Parameter Estimation And Uncertainty Analysis For Distributed Hydrologic Models

With the rapid development of computer technology,remote sensing technology and geographic information system technology,distributed hydrologic models have become important tools in hydrological research.Uncertainties stemming from input,model parameters and model structure will affect the final simulation results.Therefore,this study aims to construct a Bayesian uncertainty analysis framework to reasonably estimate predictive uncertainty over the upper Huaihe River basin.This study also proposes a new method to improve the efficiency of calibration algorithms and reduce the uncertainty of calibrated parameters.The efficiency of this method is tested over the French Creek watershed in the U.S.(1)Four latest satellite-gauge precipitation products and their hydrologic utilities are evaluated comprehensively.Satellite-gauge precipitation products further combine the advantages of satellite precipitation estimates with rain gauge data,providing great potential to be high-quality input of distributed hydrologic models.This study comprehensively evaluates four of the latest satellite-gauge quantitative precipitation estimates(QPEs)and their hydrologic utilities over the Huaihe River basin during 2003-2012.The four QPEs include NASA's TRMM 3B42V7 product,NOAA's CMORPH bias-corrected product(CMORPH CRT),CMORPH satellite-gauge merged product(CMORPH BLD)and CMORPH satellite-gauge merged product developed at the National Meteorological Information Center(NMIC)of the China Meteorological Administration(CMA)(CMORPH CMA).The results show gauge adjustment procedures and gauge density in the production of satellite-gauge QPEs greatly affect their quality of precipitation estimation.The quality of QPEs directly impacts streamflow simulations,as the precipitation biases are propagated into simulated streamflow through interaction with hydrologic processes.CMORPH CMA forced streamflow simulations even outperform those forced by the oberseved precipitation.Overall,CMORPH CMA shows great potential to improve the precipitation distribution and hydrometeorological simulations,and can serve as an alternative high quality QPE in China.(2)A new residual error model is proposed to better quantify the total uncertainty in streamflow modeling and obtain reliable hydrologic predictions.The various uncertainties in hydrological simulations can be treated in lumped manner as model residual errors.Adequately characterizing the form of residual errors through the residual error models is of great importance to obtain reliable and precise and hydrological predictions.This study compares three methods to deal with the heteroscedasticity,including the explicit linear modeling(LM)method and nonlinear modeling(NL)method using hyperbolic tangent function,as well as the implicit Box-Cox transformation(BC).Then a combined approach(CA)combining the advantages of both LM and BC methods has been proposed.In conjunction with the first order autoregressive model and the skew exponential power(SEP)distribution,four residual error models are generated,namely LM-SEP,NL-SEP,BC-SEP and CA-SEP,and their corresponding likelihood functions are applied to the VIC hydrologic model over the Huaihe River basin,China.Results show that the LM-SEP yields the poorest streamflow predictions with the widest uncertainty band and unrealistic negative flows.The NL and BC methods can better deal with the heteroscedasticity and hence their corresponding predictive performances are improved,yet the negative flows cannot be avoided.The CA-SEP produces the most accurate predictions with the highest reliability and effectively avoids the negative flows,because the CA approach is capable of addressing the complicated heteroscedasticity over the study basin.(3)The effects of different precipitation input on parameter calibration and streamflow predictive uncertatinty are studied,then the multi-satellite precipitation ensemble with BMA is used to improve predictive performance.Using the newly proposed residual error model CA-SEP within the Bayesian framework,this study further quantifies the effect of the input uncertainty on parameter calibration and streamflow predictions when applying the three global satellite-gauge precipitation products to the VIC model.The high-quality regional satellite-gauge precipitation product CMORPH CMA is also applied within the same framework to provide a benchmark.The results show that the parameter uncertainties are influenced significantly by the input of various precipitation products.There is a tradeoff between the deterministic streamflow performance and the probabilistic predictions for selecting the best input among the three global precipitation products.The streamflow uncertainty intervals for the three global precipitation products are then merged using the Bayesian Model Averaging(BMA)method.The BMA results significantly improve deterministic streamflow predictions and produce much more reliable probabilistic predictions,which even outperform the outcomes of the high-quality CMORPH CMA.Therefore,hydrologic ensemble using multiple global satellite-gauge precipitation products provides a promising and advantageous approach to support water management and decision making,especially in ungauged basins.(4)A new parameter calibration method is proposed to reduce the parameter uncertainty for for calibration problems with many parameters.Parameter optimization is an unavoidable part in the application of hydrologic models.Aimed at the potential high-dimensional calibration problems in the distributed hydrologic models,this study introduces an innovative approach,HIP-POP(HIerarchical Prioritization for Parameter OPtimization),to compile a series of expert-knowledge principles to aid in the enhancement of optimization-based calibration methods.It can overcome,or at least mitigate equifinality and make calibrated parameters have more reasonable values which better correspond to the physical reality.In addition,HIP-POP can improve the efficiency of the existed calibration algorithms by helping them converge to good objective function values faster,thus reducing the heavy computational burden.This study conducts a synthetic experiment using two distributed hydrologic models in the French creek watershed,Pennsylvania.Three HIP-POP based calibration schemes are compared to the traditional calibration scheme.Results show that HIP-POP based calibration schemes show great advantages over the traditional calibration scheme by yielding statistically significantly fewer parameters which hit boundaries of the parameter range and are not converged.In addition,HIP-POP based calibration schemes produce more reasonable parameter estimates at an accelerated rate compared to traditional optimization-based methods.The simplicity of HIP-POP allows for easy combination with most of the auto calibration algorithms and subsequent application to calibrate computationally demanding models with many parameters.