Managing multiple uncertainties in species distribution modelling

Aim

We present a methodology to address multifaceted uncertainties in species distribution models (SDMs), enhancing their robustness and providing vital insights to inform management and conservation decisions. Data uncertainties, including positional inaccuracies in historical data and absences in survey data that could be attributed to anthropogenic disturbances rather than habitat unsuitability, can compromise SDM predictions, risking the efficacy of resultant conservation strategies.

Location

While the concepts and methodologies presented hold global applicability, our case study is situated in and around the Negev Desert of southern Israel and the Palestinian West Bank, focusing on the critically endangered Be'er Sheva fringe-fingered lizard (Acanthodactylus beershebensis) that is endemic to this area.

Methods

Utilizing calculated combinations of reliable and uncertain datasets, we created diverse dataset scenarios. Pre-development distribution and habitat requirements were estimated for each scenario using a blend of statistical and machine-learning algorithms in R. Additionally, a combined scenario was modelled using hierarchical model ensembles to effectively weight data by reliability.

Results

Our innovative approach produces more robust models and reveals the impact of uncertain datasets on model predictions. Incorporating potential anthropogenic absences led to erroneous model conclusions, particularly when historical data exclusion occurred—a practice often implemented in the pursuit of model robustness.

Main Conclusions

Uncertainties in SDMs can yield incorrect conclusions, imperilling conservation efforts. Initiated by land managers, our work actively informs conservation practices. The study's global relevance provides an approach for addressing real-world challenges in estimating species distributions, advancing the application of conservation science.