An et al. (2026) Toward spatially sharper precipitation prediction via global-local frequency guidance

Identification

• Journal: npj Climate and Atmospheric Science
• Year: 2026
• Date: 2026-06-22
• Authors: Subin An, Youngwook Kim, Dongjin Cho, Yoo‐Geun Ham
• DOI: 10.1038/s41612-026-01472-y

Research Groups

• Department of Environmental Managements, Graduate School of Environmental Studies, Seoul National University, Seoul, Republic of Korea
• School of Earth and Environmental Sciences, Seoul National University, Seoul, Republic of Korea
• Environmental Planning Institute, Seoul National University, Seoul, Republic of Korea
• Institute for Sustainable Development (ISD), Seoul National University, Seoul, Republic of Korea

Short Summary

This study proposes the Wavelet-Fourier Composite Loss (WFCL) to improve precipitation prediction by jointly enforcing global spectral consistency and local phase coherence. WFCL is shown to outperform existing methods in preserving both structural realism and extreme event detection.

Objective

• Develop a composite loss framework that balances global spectral consistency and local phase characteristics for improved precipitation prediction

Study Configuration

• Spatial Scale: Global, with a grid resolution of 1° x 1°
• Temporal Scale: Daily, with a lead time of up to 3 days

Methodology and Data

• Models used: Convolutional encoder-decoder architecture with geocyclic padding and convolution-batch normalization-LeakyReLU (CBL) blocks
• Data sources: ERA5 reanalysis dataset for precipitation and lower-tropospheric specific humidity predictors

Main Results

• WFCL outperforms MSE and FACL in terms of structural quality, spatial prediction skill, and extreme event detection at a 1-day lead time
• WFCL reduces LPIPS by approximately 45%, decreases RHD by 67%, and improves CPBD by 5%
• WFCL consistently achieves lower LPC values, indicating strengthened local phase alignment while preserving amplitude

Contributions

• The study introduces the Wavelet-Fourier Composite Loss (WFCL) framework that balances global spectral consistency and local phase characteristics for improved precipitation prediction
• WFCL is shown to outperform existing methods in preserving both structural realism and extreme event detection

Funding

• This research was funded by [project name, program code, and reference]

Citation

@article{An2026Toward,
  author = {An, Subin and Kim, Youngwook and Cho, Dongjin and Ham, Yoo‐Geun},
  title = {Toward spatially sharper precipitation prediction via global-local frequency guidance},
  journal = {npj Climate and Atmospheric Science},
  year = {2026},
  doi = {10.1038/s41612-026-01472-y},
  url = {https://doi.org/10.1038/s41612-026-01472-y}
}