Spatial correlation estimates based on satellite observations corrected with the prior knowledge on sensor devices' technical characteristics

Abstract

In many empirical studies spatial correlations are used to identify the distance above which dependency is negligible, to assist the choice in locating a systematic grid of sample points in ground surveys. However, estimates are undermined by the fact that our inference is based on satellite data that are only an approximation of the ground truth, due to the presence of a series of disturbing factors like (e.g.) light scattering, presence of obstacles (like clouds), and instrument precision limitations. In this paper we introduce a procedure to correct spatial correlation estimates using prior knowledge on the satellite sensor's technical characteristics and obtain more reliable estimates. We derive an approximation of the "ground-truth" pattern of correlation as a function of the satellite-based spatial correlation and of the sensor's (user-specified) technical characteristics. We show the effects of these corrections referring to a series of illustrative examples based on theoretical calculations regarding the negative exponential correlogram. The correction efficiency relative to the classical Method-of-Moments estimator is also assessed by means of a Monte Carlo application to simulated SAR maps.