Abstract: This paper presents an application of a generalized additive model for location, scale and shape (GAMLSS) with time-varying statistical parameters to calculations of the non-stationary hydrological frequencies of annual runoff series of the Yellow River mainstream. First, we discuss the evaluation methods of model structure, parameter estimation, and fitting effect. Then, we construct a set of alternative models for non-stationary hydrological frequency calculation, based on analysis of runoff change trends of the annual series of six hydrological stations at the Yellow mainstream for the period of 1956 to 2017, and select the optimal model as featured with the probability distribution type and parameter setting that has the minimum AIC value. This model and its fitting effect are tested against the statistical characteristics of the residual series and the quantile plot of design runoffs. From the test, the optimized model manifests good fitting to the runoff series, and the design runoffs corresponding to the runoff series of each station decrease with time. Thus, we suggest that the future evaluation, allocation or operation of water resources fully consider the non-stationary effect of runoffs and carry out predictions and uncertainty assessment of water resources in changing environments.

Key words: runoff, non-stationary, change of underlying surface, Yellow River, hydrologic frequency