For years engineers tried to prevent flooding, then they realized they can’t stop nature. Now instead of trying to stop it, we try to mitigate flooding by creating spaces that can absorb a lot of water (parks along rivers are an example of this). Still, these attempts don’t always work and with increasing instability in our climate it’s getting harder to deal with more extreme flooding instances. This is where a new startup, Floodmapp, fits in. They are using machine learning and AI to improve how we understand flooding instead of the traditional physics-driven modelling.
The companyâ€™s premise is simple: We have the tools to build real-time flooding models today, but we just have chosen not to take advantage of them. Water follows gravity, which means that if you know the topology of a place, you can predict where the water will flow to. The challenge has been that calculating second-order differential equations at high resolution remains computationally expensive.
Murphy and Prosser decided to eschew the traditional physics-based approach that has been popular in hydrology for decades for a completely data-based approach that takes advantage of widely available techniques in machine learning to make those calculations much more palatable. â€œWe do top down what used to be bottoms up,â€ Murphy said. â€œWe have really sort of broken the speed barrier.â€ That work led to the creation of DASH, the startupâ€™s real-time flood model.