Convolutional Neural Network-based Statistical Post-processing of Ensemble Precipitation Forecasts
Abstract
Raw forecasts from numerical weather prediction models suffer from systematic bias and cannot be directly used in applications such as hydrological forecasting. Statistical post-processing methods can be used to remove the bias and achieve reliable ensemble forecasts. Traditional post-processing methods only use local precipitation forecasts as the only predictor, which limit their ability to extract information of raw forecasts. We develop a convolutional neural network-based post-processing method for precipitation forecasts to fully make use of spatial information and atmospheric circulation forecasts. The CNN used is similar to LeNet network. The outputs of the CNN are the parameters of the distribution of the predictands at the location to be predicted. We compare the proposed model with a state-of-the-art joint probability model and an ANN-based model. The results show that CNN-based post-processing model performs better than traditional methods in forecast accuracy, discrimination and reliability, especially for heavy rain at lead time of 2 days. The CNN-based post-processing model outperforms the joint probability model in Brier skill score by 9% at lead time of two days. Moreover, CNN-based models transcend ANN-based model by using convolution layers to extract spatial information. The results illustrate the advantages of CNN-based post-processing models to extract spatial information and different meteorological variable information to improve precipitation forecast skill.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.A15Q..06L