Evaluation of a machine learning framework to forecast storm surge

Mr Daryl Metters1

1Queensland Department Of Environment And Science, Brisbane, Australia

Machine learning is being used to achieve solutions to issues in many areas of science. Forecasting storm surge is an area of interest that has traditionally relied on parametric and numerical modelling methodologies to achieve any degree of forecasting precision. Machine learning in comparison offers computationally inexpensive means of elucidating solutions to many issues. In this study machine learning is investigated as an alternative to modelling of storm surge using numerical modelling methods only. Two methodologies were implemented using inputs of: the non-tidal residual; wind speed and direction and; atmospheric pressure. A 24-hour forecast of sea level was achieved via initially (1) forecasting wind speed and direction and atmospheric pressure using two time series machine-learning models and (2) using BoM numerically modelled forecasts of wind speed and direction and atmospheric pressure. Six machine-learning models were then used to forecast the non-tidal residual using standard learning and testing machine-learning methods extended with the two weather forecast datasets. The 24 hour forecast of sea level and storm surge was then compared to the actual sea level and storm surge for each of the two methodologies and six machine learning models. Model performance was evaluated with correlations between actual and forecast sea level and storm surge levels. Good results were achieved with the numerical model forecast inputs giving a close fit to the sea level and a good correlation with actual storm surge. The time series generated inputs failed to achieve a significant correlation. The six machine learning models varied in their performance.


Biography:
Daryl Metters is a Senior Scientist in the Coastal Impacts Unit within the Queensland Government Department of Environment and Science. Daryl has a BSc (Hons) in Marine Science from Flinders University. Daryl has worked in the National Tidal Centre, Australian Bureau of Meteorology as a Tidal analyst, and as Manager Spatial Information (Tidal services) in Maritime Safety Queensland. His work includes monitoring sea levels in Australia and the South Pacific region, data management and operational aspects of tide gauge and wave monitoring networks in Queensland.

About conferences.com.au

conferences.com.au provides delegate registration, website and app solutions, and financial management for conferences, conventions and scientific meetings.

Contact Us

Please contact the team at conferences.com.au with any questions regarding the conference.
© 2017 - 2018 Conference Design Pty Ltd. conferences.com.au is a division of Conference Design Pty Ltd.