Bivariate Splines for Ozone Concentration Forecasting

For ground level ozone prediction, we use past spatially distributed measurements and the functional linear regression model. We use bivariate splines defined over triangulations of a relevant region of the U.S.A. to implement this functional data approach in which random surfaces represent ozone concentrations. We compare the least squares method with penalty to the principal components regression approach.
We apply these two functional linear models to ground level ozone forecasting over the south eastern United States to illustrate the predictive skills of the two methods. If time allows, we will also discuss a naive extension of these models for Hurricane Path prediction where both the explanatory and response variables are random surfaces.