Spatial and temporal analysis of hemorrhagic disease in white-tailed deer in Southeast USA

Abstract

Hemorrhagic disease (HD) is a common disease transmitted by Culicoides midges in white-tailed deer that has caused significant mortality, mobidity and economic impact on recreational hunting throughout the United States. This dissertation provides a statistical analysis of the county-based spatial-temporal prediction and cluster detection of HD in white-tailed deer in southeast USA. A spatial-temporal prediction model was constructed to predict HD occurrence from 1982 to 2000. Eleven principal factors which were reduced from 42 climatic and environmental factors derived from ground-based weather stations and remotely-sensed data were used as predictor variables and HD presence/absence data for each county in the study area as the dependent variable. A generalized linear mixed logistic model was used to consider the within-subject effect of the longitudinal data. A spatial dependency term was added to the model accounting for the influence of HD occurrence of adjacent counties on a particular county. The results show that wind speed, rainfall, land surface temperature and normalized difference vegetation index (NDVI) are significant factors in predicting HD occurrence. The total prediction accuracy is 65 percent when all four factors are considered for a five state area. The prediction accuracy for individual years ranges from 27 percent to 96 percent. Remotely-sensed data prove to be informative and results in a higher prediction power than some climatic data. Kulldorff’s space-time scan statistic was applied to detect the spatial and space-time clusters in HD from 1980 to 2003. The results indicate that western and southern portions of Alabama, south of Alabama, central South Carolina, and the boundary between South Carolina and North Carolina are areas where high rate clusters of HD outbreaks occur. A maximum spatial window of 10 percent of the total population and a maximum temporal window of 25 percent of the study period are believed to be appropriate windows that include most of the clusters without leaving out subclusters. NDVI, wind speed and spatial dependency were found to be related to the HD clustering. Future study with the integration of statistical, biological, geographical information system, and remote sensing information is expected to result in a more thorough understanding of this wide spread and economically influential wildlife disease.