Uncertainty analysis of species distribution models

Area of Science:

• Ecology
• Biogeography
• Epidemiology

Background:

• Species distribution models (SDMs), such as the maximum entropy model, predict species' geographic ranges using occurrence data and environmental variables.
• Standard SDMs provide point estimates of species presence probability but lack inherent measures of uncertainty.
• Quantifying uncertainty is vital for reliable ecological predictions and public health applications, like disease vector mapping.

Purpose of the Study:

• To analytically derive the variance of maximum entropy model outputs from input data variance.
• To introduce a novel, computationally efficient method for assessing prediction uncertainty in SDMs.
• To apply the analytical method to estimate uncertainty in dengue importation probability and Aedes aegypti suitability.

Main Methods:

• Analytical derivation of the variance for maximum entropy model outputs based on input variable variance.
• Application of the analytical method to dengue occurrence and Aedes aegypti abundance data, incorporating demographic and environmental factors.
• Comparison of the analytical method's performance against the bootstrap method and Poisson point process model.

Main Results:

• The analytical method successfully derived the variance of maximum entropy model outputs, providing standard deviations for dengue importation probability and Aedes aegypti suitability.
• The analytical method demonstrated equivalence with the bootstrap and Poisson point process models under assumptions of independent point locations.
• The introduced analytical method is significantly faster than the bootstrap method for calculating uncertainty in maximum entropy models.

Conclusions:

• The developed analytical method provides a computationally efficient and accurate way to quantify uncertainty in maximum entropy SDMs.
• This approach enhances the reliability of species distribution predictions, particularly for applications in disease ecology and vector-borne disease risk assessment.
• The method offers a direct and rapid alternative to computationally intensive techniques like bootstrapping for uncertainty estimation.