Anders Malmberg, avd för matematisk statistik, Lund

A Statistical Model for Wind Fields in The North Atlantic Ocean

Abstract

Measurements of sea conditions over large areas by satellites have made
it possible to compare forecasts with the actual conditions at sea.
Marine operations and unconventional transports, which are sensitive to
sea conditions, must have reliable and actual information of these. In
this project we aim to find a model for the sea wind and use this to
update future forecasts when new satellite measurements are done.

Modeling a large-scale process like this causes many difficulties and a
traditional space-time process is not suitable. Hence we have to reduce
to number of states. This we do by adopting a physical model for the
wind process, which implies that the wind process is composed by a
finite set of principal components.

The statistical model is temporally dynamic and spatially descriptive.
This is achieved by assuming an AR(1) model for the time dynamics with
spatially colored noise. In this framework, with a measurement equation
of physical phenomena, which is described by a physical-statistical
model the Kalman filter naturally comes at hand.

Our first model will be estimated upon forecasted data only and in
discrete time.