Variable selection in large environmental data sets using principal components analysis