The change and uncertainty of environment lead to the non-stationary characteristics of hydrological series. Flood risk analysis based on consistent condition ignores the influence of climate change and human activities, and may results in the underestimation of flood risks. Here, GAMLSS model was used as the marginal distribution model and rainfall as the covariate to analyze the inconsistent frequency of the annual maximum 1-day flow and the annual maximum 15-day flood series of some hydrological stations in the main stream of Huaihe River, in order to explore the variation of flood peak and the design value of flood volume under inconsistent conditions. The results show that the designed marginal distribution values of flood peak and flood volume based on consistent condition are lower than those under inconsistent condition in the year of large flood event. Then, the joint distribution model of flood peak and flood volume was constructed under these two conditions using Copula function, which was used to analyze the co-occurrence probability of every hydrological station in the main stream of Huaihe River with a return period of 50 years. It is found that the co-occurrence probability under the consistent condition in the year of large flood event is prone to the lapse of flood risk with high peak and large volume.