Yang et al. (2025) Short-Term Frost Prediction During Apple Flowering in Luochuan Using a 1D-CNN–BiLSTM Network with Attention Mechanism

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

Identification

• Journal: Horticulturae
• Year: 2025
• Date: 2025-12-30
• Authors: Chenxi Yang, Huaibo Song
• DOI: 10.3390/horticulturae12010047

Research Groups

Not explicitly mentioned in the paper.

Short Summary

This study proposes a novel hybrid 1D-CNN-BiLSTM-Attention model, incorporating a dual attention mechanism, to enhance the prediction of early spring frost events during the Apple Flowering period, demonstrating improved classification performance and a 4-hour lead time for mitigation.

Objective

• To improve the prediction of early spring frost events during the Apple Flowering period using a novel hybrid deep learning model.

Study Configuration

• Spatial Scale: Luochuan (a specific geographical location).
• Temporal Scale: Data from 1997–2016; increased temporal resolution to 4 hours; 4-hour lead time for prediction.

Methodology and Data

• Models used: Hybrid 1D-CNN-BiLSTM-Attention model, integrating a 1D Convolutional Neural Network (1D-CNN), Bidirectional Long Short-Term Memory (BiLSTM), Self-attention mechanism, and Cross-variable Attention mechanism.
• Data sources: Observational meteorological data from Luochuan (1997–2016) including Ground Surface Temperature (GST), Air Temperature (TEM), Wind Speed (WS), and Relative Humidity (RH). Data processing involved a segmented interpolation method to increase temporal resolution and an adaptive Savitzky–Golay Filter for noise reduction.

Main Results

• The proposed hybrid model achieved higher Recall, Precision, and F1-score for frost classification compared to baseline models.
• The model demonstrated good agreement with actual frost events observed in Luochuan on 6, 9, and 10 April 2013.
• It provides a 4-hour lead time for frost event prediction, enabling timely implementation of mitigation measures.

Contributions

• Introduction of a novel hybrid deep learning architecture (1D-CNN-BiLSTM-Attention) with an integrated dual attention mechanism (Self-attention and Cross-variable Attention) for early spring frost prediction.
• Demonstrated significant improvement in frost classification performance compared to existing baseline models.
• Provides a practical 4-hour lead time for frost prediction, offering valuable guidance for agricultural decision-making and loss alleviation.
• Offers a technical reference for frost prediction during the Apple Flowering period in similar regions.

Funding

Not explicitly mentioned in the paper.

Citation

@article{Yang2025ShortTerm,
  author = {Yang, Chenxi and Song, Huaibo},
  title = {Short-Term Frost Prediction During Apple Flowering in Luochuan Using a 1D-CNN–BiLSTM Network with Attention Mechanism},
  journal = {Horticulturae},
  year = {2025},
  doi = {10.3390/horticulturae12010047},
  url = {https://doi.org/10.3390/horticulturae12010047}
}