Design flood, which plays a paramount role in reservoir construction and operation, is often calculated by the annual maximum sampling method (AMSM). However, the AMSM can only use a small amount of data and reflect the extreme value distribution of inflow flood. As a result, the annual maximum design flood or the seasonal design flood is not sufficient to formulate the reservoir operation scheme in whole life cycle. Currently, studies analyzing the sample selection method of design flood and the way to make design flood close to inflow flood are rare. To this end, an improved peaks-over-threshold method (IPOT) was designed. Together with the time-varying parameters with Poisson distribution model, we used the IPOT to calculate a new type time-varying design flood in this study and took Longyangxia Reservoir in China as a case study. Results indicate that, compared with the AMSM, the IPOT enhances the physical correlation between sample individuals by increasing data use rate and determines the optimal threshold, which avoids the influence of human factors on the sample selection. Moreover, the time-varying design flood can fully reflect the characteristics of inflow flood and combine the design flood value with time and probability to establish an inflow flood identification model. According to this model, managers can assess the probability and type of inflow flood and choose the appropriate reservoir operation scheme. Findings of this study, including the IPOT, the time-varying design flood and the inflow flood identification model are helpful for water resources planning and management.