Space-Time Prediction of Ocean Winds

The topic of this thesis is inspired by an

experiment in which a vessel, laying a submarine cable, was

provided with forecasts overlaid with satellite observations of

significant wave height. During the operation, the

vessel was close to

an adverse weather area and the personnel on board could confirm that

the forecast was not as close to the "ground truth" as the satellite

observation was. One of the outcomes of this experiment was the

suggestion to develop a method providing forecasts merged with

satellite observations. In this thesis such a method is developed for

near-surface ocean winds.

The thesis consists of four papers (Paper A-D). The contribution

of Paper A and B is the development of a statistical framework,

in which forecasts

and satellite observations in a bounded area are merged and a measure

of uncertainty is

provided.

A dimension-reduced Kalman filter is used as an emulator of

the atmospheric dynamics. This is considered in Paper A.

The method of merging Kalman filter forecasts with satellite

measurements is developed in Paper B.

Closely related to Paper A and B is the problem of modelling the

covariance structure of residuals taken as differences between

forecasts and satellite measurements. Two isotropic covariance functions

belonging to the Matern family are used. However,

neither of the functions seem

to properly model the residual field. The contribution of Paper C is an

explorative study and it forms a basis for further research.

Finally, Paper D models the dynamics of a spatio-temporal process

based on an image warping approach. Image warping models the dynamics through

the movement of a set of control points. As well as allowing affine

transformations, the model also allows for non-linear dynamics. The main

contribution of this paper is the formulation of a penalized

likelihood which is used to estimate the model.