Autoregressive time-series models
The autocorrelation function (ACF) plot (Figure S15) and statistical
tests for the monthly seroprevalence of pigs of all age groups sampled
in Gorakhpur district (n = 783) were consistent with stationarity
(Box-Ljung Χ2 = 13.2, P = 0.35; ADF = -3.6, P = 0.03;
KPSS Trend = 0.14, P = 0.05; KPSS Level = 0.28, P = 0.1). However, using
systematic combinations of seasonality and differencing, the best
fitting model, with relatively low AIC (1129), symmetrical residuals,
with no autocorrelation demonstrated in the ACF plot was a model of
order (p = 5, q = 1, P = 2, Q = 1) which incorporated seasonality (12)
and differencing (d = 1, D = 1; Table 1; Figure S16). Trend lines of
monthly total rainfall, mean minimum temperature, and mean relative
humidity showed that highest rainfall and humidity was in the second
half of the study period, and peaks in mean monthly temperature occurred
in 2016 and 2022 (Figure S17). Cross-correlation function plots
indicated no relationship between JEV seroprevalence and either monthly
total rainfall or mean minimum temperature (Figures S18 and S19) but did
indicate that JEV seroprevalence was negatively correlated with mean
relative humidity at 6 months lag (Figure S20). Humidity was included
with increasing lags in the ARIMA model (Model 4, Table 1). The model
with the lowest AIC (1004.91) incorporated mean relative humidity at 12
months lag (residuals Figure S21) suggesting an inverse relationship
between humidity and IgG seroprevalence in pigs. This likely reflects
the broad pattern of higher IgG seroprevalence in Gorakhpur in the first
half of the study period when humidity was lowest, and the reverse of
this (seroprevalence low, humidity high) in the second half of the study
period.