Statistical characterization of streamflow persistence

Streamflow forecasts are essential for a variety of water resource management decisions, including flood emergency response, water allocation for different purposes, and drought risk management, depending on the forecast time horizon. They provide vital supporting information. However, it is crucial that the forecasts possess a high level of accuracy and reliability in order to facilitate decision-making processes and provide tangible advantages to the economy, environment, and society. All forecasting methods are subject to some degree of uncertainty. This uncertainty certainly plays a significant role in introducing errors into the forecasts. The objective of this study is to contribute to the overall assessment and discussion of errors and uncertainties within the framework of streamflow forecasting. The focus here lies in the characterization of forecasting system errors, specifically examining how they vary with forecast horizon and basin size. The considerations are centered on the errors of persistence-based approach used as a forecasting model. This approach is the simplest way to make the forecast. It states that future values of streamflow time series are determined by assuming that no changes will occur between the present time and the forecasted time horizon. The statistical properties of forecast errors are examined here across space and time scales. Despite the simplicity of persistence forecasting approach, this study demonstrates that its error properties are complicated.