Assessment of bias and uncertainty of species occurrence data: metrics and methods

The increasing availability of large quantities of data on species occurrence (i.e., presence and/or presence and absence of species) to support biodiversity studies and conservation actions is not always coupled with the data quality. Indeed, species occurrence data can present forms of bias (i.e., systematic deviation from the true value) and uncertainty (i.e., dispersion of values or lack of knowledge). In the PhD thesis, I evaluated different metrics and methods to address uncertainty and bias of species occurrence data. In the first chapter, the effect of the sampling method of the presences and absences and the effect of their ratio (i.e., sample prevalence) were tested on different Species Distribution Models: Favourability and Probability-based SDMs. In the second and third chapters, I employed various metrics to evaluate the quality of species occurrence data stored in a case study database, namely the sPlotOpen database. The taxonomic, spatial and temporal bias were measured at European and habitat levels respectively with i) the completeness of the species richness, ii) the Nearest Neighbor Index, iii) the Pielou's index. Besides, the temporal uncertainty—defined as the information decay of the species occurrence over time—was quantified using a negative exponential function. Among the main results, I found that the sampling methods (i.e., random and stratified sampling) of the species occurrences had no effect on the performance of the Favourability and Probability models. The Favourability model, in contrast, exhibited lower variability and only slightly higher accuracy than Probability in the predictions of species distribution. Moreover, the metrics used to assess the dimensions of bias in species occurrence data proved to be effective, revealing heterogeneous patterns. Additionally, the analysis of temporal uncertainty identified hotspot areas across Europe. Results that highlighted the necessity of assessing data quality prior to its use in biodiversity inferences.