Abstract: Dependence variation, an important feature of hydrological variation, can be described using autoregressive (AR(p)), moving average (MA(q)), and autoregressive moving average (ARMA(p, q)) models with p and q standing for the orders of autoregression and moving average, respectively. However, application and accuracy of these models depend on effective determination of their orders. Correlation coefficient information criterion (RIC) is adopted in previous studies to determine the order of an AR(p) model, where its fitting is assessed by the mean square error and its complexity is quantified based on information entropy. This study examines the applicability of RIC criterion to MA(q) and ARMA(p, q) and evaluate its performance. For the MA(q) models (q ≤ 3) and ARMA(p, q) models (p+q ≤ 3), our statistical experiments show its accuracy is much higher than those of the BIC and AIC criteria in determining the model orders. We demonstrate it is applicable to MA(q) and ARMA(p, q) in terms of multiple observed hydrological series.

Key words: hydrological time series, dependence, correlation coefficient information criterion, moving average model, autoregressive moving average model, order determination