David Nerini
Mediterranean Institute of Oceanography, Aix-Marseille University, France


Spatial prediction of precipitation curves using temperature curves as covariate


This talk is about the construction of a statistical model for the predictions of a variable sampled on a spatial domain when we dispose of ancillary information provided by external variables.  The proposed approach owes its originality to the very nature of the data : both observed variable and covariates are data which arrive as curves. The following example will be developed as a guideline.  A collection of N = 90 sampling locations in France (essentially airport places) is available where weather stations have recorded temperature T (°C) and precipitation levels P (mm) for twenty years since 1991. We are interested in (i) reconstituting the annual profiles of both temperature and precipitation from 12 monthly mean of these variables, (ii) predicting temperature and precipitation curves at unknown location in France. However, on some sampling stations, both functional variables are observed. On other locations, only one or the other is available. The talk will be framed around the following questions :
    • What are effective statistical tools and techniques for constructing a sample of curves from discrete observed data?
    • How can we predict one functional variable, or the other, or both functions at unknown location using the sample on hand in a spatial     context?
    • How can we use the cross-information given by few stations on both curves for the prediction on a station where one variable is missing?


Time and Place

Monday May 28th, 14.15
Rossbysalen C609, Arrhenius Laboratory, 6th floor