A Novel Method to Handle the Effect of Uneven Sampling Effort in Biodiversity Databases
Figure 2
Kernel density plots of predicted probabilities of discrimination between well (dashed line) and poorly sampled units (continuous line) for NPE, STE and FIDEGAM methods.
In the scenarios of high (A) and low (B) levels of sampling exhaustiveness, the sampling units were categorized as well and poorly sampled according to the number of records (see Appendix S1), whereas, when the true richness was known (C), the true sampling completeness (see equation 4 on text) was used as a categorization criterion. Probabilities were calculated according to ROC-GLM regression models.