Testing Of HBV Model Parameters And Sensitivity Analysis In Simulating Hydrological Dynamics Of Pailugou In Qilian Mountains

Understanding the catchment hydrological daynamics and processes is fundamental to predicting and evaluating the role of forest in water conservation, and model simulation is an important approach for describing the pattern of hydrological daynamics. However, the precision of simulation and prediction by using hydrological models is highly dependent of model parameterization and there exist uncertainties. Therefore, how to select the suitable model parameters for a given area is a prerequisite for assuring the accuracy of hydrological modeling and predictions.In this study, the conceptual semi-distributed model HBV is applied in a small catchment, Pailugou, in the upper Heihe River basin in Qilian Mountains to simulate daily runoff daynamics over a period of four years (from2000through2003). Firstly, the Regional Sensitivity Analysis (RSA) is used to analyze the sensitivity of the HBV model parameters with objective functions based on statistical measures, runoff signatures, and combinations of different objective functions. Secondly, the Generalized Likelihood Uncertainty Estimation (GLUE) method with coverage, median width and cumulative runoff differences is used to analyze the uncertainty of modeled runoff. Finally, the Dynamic Identifiability Analysis (DYNIA) is used to analyze the changes in the information content of the observational runoff data when being used for identification of various model parameters, the behavours of parameters optima over the entire course of simulations, and the satisfaction of model structrues.Results of RSA show that the parameter sensitivity varies with the objective functions used; most of the model parameters are highly sensitive when runoff signatures or combinations of different objective functions are used, which reduce the model parameter uncertainty to some degree. Results based on GLUE show that the HBV model is generally capable of simulating the runoff in the Pailugou catchment, although it underestimates the peak runoffs. Most parameters show higher parameter identifiability and lower model uncertainty when runoff signatures or the combined objective functions are used. Results of DYNIA reveal that model parameters have key periods that show higher identifiability and play a crucial role in representing the predicted runoff. Among the21parameters in the HBV model, the optima for11of them (i.e. catchment parameters PCALT, TCALT; snow routine parameters CFMAX, SFCF; soil routine parameters FCshrub, LPshrub;response routine parameters PERC, UZL, K2; routing routine parameter MAXBAS) do not shift over the time domain and remain constant over the course of simulations; the optima for five parameters (i.e. snow routine parameters TT; soil routine parameters FCforest,BETAshrub, BETAgrass; response routine parameter K1) vary within10%of the original parameter ranges, indicating the possibility of a relatively stable and time-invariant parameter identification; and the optima for other five parameters (snow routine parameter CFR; soil routine parameters LPforest, BETAforest,FCgrass; response routine parameter Kl) shift over the time domain and are very difficult to identify.In summary, the HBV model is highly capable of simulating the runoffs at the Pailugou catchment outlet. Objective functions based on runoff signatures or combinations of different objective functions can improve parameter sensitivity and reduce the model parameter uncertainty, which can be used as a priority objective function for model calibration. Sixteen out of the total21parameters are proven to be well identified, but the identifiability for five of them, including CFR, LPforest, BETAforest, FCgrass, and Kl, appears to be very difficult, suggesting that some modifications are needed for improving the model structure in applictions concerning specific catchements or watersheds.