A generative adversarial network approach to (ensemble) weather prediction

We use a conditional deep convolutional generative adversarial network to predict the geopotential height of 500 hPa pressure level, two-meter temperature and total precipitation for next 24 h over Europe. The proposed models are trained on 4 years ERA5 reanalysis data from with goal associated meteorological fields in 2019. forecasts show good qualitative quantitative agreement true temperature, while failing precipitation, thus indicating that weather based alone may be possible specific parameters. further Monte-Carlo dropout develop an ensemble prediction system purely learning strategies, which is computationally cheap improves skill forecasting model, by allowing quantify uncertainty current forecast as learned model.