Spatio-temporal modelling of bluetongue virus distribution in Northern Australia based on remotely sensed bioclimatic variables

Abstract

The presence of Bluetongue virus (BTV) in Northern Australia poses an ongoing threat for animal health and although clinical disease has not been detected in livestock, it limits export of livestock from the infected areas. BTV presence is governed by variable environmental conditions, which influence vector and host habitats. The National Arbovirus Monitoring Program (NAMP) was established to determine the extent of virus activity and control the risk of infection spread. Groups of young cattle, previously unexposed to infection, are regularly tested to detect evidence of transmission. This approach is labour and cost intensive and difficult to operate in the remote areas of Northern Australia. The resulting data are therefore characterised by spatial and temporal gaps. The aim of this research is to assess the use of remotely sensed environmental and climatic data as a means of predicting the distribution of BTV seroprevalence throughout Northern Australia to complement conventional surveillance.Environmental factors relating to the viruses’ host and vector habitats and the transmission cycle of BTV have been identified based on the extensive review of virus ecology. Different data sources have been assessed to provide sufficient spatial and temporal coverage for the definition of spatio-temporal environmental variables that can be used to explain and predict the distribution of BTV. Following this assessment, satellite data products from the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Tropical Rainfall Measuring Mission (TRMM) were acquired for the Pilbara in Western Australia, and the Northern Territory. These were reprojected and processed into spatio-temporal variables for the period between the years 2000 and 2009. Due to uncertainty in the precision of the geographic location and timing of animals tested for seropositivity, summary statistics of bioclimatic variables were generated at the station (i.e. property) level for each year. Different combinations of these variables, including vegetation greenness and phenology, land surface temperature and precipitation were screened for correlation with BTV presence using a Generalised Additive Model approach. A final model was developed to predict the presence or absence of BTV seropositivity on the basis of statistical significance of the remotely sensed predictor variables, and informed by knowledge of virus ecological principles.The model, based on the maximum seasonal Normalised Difference Vegetation Index (NDVI), and mean and maximum land surface temperature variables provided excellent discriminatory ability and the basis for the generation of prediction maps of BTV seropositivity for the first eight years. Besides internal assessment, the model’s predictive capabilities were validated using monitoring data from the season 2008/09.It has been demonstrated that the predictions are useful in complementing complement NAMP surveillance by identifying areas at higher risk for seropositivity in cattle, which aids planning of livestock movement and further monitoring activities. Uncertainty in the model was attributed to the spatio-temporal inconsistency in the precision of the available serosurveillance data. The discriminatory ability of models of this type could be further improved by ensuring that exact location details and date of NAMP BTV test events are consistently recorded.