Fig 1.
Spatial and temporal distribution of vector surveillance dataset used in model.
(A) Points show the geographical sampling locations (n = 57) of the C. tritaeniorhynchus records across India*, with occurrence-only records coloured orange (n = 74), records which included occurrence and abundance data in green (n = 266), and pseudoabsence records in purple (n = 20). Stacked barplots show the temporal distribution of the total vector occurrence (orange) and abundance data (green) used in the analysis per month (B) and year (C). *Abbreviations for Indian states and union territories: AP—Andhra Pradesh, AR—Arunachal Pradesh, AS—Assam, BR—Bihar, CH–Chandigarh, CT- Chhattisgarh, DD—Daman and Diu, DL—Delhi, DN—Dadra and Nagar Haveli, GA–Goa, GJ–Gujarat, HP—Himachal Pradesh, HR—Haryana, JH—Jharkhand, JK—Jammu and Kashmir, KA—Karnataka, KL–Kerala, MH—Maharashtra, ML—Meghalaya, MN—Manipur, MP—Madhya Pradesh, MZ—Mizoram, NL—Nagaland, OR—Odisha, PJ—Punjab, PY—Puducherry, RJ—Rajasthan, SK—Sikkim, TL–Telangana, TN–Tamil Nadu, TR—Tripura, UP—Uttar Pradesh, UT—Uttarakhand, WB–West Bengal. Source of base layer https://gadm.org.
Table 1.
Model selection results for models of increasing complexity.
The table details the structure of the joint-likelihood models and their corresponding within-sample predictive accuracy assessed on Watanabe-Akaike Information Criterion (WAIC) values. Best models were selected based on minimising WAIC while adhering to model assumptions. Out-of-sample predictive accuracy was compared using mean absolute error (MAE) statistic for random cross validation. Fixed effects included two-month lagged precipitation, proportion of land under agricultural use in 1km radius and district-level measures for annual number of rice crops and total rice area cultivated and rice produced per year. Mean temperature was included as a second-order random walk function to represent the nonlinear relationship between temperature and vector population dynamics. Non-environmental effects considered were for month (M) and state-level spatial (ST) effects specified as a BYM model and study-level (S) random effects.
Fig 2.
Spatiotemporal correlates of JE vector abundance across India averaged over the period 1990–2012.
Map to show predicted C. tritaeniorhynchus abundance (maximum annual value) and vector seasonality (intra-annual variance in abundance) (A). These measures were calculated from the scaled abundance predictions and ranged from 0 to 7 logscale for maximum abundance and 0 to 3 logscale for seasonality. The map displays areas of high perennial vector abundance as orange, high seasonal vector abundance as pink, low perennial vector abundance as green and low seasonal vector abundance as blue. The fixed-effect parameter estimates and 95% credible intervals for the joint likelihood model (B) show that vector abundance is strongly influenced by climatic and land use variables. The nonlinear relationship between monthly mean temperature and vector abundance for the observed range of temperatures (C) where 95% CI is shown shaded and peaks at around 23°C and then declines. The reported thermal minima (9.5°C) for important Culex species life history traits [19] is indicated with a dashed line. Source of base layer https://gadm.org.
Fig 3.
Predicted seasonal abundance of C. tritaeniorhynchus across India for the period 1990–2012.
Average vector abundance (logscale) for the (A) winter (October to February), (B) summer (March to May) and (C) monsoon (June to September) seasons. The figure legend is scaled from 0 to 7 logscale, with light yellow colours signifying low vector abundance and dark purple emphasising high abundance. Uncertainty in predictions was estimated from standard deviation (range 0–2 SD) and is represented in the maps by transparency, (high uncertainty is more transparent). The black circles represent the location and magnitude (i.e., number of cases) for JE human outbreaks per season during the period 2009–2015 across India [68]. Source of base layer https://gadm.org.