Application of a Bayesian structural time series model for evaluating 11-year variation in pH in the headwaters of the Tama River, Japan

Understanding the temporal variations that occur in river water pH is vital for the conservation of aquatic organisms in riverine ecosystems. Temporal variations in river water pH are caused by long-term inter-annual trends and short-term periodic or nonperiodic fluctuations. However, statistical methods that explicitly and quantitatively distinguish these components in time series pH data analyses have not been applied. Using a Bayesian structural time series model, we decomposed the 11-year (132-month) variations in pH in the five Tama River headwaters into trend and periodic (seasonal) variation components, nonperiodic temporary fluctuations caused by water temperature changes, and variations not explained by these factors. The model revealed that the trend and periodic components and the influence of water temperature were unclear in the time series changes in the pH of all five rivers. We concluded that the observed time series changes in the pH of the five rivers analyzed were mostly the result of observation errors and random walk caused by accumulated noise (process error). Bayesian structural time series models like this model have several advantages in the analysis of time series environmental data. For example, it does not require the assumption that the trend and periodic components are temporally constant. It is flexible and will facilitate future limnological studies particularly when the conditions surrounding freshwater environments are unlikely to remain constant under climate change.