One-sided Multivariate Testing and Environmental Monitoring

Abstract

In environmental monitoring the quality assessment is essentially one-sided. For example quality of air decreases as one or more air pollutants increase. Statistical multivariate monitoring is then concerned with the problem of detecting some non-random one-sided shifts of one or more among various pollutant levels. In this paper, after discussing some change models useful in environmental monitoring, a generalized likelihood ratio test statistic is proposed both for retrospective testing and for on-line change detection. The test is then compared to the approach based on union-intersection of one-sided univariate tests.