Estimating spatial structure from uncertain sensor data

In a recent paper jointly published with colleagues from ETH Zürich and HERE Technologies Switzerland we propose two different ways to integrate the reliability of observations with Moran’s I, a measure of spatial autocorrelation. The proposals made are tested in the light of two case studies, one based on real temperatures and movement data and the other using synthetic data. The results show that the way reliability information is incorporated into the Moran’s I estimates has a strong impact on how the measure responds to volatile available information. It is shown that absolute reliability information is much less powerful in addressing the problem of differing contexts than relative concepts that give more weight to more reliable observations, regardless of the general degree of uncertainty. The results presented are seen as an important stimulus for the discourse on spatial autocorrelation measures in the light of uncertainties. Read the full paper