Abstract: Due to the influence of model parameters and observational data, hydrological models usually have significant uncertainties. Quantitative analysis of the uncertainty of snowmelt runoff simulation can control the uncertainty of the simulation results and improve the predictive performance of the model, which is of great significance for the scientific management of water resources in cold and arid mountainous regions. In this study, the parameter uncertainty analysis of snowmelt runoff model based on Markov chain Monte Carlo method (MCMC) is carried out in the Cele River basin in southern Xinjiang as the study area. Aiming at the difficulties such as zero flatted error of the likelihood function and the residual structure in MCMC, a mixed likelihood function with a combination of binomial and normal distributions is used to describe the residual structure, and the results are compared and analyzed with those of the traditional Gaussian likelihood function. Model evaluation indicators such as the Root Mean Square Error, coefficient of determination and prediction interval coverage show that the uncertainty analysis results based on the mixed likelihood function have better prediction performance. Therefore, the mixed likelihood function can more reasonably portray the structure of the simulated residuals of snowmelt runoff, significantly improve the prediction performance of hydrological models, and is conducive to the implementation of scientific and accurate water resource management and protection.

Key words: uncertainty in hydrological modeling, mixed likelihood function, snowmelt runoff model, MCMC, zero flatted error