An ocean   of data

Dion Häfner

NBI Copenhagen

A naïve perspective on rogue wave prediction

$ whoami

  1. I like to do science with computers
  2. main author of the first (?) high-performance, pure Python GCM
  3. not a wave expert
  4. (also a person)

The Physics

[McAllister, 2019]

The Physics

Causes of rogue waves:

  • Non-linear instabilities (e.g. BFI)
  • Crossing sea states
  • Wind-wave interactions
  • Wave-current interactions
  • Non-linear focusing
  • Linear statistics

[Adcock, 2014]

Wave Statistics

A Gaussian sea?

Wave Statistics

A Gaussian sea?

Wave Statistics

Nonlinear seas

[Gemmrich, 2011]

Hypotheses

There are no freaks, our statistics are off

 

 There are freaks, normal waves should break before reaching extreme heights

 There are freaks, special mechanisms are required to transfer that much energy into one wave

[Fedele, 2016]

Correlations

Sea state

[Cattrell, 2018], [Janssen, 2009]

Weather

[Cavaleri, 2016]

2D wave field

[Cousins, 2016]

Local features

... but what has the most predictive power?

A model

ML
Neural network
Gradient boosting

Random forest

Prediction
Rogue wave likelihood

Data

Raw time series

Sea state

Weather

Physics

(Wave model)

Spectra

Directional spread

Possible pitfalls

Ask a silly question, get a silly answer

 

Rogues are more likely during a storm

Possible pitfalls

Too little data

 

P=0

Possible pitfalls

Does not generalize    [Cattrell, 2018]

 

We find that frequency of occurrence of rogue waves and their generating mechanism is not spatially uniform, and each location is likely to have its own unique sensitivities [...] We conclude that forecastable predictors of rogue wave occurrence will need to be location specific and reflective of their generation mechanism.

So, why ML?

  • Shed light on relative importance of effects
  • Quantitative predictions
  • Weights should yield some physical insight
  • Might generalize (fingers crossed)

References

  • McAllister, M. "How scientists recreated a monster wave that looks like Hokusai’s famous image." https://theconversation.com/how-scientists-recreated-a-monster-wave-that-looks-like-hokusais-famous-image-110304
  • Gemmrich, J., and C. Garrett. "Dynamical and statistical explanations of observed occurrence rates of rogue waves." Natural Hazards and Earth System Sciences 11.5 (2011): 1437-1446.
  • Cavaleri, Luigi, et al. "The Draupner wave: A fresh look and the emerging view." Journal of Geophysical Research: Oceans 121.8 (2016): 6061-6075.
  • Cattrell, A. D., et al. "Can rogue waves be predicted using characteristic wave parameters?." Journal of Geophysical Research: Oceans 123.8 (2018): 5624-5636.
  • Adcock, Thomas AA, and Paul H. Taylor. "The physics of anomalous (‘rogue’) ocean waves." Reports on Progress in Physics 77.10 (2014): 105901.

References

  • Cousins, Will, and Themistoklis P. Sapsis. "Reduced-order precursors of rare events in unidirectional nonlinear water waves." Journal of Fluid Mechanics 790 (2016): 368-388.
  • Fedele, Francesco, et al. "Real world ocean rogue waves explained without the modulational instability." Scientific reports 6 (2016): 27715.
  • Janssen, P. A. E. M., and Jean Bidlot. On the extension of the freak wave warning system and its verification. European Centre for Medium-Range Weather Forecasts, 2009.

An ocean of data

By Dion Häfner