A long-lead forecast model for the prediction of shelf water oscillations along the Caribbean coast of the island of Puerto Rico

It is a well-established fact that shelf waters along the south coast of Puerto Rico oscillate at the seiche mode period of approximately one-hour. Under the right conditions the amplitude of these events equals that of the diurnal tide. At several locations worldwide this type of coastal water oscillation has extremely large amplitude, causes considerable damage and is also the focus of current research. Even though the precise forcing of this phenomenon at this particular location is not yet completely understood, it is believed that packets of solitary waves impinging against the steep slope could be forcing the shelf waters into motion. A long-lead forecast model, which is introduced in this paper, is being made available for the prediction of these coastal waters oscillations. The model will facilitate the establishment of an adequate data collection schedule for the study of these events. The one-hour oscillations of the shelf waters are conspicuously modulated and render a narrow band signal within the tide spectral record. Different techniques were used to identify the main spectral components of the signal envelope. These components were found to be nearly orthogonal and formed the basis for the formulation of a long lead seasonal forecast model. In order for the predictor to be an effective forecast tool the phase of each harmonic constituent was carefully calibrated in the time domain with respect to the different lunar cycles. The weights of the model were assigned using least squares. Stepwise regression and principal component analysis were used to manage the inclusion of constituents into the predictor. Although the main model is linear in nature the extreme excursions in amplitude were tackled by the formulation of a non-linear predictor