Cell concentrations measured by qPCR were log transformed and each data point was plotted as the average log cells 100 mL⁻¹ versus incubation time (days). Error bars on each symbol represent the 95% confidence interval of duplicate PCR measurements. Decay curves of log transformed data were calculated by regression analysis using GInaFiT (Geeraerd and Van Impe Inactivation Model Fitting Tool; Geeraerd et al. 2005). The model that best fits the data was selected based on decay rate coefficients, root mean squared error and R² values. A biphasic inactivation model (Cerf 1977) was found to be the best fit for all the river water data sets, but for certain data sets from autoclaved river water (as shown on Table 1) a log linear inactivation model (Bigelow & Esty 1920) was applied. The model decay curves were used to calculate decay rate coefficients, R² values and 1 log decimal reduction times (T₉₀) (Table 1). Multivariate analysis of variance (ANOVA) statistical analysis of experimentally measured bacterial concentrations was used to determine if there was a significant difference in cell persistence among temperatures based on the entire data sets including results from all sampling days. Any statistical probability equal to or less than α = 0.05 was considered significant.