Wang et al. (2026) Improving Daily Rainfall Downscaling Over Taiwan's Complex Terrain With Surface‐Wind‐Informed Deep Learning Model

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: International Journal of Climatology
• Year: 2026
• Date: 2026-10-04
• Authors: Yi‐Chi Wang, Chia‐Hao Chiang, Chiung-Jui Su, Ko‐Chih Wang, Wan‐Ling Tseng, Hsin‐Chien Liang
• DOI: 10.1002/joc.70607

Research Groups

• Department of Civil Engineering, National Taiwan University
• Institute of Atmospheric Sciences, National Central University

Short Summary

This study evaluates a deep neural network model to bias-correct and downscale ERA5 daily rainfall in Taiwan using multi-head attention, improving the reproduction of high-resolution gridded rainfall fields for regions with steep terrain.

Objective

• Investigate the effectiveness of an encoder-decoder deep neural network with multi-head attention in correcting biases and downsampling ERA5 daily rainfall to match Taiwan's 5-km TCCIP gridded observations.

Study Configuration

• Spatial Scale: Regional (Taiwan)
• Temporal Scale: Daily, seasonal, and interannual

Methodology and Data

• Models used: Encoder–Decoder with Attention (EDA) model
• Data sources:
  • ERA5 daily rainfall
  • TCCIP gridded observations (5-km resolution)
  • 10-m winds
  • Coarsened TCCIP rainfall
  • 5-km topography

Main Results

• Improved placement of orographic rainfall
• Reduced low-intensity wet bias in ERA5
• Better reproduction of RX1day, RX10mm, CDD, and interannual variability
• Largest gains occur in seasons dominated by monsoon and typhoon forcing

Contributions

• Original value lies in demonstrating the effectiveness of attention-based statistical downscaling for regions with steep terrain
• Improves upon existing bias correction methods (BCSD) in reproducing high-resolution gridded rainfall fields

Funding

• This research was funded by [project name], [program name] (reference code: [code])

Citation

@article{Wang2026Improving,
  author = {Wang, Yi‐Chi and Chiang, Chia‐Hao and Su, Chiung-Jui and Wang, Ko‐Chih and Tseng, Wan‐Ling and Liang, Hsin‐Chien},
  title = {Improving Daily Rainfall Downscaling Over Taiwan's Complex Terrain With Surface‐Wind‐Informed Deep Learning Model},
  journal = {International Journal of Climatology},
  year = {2026},
  doi = {10.1002/joc.70607},
  url = {https://doi.org/10.1002/joc.70607}
}