Abstract (EN)

We first analyze some atmospheric parameters measured in 2004 at the top of La Soufrière volcano: wind speed, pressure and temperature. Next, we study the stochastic process of the wind speed modulus which is of interest for prediction of pollution induced by the plume. We fit to the data various models : FARIMA, periodic trend plus multifractional Brownian noise, multifractal continuous cascades. The estimations show that they evoluate during some transition phases in response to environment changes. Thus, we finally suggest, for future works, some settings that should be more appropriate: time series with regime changes and stochastic processes in random environment.