A new artificial intelligence system is being designed by the University of Waikato to predict the impact of future flooding events in the Coromandel region.
It is being developed as part of the TAIAO project and will be one of only a handful of systems in operation in New Zealand using artificial intelligence (AI) machine learning to predict floods and the effects of flooding.
Senior researcher at the University of Waikato AI Institute and senior regional hazard advisor at Waikato Regional Council, Dr Phil Mourot says the idea for the project first came about after the Queen’s Birthday weekend flooding in the Coromandel in 2020.
“Currently, there is no accurate way to make predictions about the impact of future weather events. When an event does occur, all we can really do is monitor the situation and react to it. During the 2020 Queen’s Birthday weekend floods, the inability to accurately forecast the severity of the flooding hindered the ability of civil defense and regional and local councils to prepare for it,” Dr Mourot says.
“After seeing this, we decided to try and use AI machine learning to solve this issue and make it easier for councils, civil defense and even farmers and growers to prepare for the impacts of flooding.”
Although it is not the first of its kind, Dr Mourot says the key point of difference is the multiple data streams the system uses to make predictions that makes it unique.
“Unlike others, this system uses a range of data sources provided to us by project partners such as MetService and MetOcean and the Waikato Regional Council including rainfall measures, river water levels and rain gauge catchment data to make predictions.”
This data is fed into an AI Machine which learns by identifying patterns in historic events and using the information to predict future events. While the system is still in its development phase, Dr Mourot says the future is exciting.
“Right now, we have been focused on finding and integrating new data sources into the system and then testing the accuracy of the system against past flooding events and hypothetical future events. The challenge now is to get it to the point where it can make predictions that are one hour, three hours, six hours or even 12 hours into the future in real time.”
The team plans to have a prototype ready to roll out in the Coromandel by 2023. Postdoctoral research fellow at the University, Dr Nick Lim says if the prototype is successful, the system will eventually be rolled out across the Waikato and then onto other regions across New Zealand.
“The flood prediction system is designed to easily integrate into existing regional and local council processes for managing weather events. However, because the type and frequency of weather events are different from area to area and due to regional differences in methodology for dealing with weather events, the programme has to be tailored to each region,” Dr Lim says.
“We have a lot of work ahead of us, but I am excited for what the future of this project will bring.”
The flood prediction project is one of several underway as part of the TAIAO project which aims to improve capability and reproducibility in New Zealand environmental data science.