A multivariate Gaussian scan statistic forspatial data

A new spatial scan statistic is proposed for multivariate data indexed in space. Such as many other scan methods, it relies on a generalized likelihood ratio but it also takes into account the correlations between variables. This spatial scan test seems to be more powerful than the independent version, whatever the level of correlation between variables. We apply this method to a data set recording the levels of pollutant metals in the area of Lille, France.