The use of habitat analysis in the control of wildlife tuberculosis in New Zealand

A case-control study of 132 possum capture sites was used to identify habitat factors associated with the probability of capturing TB possums. Two unconditional logistic regression models were produced. Slope of the land and factors associated with the quality of dens were associated with the probability of capturing multiple TB possums. Slope of the land, height of trees, percent cover of vegetation in the height range 0.3 – 2 metres and percent of the ground covered in vascular vegetation were associated with the probability of capturing single TB possums. A second case-control study at the larger scale of farms was conducted to identify possum habitat, geographic and farm factors associated with the probability of tuberculous possums being present on a farm, using random effects logistic regression modelling. Variables were included in the model to account for spatial autocorrelation patterns in the data and a variogram analysis was conducted on residuals to test for any significant spatial autocorrelation that was not accounted for in the model. A geographic information system (GIS) was used to combine and analyse digital geographic data sets to produce variables for the statistical analysis. Fragstats, a habitat pattern analysis programme, was used to produce variables representing different features of possum habitat on farms. Distance from the coastal forest, total area of the farm, area of pine forest, area of manuka/pasture and proportion of the farm covered with podocarp/broadleaved species were the significant variables in the final model. An automated classification procedure was used to generate a vegetation map from a SPOT3 multi-spectral satellite image for this study. The resulting raster map comprised 8 vegetation classes, including 3 tree classes, 4 scrub classes and a pasture class. This map had a high degree of spatial accuracy and more detailed vegetation classes than existing digital data available in New Zealand. However, some categories contained a heterogeneous range of plant species which limited the discriminatory power of models using this vegetation data. Spatial clustering of percent TB-positive years, five-year cumulative incidence and space-time clustering of annual cumulative incidence of TB in cattle were analysed for farms in an area where possums had been infected with TB for more than 30 years, using spatial and spacetime scan statistics. A comparison of farms included in the most significant high-incidence rate and low-incidence rate clusters for both percent TB-positive years and the five-year