Abstract
Although numerical weather prediction models are effective for short-term precipitation forecast, it is difficult to avoid prediction errors. To correct the numerical weather prediction model output, it might be effective to apply deep learning techniques, which have led to remarkable achievements in processing big data.
In this study, we developed a deep learning method using a convolution neural network (CNN) to correct numerical weather prediction model outputs. This method outputs the distribution of rainfall amount, and can be used as precipitation guidance.
Data augmentation and data selection (rejection of no rainfall data) were applied. As input data, the vertical near-surface wind speed and precipitation predicted by the CReSiBUC numerical weather prediction model in the Keihanshin region of Japan between August 1 and 31, 2001, were used. As training data, we used the observed rainfall based on the Radar/Raingauge-Analyzed Precipitation provided by the Japan Meteorological Agency. The observed data were interpolated into the same meshes using the nearest neighbor method simultaneously with the prediction results.
The two-dimensional rainfall distributions before and after being corrected by the deep-learning method were compared with the observed data from the Radar Raingauge Analyzed Precipitation. Inputting the vertical near-surface wind speed into the deep-learning model improved the accuracy during a typhoon rainfall event. We found that introducing the deep-learning technique improved the accuracy of predicating precipitation over a large spatial scale during a typhoon, although this technique did not improve the accuracy during a localized heavy rainfall event.
