Co-clustering of multivariate functional data for the analysis of air pollution in the South of France

Nowadays, air pollution is a major threat for public health with clear relationships many diseases, especially cardiovascular ones. The spatiotemporal study of great interest governments and local authorities when deciding alerts or new city policies against increase. aim this work to profiles environmental data collected in the south France (Région Sud) by agency AtmoSud. idea better understand exposition pollutants inhabitants on large territory important differences term geography urbanism. gather recording daily measurements five variables, namely, three (PM10, NO2, O3) two meteorological factors (pressure temperature) over six years. Those can be seen as multivariate functional data: quantitative entities evolving along time which there growing need methods summarize them. For purpose novel co-clustering model defined. based latent block assumes each co-cluster probabilistic distribution principal component scores. A stochastic EM algorithm, embedding Gibbs sampler, proposed inference well selection criteria choosing number co-clusters. application algorithm Région Sud allowed divide region, composed 357 zones, into macroareas common exposure pollution. We showed that vary accordingly seasons, patterns are similar during years studied. These results used develop specific programs reduce at macroarea level identify periods year high peaks order set up prevention programs. Overall, approach powerful resource analyse intrinsic structure variables long time.