.png){kind=icon}
.png){kind=icon}
Return
3D U-Net correction of Northwest Pacific sea surface wind field forecasts incorporating wind-pressure relationship constraints
.png){kind=icon}
PRE
AI
Z
X
Z
DOI：10.1016/j.atmosres.2026.108939.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate sea surface wind forecasting is crucial for preventing marine disasters, ensuring maritime safety, and aiding oceanographic research. However, biases persist in wind field forecasts from the existing Global Forecast System (GFS) over the Northwest Pacific. To enhance GFS wind field forecast accuracy in this region, this study proposes an artificial-intelligence-based correction model. Specifically, the equations governing atmospheric motion (describing the wind-pressure relationship) are integrated into a three-dimensional U-Net as a physical loss term alongside a regression loss, and multi-objective optimization is performed. The proposed physical constraint correction model effectively improves the GFS wind field forecast at various lead times. In the case of 0-120-h forecasts using data from 2023, the optimal UVPS_PE scheme (where U, V, P, S, and PE denote the loss terms for the 10-m zonal wind, meridional wind, mean sea level pressure, 10-m wind speed, and wind-pressure constraints, respectively) reduces the mean absolute errors of wind speed and wind direction from 1.375 m/s to 1.148 m/s (16.509% reduction) and from 23.948 degrees to 21.399 degrees (10.644% reduction). Further analysis indicates that the optimal correction scheme achieves superior performance across the full year, as well as during both tropical cyclone and non-tropical cyclone periods, verifying its robust correction capability under different weather regimes. This study verifies the effectiveness of integrating physical constraints into deep learning methods to correct numerical weather predictions, providing valuable insights for developing physics-informed artificial intelligence approaches in atmospheric and oceanic modeling.
Keywords:
Sea surface wind
Artificial-intelligence-based forecast correction
Wind-pressure relationship
Physical constraint
Journal
IF:
4.4
Papers:
1.1K
Citations:
2.2W
Organization
Cited Papers