Representation of animal distributions in space: how geostatistical estimates impact simulation modeling of foot-and-mouth disease spread

Abstract

Modeling potential disease spread in wildlife populations is important for predicting, responding to and recovering from a foreign animal disease incursion. To make spatial epidemic predictions, the target animal species of interest must first be represented in space. We conducted a series of simulation experiments to determine how estimates of the spatial distribution of white-tailed deer impact the predicted magnitude and distribution of foot-and-mouth disease (FMD) outbreaks. Outbreaks were simulated using a susceptible-infected-recovered geographic automata model. The study region was a 9-county area ( 24 000 km(2)) of southern Texas. Methods used for creating deer distributions included dasymetric mapping, kriging and remotely sensed image analysis. The magnitudes and distributions of the predicted outbreaks were evaluated by comparing the median number of deer infected and median area affected ( km2), respectively. The methods were further evaluated for similar predictive power by comparing the model predicted outputs with unweighted pair group method with arithmetic mean (UPGMA) clustering. There were significant differences in the estimated number of deer in the study region, based on the geostatistical estimation procedure used ( range: 385 939 - 768 493). There were also substantial differences in the predicted magnitude of the FMD outbreaks ( range: 1 563 - 8 896) and land area affected ( range: 56 - 447 km2) for the different estimated animal distributions. UPGMA clustering indicated there were two main groups of distributions, and one outlier. We recommend that one distribution from each of these two groups be used to model the range of possible outbreaks. Methods included in cluster 1 ( such as county-level disaggregation) could be used in conjunction with any of the methods in cluster 2, which included kriging, NDVI split by ecoregion, or disaggregation at the regional level, to represent the variability in the model predicted outbreak distributions.