Summary: Workshop on Correlated Extreme Events

Lead Organizers: Colin Raymond, Radley Horton

Steering Committee: Amir AghaKouchak, Olivia Martius, Thomas Wahl, Jakob Zscheischler, Suzana Camargo, Alex Ruane, Adam Sobel, Michael Oppenheimer, Noah Diffenbaugh, Sonia Seneviratne

This workshop took place at Columbia University (New York, USA) on May 28-31. It was designed to bring together the communities of researchers studying extreme events connected to each other through some physical mechanism. A particular aim was to draw on core disciplines that have used correlated frameworks – for example, multivariate extremes in the hydrological community, or sequential and contingent extremes in the natural-hazards community — and thus to build a more integrated sense of the importance of ‘correlation’ or ‘compounding’ across all climate extremes. Exploring the current thinking and future directions of how to optimally conduct impacts-driven correlated-extremes research was another major theme. The first sessions were structured around a division of events into compound/multivariate, concurrent, and sequential or persistent types, with the latter sessions aiming to integrate these into traditional disciplines of study (e.g. drought or tropical cyclones) as well as into broader societal decision-making.

Approximately 175 people attended the workshop, representing 20 countries. About 2/3 were climate scientists, the remainder being social scientists, company representatives, or government officials. A total of 90 presentations (42 talks and 48 posters) were presented, including 44 by women, 40 by early-career researchers, and 7 by developing-country researchers. The website contains links to the full video recordings of the workshop, and to many of the slides and poster pdfs.

Two keynotes underscored the growing prominence of correlated thinking for hazards that can be understood as resulting from multiple connected drivers, and the simultaneous need to recognize that climate information is most useful to practitioners when generated, packaged, and presented in particular ways. Connection (or correlation, if the linkage is physically based) raises the likelihood that two hazardous events will occur simultaneously or within a short window of time, and these correlation coefficients are themselves variable with climate change (e.g. the intensifying interaction between continental drought and heat). From an impacts perspective, self-consistent ‘storylines’ are often preferable to overlapping sets of probabilities, and specific thresholds often mark the need for qualitative changes in strategy. Additionally, a particular adaptation measure’s usefulness is best measured by the ease with which it can be enhanced in the future, and the complexity and cost with which it is associated, rather than merely its effectiveness in the climate realm over some evaluation period. 

The first several sessions covered recent physical-science advances in multivariate (which we termed 'compound') and concurrent extremes, expanding on topics covered in a 2017 WCRP-sponsored workshop. Major themes included the nuanced and sometimes non-monotonic effects across combinations of variables, especially under climate change where conditions will soon be far from 20th-century baselines. Several speakers described the necessity of sophisticated modeling, due to the poor spatial coverage and temporal record of observed variables other than temperature and precipitation, as well as the rarity of certain events. Talks ranged from statistical modeling techniques to fluvial-pluvial flooding to more-exotic combinations of drivers, such as tropical-cyclone winds buffeting the drought-stricken Iberian Peninsula and exacerbating wildfires. The Katia-Irma-Jose sequence of Atlantic tropical cyclones was cited as encapsulating how ‘types’ of correlation are often more a matter of perspective or convenience than a physically meaningful distinction. In the mid-latitudes, recurrent Rossby waves were discussed as contributing strongly to persistent and geographically specific summer heat and extratropical cyclones, though the ultimate origin of these wave patterns remains opaque.

The second day moved into other areas, such as sequences of the same or different types of event, or exceptionally persistent events. Projected increases in persistence were revealed for U.S. temperature/air pollution compound episodes and for mid-latitude consecutive precipitation days. The importance of conducting such assessments with regard to seasonal, subregional, and meteorological conditions, as appropriate for the processes underlying a given correlated extreme, was emphasized.

Splitting into eight breakout groups provided the opportunity for diverse crowdsourced identification of themes, among which were the potential benefits of closer collaboration with engineering organizations and the need for both climate scientists and policy-makers to use metrics that accurately reflect societal impacts and values.

Two final sessions considered a variety of perspectives on correlated extremes from sectoral experts, and estimates of how these affect systemic risk to global interconnected networks such as food, business supply chains, and ecology. The complex nature of these systems can either exacerbate or buffer shocks from correlated extremes, depending largely on the human-management side of the equation. A few presentations gave hope for increased predictability, such as by highlighting the important role played by ENSO cycles in modulating concurrent crop failures, or by increasing the utility of climate information — such as through learning from water managers that identifying types of years which lie within the current climate space, but for which there are no good historical analogues, is a valuable piece of data for them. Where, when, and on whom the impacts of correlated extremes fall are all telling predictors of their severity, with vulnerability often being compound in a demographic sense as well as a physical one. There was often a two-sides-of-the-same-coin feel to the discussion; for example, in the realization that nonlinearity of impacts is nearly universal, but that it derives in some cases primarily from a physical response (waterlogged soil becoming more susceptible to flooding from subsequent rain), while in other cases from a societal inability to muster enough financial or political resources in the face of repeated extreme events.

An overarching conclusion of the workshop was that risk management requires knowledge of the full spectrum of possibilities, organized into a coherent framework that is digestible by impacts decision-makers through their normal deliberation processes. Correlated extremes are rarely (if ever) incorporated into current policy guidelines for infrastructure, development, etc., indicating that significant effort will be required to translate even our existing scientific understanding. However, a positive realization was that minor but frequent updates to guidelines are not only more politically feasible, but operationally desirable in terms of being able to track changing probabilities of the occurrence of particular correlated extremes as estimations and emissions scenarios are updated.

This summary post was written by Colin Raymond.