Published April 3, 2022 | Version v1

Multiscale GWR needs looking at: searching, looking for LOVE (low optimal value evaluations)

Description

GWR is a commonly used technique for understanding process spatial heterogeneity. Recently Multi- Scale GWR (MS-GWR) has been proposed as the default GWR as it allows the scale of individual response-to-predictor relationships to vary spatially, rather than a “best on average” scale with standard GWR. This work describes problems with the back-fitting algorithms used to search through the solution-space used in all MS-GWR implementations. It compares the results of these with a linear search and shows current MS-GWR searches to be sub-optimal. gwverse proposes a new modular structure for GWR that incorporates a function factory approach. A number of areas of future work are described in order to create a new MS-GWR module in gwverse.

Files

GISRUK_2022_paper_65.pdf

Files (248.2 kB)

Name Size
md5:0dc8a5d35c66e6a712185c6607c29e7c
248.2 kB Preview Download