Jin et al. (2026) Improving FY-4B Satellite Precipitation Retrieval over Coastal Complex Terrain of Eastern China: Deep Learning Approaches with Multi-Source Underlying Surface Data

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

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-07-19
• Authors: Xi Jin, Zuodong Yang, Shoujuan Shu, Meiying Dong, Huiyan Xu, C ZHANG, Xiayi Lang, Hong Yuan, Hangfeng Shen
• DOI: 10.3390/rs18142397

Research Groups

Not specified in the provided text.

Short Summary

This study develops a deep learning framework to improve precipitation retrieval from FY-4B satellite data by integrating underlying-surface information. The findings demonstrate that incorporating topographic and land-cover data enhances precipitation detection, with the degree of improvement depending on the specific neural network architecture used.

Objective

• To evaluate and quantify the contribution of underlying-surface information (DEM, slope, surface roughness, and LULC) to the accuracy of precipitation retrieval using FY-4B multi-channel infrared brightness temperatures and cloud-top temperature (CTT).

Study Configuration

• Spatial Scale: Regional (specifically focusing on complex regions with varying topography and land cover).
• Temporal Scale: Not specified.

Methodology and Data

• Models used: DS-UNet, SmaAt-UNet (lightweight architectures), U-Net, and Attention U-Net (computationally intensive architectures).
• Data sources:
  • Satellite: Fengyun-4B (FY-4B) infrared brightness temperatures and cloud-top temperature (CTT).
  • Surface Data: Digital Elevation Model (DEM), topographic factors (slope, surface roughness), and land use/land cover (LULC).
  • Validation Data: Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG) and rain-gauge observations.

Main Results

• Incorporating underlying-surface information improves the overall detection of precipitation.
• Model architecture significantly influences the utility of the input data:
  • Lightweight models (DS-UNet, SmaAt-UNet) show higher sensitivity to discrete LULC boundaries, providing better spatial constraints.
  • Deeper models (U-Net, Attention U-Net) are more effective at encoding continuous topographic gradients.
• Simple combinations of DEM and LULC are not always effective due to spatial scale mismatches and thermodynamic forcing; however, integrating multiple surface factors simultaneously alleviates these conflicts and improves retrieval skill.

Contributions

• Demonstrates that the integration of surface-based auxiliary data with satellite infrared data enhances precipitation monitoring in complex regions.
• Identifies a critical relationship between the choice of deep learning architecture and the type of surface information (discrete vs. continuous) used for retrieval.

Funding

Not specified in the provided text.

Citation

@article{Jin2026Improving,
  author = {Jin, Xi and Yang, Zuodong and Shu, Shoujuan and Dong, Meiying and Xu, Huiyan and ZHANG, C and Lang, Xiayi and Yuan, Hong and Shen, Hangfeng},
  title = {Improving FY-4B Satellite Precipitation Retrieval over Coastal Complex Terrain of Eastern China: Deep Learning Approaches with Multi-Source Underlying Surface Data},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18142397},
  url = {https://doi.org/10.3390/rs18142397}
}