Accounting for spatial and environmental sampling bias in a species distribution model of Aedes vexans (Diptera: Culicidae)

Abstract

Abstract This study evaluates the potential of species distribution models to predict habitat suitability for the floodplain mosquito Aedes vexans , with a particular focus on ephemeral breeding habitats such as flood-prone areas. These habitats are essential for oviposition, yet have not been explicitly incorporated as predictors in existing modelling approaches. Flood hazard was included as an environmental predictor to represent the species’ dependence on transient water bodies. However, initial model runs showed only a marginal contribution of this predictor to overall model performance. We demonstrate that this weak contribution was not due to the ecological irrelevance of ephemeral habitats, but instead resulted from strong sampling biases in the occurrence data, which were predominantly concentrated in urban environments. To address this issue, we applied spatial thinning and environmental filtering to reduce spatial clustering and the overrepresentation of environmentally similar regions. Using Maxent, we compared candidate models with and without bias correction across multiple environmentally stratified test datasets. Our results show that correcting for sampling bias improved model accuracy more than the inclusion of an additional habitat-relevant predictor alone. The final model produced more reliable suitability predictions, particularly in flood-prone and previously under-sampled areas.