Abstract
While widely recognized as one of the most substantial weather forecasting methodologies, Numerical Weather Prediction (NWP) usually suffers from relatively coarse resolution and inevitable bias due to tempo-spatial discretization, physical parametrization process, and computation limitation. With the roaring growth of deep learning-based techniques, we propose the Dual-Stage Adaptive Framework (DSAF), a novel framework to address regional NWP downscaling and bias correction tasks. DSAF uniquely incorporates adaptive elements in its design to ensure a flexible response to evolving weather conditions. Specifically, NWP downscaling and correction are well-decoupled in the framework and can be applied independently, which strategically guides the optimization trajectory of the model. Utilizing a multi-task learning mechanism and an uncertainty-weighted loss function, DSAF facilitates balanced training across various weather factors. Additionally, our specifically designed attention-centric learnable module effectively integrates geographic information, proficiently managing complex interrelationships. Experimental validation on the ECMWF operational forecast (HRES) and reanalysis (ERA5) archive demonstrates DSAF's superior performance over existing state-of-the-art models and shows substantial improvements when existing models are augmented using our proposed modules.
Figure 1: (a) and (b) Comparison of Correction, pure Downscaling and Our Tasks for 2m Temperature: Illustration of specific differences in
The datasets utilized in this study are derived from the European Centre for Medium-Range Weather Forecasts (ECMWF) operational forecast (HRES) and reanalysis (ERA5) archive. For regional NWP downscaling, we construct a real-world dataset called Huadong, covering the East China land and sea areas. In this dataset, HRES data is employed as the predictive data, while ERA5 reanalysis data serves as the ground truth.
- HRES: https://confluence.ecmwf.int/display/FUG/HRES+-+High-Resolution+Forecast
- ERA5: https://cds.climate.copernicus.eu/cdsapp#!/home
Dataset Details. The Huadong dataset encompasses a latitude range from sp
), 2m temperature (2t
), 2m dewpoint temperature (d2m
), skin temperature (skt
), 10m u component of wind (10u
), 10m v component of wind (10v
), 100m u component of wind (100u
), and 100m v component of wind (100v
).
-
Data processing:
data_processing/ERA5.py
anddata_processing/HRES.py
-
2x task:
2x_task/main.py
-
4x task:
4x_task/main.py
In this work, we approximate the DSAF/2x_task/train.py
for the 2x task and DSAF/4x_task/train.py
for the 4x task. The
The fourth-order different difference kernels with the shape of