Chen et al. (2026) MDE-UNet: A Physically Guided Asymmetric Fusion Network for Multi-Source Meteorological Data Lightning Identification

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

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-03-29
• Authors: Yu Chen, Yuge Han, Yujian Zhang, Yi Liu, Lin Song, Jialei Wang, Xinjue Wang, Qilin Zhang
• DOI: 10.3390/rs18071027

Research Groups

Not provided in the text.

Short Summary

This paper proposes a novel lightning identification network that integrates multi-source meteorological data by addressing challenges like modal competition, data sparsity, and class imbalance, resulting in improved hit rates and reduced false alarms for lightning strike area identification.

Objective

• To develop a high-precision lightning identification network that effectively utilizes multi-source heterogeneous meteorological data (satellite and radar) by overcoming challenges such as dimensional imbalance, modal competition, model training bias from sparse lightning samples, and an imbalance between false alarms and missed detections.

Study Configuration

• Spatial Scale: 2 km grid scale
• Temporal Scale: Not specified in the provided text.

Methodology and Data

• Models used: A lightning identification network guided by physical priors and constrained by supervision, incorporating:
  • Physical prior-guided asymmetric radar information enhancement mechanism.
  • Multi-source multi-scale feature fusion module.
  • Weighted sliding window–multilayer perceptron (MLP) enhanced decoding unit.
  • Asymmetrically weighted BCE-DICE loss function.
• Data sources: Multi-source meteorological data, specifically satellite data and radar data.

Main Results

• The proposed method effectively increases the hit rate while substantially reducing the false alarm rate.
• It enables efficient utilization of multi-source data and high-precision identification of lightning strike areas.

Contributions

• Introduction of a physical prior-guided asymmetric radar information enhancement mechanism to address modal competition between high-dimensional satellite and low-dimensional radar data.
• Construction of a multi-source multi-scale feature fusion module for coupling multi-scale physical features at a 2 km grid scale, improving lightning signal detection.
• Development of a weighted sliding window–MLP enhanced decoding unit to smooth spatially discrete noise and ensure spatial continuity in reconstructed results.
• Design of an asymmetrically weighted BCE-DICE loss function to mitigate model bias caused by severe class imbalance and achieve synergistic optimization of feature enhancement and directional suppression.

Funding

Not provided in the text.

Citation

@article{Chen2026MDEUNet,
  author = {Chen, Yu and Han, Yuge and Zhang, Yujian and Liu, Yi and Song, Lin and Wang, Jialei and Wang, Xinjue and Zhang, Qilin},
  title = {MDE-UNet: A Physically Guided Asymmetric Fusion Network for Multi-Source Meteorological Data Lightning Identification},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18071027},
  url = {https://doi.org/10.3390/rs18071027}
}