Three Non-Linear Statistical Methods for Analyzing PM10 Pollution in Rouen Area

Abstract

The aim of this paper is to illustrate three modern statistical methods through a case study, which arises from a scientific collaboration between Air Normand on the applied side and Université Paris Descartes and the Institut National des Sciences Appliquées (INSA) in Rouen on the academic side. The problem is to analyze PM10 pollution during 2004-2006 in Rouen area using six different monitoring sites and to quantify the effects of variables of different types, mainly meteorological versus other pollutant measurements. Three methodologies – random forests, mixtures of linear models, and nonlinear additive models – are used in the analysis. In addition to the statistical interest of the study, we give detailed software oriented results and complete code using three freely available R packages.