Incorporating a local-statistics-based spatial weight matrix into a spatial regression model to predict the distribution of invasive Rosa multiflora in the Upper Midwest

Abstract

In this study, we extended the spatial weight matrix defined by Getis and Aldstadt (2004) to a more general case to map the distribution of Rosa multiflora, an invasive shrub, across the Upper Midwest counties in a spatial lag model (SLM) context. Both the simulation study and the application to the invasion data of invasive Rosa multiflora collected in 2005-2006 proved that the modified spatial weight matrix outperforms its original case and the contiguity- and nearest distance- based spatial weight matrices in diagnostic statistics and resultant invasion maps. The geographical distribution of Rosa multiflora in the Upper Midwest was significantly associated with latitude; local clusters (groups of counties) of high abundance/presence of Rosa multiflora were significantly determined by TRPF (a ratio of road density to percentage of forest cover at the county level), a variable reflecting the intensity of human disturbance. Both the multiple linear regression model and the SLM models with the original spatial weight matrix and contiguity- and nearest distance- based spatial weight matrices incorporated tended to underestimate the effect of forest type (community) on multiflora rose. As a conclusion, the SLM model incorporating the modified spatial weight matrix has potential applications in mapping spatial data with strong clustering patterns and estimating spatial autocorrelation structure and covariate effects in ecological studies.

Citation

Yu, Weiming; Fan, Zhaofei; Moser, W. Keith. 2017. Incorporating a local-statistics-based spatial weight matrix into a spatial regression model to predict the distribution of invasive Rosa multiflora in the Upper Midwest. Mathematical and Computational Forestry and Natural-Resource Sciences. 9(2): 17-29.