Computational Improvements to Multiscale Geographically Weighted Regression

Keywords: multi scale, geographically weighted regression, multiscale, local modelling, spatial analysis, parallel computing.
Session Type: Virtual Paper
Day: Wednesday
Session Start / End Time: 4/7/2021 08:00 AM (Pacific Time (US & Canada)) - 4/7/2021 09:15 AM (Pacific Time (US & Canada))
Room: Virtual 18

Abstract

Geographically Weighted Regression (GWR) has been broadly used in various fields to model spatially non-stationary relationships. Multiscale Geographically Weighted Regression (MGWR) is a recent advancement to the classic GWR model. MGWR is superior in capturing multi-scale processes over the traditional single-scale GWR model by using different bandwidths for each covariate. However, the multi-scale property of MGWR brings additional computation costs. The calibration process of MGWR involves iterative back-fitting under the additive model (AM) framework. Currently, MGWR can only be applied on small datasets within a tolerable time and is prohibitively time-consuming to run with moderately large datasets (greater than 5,000 observations). In this paper, we propose a parallel implementation that reduces both memory footprint and runtime (up-to hundreds of times speed-up) to allow MGWR modelling to be applied to large datasets. This development enhances the accessibility of MGWR for new applications to explore multi-scale spatial heterogeneity but also brings the possibility of much larger scale local multi-scale analysis. The method introduced in this paper have been integrated into the mgwr python package and the MGWR 2.0 software, both of which are open-source and widely distributed.