Predicting the ecological niches of Aedes aegypti s.l. using Maximum Entropy in Kenya

Samuel K. Muchiri^1,2*Moses M. Musau²Paul Mwaniki³Fridah Kirimi¹Nathan O. Agutu¹Emelda A. Okiro^2,4Simon Dellicour^5,6,7Robert W. Snow^2,4

• ¹Department of Geomatic Engineering and Geospatial Information Systems, Jomo Kenyatta University of Agriculture and Technology, Nairobi, Kenya
• ²Population and Health Impact Surveillance Group, KEMRI-Wellcome Trust Research Programme, Nairobi, Kenya
• ³Data and Statistics, KEMRI-Wellcome Trust Research Programme, Nairobi, Kenya, Nairobi, Kenya
• ⁴Centre for Tropical Medicine and Global Health, Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom
• ⁵Spatial Epidemiology Lab (SpELL), Université Libre de Bruxelles, Brussels, Belgium, Brussels, Belgium
• ⁶Department of Microbiology, Immunology and Transplantation, Rega Institute, KU Leuven, Leuven, Belgium
• ⁷Interuniversity Institute of Bioinformatics in Brussels, Université Libre de Bruxelles, Vrije Universiteit Brussel, Brussels, Belgium

Background: To evaluate the human population at risk of arboviral illnesses and improve vector and disease surveillance, it is crucial to model the probability of occurrence of impactful mosquitoes such as Aedes aegypti sensu lato (s.l.) which transmits dengue and Chikungunya etc. While majority of studies on Aedes distributions have focused on global ecological niche modelling (ENM), there is need to build local vector niche models using national data to design targeted vector surveillance and control strategies. Here, we built a spatial inventory of Aedes aegypti s.l. and applied a national-wide ENM approach to predict the probability of occurrence of Ae. aegypti s.l. across Kenya. Methods: Occurrence data on Aedes aegypti s.l. from 2000 to 2024 were assembled from the Global Biodiversity Information Facility (GBIF), Walter Reed Biosystematics Unit's (WRBU) VectorMap, and online literature searches. A maximum entropy approach was used to predict Ae. aegypti s.l. probability of occurrence in Kenya for 2024 at ~5 x 5 km resolution, using the occurrence data assembled and environmental covariates: population density, daytime and nighttime land surface temperature (LST), enhanced vegetation index (EVI), elevation, and land cover. Model performance was evaluated using the area under the curve (AUC) metric. Results: A total of 291 unique locations reported positive identification of Ae. aegypti s.l. Population density, daytime and nighttime LST were the most influential predictors. The models predicted high probabilities of occurrence of Ae. aegypti s.l. along the coast, northeastern and western Kenya, and in urban centres, while lower probabilities were predicted in sparsely populated areas. The models achieved a mean AUC value of 0.732 (0.653-0.779), indicating a moderate performance. Conclusion: The predicted distribution of Ae. aegypti s.l. can guide vector surveillance in high-risk areas and help identify populations at risk of arboviral diseases like dengue fever and Chikungunya, aiding in future outbreak preparedness.