Storm tide forecasting with machine learning

Forecasting sea level up to 24 hours in advance during extreme weather events

Current technology can accurately predict normal tide levels, but not storm surge levels.

Partnering with Deloitte, the Coastal Impact Unit (CIU), as part of the Department of Environment and Science (DES) Accelerating Science Delivery Innovation (ASDI) program, conducted a proof of concept to determine if machine learning could help.

If we could predict storm surge levels more accurately, governments, the private sector and communities, could better prepare for evacuation and recovery during extreme weather events.

The team looked at the technologies available to support machine learning, and whether machine learning could improve predictions.

They used a cloud-based Amazon Web Services (AWS) machine learning environment, provided by the High Performance Computing support team in the Department of Natural Resources and Mines and Energy (DNRME), which meant the support team could directly manage the virtual environments. This helped the project deliver findings within a short, 6-week timeframe.

The team approached the problem using Agile methodology (similar to that used by the Testing within Government (TWiG) program). They developed a problem statement, conducted discovery and pitch sessions, held daily stand-ups, and presented regular showcases on their progress which gave others in DES the opportunity to contribute and give feedback.

It was a fantastic experience having the opportunity to work with a broad array of stakeholders across the Coastal Impacts Unit and ASDI teams to apply advanced machine learning techniques in a novel way. The outcomes of the PoC served to further underscore the power of data and advanced analytics in helping to solve challenging, real world problems.

Will Dodd, Director, Consulting; Deloitte

CIU were able to prove that the machine learning models delivered by Deloitte could address the original requirement for more accurate storm surge prediction. However, to progress any one of these models into a usable forecast system, more work would need to be done.

The team gained important foundational knowledge of machine learning through this proof of concept, which they’ll continue to investigate and advocate for in forecasting and service delivery.

Read the lessons we learnt from this initiative.