Python Machine Learning Conference & GeoPython 2020

»Hybrid Downscaling of Swells in Small Islands«
2020-09-22, 15:05–15:25, Room 2

A hybrid coastal modelling method to downscale wave climate in small Pacific islands has been developed from different Python modules. User-friendly Jupyter Notebooks are also included in the results visualization.

Accurate wave hindcasts and forecasts in small islands are essential for reducing its susceptibility to coastal flooding due to distant-source wind waves. Existing global wave hindcasts are usually the boundary condition for the dynamic downscaling, using SWAN as the generation wave model. The numerical effort can be computationally very intensive, being useful to analyze this problem by means of hybrid models (surrogate models/ metamodels).

In this work, we propose a combination of: (a) definition of hydraulic boundary conditions based on a convolution of the incoming swells to a particular island, dealing with an unique time series of directional spectra; (b) physically- based techniques for identifying swell trains (Portilla et al 2015); (c) definition of swell families; (d) parameterization of each swell train (significant wave height, peak period, mean direction and directional spreading) in terms of parameters of enhancement and decay; (e) selection of N swells based on the maximum dissimilarity algorithm (Camus et al, 2011); (f) numerical simulation of the N swells (usually N is about 1/1000 of the total number of sea states of the wave hindcast); (g) nearshore reconstruction of wave parameters throughout the island (Camus et al 2011b); and (h) analysis of wave climate considering the swell families.

The application of this methodology to different small islands in the Pacific (Kwajalein, Majuro, Guam) shows the flexibility and robustness of this hybrid downscaling for improving our knowledge of wave climate in small islands.

To address this problem, a Python library composed by different interdependent modules has been developed. These modules include: (a) internet access to data services for long-term wave hindcast datasets; (b) machine learning methods; (c) command line interface for preparing and running SWAN model; (d) reconstruction of the time series of wave parameters throughout the island; (e) user-friendly visualization of the results. All these software codes are provided with Jupyter notebooks (Kluyver et al. 2016) allowing showing use cases of the different library’s modules. It is envisioned that the Jupyter environment will become the de-facto standard for doing and exchanging reproducible research, in particular for the coastal engineering community, in the following years.

REFERENCES

Camus, P., Mendez, F. J., Medina, R. & Cofiño, A. S. Analysis of clustering and selection algorithms for the study of multivariate wave climate. Coast. Eng. 58, 453–462 (2011). Camus, P., Méndez, F.J., Medina, R., (2011b). A hybrid efficient method to downscale wave climate to coastal areas. Coastal Engineering 58 (9), 851–862. Portilla-Yandún, J., L. Cavaleri, and G. Ph. Van Vledder (2015) Wave spectra partitioning and long term statistical distribution. Ocean Modell., 96, 148–160, doi:10.1016/j.ocemod.2015.06.008. Kluyver, T., et al (2016). Jupyter Notebooks - a publishing format for reproducible computational workflows. ELPUB . doi:10.3233/978-1-61499-649-1-87