Multi-Scenario Species Distribution Modeling.

Area of Science:

• Ecology
• Computational Biology
• Environmental Science

Background:

• Correlative species distribution models (SDMs) are vital for predicting insect habitats.
• Discrepancies and uncertainty in SDM predictions are significant concerns in ecological research.

Purpose of the Study:

• To investigate how different modeling components affect SDM prediction accuracy.
• To identify sources of discrepancy and uncertainty in SDM predictions.

Main Methods:

• Factorial study analyzing species-training-datasets, predictor variables, dimension-reduction methods, and model types.
• Evaluation of model performance using various metrics.

Main Results:

• Model type was the primary driver of variation in species distribution predictions.
• Machine-learning models generally outperformed other types when components were constant.
• Hierarchical Non-Linear Principal Components Analysis improved model performance.
• Standard performance indices may not reliably identify the best model if performance variation is small.

Conclusions:

• Discrepancies in SDM results can stem from component selection, not necessarily a lack of robustness.
• Robust model evaluation methods are needed.
• Multi-scenario modeling can mitigate errors and improve understanding of prediction uncertainty.