A Bayesian maximum entropy model for predicting tsetse ecological distributions.

Area of Science:

• Ecology
• Epidemiology
• Geospatial Analysis

Background:

• African trypanosomiasis is a tsetse-borne disease impacting humans and animals across Sub-Saharan Africa.
• Accurate spatial and temporal understanding of tsetse fly habitats is crucial for disease surveillance and risk management.
• Existing remote sensing data present temporal lags and coarse resolution, hindering effective disease control modeling.

Purpose of the Study:

• To develop a heuristic for identifying fine-resolution tsetse habitats in future time periods and data gaps.
• To mitigate the temporal lag issue in remote sensing data for trypanosomiasis control.
• To provide a scalable and open-access model for predicting tsetse distributions.

Main Methods:

• Introduced a generalizable, open-access Tsetse Ecological Distribution (TED) model.
• Developed a geospatial Bayesian Maximum Entropy (BME) prediction model trained on TED output data.
• Utilized cluster and parallel computing with Monte Carlo analysis for optimizing BME computations on a large dataset (over 2 billion data points).

Main Results:

• The BME kriging analysis achieved 74.8% prediction accuracy for maximum suitability extent in Kenya.
• The BME kriging analysis demonstrated 97% accuracy in forecasting tsetse distribution across Kenya.
• The study successfully analyzed a large dataset at a finer resolution and larger spatiotemporal scale than previously possible.

Conclusions:

• The BME model provides a reliable solution for forecasting future tsetse distributions, enabling proactive control strategies.
• This approach addresses the temporal data gap in rainfall predictions and delayed remote sensing data processing.
• Open-sourced GEE-TED and BME libraries promote reproducibility and future updates with new data.