Research On Uncertainties In Hydrological Simulations And Flood Frequency Analysis

Uncertainty represents the differences between reality and estimate. It is very important to carry uncertainty study out because it is the basic requirement if the estimate will be applied in the real world. Thus, in engineering hydrology, uncertainty analysis has become a very important research topic. Many studies have been conducted, and a lot of uncertainty analysis approaches have been developed. Uncertainties have been existing in engineering hydrology in many aspects:for example, the uncertainty in rainfall-runoff data, parameter uncertainty in hydrological models, the data uncertainty in flood frequency analysis, the probability distribution uncertainty in flood frequency analysis, et.al. In this study, the uncertainties in some remote sense based data and flood frequency will be investigated. The main findings from this study are as below.First, the uncertainty from the global land data assimilation system data (GLDAS/Noah), including the input, output, seasonal and spatial distributions, are investigated. A distributed biosphere hydrological model is employed, which enables the evaluation of the output of the global land data assimilation system and the simulations of seasonal and spatial distributions of fluxes. The distributed biosphere hydrological model is calibrated with flow observations and MODIS land surface temperatures to provide accurate water and energy cycle simulations. GLDAS/Noah air temperatures and humidity agree well with observations, but GLDAS/Noah overestimates downward solar radiation and wind speed. Two correction functions are developed for downward solar radiation and wind speed. The accuracy of discharges and LSTs is improved after corrections. The simulated seasonal and spatial distributions of water and energy fluxes and states show high accuracy using corrected GLDAS/Noah data.Second, the uncertainties of six fine-resolution precipitation products, including precipitation radar, infrared, microwave and gauge-based products using different precipitation computation recipes, is evaluated using statistical and hydrological methods in northeastern China. In addition, a framework quantifying uncertainty contributions of precipitation products, hydrological models and their interactions to uncertainties in ensemble discharges is proposed. The investigated precipitation products are TRMM3B42, TRMM3B42RT, GLDAS/Noah, APHRODITE, PERSIANN and GSMAP-MVK+. Results show APHRODITE has high accuracy at a monthly scale compared with other products. Interactions between precipitation products and hydrological models can have the similar magnitude of contribution to discharge uncertainty as the hydrological models. A better precipitation product does not guarantee a better discharge simulation because of interactions.Third, the uncertainty quantification problem of future extreme flood predictions under climate change is investigated. For this purpose, this paper develops a new assessment framework, which uses a variance-based sensitivity analysis approach to explicitly quantify influences of each uncertainty source and their combined effect. The Long Ashton Research Station Weather Generator (LARS-WG) approach is used to downscale multiple general circulation models (GCMs), and the dynamically dimensioned search approximation of uncertainty approach is used to quantify hydrological model uncertainty. Extreme floods in northeast China are studied. Six GCMs and three emission scenarios (AIB, A2 and B1) are considered. Results show that the 500-year flood could increase by 4.45% in 2046-2065 and by 6.35% in 2080-2099. It is also found that the combined effect of GCMs, emission scenarios and hydrological model could have a larger influence than emission scenarios alone.Fourth, a bivariate cost-benefit analysis approach for design flood estimation is developed. In this approach, the dependence between design variables is incorporated into cost-benefit analysis through a copula function. A new total cost criterion, which integrates a copula function, flood damage function and return period, is proposed for bivariate design flood selection. A case study with 54-year observed data is used to study the uncertainty resulting from data dependence. The results show that the design flood peak and volume values from the bivariate cost-benefit analysis are smaller than those from the univariate cost-benefit analysis. The proposed new approach can provide consistent flood peak and volume estimation in terms of total cost, compared with the univariate approach where a design flood corresponds to two different total costs.Fifth, a holistic and coherent framework to allow for realistic design flood estimations under multiple uncertainties is developed. This approach effectively combines aleatory and epistemic uncertainties from data, probability distribution functions, and parameters on the basis of the Dempster-Shafer theory. It also presents upper and lower bounds of total cost faced by decision makers when selecting a design flood. In addition, a robustness criterion for decision support in design flood selection is proposed. The design flood corresponding to the smallest minimum total cost can tolerate lower uncertainties, thus is not robust. With an increasing design flood magnitude, more uncertainties can be tolerated while still guaranteeing the calculated total cost varies only slightly, thus the robustness increases, but the minimum total cost increases as well. Between total cost and robustness, there is a clear trade-off which decision makers need to balance in the decision making process.