Uncertainty analysis of hydrological model parameters based on the bootstrap method: A case study of the SWAT model applied to the Dongliao River Watershed, Jilin Province, Northeastern China

As an important tool for the description and analysis of hydrological processes, the watershed hydrological model has been increasingly applied to watershed hydrological simulations and water resource management. However, in most cases, model parameters are only determined in a calibration scheme which fits the modeled data to observations, thus significant uncertainties exist in the model parameters. How to quantitatively evaluate the uncertainties in model parameters and the resulting uncertainty impacts on model simulations has always been a question which has attracted much attention. In this study, two methods based on the bootstrap method (specifically, the model-based bootstrap and block bootstrap) are used to analyze the parameter uncertainties in the case of the SWAT (Soil and Water Assessment Tool) model applied to a hydrological simulation of the Dongliao River Watershed. Then, the uncertainty ranges of five sensitivity parameters are obtained. The calculated variation coefficients and the variable parameter contributions show that, among the five parameters, ESCO and CN2 have relatively high uncertainties: the variation coefficients and contribution rates are 23.98 and 70%, 14.43 and 18%, respectively. The three remaining parameters have relatively low uncertainties. We compare the two uncertainty ranges of parameters acquired by the two bootstrap methods, and find that the uncertainty ranges of parameters acquired by the block bootstrap are narrower than those acquired by the model-based bootstrap. Further analysis of the effects of parameter uncertainties on the model simulation reveals that the parameter uncertainties have great impacts on results of the model simulation, and in the model calibration stage 60%-70% of runoff observations were within the corresponding 95% confidence interval. The uncertainty in the model simulation during the flood season (i.e. the wet period) is relatively higher than that during the dry season.