Accuracy of climate-based forecasts of pathogen spread

Area of Science:

• Ecology
• Epidemiology
• Computational Biology

Background:

• Species distribution models (SDMs) are crucial for predicting pathogen spread.
• Emerging pathogens often exist in non-equilibrium states, challenging traditional SDM assumptions.
• Accurate modeling is vital for managing emerging infectious diseases in wildlife.

Purpose of the Study:

• To evaluate the performance of different SDM approaches for non-equilibrium pathogens.
• To identify optimal modeling strategies for predicting the spread of emerging infectious diseases.
• To assess the impact of data time-series length on predictive accuracy.

Main Methods:

• Utilized time-incremented data subsets for training and testing SDMs.
• Applied multiple SDM algorithms, including boosted regression trees, random forests, and MaxEnt.
• Evaluated model performance using metrics such as AUC, kappa, false negative rate, and Boyce index.

Main Results:

• Boosted regression trees and random forests demonstrated superior predictive performance.
• MaxEnt also showed strong results, closely following the top-performing models.
• Model predictive accuracy generally increased with longer time-series data used for training.

Conclusions:

• Specific SDM algorithms (boosted regression trees, random forests) are effective for modeling emerging pathogens.
• The duration of the time-series data significantly influences the accuracy of pathogen spread predictions.
• These findings aid in rapidly determining the potential range of emerging diseases and selecting appropriate models for early-phase management.