How do scientists assess the impact of human actions on extreme weather events?

In the immediate aftermath of an extreme weather event, a common question arises around how much human-caused climate change contributed to its likelihood or severity.

For the past two decades, the rapidly growing field of extreme event attribution means that scientists have increasingly been able to answer this question.

However, carrying out these assessments and publishing them in peer-reviewed journals typically takes a year or longer. This means that the findings are only available long after the event has ended and it is merely a distant memory in the public mind.

To tackle this problem, we established the World Weather Attribution (WWA) initiative in 2015. This collaboration – between climate scientists in the UK, Netherlands, France and beyond – works to rapidly quantify the role of climate change within a few days or weeks after an extreme event. This means we are providing information while it is still highly relevant for the public and disaster recovery planners.

Since our first study of the European heatwave in the summer of 2015, we have carried out more than 30 individual assessments. In a new paper, published in Climatic Change, we look back at the process we developed and discuss some of the lessons we have learned.

Our main realisation from the past five years is that the actual attribution assessment is actually just one step out of eight – and it is probably the easiest one of the whole set.

In this piece, we look at what is involved in each individual step of the process:

The sheer size of the Earth means that extreme weather is typically occurring somewhere almost every day. So, which of these events merit an attribution study?

At WWA, we try to prioritise events that have had a large impact or that have provoked a strong discussion in society, so that the answers will be useful for a large audience. These are often events for which the Red Cross – a WWA partner – issues international appeals.

On occasions, smaller events closer to home or even meteorological records that did not affect many people also seem to generate enough interest to warrant assessment.
We explicitly do not include the expected influence of climate change on the event as the trigger criteria. A result that an event was not affected by climate change – or even became less likely – is just as useful scientifically as one where the probability increased.

Defining the event turned out to be both much harder and more important than we initially thought.

As an example: the first published extreme event attribution study analysed the European summer heatwave of 2003. It took as its event definition a European-wide seasonally averaged temperature, whereas the impacts had been tens of thousands of deaths in the 10-day hottest period in cities. We try, therefore, to define the events as close to the impacts as possible.

It should be noted that, in practice, finding out what really happened during the event is not easy. An example is given in the figure below, which shows the very different estimates of the highest three-day rainfall around Houston due to Hurricane Harvey in 2017 from different observing systems.

The radar analysis (c) is probably the most reliable option, but the dataset only goes back to 2005. The CPC gridded analysis (b) only uses stations that report in real-time and therefore underestimates the extreme weather event. The satellite analysis (d) diverges strongly from the others. Here, we chose station data (a) as the most reliable dataset with a long time series – more than 100 years for many stations.

Observed maximum three-day averaged rainfall over coastal Texas for January-September 2017 (mm/dy). a) GHCN-D v2 rain gauges, b) CPC 25 km analysis, c) NOAA calibrated radar (maximum in August 25–30), d) NASA GPM/IMERG satellite analysis. Source: van Oldenborgh et al. (2017).

The observed data we use is an essential part of an attribution assessment and gives us two pieces of information: how rare the extreme weather event is in the current climate and how much this has changed over the observed record.

The probability of the event in the current climate is very important to inform policymakers whether this is the kind of extreme that infrastructure should be able to handle or not. As an example, the floods that paralysed Jakarta in Indonesia in January 2014 turned out to be caused by a rainfall event with a “return period” – that is, how often an event of that size would be expected – of 4-13 years.

This is an extreme weather event that is not particularly rare, which suggests the city has a very high vulnerability to flooding.

Conversely, the floods in Chennai in December 2015 were caused by rainfall with an estimated return period of 600 to 2,500 years, which implies that the event might be too rare to expect defence mechanisms to hold out in such circumstances.

However, a warming climate means that types of events with historically very large return periods are being seen more frequently. This means that infrastructure that has been designed for the past climate may be overwhelmed more than expected.

To compute how much the likelihood of the extreme weather event has changed over the period with observations, we fit these to a mathematical function called an “extreme value distribution” that changes with global warming. This gives the change in probability and severity for an event as observed due to all influences, including global warming.

Observations alone cannot be used to link – or “attribute” – a trend in extreme weather events to global warming (or natural influences on the climate). For this, we need climate models.

Each individual climate model will have strengths and weaknesses and we can only use those that realistically simulate the extremes for the location we are investigating.

In practice, we use the following three criteria to select a collection of climate models for the assessment:

As climate models are imperfect representations of reality, we demand at least two – and preferably more models – to be good enough for the attribution analysis in order to conduct a study.

The next step is the actual attribution analysis. For each model, we compute how much more likely or intense the extreme weather event has become due to human-caused emissions of greenhouse gases and aerosols.

This can be done in one of two ways. One method is to run two sets of model simulations – one for the current climate and one for a “counterfactual” world without human influence on the climate. We can then identify how many severe weather events match the one we are assessing in each set. The difference between the counts in both worlds gives how much more or less likely the extremes have become with global warming.

The second method is to use existing simulations, such as the historical and future climate model runs from the global Coupled Model Intercomparison Project. These simulations can then be analysed in exactly the same way as the observations (see point 3).

This means we can estimate whether there is a change in probability and severity because of potential drivers – such as climate change – but also land use changes or any other climate forcings the model includes. In the models, we know which forcings they contain (such as greenhouse gases and aerosols) and which they do not (for example, some local feedbacks) and we have a lot more data than in the observations.

Under either method, this attribution step provides an estimate from each model for the change in likelihood for the extreme weather event as a result of climate change. Similarly, the approaches can quantify the change in intensity for an event of a particular probability.

The next step is to combine the information from the observations and multiple models into one overarching statement of how the probability and intensity of the physical extreme weather event – that is, the hazard – has changed.

This is not trivial and requires estimates on how good we think the models are in describing the extremes. Part of that information can be derived from the spread of the models, but part has to be a judgment based on how realistic the extremes are simulated in the model.

A weather-related disaster happens due to a combination of three factors: a hazard (the meteorological extreme), exposure (the people in harm’s way) and vulnerability (how well people and ecosystems are able to cope with the hazard).

As well as the hazard, we consider it essential to discuss the vulnerability and exposure in an attribution study. Not only do these combine with the changes in the physical extremes (computed in the previous steps) to determine the impact of the extreme weather, but they may have significant trends themselves.

For example, our study of the drought in São Paulo, Brazil in 2014-15 found that it had not been made more severe from climate change. However, our analysis showed that the increase of population of the city by roughly 20% in 20 years, and the even faster increase in per capita water usage, had not been addressed by appropriate updates in the storage and supply systems.

Hence, in this case, the trends in vulnerability and exposure were the main driver of the significant water shortages in the city.

The final step is communicating the results to a range of audiences. This means explaining the findings in a scientifically accurate way that is also useful and understandable for the intended audience.

We found three layers of communication are necessary, relating to three groups of users of the results of our studies: scientists; policymakers and emergency management agencies; and media outlets and the general public.

In presenting results to the scientific community, we always publish a scientific report that documents the attribution study in sufficient detail for another scientist to be able to reproduce the results. If there are new methodological elements to the analysis, we also commit to submitting the study to a journal to undergo full peer-review.

For policymakers, humanitarian aid workers and other non-scientific professional audiences, we found that the most effective way to communicate attribution findings in written form is briefing notes that summarise the key points from the physical science analysis, elaborate on the vulnerability and exposure context and then provide specific recommended next steps to increase resilience to this type of extreme weather event.

Finally, to help reach the public via the media, we will typically prepare a press release and/or website news item that communicates the primary findings of the study. In addition to the physical science findings, these press releases typically provide a very brief, objective description of the non-physical science factors that contributed to the extreme weather event.

Over the past five years, we have found that the steps outlined above have allowed us to provide a robust message on how extreme weather events are – and are not – being influenced by climate change.

And by reacting in a matter of days or weeks, we have been able to inform key audiences with a solid scientific result swiftly after an extreme weather event has occurred – when the interest is highest and results most relevant.

The results of attribution studies are useful for informing risk reduction for future extreme weather events, and also for raising awareness about the rising risks in a changing climate and thus the importance of reducing greenhouse gas emissions.

Importantly, the results are relevant simply because the question is often asked – and if it is not answered scientifically, it will be answered unscientifically.