A generative adversarial network approach to (ensemble) weather\n prediction

2020/06/13 by Alexander Bihlo, Bihlo, Alexander · 1 citation
Earth and Planetary Sciences · Physics and Astronomy · Environmental Science · #Meteorological Phenomena and Simulations #Model Reduction and Neural Networks #Climate variability and models

paper · pdf · doi:10.48550/arxiv.2006.07718

Abstract

We use a conditional deep convolutional generative adversarial network to\npredict the geopotential height of the 500 hPa pressure level, the two-meter\ntemperature and the total precipitation for the next 24 hours over Europe. The\nproposed models are trained on 4 years of ERA5 reanalysis data from 2015-2018\nwith the goal to predict the associated meteorological fields in 2019. The\nforecasts show a good qualitative and quantitative agreement with the true\nreanalysis data for the geopotential height and two-meter temperature, while\nfailing for total precipitation, thus indicating that weather forecasts based\non data alone may be possible for specific meteorological parameters. We\nfurther use Monte-Carlo dropout to develop an ensemble weather prediction\nsystem based purely on deep learning strategies, which is computationally cheap\nand further improves the skill of the forecasting model, by allowing to\nquantify the uncertainty in the current weather forecast as learned by the\nmodel.\n

Cited by

• An intercomparison of generative machine learning methods for downscaling precipitation at fine spatial scales

Related

• Precipitation nowcasting with generative diffusion models
• Deep Learning and Foundation Models for Weather Prediction: A Survey
• Deep learning for post-processing global probabilistic forecasts on sub-seasonal time scales
• Probabilistic weather forecasting with machine learning
• Generative ensemble deep learning severe weather prediction from a deterministic convection-allowing model