Section: New Results

Non–homogeneous Markov switching auto–regressive models for wind time series

Participants : Valérie Monbet, Julie Bessac.

This is a collaboration with Pierre Ailliot (UBO) and Françoise Pène (UBO).

We proposed [20] non–homogeneous Markov switching auto–regressive models for bivariate wind time series considering Cartesian coordinates on one hand and polar coordinates on the other hand. In non–homogeneous models, the transitions depend on the wind direction at the previous time. At the location of interest, wind is rotating more often clockwise but wind direction may also oscillate around two prevailing directions (northeast for anti–cyclonic conditions and southwest for cyclonic conditions). These features induce respectively some cycles which can be seen in the second order structure and modes in the marginal distribution. In broad outline, non–homogeneous transitions help the process to stay in the same weather regime when the wind direction is close to the prevailing directions and lead to sojourn duration in the regimes which are not geometric.