Evaluation of flood season segmentation using seasonal exceedance probability measurement after outlier identification in the Three Gorges Reservoir

Abstract

Flood season segmentation, which partitions an entire flood season into multiple subseasons, constitutes a considerable water resources management task. Moreover, the risks associated with various schemes for flood season segmentation should be evaluated. Preliminary analysis in this study used the principal component based outlier detection (PCOut) algorithm to identify possible outlying observations to reduce the uncertainty involved in flood season segmentation. Then, a quantitative measurement, the seasonal exceedance probability (SEP), was proposed to evaluate various segmentation schemes. The SEP quantifies the risk that the maximum observation occurs outside the main flood season. Several findings were derived based on a case study of China’s Three Gorges Reservoir (TGR) and daily streamflow records (1882–2010). (1) The PCOut algorithm was found effective in identifying outliers, and the estimation uncertainty of the segmentation evaluation due to outliers decreased when the end date of main flood season (EDMFS) was postponed. (2) The proposed SEP measurement was shown capable of supporting quantitative evaluation of the segmentation schemes in the flood season. (3) The current flood segmentation scheme based on an EDMFS of September 10 is sufficiently safe for the TGR. The findings of this study could help in the proper operation of the TGR.