Projet de recherche S0/09/033 (Action de recherche S0)
Context and objectives
Bluetongue is a non-contagious infectious disease transmitted by biting midges (Culicoides species). The main vector is Culicoides imicola but other species belong the Culicoides obsoletus complex. The disease causes high mortality in sheep, whilst cattle are host but rarely show symptoms for the serotypes found in the Mediterrenean basin. The midges are very small and are suspected to spread over long distances by wind. This hypothesis has mainly been tested qualitatively and not quantitatively with the exception of Bishop (2001) and Alba et al (2004).
The objectives of the projects were twofold. The first objective was to determine the distribution of the possible vectors based on modelling. The aim was to contrast several modelling techniques as well as to determine what the minimum sampling size necessary for modelling should be.
The second objective was to characterise the possible spread of the vector by wind on a quantitative basis.
Methodology
• In order to determine the distribution of the possible vectors, three different modelling techniques were used: logistic regression, artificial neural networks and ecological niche factor analysis. Several predictor variables were derived from a one-year time series of MODIS data. This MODIS data was quality checked prior to modelling. In a first step, all predictor variables were used ‘blindly’ without further regard as to correlation. Secondly only non-correlated variables were used. All results were compared based on the area under curve value of the receiver operating characteristic. A split sample approach was used: a training set was used to build the model and an independent test set was used to validate the outcome. For each split rate, modelling technique and correlated/non correlated approach the procedure was repeated thirty times. The mean model as well as the standard deviation was determined.
• Horizontal isentropic wind trajectories were calculated based on the procedure by Codina (2001). For each outbreak point forward and backward trajectories were determined at 4 different pressure levels as well as at 10m height 4 times per day. They were calculated starting from 8 days prior to the day of outbreak until 6 days after the outbreak.
Results
Artificial neural networks are significantly better than LR models, especially when the correlation between the variables is not taken into account. ENFA gives much more detail in the areas where the habitat is more suitable for the vector than in the areas where the habitat is found to be unsuitable. The models are well suited within the study area but spurious error occur outside the study area. LR/ANN models require absence data in order to function and the artificial creation of these negative points based on literature introduced a bias. However, when this was compared to the outcome of the ENFA approach, which does not need absence data, this bias tended to be low.
The wind trajectories show a high correlation with the different stages in the epidemics in Greece and Bulgaria. For each of the three years, we could find that based on the different serotypes of bluetongue, underlying wind patterns might explain the evolution of the epidemics.
Products and services
Based on the results of the wind trajectories, a service has been developed to integrate wind trajectory modelling in the analysis of bluetongue epidemics. This service is validated in the current epidemic in Belgium, the Netherlands and Germany.
