Characterizing larval breeding sites: combining multiple environmental conditions
To consolidate the analysis of different categories of environmental conditions, we performed another random forest model in La Lopé and Rabai, respectively. The model included scores on the first three principal components (PCs) from the physical variable analysis and scores on the first two NMDS axes from the bacterial community composition analysis (i.e., predictive variables), and used them to classify the larval breeding site groups (i.e., the dependent variable). For the analysis from Rabai, we also added the microbial density and the density of all mosquito larvae. These two variables had many missing values in the La Lopé dataset and thus were excluded. The model generated a confusion matrix, which displayed the number of samples correctly or wrongly assigned to each larval breeding site group. A lower proportion of misclassification between groups suggests a stronger distinction in their environmental conditions.