The Application And Verification Of A Post-processor Of Hydrological Ensemble Prediction Based On The Mopex Data

Using the probabilistic ensemble forecasts to describe the uncertainty in the hydrological forecasts can not only improve the precision and accuracy of the forecast, but also provide more risk information than deterministic forecasts. The General Linear Model (GLM) is(?) statistical model to adjust the biased hydrological simulations generated by the hydrological models. It can extract the characteristics and trends of the streamflow changes from historical observations and corresponding simulations and generate an ensemble forecast for the future hydrological events. This post-processor can produce reliable hydrological ensemble forecast, retain the skills of the original forecast, reduce the errors in the simulations and obtain hydrological simulations with more accuracy.In this thesis, various experiments are designed to test the performance of the GLM model. On one hand, the GLM model processed different data sets with different error levels and different data from different time periods to verify the performance of the post-processor and its capability to reduce the biases. On the other hand, the processor tested in different configurations to examine the sensitivity to its own parameters, basins with different climatic conditions and hydrological models with different predictive capabilities.The main conclusions are as follows:(1) The post-processor of the hydrological ensemble forecast system can measure those uncertainties in the hydrologic forecast effectively, reduce the biases in the hydrological simulations, get the adjusted forecast which is more approximate compared to the observations. The adjusted ensemble forecast is superior to the raw simulation ensemble forecast in the terms of average, the standard deviation, the RMS error and some other indicators.(2) The comparison of the adjusted forecast of the uncalibrated hydrological model with the forecast generated by the calibrated model shows that the capability of the post-processor and the parameter calibration of hydrological model have similar capabilities to remove the biases in the raw forecast. Even for the forecast generated by the calibrated model, the post-processor still can make some improvement.(3) Through the comparison of the adjusted results of the calibration and verification, the performance of the GLM to process the data for different time periods is reliable and stable. The more data used, the better result obtained. However, not all the biased simulations can be improved by the GLM.(4) The tests about parameters of the GLM model show that the longer analysis period and buffer period are both helpful to the performance of the post-processor. Also it is found that the ability of the GLM is independent of the selection of the forecast dates.(5) The results further indicate that the GLM shows better performance in the wetter basins. When applied to the hydrological model with different forecast abilities, the post-processor shows similar capabilities on the error correction.