Characterizing larval breeding sites: physical variables
We measured 11 physical variables for each larval breeding site in La Lopé, Gabon, and five additional variables in Rabai, Kenya (Table S1 in Appendix). These variables describe the size of the larval breeding site (e.g., diameters, circumference, surface area, volume, container depth, water depth, etc.), ambient environments (temperature, relative humidity, and canopy coverage), and water characteristics (pH, conductivity, salinity, water temperature, and total dissolved solids). Methodological details were summarized in Table S1 in the Appendix.
After removing eight sites with excessive missing data, we performed a principal component analysis (PCA) to summarize all physical variables. The multivariate differences between larval breeding site groups, habitats, and Ae. aegypti presence status were tested by multiple response permutation procedure (MRPP) with 999 permutations. The p-values for multiple comparisons were adjusted using the Holm method. We also identified variables that are most differentiated in each comparison by ranking variable importance using a random forest algorithm in R package randomForest v4.6-14 (Liaw & Wiener, 2002). Random forest is a decision-tree based classification algorithm that works well with small sample size and correlated variables (Qi, 2012).