D2.3 Biodiversity impact estimates and documentation for indicators on multi-dimensional biodiversity aspects ready for use in WP3

This deliverable presents several improvements to the biodiversity models when comparing to the deliverable 2.2. This includes increasing the searches in the random forest models to create the vegetation maps and the biodiversity dimensions, as well as formally testing a new methodology that minimizes the effects of sampling bias on biodiversity estimates. This improvement in the random forest method improved model prediction accuracy for species richness and endemism. Furthermore, the use of the new approach to minimize the effects of sampling bias proved to be very effective in generating more realistic results in richness and endemism models. Importantly, we use our models in this this deliverable to explore how anthropogenic impact variables were related to decreases in observed biodiversity. This information will be used to derive characterization factors for use in WP6. Specifically, we tested the variables: type of land use (urban areas, croplands, pastures, and native vegetation), cropland intensity, intensity of cattle harvesting and number of fire events. The results presented here represent an important advancement in biodiversity models by allowing predictions that are less influenced by sampling bias and more accurate. These results provide a valuable further step in describing human-induced changes in vegetation and biodiversity patterns across both South America and Africa. Factors such as croplands, livestock pastures and the intensity of these land uses are significantly related to biodiversity loss. The biodiversity impact estimates, and the documentation provided in this deliverable will provide direct inputs in upcoming CLEVER work in WP3 and WP6