M-estimation for linear models with spatially-correlated errors

When a linear model is used to analyze spatially correlated data, but the form of the spatial correlogram is unknown or difficult to specify, it is not straightforward to carry out valid statistical inference on regression parameters. We derive the asymptotic distributions for a class of M-estimators, which includes the least squares and the least absolute deviation, so as to provide the basis for valid large-sample inference even when the spatial correlation structure is not taken into account in estimating the regression coefficients.