Department Seminar of Yuval Yevnin - On deep learning and numerical wave forecasting models

09 November 2022, 14:00 - 15:00 
הפקולטה להנדסה 
0
Department Seminar of Yuval Yevnin - On deep learning and numerical wave forecasting models

 

 

 

School of Mechanical Engineering Seminar

Wednesday, November 9, 2022 at 14:00
Wolfson Building of Mechanical Engineering, Room 206

On deep learning and numerical wave forecasting models

Yuval Yevnin

PhD student of Prof. Yaron Toledo

School of mechanical engineering

Ocean waves have been affecting human life for millennia, from the early days of hunter-gatherer tribes fishing, through the explorations of the Viking age to modern trade and warfare. In most of those years the understanding and prediction ability of the seas and oceans was intuitive at best. This started to change during and after the second world war, as modern operational wave forecasting models were developed. The third generation of these stochastic numerical models are based on the wave action equation, and have been in use since the late 1980’s.

In this talk we discuss both improvements to the currently used operational models, as well as what we see as the next big step in ocean wave forecasting – the use of deep learning models.

First, an addition of a source term accounting for bottom wave reflection to WAVEWATCH III operational forecasting model is presented. This source terms theoretical background and derivation were extended from previous work. Next, it was implemented and was shown to improve the forecast in the near-shore area and in shallow water.

Second, a deep learning model for short-term forecasting of ocean wave height was developed. The model utilizes in-situ buoy measurements and mid-range operational model forecasts as input and predicts the short-term wave height at the buoy location. The model was shown to improve the forecast RMSE by as much as 77% for one hour horizon and by ~12% for up to 12 hours. In Addition, the model was also shown to be transferable to buoys at other locations without further training.

Finally, an advanced deep learning model for improved wind and consequent wave forecasts is presented. The model improves wind velocity magnitude forecast by ~10% and by using the improved wind in an operational wave model, a similar 10% improved wave height is achieved. The model can be localized in space and time, which is shown to produce 35% improvement in forecasting wave height at the Aegean Sea during the months of May to Septembers, when the Etesian wind is dominant.

Join Zoom Meetin https://us02web.zoom.us/j/82108132163?pwd=Z2h4UzNzUS9mbXplT0lMU1pZenFEQT09

Tel Aviv University makes every effort to respect copyright. If you own copyright to the content contained
here and / or the use of such content is in your opinion infringing, Contact us as soon as possible >>