Floods are among the most devastating natural hazards, claiming lives and damaging infrastructure.
The question of how we can be prepared for these extreme events quickly reaches an almost philosophical level: First of all, what is an extreme event? Second, how can we know what the future will bring?
For the last century hydrologists have relied on statistical concepts, which are based on observed streamflow, to assess flood hazard but this approach opens many other questions: Do we believe that floods follow a statistical distribution? How long do we have to observe streamflow to make predictions into the future? And how do we deal with changing atmospheric conditions because of climate change?
From a mathematical standpoint we cannot perform sound extreme value statistics with little data and in a changing climate. That’s where creativity – and counterfactual scenarios – come in.
Counterfactual scenarios: Thinking of what COULD have happened
The smaller the catchment, the less likely that the position of the catchment coincided with an extreme rainfall event within our observational period. If an extreme flood hits this catchment, we are usually surprised and call this event unprecedented.
But again, this perception is only based on our short horizon of past experiences. If by now, you think that this way of assessing flood hazard is a rather strange way of doing it, be relieved because many hydrologists agree (see e.g. this paper).
We can be more creative instead of just waiting like sitting ducks for the next extreme flood to happen. Detailed observations of rainfall events, e.g. by weather radar, and powerful hydrological models enable us to simulate a large number of hypothetical events (counterfactuals) that did not happen in the past but might have happened.
Have you ever asked yourself how a rainfall event that caused a flood elsewhere would have affected your area? This idea is the basis of counterfactual thinking. The idea has been around for almost a century (see a good review here) but really picks up with nowadays computational possibilities.
A bridge destroyed by floods in the Himalayas. Find the person for scale. Image credit: Paul Voit.
In regard to flood risk management we are particularly interested in “downward counterfactuals”, scenarios with a worse outcome than they had in reality. With this pro-active approach we are able to broaden our data basis, find the range of extreme floods and adapt flood risk management accordingly.
For example, based on counterfactual studies we could show that the devastating flood that hit Western Germany in 2021 and was the costliest natural disaster in German history could have hit the region even worse, had the rainfall been slightly shifted. (By the way, one month before a similarly extreme rainfall event happened east of Berlin and had no impact, because the rain fell in a rather flat area with sandy soils and few inhabitants.)
And one does not need to look far, to find similar near-miss events that reflect the extreme 200-year design flood as we found in another study. In the USA, flood risk management is now moving towards the use of counterfactuals, rather than solely relying on extreme value statistics.
Leveraging counterfactual scenarios for science communication
Another advantage of the counterfactual approach is public communication.
Currently, a lot of communication relies on statistical methods and the infamous “return periods”, which often cause confusion and mistrust. (“We already had a 100-year flood last year, why now another one? Your science is nonsense!”)
Two 100-year floods (Jahrhunderthochwasser) in 11 years, critically marked at a restaurant on the Danube close to Vienna. Image credit: Paul Voit.
In contrast, explaining to stakeholders that a recently observed rainfall event could have affected their interests or assets much more strongly, had it just happened in a slightly different position is much easier and leads to broader acceptance.
However, flood risk assessments based on counterfactuals will be only meaningful if the counterfactual scenarios are plausible. To which extent can rainfall events be shifted? What are realistic scenarios?
I think we are on the right track and these are the questions which we will have to tackle in future research projects to harness counterfactuals as scenarios for hydrological modelling and powerful public communication tools.
