Presentation of the paper: Statistical methods for massive data observed over space and time

Speaker: Emilio Porcu, Universidad de Valparaiso

Abstract: statistical methods for the analysis of massive space--time datasets are on high demand. Meteorological variables and air pollutants are densely observed over space and time. For n sites and k temporal instants where observations are available, the order of computation for the likelihood is O(n^3k^3). Thus, it is necessary to propose statistical methods that allow a considerable gain in terms of computation whilst preserving a certain level of statistical efficiency. In this talk, we present some classes of space--time covariances for multivariate random fields having the important feature of being compactly supported, that is they are identically zero outside a finite radius. Features of such a construction are then illustrated through real as well as artificial data analysis.