Risk surface mapping through multivariate geographical dependences by hierarchical Bayesian models