Artificial neural network post-processing of ensemble streamflow forecasts for Huai-he and Rhine river
2019
Online
Elektronische Ressource
Along with the development of streamflow forecasting, ensemble streamflow prediction (ESP) has been widely applied to account for inherent uncertainties in the forecasts. Mostly, uncertainties are caused by data (including missing observations and observation errors), problems of model structure, and lack of knowledge. Therefore, post-processing methods are adopted to reduce bias and improve accuracy of ESP results, hence reducing uncertainties of ESP. In this work, Huai river (a natural river located in South of China) and Rhine river (a European river beginning in Switzerland and emptying in the North Sea, the Netherlands) are selected as case studies. Xixian station of Huai river and Lobith station of Rhine river are used as forecasting locations. The lead time analysed is 24 hours as this is considered the minimum lead time of interest for flood early warning at Xixian station. The same lead time was then analysed for Lobith station. The post-processing method studied in this work is Artificial Neural Network (ANN). It is used to improve the ensemble forecasted streamflow based on training and testing data sets of observed and forecasted streamflow. As for the ensemble streamflow predictions of Huai river, they are generated by XAJ operational hydrological model of the Huai River Commission, which uses the ensemble precipitation forecasts of GFS model post-processed with hydroscedastic logistic regression as initial condition. The main challenge addressed is how to build the ANN model for post-processing. At the start of this research, there was not a suitable ANN model for post-processing streamflow forecasts of the Huai river. Thus, one of the objectives of this research was to look for a good input method of ANN. In order to achieve this target, four experiments on Huai river are implemented in this research firstly. And then, the model which has the best performance is applied on Rhine river. The first experiment inputs the original ESP data directly, which
Titel: |
Artificial neural network post-processing of ensemble streamflow forecasts for Huai-he and Rhine river
|
---|---|
Link: | |
Veröffentlichung: | 2019 |
Medientyp: | Elektronische Ressource |
Schlagwort: |
|
Sonstiges: |
|