Characterization of annual river flow time series and interpretation of the relative performance of time series forecasting methods

A large number of annual river flow time series is characterized by computing five descriptive features. These features are the coefficient of variation, the lag-1 sample autocorrelation, the Hurst parameter of the fractional Gaussian noise process, and the trend strength and spectral entropy of the time series. Five individual time series forecasting methods (i.e., the persistent method, an autoregressive fractionally integrated moving average – ARFIMA model, the simple exponential smoothing model, the complex exponential smoothing model and the Prophet model) and their simple combinations are also employed for delivering one-step ahead forecasts, and the quality of the latter is assessed. Based on the annual river flow characterizations and the forecasting performance scores, the possibility of case-informed integrations of diverse hydrological forecasting methods within systematic frameworks is algorithmically investigated and discussed. The related investigations encompass linear regression analyses, which aim at finding interpretable relationships between the values of a representative forecasting performance metric and the values of selected river flow statistics. We find only loose (but not negligible) relationships between the formed variable sets. These relationships could be exploited for improving (to some extent) future forecasting applications. The results of our big data forecasting experiment are lastly exploited for characterizing –in relative terms− one-year ahead river flow predictability.