 |
 |
Date: Friday, 19th November 1999 Time: 3.15 Room: Lecture Theatre C, James Clerk Maxwell Building, King's Buildings Speaker: Chris Glasbey (Biomathematics and Statistics Scotland, Edinburgh) Title: Time series models for weather data Abstract: The modelling of weather data poses many interesting challenges for statisticians. For example, what time series models are appropriate for rainfall and solar radiation, which have marginal distributions that are highly non-Gaussian? In this talk, we take radically different approaches for the two variables. By applying a monotonic transformation to rainfall data, marginal normality is achieved. This defines a latent Gaussian variable, with zero rainfall corresponding to censored values below a threshold. For solar radiation data, a new form of nonlinear autoregressive times series is proposed, by specifying joint marginal distributions at low lags to be multivariate Gaussian mixtures. The model is also a special case of multiprocess dynamic linear models. For both models, we consider issues of estimation and usage. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%