Projects‎ > ‎

Extreme weather event real-time attribution machine

As greenhouse gases continue to accumulate in Earth’s atmosphere, the resultant warming of the climate system changes the 
nature of extreme weather events (EWEs). Every EWE has a contribution from natural variability as well as an anthropogenic contribution resulting from climate change. Through this project our team is conducting the research necessary to develop an Extreme Weather Event Real-time Attribution Machine (EWERAMwhere, within a day or two of an EWE having occurred over New Zealand, and in response to media questions about the role of climate change in that event, rather than generic statements, scientifically defensible data will be available to inform quantitative statements about the role of climate change in both the severity and frequency of the event. For example, we want to be able to make statements such as ‘this rainfall event was 27% more severe than it would have been had there been no anthropogenic climate change’ or ‘in pre-industrial times these sorts of events would have happened once every 65 years but now, because of anthropogenic climate change, are expected to happen every 35 years’. Such statements will then be broadly disseminated to the New Zealand public through standard Meteorological Service of New Zealand communication channels. 

Our project builds on research that has emerged over the past decade on the detection and attribution of EWEs. Previous detection and attribution systems mainly conducted retrospective analyses on e.g. drought events, up to a year after the termination of the event. Near real-time, scientifically defensible, attribution of such events to human-induced changes in climate not only satisfies the public’s desire to know, but can also highlight the future risks of such events to emergency managers, regional planners, the insurance industry, and policy-makers at all levels of government. We are building such a near real-time capability on a foundation of improved understanding of EWE attribution. The research is providing deeper insight into, and confidence in, the many risk calculations that underpin New Zealand’s building codes; land, water, health and flood management; insurance; transport networks, and many additional aspects of daily life. We can be confident that climate warming has increased the moisture-holding capacity of the atmosphere, and evidence for increased severity of extreme rainfall is accumulating. Furthermore, atmospheric circulation changes will affect the nature of EWEs in complex ways. While this creates a greater attribution challenge, for New Zealand in particular, accounting for changes in dynamics is essential to communicating a coherent picture of EWE attribution to stakeholders. 

The simplest way to attribute a specific EWE to the effects of climate change is to pose the question ‘if a weather system of exactly this kind (the same synoptic weather situation) had occurred in pre-industrial times, how more or less severe would it have been?’. MetService conducts continual simulations of the current weather as part of its weather forecasting operations. A large ‘ensemble’ of weather model simulations is run on a supercomputer to capture the inherent uncertainty in our knowledge of the initial conditions that precede the event (so-called ‘factual’ simulations – also called ANT (anthropogenic ) simulations. As part of our project we are conducting a similar ensemble of simulations but with sea surface temperatures, atmospheric temperatures and atmospheric humidity values modified to mimic what they would have been under pre-industrial conditions (so-called ‘counterfactual’ simulations – also called NAT (natural) simulations). We can do this by sourcing pre-industrial model simulations from the IPCC model simulation archives, and from weather@home simulations of New Zealand’s weather under pre-industrial conditions. By comparing the ensembles of factual (ANT) and counterfactual (NAT) simulations, we can make inferences about changes in the severity of an extreme event (see Figure 1) and about the fraction of attributable risk (FAR) for that EWE. In Figure 1, the PDF is the probability density function of some climate variable (e.g. rainfall) showing how a shift in the distribution to the right creates a large increase in the probability of exceedance of some threshold value that would be considered extreme.

This is considered to be a highly ‘conditioned’ approach – it assumes that synoptic situations, identical to the target event, would have occurred in the past, and that may not necessarily be the case. This would be considered a somewhat ‘idealized’ approach. Making statements about expected changes in the likelihood of such events, or making statements about the role of climate change in affecting the underlying state of the atmosphere that drives the evolution is such EWEs, is far more challenging. This project calls on the combined skills and expertise of a team of researchers from five institutions across New Zealand, viz.: Bodeker Scientific (lead organisation), MetService, NIWA, Victoria University of Wellington, and the University of Canterbury.

We know that the applicability of the highly conditioned approach is quite limited; a goal of the project will be to conduct the research to understand exactly under what situations the highly conditioned approach does provide value. Answering the question of how the likelihood of similar events has changed as a result of climate change faces two challenges, (i) what does ‘similar’ mean? and (ii) how do you identify such ‘events’ in climate model simulation output so that you can count them? We are going to trial machine-learning approaches for event identification. Finally, achieving all of this in real-time, will be a massive challenge. The whole system needs to be highly automated and various components will need to rely on advanced data processing methods so that inferences can be made without significant investment of someone’s time.

A poster/one-page flyer that gives an overview of the project and its first results can be found here.