Pathogen seasonality and links with weather in England and Wales: a big data time series analysis

Cherrie, Mark P. C.; Nichols, Gordon; Iacono, Gianni Lo; Sarran, Christophe; Hajat, Shakoor; Fleming, Lora E.

For the second stage, we aggregated the pathogen incidence data to monthly resolution so that they were able to be merged with the weather variables previously processed to monthly values by the National Climate Information Centre. The time series' for each of the weather variables was shown to be stationary (no significant trend from year to year) by using the Augmented Dickey-Fuller (AF) test (p < 0.05) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test (p > 0.05). We tested each pathogen time series in the same way. Some were found to be non-stationary and differenced (once or twice, depending on results of AF and KPSS tests). Cross correlation coefficients were generated between cases and weather variables for the month that they were recorded and then by the meteorological values lagged by 1 month. The correlation coefficients were then used as input to the K-means clustering method. Two clusters were generated in order to narrow the focus on those correlated with weather. The terminology for discussing the correlation coefficients was as follows: very weak (r = 0-0.19), weak (r = 0.20-0.39), moderate (r = 0.40-0.59), strong (0.60-0.79) and very strong (r = 0.80-1.00). Seasonality and weather correlation results were summarised and discussed in terms of differences between weather variables and within the most common genus for which serotypes were available (Salmonella).