Random forest analysis of impact of abiotic factors on Culex pipiens and Culex quinquefasciatus occurrence

Authors: ARORA, ARINDER - University Of Florida; SIM, CHEOLHO - Baylor University; SEVERSON, DAVID - University Of Notre Dame; Kang, Dave

Submitted to: Frontiers in Ecology and Evolution
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/20/2021
Publication Date: 1/27/2022
Citation: Arora, A.K., Sim, C., Severson, D.W., Kang, D.S. 2022. Random forest analysis of impact of abiotic factors on Culex pipiens and Culex quinquefasciatus occurrence. Frontiers in Ecology and Evolution. 9. Article 773360. https://doi.org/10.3389/fevo.2021.773360.
DOI: https://doi.org/10.3389/fevo.2021.773360

Interpretive Summary: The Culex pipiens complex of mosquitoes are significant vectors of several pathogens resulting in infectious human diseases in North America, including but not limited to West Nile encephalitis, Rift Valley Fever, and Lymphatic filariasis. Among this complex are Culex pipiens form pipiens and Culex quinquefasciatus. While morphologically similar, the mosquitoes exhibit unique life histories that suit them uniquely to divergent niches, wherein C. pipiens can thrive despite the cold winters of the northern USA and C. quinquefasciatus is able to survive periods of drought typical in the southern states. Here, we used machine-learning algorithms to model and determine which environmental parameters best explain mosquito occurrence in historical trapping data across the continental United States. The results of this study will improve vector management programs by explaining which environmental variables will provide the most accurate predictions of mosquito presence at a given site. These predictions will lead to improved human health in the U.S. and abroad.

Technical Abstract: We use random forest models and found that environmental factors can be used to explaine between 71-97% of the presence or absence of the two mosquitoes based on historical climatic data. We find that the model fit remains accurate after bootstrapping the 25% of the climate data set aside for testing from before 2017, as well as data from 2017. We further used partial dependence plots to determine how interactions between climate parameters impact the accuracy of the trained models. The results of this study will improve vector management programs by explaining which environmental variables will provide the most accurate predictions of mosquito presence at a given site.