An Analysis of a Univariate Multi-Lag Auto-Regressive Stochastic Streamflow Model

Abstract

An analysis of a single-site auto-regressive multiple correlation is presented for the purpose of justifying a streamflow generator model. The multi-lag analysis as presented by Fiering and Jackson (1971) is studied through its application to two streamflow records and through the use of different correlation coefficients. The simple hydrologic model by Fiering (1967) is also studied to determine its ability to mimic the actual flow instabilities . Various model parameters are tested and their effect on the general characteristics of the resulting generator are determined.

This analysis yielded the following conclusions: The Fiering model for estimate correlation between flows does not mimic actual correlograms in the general sense. Computer modeling presents instabilities in generation especially for AR models with the flows lagged greater than seven. Care must be taken in the choice of a random number generator so that the proper statistics will be preserved. Data generation should only be made after a careful analysis of the stochastic part of the model because short term generations are greatly affected by this part of the model.