Pseudo best estimator by a separable approximation of spatial covariance structures

TL;DR

This paper introduces a computationally efficient pseudo best estimator for spatial regression models that approximates covariance structures with separable functions, outperforming least squares and approaching the accuracy of the generalized least squares.

Contribution

The paper proposes a novel pseudo best estimator using separable covariance approximations, reducing computational costs while maintaining high accuracy in spatial regression analysis.

Findings

PBE outperforms LSE in accuracy even with isotropic Matérn covariance.

PBE is computationally more efficient than GLSE for large datasets.

Monte Carlo simulations validate the effectiveness of PBE.

Abstract

We consider a linear regression model with a spatially correlated error term on a lattice. When estimating coefficients in the linear regression model, the generalized least squares estimator (GLSE) is used if the covariance structures are known. However, the GLSE for large spatial data sets is computationally expensive, because it involves inverting the covariance matrix of error terms from each observations. To reduce the computational complexity, we propose a pseudo best estimator (PBE) using spatial covariance structures approximated by separable covariance functions. We derive the asymptotic covariance matrix of the PBE and compare it with those of the least squares estimator (LSE) and the GLSE through some simulations. Monte Carlo simulations demonstrate that the PBE using separable covariance functions has superior accuracy to that of the LSE, which does not contain the information of the spatial covariance structure, even if the true process has an isotropic Mat\'ern covariance function. Additionally, our proposed PBE is computationally efficient relative to the GLSE for large spatial data sets.