Impact of biased sampling effort and spatial uncertainty of locations on models of plant invasion patterns in Croatia

Very frequently biological databases are used for analysing distribution of different taxa. These databases are usually the result of variable sampling effort and location uncertainty. The aim of this study was to test the influence of geographically biased sampling effort and spatial uncertainty of locations on models of species richness. For this purpose, we assessed the pattern of invasive alien plants in Croatia using the Flora Croatica Database. The procedure applied in testing of the sensitivity of models consisted of sample area sectioning into coherent ecological classes (hereinafter Gower classes). The quadrants were then ranked based on sampling effort per class. This resulted in creation of models using varying numbers of quadrants whose performance was tested with independent validation points. From this the best fitting model was determined, as well as a threshold of sampling effort. The data from quadrants with sampling effort below the threshold were considered too unreliable for modelling. Further, spatial uncertainty was simulated by adding a random term to each location and re-running the models using the simulated locations. Biased sampling effort and spatial uncertainty of locations had similar effects on model performance in terms of the magnitude of the affected area, as in both cases 7% of the quadrants showed statistically significant deviations in alien plant species richness. The model using only on the quandrants with the highest 35% quantile sampling effort best balanced the sampling effort per quadrant and overall geographical coverage. It predicted a mean number of 3.2 invasive alien plant species per quadrant for the Alpine region, 5.2 for the Continental, 6.1 for the Mediterranean and 5.3 for the Pannonian region of Croatia. Thus, the observational databases can be considered as a reliable source for species richness models and, most likely, for other types of species distribution models, given that their limitations are accounted for in the data selection process. In order to obtain precise estimates of species richness it is required to sample the whole range of ecological conditions of the study area.