Flood models tested in the first collective validation captured between 52 and 97% of flooded areas for two flood events and three areas in Nigeria and Mozambique. This result was “better than expected”, according to the research team.
“These models have shown a level of performance that allows them to be used for large-scale questions, like identifying the regions of highest flood risk globally,” says Mark Bernhofen of the University of Leeds, UK. “The majority of the models performed fairly well.”
In the decade leading up to 2015, floods killed an estimated 157,000 people worldwide, and affected 2.3 billion in total. As climate changes, the number of people exposed to river flooding over the next three decades is likely to rise by over 30%.
Every model of climate impact depends on a broader climate model to drive it. Whereas climate models have been compared and validated since the late 1980s, flood models have existed only for the last decade or so. Until now, their validation has taken place individually, using different methodologies. “The different results are difficult to compare,” says Bernhofen.
To overcome this problem, Bernhofen and colleagues from the UK, Japan, the Netherlands and Italy tested six global flood models using the same technique, for flooding in Lokoja and Idah in Nigeria in 2012, and Chemba in Mozambique in 2007. The researchers compared the models’ predicted flood areas against historical satellite data over millions of pixels, each about 90 sq. m in size.
The 2007 floods in Mozambique affected 130,000 people while those in Nigeria in 2012 affected 4 million people.
According to a widely used “critical success index”, which rates a model from zero – worst at representing reality – to one, the six flood models in the validation ranged from 0.45 to 0.7. This was better than the team expected.
The models performed best for the 2012 flooding in Lokoja, a city at the confluence of the Niger and Benue rivers that has a well-defined floodplain. The worst performance was for the same flood event downstream from Lokoja in the Idah region, where the floodplain is large, flat and has many interconnecting channels. “That makes modelling the flow of water a lot more complicated,” says Bernhofen.
Although the global flood models are relatively young, Bernhofen believes they’ve benefitted from the many decades of detailed flood modelling at smaller scales. He hopes this analysis highlights the effectiveness of the validation procedure, and helps people “make an informed decision about the most appropriate model to use and where”.
“The goal now is to expand this validation to include more flood events in more regions, and broaden the skills-measures we use, so that we can get a truly worldwide view of the performance of these global flood models,” he says.