, the argument being that their behaviour is easier to understand than that of the most complex models. However, the difficulties in understanding mainly pertain to how to determine causal inference in a highly nonlinear system with strong dynamical feedbacks, which is an issue for the simplified models as well. Moreover, experience has shown that the simplifications introduced in such models (e.g. neglect of moist processes, or physically inconsistent parameterizations) can lead to spurious results. In WP1, a new approach to tackle this question is being promoted where the most complex models are used in simplified settings (e.g. no land surface). This provides stronger explanatory power (by reducing the number of relevant factors) whilst retaining an experimental pathway to the full system, and is illustrated in Polichtchouk & Shepherd (2016).

The diagnosis of climate model biases is hampered by the strong dynamical feedbacks that amplify or even obscure the source of the biases. In weather forecasting, model biases are more readily identified because these feedbacks do not have time to develop as strongly. However, most climate models are not configured to also produce weather forecasts. WP1 has overcome this limitation in the case of surface drag processes by promoting a novel nudging methodology, whereby the free atmosphere is constrained by observations whilst the boundary layer is free to respond to model perturbations. The method is illustrated in van Niekerk et al. (2016) and has recently been proposed as an experimental protocol within the World Meteorological Organization.

The traditional approach to quantifying uncertainty in climate model projections, e.g. in the IPCC Assessment Reports, is to take the ensemble mean as the central estimate, and the spread as a measure of uncertainty. This is widely admitted to be unfounded but is still the standard practice. WP2 is pioneering new approaches to this problem. In Shepherd (2016 CCCR) it was argued that given the very different nature of the uncertainties associated with thermodynamic and circulation-related aspects of climate change, those uncertainties ought to be treated differently. In Zappa & Shepherd (2017), we promote a novel ‘storyline’ approach to representing the circulation-related aspect of the uncertainty. The epistemological basis for the storyline approach is laid out in Shepherd (2019).

The traditional approach to extreme-event attribution is to assess whether the risk of the event has been increased by climate change. This is very challenging for extremes involving the atmospheric circulation or storms, generally leading to no result. Trenberth, Fasullo & Shepherd (2015) advocated an alternative and novel framing of the question, taking the meteorological event as given and asking whether its impacts were increased by known thermodynamic aspects of climate change, concerning which there is reasonably high confidence.

To understand natural internal variability, the standard approach has been to examine the behaviour of anomalies relative to the long-term average, using Granger causality. However Granger causality relies on statistical stationarity, which is not a valid assumption for the climate system. WP3 has shown the pitfalls of this approach (Byrne et al. 2016), and Byrne et al. (2017) developed novel statistical approaches that take proper account of interannual variations in the seasonal cycle.

Inter and cross disciplinary developments

I have been raising the awareness of the importance of atmospheric circulation as a source of uncertainty in climate change — the focus of this project — in many seminars and invited talks at conferences. One such talk led to an invitation to write a peer-reviewed Perspective for Nature Geoscience (Shepherd 2014), a high-impact inter-disciplinary journal. The article was accordingly aimed at non-specialists, and seems to have struck a chord as it is being very highly cited.

Related to this, I have been involved in visioning within the World Climate Research Programme, leading to two more of these peer-reviewed Perspectives in inter-disciplinary Nature family journals (though in these cases, they are community papers). One concerns the coupling between clouds, circulation and climate sensitivity, and was published in Nature Geoscience (Bony et al. 2015); the other concerns dynamical linkages between Arctic warming and midlatitude weather, and was published in Nature Climate Change (Overland et al. 2016). A third piece of this type arose from my involvement in a COST action on bias correction and downscaling, and was published in Nature Climate Change (Maraun et al. 2017).

Extreme-event attribution is becoming a very important topic in climate science, both for public communication and for adaptation planning. I have become very interested in extreme events connected with atmospheric circulation, because they challenge the robustness of our understanding of the physical mechanisms of climate change. This has led to a number of inter-disciplinary activities.

I was invited to write a short article on extreme events by the two highest-profile inter-disciplinary science journals in the world: a News & Views piece on heat waves in Nature (published in July 2015), and a Perspective on Arctic-midlatitude weather linkages in Science (published in September 2016). Both pieces are non-technical, non-peer-reviewed, and aimed at a broad audience.

I was also invited to write a review paper on extreme event attribution, published in early 2016 in the inter-disciplinary review journal Current Climate Change Reports (see previous section). This paper developed a common framework within which the alternative framing of the climate-change event-attribution question proposed in the Nature Climate Change Perspective by Trenberth et al. (2015), and discussed in the previous section — namely, taking the meteorological event as given and asking whether its impacts were increased by known thermodynamic aspects of climate change — could be seen as a limiting case of the traditional risk-based framing. This view seems to have become accepted by the event-attribution community and is helping to break down the polarization that had developed in recent years.

I was asked to join a Committee of the US National Academy of Sciences writing a peer-reviewed assessment of Extreme Event Attribution in the Context of Climate Change. This was a major, high-profile undertaking, with the report released in March 2016 including briefings on Capitol Hill and at the White House (see Dissemination). My role in the report mainly had to do with the chapter on Framing (of the question), which was the most inter-disciplinary subject of the entire report.

I held an inter-disciplinary workshop on storyline approaches to the representation of uncertainty in physical aspects of climate change, which led to an inter-disciplinary publication from the workshop participants (Shepherd et al. 2018).

Finally, I have become very interested in how the language used to convey scientific uncertainty affects people’s perception of climate change, and initiated a collaboration with a team of psychologists. This has led to one publication (Løhre et al. 2018) with another in progress.

WP1: Diagnosing model error