Keywords
predictability, time series, long memory, arfima model, ar model
Start Date
1-7-2008 12:00 AM
Abstract
Predictability is an important aspect of the dynamics of hydrological processes. The predictability of streamflow processes can be estimated based on the multivariate approach, which takes multiple explanatory variables into consideration, or, based on a univariate time series approach, which measures the predictability on the basis of univariate streamflow itself. In this study we investigate the predictability of 31 daily average discharge series with different drainage areas observed in 8 river basins in Europe and northern America on the basis of univariate time series approach. The results show that, although the existence of long memory is detected in the daily streamflow processes, the predictability of the streamflow process is more dominated by short-range autocorrelations than by the existence of long-memory in the streamflow process; the predictability is positively related to watershed scale, that is, the larger the watershed scale, the better the predictability of the streamflow process, and this kind of relationship mainly stems from the positive relationship between autocorrelation and basin scale. Because of the impacts of many factors, the predictability is dynamic rather than invariable, and there are many uncertainties present in the estimation of streamflow predictability.
Measuring Predictability of Daily Streamflow Processes Based on Univariate Time Series Model
Predictability is an important aspect of the dynamics of hydrological processes. The predictability of streamflow processes can be estimated based on the multivariate approach, which takes multiple explanatory variables into consideration, or, based on a univariate time series approach, which measures the predictability on the basis of univariate streamflow itself. In this study we investigate the predictability of 31 daily average discharge series with different drainage areas observed in 8 river basins in Europe and northern America on the basis of univariate time series approach. The results show that, although the existence of long memory is detected in the daily streamflow processes, the predictability of the streamflow process is more dominated by short-range autocorrelations than by the existence of long-memory in the streamflow process; the predictability is positively related to watershed scale, that is, the larger the watershed scale, the better the predictability of the streamflow process, and this kind of relationship mainly stems from the positive relationship between autocorrelation and basin scale. Because of the impacts of many factors, the predictability is dynamic rather than invariable, and there are many uncertainties present in the estimation of streamflow predictability.
