DefinePK

Abstract

The development of local forms of spatial analysis has been the subject of intense research over last decade. The condition that observations are independent is not frequent in spatial data analysis. Parameter drift is often recognized when a model is calibrated separately across locations in space. In this study, we propose a local calibration procedure for handling varying parameters estimates of an origin-constrained spatial interaction model. In this context, the estimates of local parameters depend both on origins and destinations and a four dimensional space is involved. A suite of local parameters can be obtained by the maximisation of a weighted maximum likelihood function, exploiting the same principle of Geographically Weighted Regression (GWR) approach. Generally, the function for the weighting scheme in GWR uses only distance (geographical space) to determine weights. We propose a modified version of weighting function which takes into account both the spatial distance and a function of strength of connection between two specific destinations. The performance of different weighting functions for modelling the effects of spatial heterogeneity on flow data are compared and evaluated.