Developing spatio-temporal prediction models for arbovirus activity in nothern Australia based on remotely sensed bioclimatic variables

Abstract

Vector-borne diseases pose an ongoing threat to public and animal health in the north ofAustralia. A number of surveillance programs are in place to determine the extent of virus activityand control the risk, but these are labour- and cost intensive while producing data with largetemporal and spatial gaps. Using the example of Bluetongue virus, the aim of this study was toinvestigate the potential of remotely sensed variables to facilitate the development of area-widepredictive models that complement traditional surveillance activities.Bioclimatic variables were derived for the Northern Territory from MODIS and TRMM remotesensing data products covering a period of nine years. Spatial and temporal uncertainty in thesurveillance data required the annual aggregation of environmental variables on a pastoralproperty level. Generalized Additive Models (GAM) were developed based on variables such asNDVI and land surface temperature to produce annual prediction maps of virus activity. Externalvalidation showed that the model correctly predicted 75% of the results from cattle stations testedfor Bluetongue. Remaining uncertainty in the model can be mainly attributed to the spatio-temporalinconsistency of the available surveillance data.This case study has developed a cost-effective approach based on a set of robustenvironmental predictors that facilitate the generation of arbovirus prediction maps soon after thepeak of risk for infection. While this research focused on Bluetongue Virus, we see a large potentialto expand the method to other areas and viruses particularly in view of the increasing populationsin Northern Australia.