D3121: Improving methods for proximity analysis in spatial epidemiology

Spatial configuration affects processes occurring in natural and managed landscapes. Pest dispersal and vectored pathogen transmission are distance-dependent processes: as in Tobler’s first law of geography, points nearer in space have greater influence on each other. Experimental studies of dispersal and vectoring are limited by complexity of landscapes. Observational studies are limited by analytical methods that must be improved to better represent distance-dependent effects. The analytical practice of calculating proportional composition of landscape variables in concentric rings around a point of interest is problematic for two reasons: 1) it does not correct for the relationship between ring distance and area, and 2) it requires arbitrary quantization of distance. I explore consequences of these issues using simulation and model recovery analyses, and develop an improved method for estimating and ranking influence of cover types in realistically heterogeneous landscapes. This improved proximity analysis method does not rely on arbitrary distance categories, thus negating the need to correct for the area of circles involved in correlative analysis. I show that the distance/area relationship causes biased estimates when using proportional ring composition as an input variable for proximity analysis. The simulation generates fields of a prescribed size and autocorrelation according to given relationship parameters, and the performance of different analytical techniques in recovering these parameter values is then compared. The objective of this work is to improve the accuracy of distance-weighting parameter estimation, to increase the utility of predictive models of pest abundance and pathogen transmission when used to generate projections for heterogeneous landscapes.