High-Resolution Mapping of Environmental Suitability for Reported Ebola Spillover Events in Uganda to Inform Surveillance and Preparedness.

Abstract

Background: Uganda experiences recurrent Ebola disease (EBOD) outbreaks, yet national suitability maps often pool virus species or are too coarse for district-level planning. We aimed to generate species-resolved, 1 km relative environmental suitability surfaces for reported Ebola spillover events in Uganda; quantify the contribution of environmental, anthropogenic, and reservoir-related predictors to modeled suitability; and translate continuous outputs into operational prioritization tiers for district surveillance and preparedness.

Methods: We compiled 71 laboratory-confirmed reported Ebola event localities in Uganda from 2000 to 2023, defined as the most likely documented locality of initial human infection where available, and paired them with 11 minimally collinear predictors. Pooled and virus-specific MaxEnt models were tuned in ENMeval by comparing feature-class sets across β = 0.5 to 3.0, ranked by AICc, and screened for acceptable omission at the 10% training-presence threshold. Performance was estimated using fourfold spatial block cross-validation. Pooled cloglog suitability values were thresholded at the 10% training-presence value and stratified into four operational prioritization tiers.

Results: Tuning selected LQHP with β = 1.5 for the pooled model. Under spatial block cross-validation, pooled performance was Test AUC = 0.845 (0.825 to 0.862) and Test Boyce = 0.58 (0.52 to 0.63), with Test OR10 = 0.101 (0.092 to 0.109). The operational threshold was cloglog = 0.084, classifying 25.4% of Uganda as suitable within the modeling framework. Tier 1, defined as the highest-suitability decile among suitable pixels (cloglog ≥ 0.65), covered 5.8% of land area yet contained approximately 12.6 million residents (29%). High-suitability areas were clustered in the Albertine Rift escarpment belt, the Kampala, Wakiso, Mukono peri-urban cluster extending along major corridors, and a northern Lake Kyoga shoreline crescent. Species-specific outputs were interpreted as exploratory secondary analyses. SUDV broadly reproduced the pooled modeled suitability geography, EBOV showed elevated suitability along the western border corridor, and BDBV suitability was concentrated around the Albertine escarpment. Checkerboard2 sensitivity produced a small AUC increase (ΔAUC = +0.006) and high tier agreement.

Conclusions: Relative environmental suitability for reported Ebola spillover localities in Uganda is concentrated in compact high-priority belts where elevated modeled suitability overlaps with large resident populations. Tier 1 and Tier 2 areas provide a focused basis for strengthening event-based surveillance, early detection, laboratory referral, infection prevention and control readiness, rapid response capacity, One Health field investigation, and community engagement.