Drivers of systematic bias in alien plant species distribution data

Abstract

Among the main challenges in modelling biological invasion is a lack of valid data on the absence of invasive species. Absence data are important for assessing the reliability of models, but multiple surveys at a location are needed. In practice, omission errors are more frequent than commission errors. We therefore quantified how eliminating potentially biased areas from invasive species distribution models (iSDMs) affected the models' performance, and we assessed how the distribution of biased areas correlated with environmental factors. We hypothesized that for neophytes, the distribution of biased areas corresponds to specific land relief and/or particular landscape and land use, but not the density of roads and urbanized areas. The data on neophytes were obtained from a distribution atlas covering approximately 31,000 km² in Central Europe overlaid with a 2 × 2 km square grid. One hundred fifty-three species were used for modelling neophyte richness, and negative residuals from the model were assumed to indicate biased squares. Twenty invasive species were used as an independent dataset for testing the effect of excluding the biased squares on iSDM performance. The exclusion of biased squares increased the iSDM performance from an area under the curve value of 0.73 to 0.78. The best results were obtained by excluding 30 % of the squares from the original dataset. The presence of damp sites explained the distribution of biased squares; the density of roads and urbanized areas had no impact. The applied method allows distinguishing biased, plausibly undersampled squares in a species distribution atlas, the exclusion of which significantly improves iSDM performance. The results suggest that the commonly observed low sampling effort in areas distant from communication routes and urbanized areas was not crucial in modelling invasive species distribution, which can be related to smaller neophyte richness in remote areas resulting from low propagule pressure.

Keywords: Invasive species distribution atlases; Invasive species modelling; Sampling bias; Sampling effort.