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

On the use of Functional Data Analysis (FDA) in descriptive meteorology and oceanography


Functional Data Analysis (FDA) refers to the branch of statistics which develops methods for analyzing data sets of variables indexed along a continuum. And most data in environmental sciences arrive as sampled curves. Temperature profiles, size distribution of organisms, satellite images, are such examples of data which can be handled as functions with argument being time, space, or any other real variable.  Since the famous monograph of Ramsay and Silverman (2005), many theoretical efforts have been conducted to generalize the usual multivariate methods (PCA, CCA, linear model, kriging,...) when data lie in functional spaces. Surprisingly, there are still few applications on real data. This talk purposely deals with some statistical analysis of environmental data which integrate the functional nature of the data. Through applications in oceanography and meteorology, we present some mathematical tools dedicated to dimension reduction and to the regression of a functional variable using functional covariates. We illustrate the fact that the use of functional methods owns many advantages such as (i) including curves shape into the analysis, (ii) sweeping out variability of sampling devices by data smoothing steps, (ii) fixing sampling design problems by constructing continuous data from raw data, and (iv) integrating successive derivatives of the functional object into the analysis.

Time and place
Wednesday 1 June 2016, 15.00
Room C609, Arrhenius Laboratory, 6th floor