Scholarlink

<p>Precipitation prediction is critical for water-related hazard preparedness, and deep learning has shown strong potential. However, conventional mean squared error (MSE) training oversmooths high-frequency variability, producing blurred precipitation fields. Although frequency-domain losses such as Fourier Amplitude and Correlation Loss (FACL) enhance global spectral consistency, they provide limited constraints on local phase coherence. To address this, we propose the Wavelet-Fourier Composite Loss (WFCL), integrating Wavelet Amplitude and Correlation Loss (WACL) based on the Dual-Tree Complex Wavelet Transform with FACL to jointly enforce global spectral alignment and localized phase coherence. Experiments with a U-Net show that, for 1-day lead time prediction, WFCL reduces Learned Perceptual Image Patch Similarity from 0.1964 to 0.1079 relative to MSE, improves Local Phase Coherence by approximately 51%, and decreases Regional Histogram Distance by about 23% locally and 37% globally compared with FACL. Furthermore, WFCL maintains stable and superior performance across 1–3 day lead times during 2018–2023, demonstrating its effectiveness in preserving both local phase coherence and global spectral consistency in short-range precipitation forecasting.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Toward spatially sharper precipitation prediction via global-local frequency guidance

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月22日

查看全文
  • Subin An,
  • Youngwook Kim,
  • Dongjin Cho,
  • Yoo-Geun Ham

摘要

Precipitation prediction is critical for water-related hazard preparedness, and deep learning has shown strong potential. However, conventional mean squared error (MSE) training oversmooths high-frequency variability, producing blurred precipitation fields. Although frequency-domain losses such as Fourier Amplitude and Correlation Loss (FACL) enhance global spectral consistency, they provide limited constraints on local phase coherence. To address this, we propose the Wavelet-Fourier Composite Loss (WFCL), integrating Wavelet Amplitude and Correlation Loss (WACL) based on the Dual-Tree Complex Wavelet Transform with FACL to jointly enforce global spectral alignment and localized phase coherence. Experiments with a U-Net show that, for 1-day lead time prediction, WFCL reduces Learned Perceptual Image Patch Similarity from 0.1964 to 0.1079 relative to MSE, improves Local Phase Coherence by approximately 51%, and decreases Regional Histogram Distance by about 23% locally and 37% globally compared with FACL. Furthermore, WFCL maintains stable and superior performance across 1–3 day lead times during 2018–2023, demonstrating its effectiveness in preserving both local phase coherence and global spectral consistency in short-range precipitation forecasting.