An ARMA-type section model for average ten-day streamflow synthesis

Abstract

Most of the rivers in Taiwan are short and run on a steep slope due to the island's topography. Because of the weak correlations of streamflow in time and the occurrence of extreme events such as typhoons, classical autoregressive-moving average (ARMA) models have difficulties in forecasting and synthesizing the average 10-day streamflow in Taiwan. In this study, the synthesis of the average 10-day streamflow of the Tanshui River in Taiwan is accomplished by a section model. The model divides the year-round streamflow records into several sections according to their distinguishable patterns, and each section is modeled by a separate ARMA model. For parameter control, a heuristic grouping procedure, based on statistical inference of the random noise part, is used to separate a year into a minimum number of sections. The section separation procedure follows the general precipitation pattern in a year. The case study results indicate high statistical agreement between synthesized series and historical records. Additionally, a new procedure, extended autocorrelation function (EACF), is introduced and applied in this study to assist in model identification.