%0 Journal Article %1 espeholt_deep_2022 %A Espeholt, Lasse %A Agrawal, Shreya %A Sønderby, Casper %A Kumar, Manoj %A Heek, Jonathan %A Bromberg, Carla %A Gazen, Cenk %A Carver, Rob %A Andrychowicz, Marcin %A Hickey, Jason %A Bell, Aaron %A Kalchbrenner, Nal %D 2022 %J Nature Communications %K Atmospheric Climate Computational Earth Statistics, and ecomodelling modelling, science science, system %N 1 %P 5145 %R 10.1038/s41467-022-32483-x %T Deep learning for twelve hour precipitation forecasts %U https://www.nature.com/articles/s41467-022-32483-x %V 13 %X Existing weather forecasting models are based on physics and use supercomputers to evolve the atmosphere into the future. Better physics-based forecasts require improved atmospheric models, which can be difficult to discover and develop, or increasing the resolution underlying the simulation, which can be computationally prohibitive. An emerging class of weather models based on neural networks overcome these limitations by learning the required transformations from data instead of relying on hand-coded physics and by running efficiently in parallel. Here we present a neural network capable of predicting precipitation at a high resolution up to 12 h ahead. The model predicts raw precipitation targets and outperforms for up to 12 h of lead time state-of-the-art physics-based models currently operating in the Continental United States. The results represent a substantial step towards validating the new class of neural weather models.

@article{espeholt_deep_2022, abstract = {Existing weather forecasting models are based on physics and use supercomputers to evolve the atmosphere into the future. Better physics-based forecasts require improved atmospheric models, which can be difficult to discover and develop, or increasing the resolution underlying the simulation, which can be computationally prohibitive. An emerging class of weather models based on neural networks overcome these limitations by learning the required transformations from data instead of relying on hand-coded physics and by running efficiently in parallel. Here we present a neural network capable of predicting precipitation at a high resolution up to 12 h ahead. The model predicts raw precipitation targets and outperforms for up to 12 h of lead time state-of-the-art physics-based models currently operating in the Continental United States. The results represent a substantial step towards validating the new class of neural weather models.}, added-at = {2023-07-31T08:05:54.000+0200}, author = {Espeholt, Lasse and Agrawal, Shreya and Sønderby, Casper and Kumar, Manoj and Heek, Jonathan and Bromberg, Carla and Gazen, Cenk and Carver, Rob and Andrychowicz, Marcin and Hickey, Jason and Bell, Aaron and Kalchbrenner, Nal}, biburl = {https://www.bibsonomy.org/bibtex/2bf8784d68964522be411bc3133d9cee9/jascal_panetzky}, copyright = {2022 The Author(s)}, doi = {10.1038/s41467-022-32483-x}, file = {Full Text PDF:/Users/pascal/Zotero/storage/JHFN8HPK/Espeholt et al. - 2022 - Deep learning for twelve hour precipitation foreca.pdf:application/pdf}, interhash = {24a12a7c62d61914a4692a613f6d893f}, intrahash = {bf8784d68964522be411bc3133d9cee9}, issn = {2041-1723}, journal = {Nature Communications}, keywords = {Atmospheric Climate Computational Earth Statistics, and ecomodelling modelling, science science, system}, language = {en}, month = sep, note = {Number: 1 Publisher: Nature Publishing Group}, number = 1, pages = 5145, timestamp = {2023-07-31T08:07:14.000+0200}, title = {Deep learning for twelve hour precipitation forecasts}, url = {https://www.nature.com/articles/s41467-022-32483-x}, urldate = {2023-07-18}, volume = 13, year = 2022 }