Busker's research shows that weather forecasts for extreme rainfall in Europe are accurate enough to significantly reduce flood damage. The researcher examined rainfall forecasts issued prior to the 2021 flood in Limburg, western Germany and Belgium. 'We see that the warning indicators pointed to the Ahr valley two days in advance. Unfortunately, there were more than a hundred casualties at this location'.
When do you issue a warning?
Much research has already been done on how to better predict this extreme weather, but how do you determine when to issue a warning, such as a red weather alert? Should we take measures in advance, even if it costs a lot of money? 'We need to be able to better assess in which situations weather extremes could have major societal impacts,' says Busker. 'Based on this, we should carefully consider whether and when to take action.'
People do need to be able to trust such warnings: too many false alarms erode the credibility of forecasts. This leads to a reluctance to take action.
Busker's research shows that rainfall forecasts several days in advance can trigger effective actions, such as temporarily raising the dike, protecting a house with flood barriers, or even evacuation. However, forecasts longer than four days ahead are no longer accurate enough to act on in large parts of Europe. Really costly actions should only be taken one to two days in advance.
Making predictions with AI
Drought can also cause major impacts. We can often predict droughts better and earlier than floods. In East Africa, Busker and his colleagues investigated how warning systems can mitigate harmful consequences of drought, such as famine and water shortages. 'Often we can see a drought coming weeks to months in advance. Based on that, emergency measures can be taken in advance,' says Busker.
Some impacts of drought are more difficult to predict, such as the onset of a famine. For this, the researchers have developed a model based on Artificial Intelligence (AI), capable of predicting food insecurity in Kenya, Somalia and Ethiopia. 'Machine learning is enormously powerful, but it has one major requirement: enough data must be available, from both normal and extreme conditions, to 'train' the model. We see that precisely the biggest famines, which are infrequent, are harder to predict as a result.'
New warning systems
Worldwide, investments are currently being made in new warning systems. The UN has launched the Early Warnings for All initiative, with the aim that everyone on earth will have access to a warning system by 2027. The Royal Netherlands Meteorological Institute (KNMI) is developing a separate centre for weather warnings: the Early Warning Centre. 'It is important that the latest research insights are integrated into all these initiatives, to maximise the potential of warning systems.'
Busker will defend his thesis on 7 November.