Analysis of flood forecasting accuracy for mainstream and tributaries of upper Changjiang River

The effective application of flood forecasting information is pivotal for enhancing reservoir regulation efficiency, with the analysis of forecast accuracy and error distribution patterns serving as the core prerequisite for its practical implementation. This study systematically evaluated forecast accuracy and error distribution characteristics with 1 to 10 days forecast period based on a decade of data from seven critical hydrological control sections in the upper mainstream and tributaries of Changjiang River Basin (data for the Cuntan and Three Gorges Sections covered the period from 2011 to 2023, while the data for other five sections were from 2017 to 2023), focusing on continuous runoff processes, graded discharge magnitudes, and flood event processes. The results demonstrated that the mainstream forecasting accuracy was generally superior to that of the tributaries, with the Three Gorges Section achieving a qualified rate exceeding 85% for lead times of 1 to 5 days, representing the optimal forecasting performance. In graded discharge forecasting, large-magnitude flow predictions at other sections outperformed medium-small magnitudes, except for the Three Gorges Section. For flood events at the Three Gorges, both peak flow and flood volume forecast errors remained within 20% across a 5-day lead time. Furthermore, hypothesis testing indicates that errors in continuous runoff processes do not follow conventional distributions, although errors for specific discharge levels conform to a normal distribution. Therefore, conventional distributions cannot fully represent probabilistic forecast errors, while modeling error distributions based on the maximum entropy principle can provide a decision-making basis for risk-informed reservoir operation.