Uncertainty Analysis In Runoff Simulation Considering The Impact Of Input Data Derived From Climate Models

As hydrological model is the simplified description of the actual hydrological processes, it cannot capture every aspects of the real world. Uncertainty analysis of the runoff simulated by hydrological models has become a necessary component in runoff prediction. The sources of uncertainties usually arise from the input data, parameters of hydrological models and the model structure. When using climate data derived from GCMs as the input data, the input uncertainty is not only caused by the observation error, but also the uncertainties in climate models, downscaling methods and emission scenarios. These uncertainties will directly affect the precision of the input data, and thus affect the performance of simulated runoff. Therefore, besides the uncertainty analysis of parameters and the structure in hydrological models, the input uncertainties with respect to the climate models were taken into consideration as well. The main contents and results in this paper were summarized as follows:(1) In order to check the sensitivities of the parameters in three hydrological models, i.e. Xinanjiang model, SMAR and SIMHYD, GLUE method was used to get the scatter plots of likelihood function. According to the results of scatter plots in two study basins, parameters of three models could be classified into three groups:non-sensitive parameters; sensitive parameters; regional sensitive parameters.(2) Since the BMA is a method that can combine the forecasts of different models together to generate a new forecast expected to be better than any individual model’s forecast, and also has the ability to provide a uncertainty interval of the quantity to be forecasted, three hydrological models were employed in the investigation of two BMA schemes in this research to see if the BMA could improve the prediction reliability. The first BMA scheme was to calibrate each of the three models under the same Nash-Sutcliffe efficiency objective function, thus providing three different forecasts for the BMA combination. In the second BMA scheme, three different objective functions other than Nash-Sutcliffe efficiency were adopted, each of which is targeted for simulating different parts of flows, i.e. low flow, medium flow, and high flow. All three models were respectively calibrated for each of three objective functions to obtain the optimized parameter sets. As the same model with the different optimized parameter sets would give rise to different forecasts, thus in the second BMA scheme, there were nine different forecasts used for the BMA combination. For each of individual member model as well as both BMA combination schemes, the Monte Carlo method was used to infer the probability distribution of the quantity to be forecasted and determine prediction uncertainty intervals. Then, the model efficiency and uncertainty of each member model and two BMA combination schemes were assessed and compared.(3) The uncertainties in rainfall simulation when considering the impact of climate models were analyzed. The output data of three GCMs, i.e. BCCR-BCM2.0, CSIRO-MK3.0and GFDL-CM2.0under the scenario of20C3M was firstly prepared. For each GCM, three downscaling methods were employed to downscale the rainfall data from the global scale to the regional scale. At last, based on BMA method, the uncertainty from climate models and the uncertainty from downscaling methods were discussed respectively.(4) For the comprehensive uncertainty analysis of simulated runoff considering the input data uncertainty arose from climate models and downscaling methods, two BMA programs were firstly proposed in this paper. In the first BMA program, three GCMs and three downscaling methods were combined to construct nine sets of rainfall as input data. Therefore, nine sets of simulated runoff were obtained by Xinanjiang model for BMA combination. Because BMA method was only used once in the process of runoff averaging, the first BMA program was named "One-level BMA". In the second BMA program, the first step of constructing nine sets of rainfall was the same as the first BMA program. Secondly, nine-set rainfall were weighted averaged by BMA method to get a set rainfall called BMA(9) rainfall. Thirdly, BMA(9) rainfall was taken as input data respectively for three hydrological models to derive three sets of simulated runoff. At last, BMA method was used for the second time in the process of runoff averaging. As BMA method was used twice in the second BMA program, the second program was named "Two-level BMA". The BMA program which has a better performance in runoff simulation was chosen to be used in the future prediction.(5) Based on the BMA weights calculated by Two-level BMA, both averaged rainfall and averaged runoff under three climate scenarios in the future could be computed easily. Using log-normal distribution to fit the frequencies of averaged rainfall and averaged runoff of three scenarios, the differences of three climate scenarios in predicting the future rainfall and runoff were distinguished.