M-Estimator of the drift coefficients in a spatial linear model

Kriging as an interpolation method, uses as predictor a linear function of the observations, minimizing the mean squared prediction error or estimation variance. Under multivariate normality assumptions, the given predictor is the best unbiased predictor, and will be vulnerable to outliers. To overcome this problem, a robust weighted estimator of the drift model coefficients is proposed, where unequally spaced data may be weighted through the tile areas of the Dirichlet tessellation.