Assessing uncertainty and performance of ensemble conservation planning strategies

Systematic conservation initiatives attempt to cater to the needs of many species via the integration of multiple species distribution models (SDMs), or via the integration of Systematic Conservation Planning (SCP) software, such as Zonation. Unfortunately, due to limited data and knowledge, it is often difficult to select the most suitable model for specific species, let alone an appropriate ensemble modeling method for multiple species. In general, model selection criteria are based on either model performance or consensus. The former integrates the highest-performing SDM for all focal species, whereas the latter integrates multiple SDM outputs based on consensus. While higher-performing ensemble models presumably identify high-quality habitats better, many have argued that high consensus ensemble models have less uncertainty originating from sporadic model variability. This study develops and validates seven ensemble-modeling strategies for integrating outputs of the systematic conservation tool Zonation. First, we considered the distributions of 11 bird species via 100 runs of five SDMs across Taiwan. Second, we evaluated the local and global uncertainty of all five models. Third, we used Zonation to obtain conservation priorities. We then used Principal Component Analysis (PCA) to quantify different sources of uncertainty. Finally, we used independent third-party habitat data to validate each strategy. On average, the 'best model' strategy (based on the highest AUC value) performed best. Based on our modeling exercise we present a comprehensive framework for conservation prioritization, validation and the quantification of uncertainty intrinsic to SDMs according to different conservation scenarios and goals.