Paul Harris and Chris Brunsdon ran a short workshop at the 7th Channel Network Conference on July 10th 2019 which was hosted at Rothamsted.
Spatial statistics is an ever-expanding discipline providing analytical techniques for a wide range of disciplines in the natural and social/economic sciences. In this workshop, we’ll outline techniques from a particular branch of spatial statistics, termed geographically weighted (GW) models. GW models suit situations when data are not described well by some global model, but where there are spatial regions where a localised calibration provides a better description. The approach is moving window based, where localised models are found at target locations calibrated with weighted data subsets. Outputs are mapped and spatially-interrogated to provide insight into the nature of the data’s spatial heterogeneity. Core GW techniques include: GW summary statistics, GW principal components analysis, GW regression, GW generalised linear models and GW discriminant analysis (Gollini et al. 2015; JSS 63(17):1:50). This workshop will focus on GW regression, illustrated with applications in agriculture.
At the beginning of December we uploaded an updated version of GWmodel to CRAN. This is version 2.0-1. While there have been some minor cosmetic changes the major improvement is that some of the more computationally intensive functions have been recoded in C++. This means that many GWmodel functions will now run noticeably faster than they did in previous versions. This is now the version that is installed by the install.packages() and update.packages() functions.
There have been two GW modelling workshops in the first 6 months of 2016.
The first took place at the University of Sheffield on March 22nd and 23rd. This Advanced Spatial Analysis workshop was organised by the Applied Quantitative Methods Network, and entitled “Modelling Spatial Heterogeneity with Geographically Weighted Models using R”. The course was run by Chris Brunsdon, Paul Harris and Martin Charlton. We covered geographically weighted summary statistics, geographically weighted regression, geographically weighted principal components analysis, and further issues in spatial models, including dealing collinear data using locally compensated models.
The second was hosted by the International Institute for Applied Systems Analysis (IIASA) at the Schloß Laxenburg, |Austria, on May 24th and 25th. We covered some fairly advanced material, including geographically weighted principal component analysis, REML models, GW discriminant analysis, and GW contingency tables, and some issues in spatial indexing to deal with big data. The workshop was run by Paul Harris, Chris Brunsdon and Martin Charlton, with enthusiastic participation from researchers at IIASA. Each of us probably ate our own body weight in Wiener Schnitzeln… We stayed in the Hotel Prinz Eugen, which is a short walk to the bus station at Wien Hbf, for the 30 minutes ride on the No 200 bus to Laxenburg.
We’ve had some server problems recently, so http://gwr.nuim.ie has been intermittent. This gave Chris Brunsdon and myself a chance to reflect on the nature of the website. A few changes are in hand.
We’re actually about the GWmodel R package, so we’ll change from Geographically Weighted Regression to Geographically Weighted Modelling in the not too distant future. The team that keeps all this in order is:
The Core Group
Chris Brunsdon: National Centre for Geocomputation, Maynooth University, Ireland
Martin Charlton: National Centre for Geocomputation, Maynooth University, Ireland
Paul Harris: Rothamsted Research, North Wyke, Devon, United Kingdom
Binbin Lu, School of Remote Sensing and Information Engineering, Wuhan University, China
Tomoki Nakaya, Department of Geography, Ritsumeikan University, Japan
They are assisted by a group of Associates:
Lex Comber, School of Geography, University of Leeds, United Kingdom
Richard Harris, School of Geographical Sciences, University of Bristol, United Kingdom
Guanpeng ‘Gavin’ Dong, Sheffield Methods Institute, University of Sheffield, United Kingdom
Chris Brunsdon and Martin Charlton attended the International Summer School on Spatial Structures and Dynamics, held at the Villa Finaly, Florence 14th-19th July 2014. Participants included 18 instructors together with 36 postgraduates and early career researchers. Chris presented the opening plenary, on Geocomputation and Social Science during the morning of Monday 14th July. In the afternoons of Monday and Tuesday we ran tutorials on Spatial Statistics to a smaller group of 18 students in each session. During each 3 hour tutorial the students were introduced to geographically weighted summary statistics, and geographically weighed regression, with an extensive practical based around GW summary statistics. The latest version of GWmodel and R was used for this, running on Mac and Windows laptops.
Participants in Monday afternoon’s tutorial
Participants in Tuesday afternoon’s tutorial
Chris presenting at the plenary on Monday morning.
The Summer school was organised by Arnoud Banos of the Laboratoire d’Excellence Dynamiques Territoriales et Spatiales (LabEx DynamiTe) which is part of the University of Paris 1 Sorbonne. There are some web pages about the workshop here; as well as some general information, there are photographs, videos, and some of the various presentations.
The Villa Finaly is a remarkable institution; originally built in the early 15th century, it was left to the Université de Paris in 1953 by the then owners, the Landau-Finaly family. In the 1990s the 13 Universities which now compose the Université de Paris, as common owners, decided to restore the building to something of its original splendour. More detail about the Villa is available.
An updated version of the GWR Windows application is now available for download. Version 4.08 incorporates a number of changes including: improvements to the fitting algorithm for Gaussian semiparametric models, some bug fixes, and a revision to the manual. Click on the GWR4 FOR WINDOWS link to find 32 bit and 64 bit versions of the software.