Using the Kriging Technique for Prediction of Non-Continuous Phenomena in Unmeasured Locations: Dispelling the Myth

Abstract

The Kriging technique was designed to model continuous phenomena such as temperature, mineral deposits, sound, etc. However, it is often used in non-continuous phenomena, such as predicting road traffic, modeling the number of trips made on public transit, predicting critical crime locations, etc., which can result in the violation of established assumptions. In this way, recurrent confusion lies in the equivocal association between the level of measurement of a continuous random variable and the erroneously assumed continuous nature of the phenomenon under study. Thus, this study aims to demonstrate how problematic using Kriging is in non-continuous phenomena, mainly in transportation studies, and to present the Geographically Weighted Regression (GWR) as a robust competitor to this task. The results of two case studies using the variables “Households Income” and “Car Trip Rate” in the city of São Paulo, Brazil, showed some problems with the Kriging technique when there are few sampled points and very similar results between Kriging and GWR when there are many sampled points, being the latter much simpler to do.