Usutu virus induced mass mortalities of songbirds in Central Europe: Are habitat models suitable to predict dead birds in unsampled regions?

Abstract

The Usutu virus (USUV) is a mosquito-borne flavivirus closely related to the better known West Nile virus, and it can cause mass mortalities of song birds. In the present paper, a dataset of georeferenced locations of USUV-positive birds was compiled and then used to map the geographical distribution of suitable USUV habitats in Central Europe. Six habitat models, comprising BIOCLIM, DOMAIN, maximum entropy model (MAXENT), generalized linear model (GLM), boosted regression trees model (BRT), and random forests model (RF), were selected and tested for their performance ability to predict cases of disease in unsampled areas. Suitability index maps, a diagram depicting model performance by the Area Under the Curve (AUC) vs. the True Skill Statistic (TSS), and a diagram ranking sensitivity vs. specificity as well as correct classification ratio (CCR) vs. misclassification ratio (MCR) were presented. Of the models tested GLM, BRT, RF, and MAXENT were shown suitable to predict USUV-positive dead birds in unsampled regions, with BRT the highest predictive accuracy (AUC = 0.75, TSS = 0.50). However, the four models classified major parts of the model domain as USUV-suitable, although USUV was never confirmed there so far (MCR=0.49 to 0.61). DOMAIN and especially BIOCLIM can only be recommended for interpolating point observations to raster files, i.e. for analyzing observed USUV distributions (MCR = 0.10). Habitat models can be a helpful tool for informing veterinary authorities about the possible distribution of a given mosquito-borne disease. Nevertheless, it should be taken in consideration, that the spatial and temporal scales, the selection of an appropriate model, the availability of significant predictive variables as well as the representativeness and completeness of collected species or disease cases may strongly influence the modeling results.

Keywords: Bioclimatic variables; Disease mapping; Model comparison; Species distribution; Suitability index; Vector-borne.