In many ways, geographically weighted regression (GWR), which is a multivariate regression technique performed at the kernel scale, is an ideal method to evaluate relationships among spatially nonstationary variables but the results are notoriously unstable. Typical multivariate regression techniques are often applied to the whole dataset, which results in global estimates of parameter relationships or necessitates manual subsetting of the data to study local relationships. We present a modified application of GWR executed in a Monte Carlo style. Our implementation of GWR uses random subsets of the spatial data to estimate parameter coefficients and study their distribution through multiple iterations. This technique provides more robust estimates of parameter coefficients. Here, we discuss an application of this methodology to a channel shape prediction problem in the Gulf of Alaska where we rank the influence of several competing processes on a measured parameter.


Follow us on Social Media!
Twitter:   / esri  
Facebook:   / esrigis  
LinkedIn:   / esri  
Instagram:   / esrigram  

The Science of Where: http://www.esri.com