Skilful nowcasting of extreme precipitation with NowcastNet
Problem
Extreme precipitation is a considerable contributor to meteorological disasters, and existing nowcasting methods suffer blur, dissipation, intensity or location errors; physics-based methods miss chaotic dynamics while data-driven methods violate physical laws.
Why it matters
Skilful high-resolution nowcasting is needed to mitigate the socioeconomic effects of extreme precipitation disasters.
What the AI did
NowcastNet unifies physical-evolution schemes and conditional-learning methods in a neural-network framework with end-to-end forecast error optimization to produce extreme-precipitation nowcasts from radar observations.
Results
Physically plausible nowcasts with sharp multiscale patterns over 2,048 km × 2,048 km regions with lead times up to 3 h, based on radar observations from the USA and China.
Impact
Evidence supports skilful nowcasts up to 3 hours ahead over large regions; disaster mitigation itself is not demonstrated.
Record
| Venue | Nature |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Yuchen Zhang, Mingsheng Long, Kaiyuan Chen, Lanxiang Xing, Ronghua Jin, Michael I. Jordan, Jianmin Wang |
| Citations | 451 (OpenAlex, at import time) |
| References | 26 |
| Model or method | NowcastNet |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41586-023-06184-4 |
| Open access | hybrid |
| Keywords | nowcasting, extreme precipitation, neural network, radar, weather forecasting |
Importance rating
major — millions of lives, an industry or a field
Abstract states extreme precipitation is a considerable contributor to meteorological disasters with great need to mitigate socioeconomic effects; skilful nowcasting over USA/China affects millions.