Dealing with noisy absences to optimize species distribution models: an iterative ensemble modelling approach

Area of Science:

• Ecology
• Conservation Biology
• Computational Biology

Background:

• Species distribution models (SDMs) are crucial tools in ecology and conservation.
• Model accuracy is often compromised by non-environmental (noisy) absence data.
• Existing methods struggle to effectively address these data limitations.

Purpose of the Study:

• To introduce an Iterative Ensemble Modeling (IEM) approach to improve SDM reliability.
• To address the challenge of noisy absences in species occurrence datasets.
• To enhance the predictive accuracy of ensemble modeling.

Main Methods:

• Developed an iterative ensemble modeling (IEM) approach.
• Used outputs from classical ensemble models (EM) to update raw occurrence data iteratively.
• Stabilized predictions by repeating the EM process until convergence.
• Validated IEM using virtual species and compared it against classical EM.

Main Results:

• IEM demonstrated rapid convergence and increased consensus among different model predictions and datasets.
• Significantly improved prediction reliability (Kappa, TSS, well-predicted sites) with high levels of noisy absences compared to EM.
• Reduced biases in species prevalence estimates.
• Effectively handled noisy absences by simulating presences during iteration.

Conclusions:

• IEM enhances the reliability and accuracy of species distribution models.
• The method is particularly effective for difficult-to-detect species or those not in environmental equilibrium.
• IEM offers a promising solution for improving SDM predictions with imperfect occurrence data.