Applying artificial intelligence to better predict real weather (IMAGE)
Caption
Scientists transform the gridded weather forecast correction problem into an image-to-image translation problem in deep learning. The block diagram indicates the architecture of a deep neural network termed CU-net (Correction U-net). Also shown is the superior performance of CU-net in correcting ECMWF-IFS forecasts of temperature, relative humidity, and wind, with a forecast lead time of 24 to 240 hours.
Credit
Advances in Atmospheric Sciences
Usage Restrictions
none
License
Original content