Anthropogenic landscape decreases mosquito biodiversity and drives malaria vector proliferation in the Amazon rainforest

abundance and: i) the forest cover (PLAND), ii) fragmentation (ED), and iii) the relationship between PLAND and ED (landscape structure). ED measured at the 0.5 km radius was grouped into three categories—low (0–1.5 m / ha x 100); moderate (1.5–2.0 m / ha x 100) and high (2.0–3.5 m / ha x 100) and was always used in posterior analysis as a random effect and as categorical variables (low, moderate and high). These followed results from [56, 63], and the equal distribution of samples categories, with the mosquito collection point as the random effect. To define the ED categories as a random effect, we employed the results of the descriptive analysis that demonstrated differences in mosquito biodiversity in each specific ED category defined here. PLAND and ED measured at every sampling unit (1km radius) were used both as continu- ous and categorical variables. The continuous was also used as a predictor, while the categori- cal entered as a random effect in the analysis. Combinations of PLAND and ED were classified in four categories of landscape (S2 Fig in S1 File) once we decided to test the hypothesis that modification in the landscape structure (herein considered to be a combination of PLAND and ED metrics in each sampling unit area) shifts in mosquito species composition and Ny. darlingi abundance. The categories were: A–PLAND � 50% and ED < 0.015 m/ha; B– PLAND � 50% and ED � 0.015 m/ha; C–PLAND > 50% and ED � 0.015 m/ha; D– PLAND > 50% and ED < 0.015 m/ha. The overview of the sample distribution according to

landscape structure analysis (categories A, B, C, D) is presented in S3 Fig in S1 File. We also measured the distance from each of the 79 sampling units to the nearest water drainage network (DW). Each network was constructed from the Digital Elevation Model (DEM) available on ESA. The spectral bands of each of the images were composed of bands 2 (Blue; 0.49 μm), 3 (Green; 0.56 μm) and 4 (Red; 0.66 μm) to generate a final-colored image for each region of the collection points. The drainage network indicates where water is most likely to accumulate according to the terrain relief but does not necessarily indicate locations of riv- ers or streams. Diversity indices The diversity indices of Margalef, Shannon, Simpson and Berger-Parker for each sampling unit were calculated by the Re´nyi series, based on Hill’s measures [64]. As diversity indexes are highly autocorrelated, we used the Shannon index in analytical associations because it has links to entropy theory [65]. Statistical analysis Data analyses were conducted in three steps: 1) descriptive statistics, and 2) analytical associa- tions. To construct the models, and prior to commencing the analytical approach, we applied extensive descriptive statistics. We assessed the sampling unit distribution according to the landscape structure and vegetation class (S3 Table in S1 File). The units that corresponded to landscape D and Open Tropical Moist Forest (OTMF) showed the lowest mean Ny. darlingi abundance and were considered the baseline group to create dummy variables for generalized

linear mixed model (GLMM) and regression analysis. For step 1, Spearman’s correlation index (at a significance level of 5%) was utilized to verify the correlation among predictor variables (diversity index, PLAND, ED, drainage network (DW), landscape structure—landscape categories A, B, C, D—vegetation class), and number of Ny. darlingi. Principal Component Analysis (PCA) was used in exploratory analyses to iden- tify associations among the number of Ny. darlingi, PLAND, ED and Shannon index variation according to the four categories of landscape structure. The non-parametric Kolmogorov- Smirnov test was employed to verify whether the variables had a normal distribution (at a PLOS ONE Anthropogenic landscape and mosquito biodiversity drives malaria vector proliferation PLOS ONE | https://doi.org/10.1371/journal.pone.0245087 January 14, 2021 6 / 25