Copula Based Post-processing Method for Hydrologic Ensemble Forecast

Hydrology forecasts often contain uncertainties from model inputs, model parameters and model structure. Post processing methods can be applied to the original hydrology ensemble forecasts to correct the bias and spread error. Most existing post-processing methods applied the Normal Quantile Transform (NQT) to transform the hydrology variables to normal distribution for convenient statistical inference. However, the NQT based algorithm suffer several problems, such as the extrapolation problem in back-transform process. In this research, a copula based post-processing method was developed. The copula function estimates the joint distribution of observation and model forecast directly, and then the conditional distribution of observation given the model forecasts could be obtained without NQT. The proposed post-processing method was tested and compared with two other popular methods based on NQT, namely the Hydrology Uncertainty Processor (HUP) and General Linear Model Post-Processor (GLMPP) using the observation and simulation dataset from the Model Parameter Estimation Experiment (MOPEX) project. The results show that the drawback of NQT based post-processing methods can be alleviated in the proposed algorithm. Some suitable conditions and suggestions on the application of copula based post-processing method for hydrology ensemble forecast were also provided.