Robust trend parameters in a multivariate spatial linear model

This article gives a robust estimator of the trend parameters in multivariate spatial linear models. This estimator is presented as an alternative to the classical one which is obtained by using cokriging. The goal focuses on improving predictions of spatial variables when data contain both atypical and high influence observations. The procedure consists of extending robust methods used in linear regression models to the multivariate spatial context. The resulting estimator belongs to the class of GM-estimators and then, it is a bounded influence estimator and it has good robust properties, in particular, a high breakdown point and a high efficiency. An illustrative example is given to show how the proposed estimator works.