Bayesian spatial modelling of disease vector data on Danish farmland
Carsten Thure Kirkeby (Lecturer)
Activity: Talks and presentations › Conference presentations
Spatially distributed ecological count data, such as abundance of disease vectors, typically cannot be modelled statistically using the Gaussian framework but require a more elaborate approach such as the generalised linear geostatistical model (GLGM).
We applied the GLGM to a dataset with counts of Culicoides disease vectors around farms in Denmark. The GLGM assumes a Poisson distribution for the dependent variable whose intensity is governed by explanatory factors such as distance to farms, and by a spatially correlated stochastic signal. We calibrated the model on counts of traps catching disease vector in 25 locations in a grassland area over 25 consecutive nights. Separate models were fitted for all nights, resulting models compared and differences interpreted and related to temporal meteorological conditions (e.g. wind speed, temperature, rainfall and moisture). All statistical analyses were conducted using the geoRglm package in R.
Results show that one of the most important spatial explanatory factors for the abundance of Culicoides vectors is the distance to host animals. Comparisons of fitted models under different meteorological conditions indicated that the weather had a marked effect on the model parameters and spatial distribution of the disease vectors.
The GLGM used in this study provided a sound representation of the data and indicated that distance to host animals and meteorological conditions are important controlling factors. Further improvements are foreseen by incorporating both temporal and spatial factors in an integrated spatio-temporal GLGM.