Invasive mosquito species and the pathogens they transmit represent a serious health risk to both humans and animals. Thus, predictions on their potential geographic distribution are urgently needed. In the case of a recently invaded region, only a small number of occurrence data is typically available for analysis, and absence data are not reliable. To overcome this problem, we have tested whether it is possible to determine the climatic ecological niche of an invasive mosquito species by using both the occurrence data of other, native species and machine learning. The approach is based on a support vector machine and in this scenario applied to the Asian bush mosquito (Aedes japonicus japonicus) in Germany. Presence data for this species (recorded in the Germany since 2008) as well as for three native mosquito species were used to model the potential distribution of the invasive species. We trained the model with data collected from 2011 to 2014 and compared our predicted occurrence probabilities for 2015 with observations found in the field throughout 2015 to evaluate our approach. The prediction map showed a high degree of concordance with the field data. We applied the model to medium climate conditions at an early stage of the invasion (2011-2015), and developed an explanation for declining population densities in an area in northern Germany. In addition to the already known distribution areas, our model also indicates a possible spread to Saarland, southwestern Rhineland-Palatinate and in 2015 to southern Bavaria, where the species is now being increasingly detected. However, there is also evidence that the possible distribution area under the mean climate conditions was underestimated.
Keywords: Citizen science; Invasive species distribution models; Machine learning; Occurrence probability; Support vector machine.