Liu et al. (2025) Ensemble modelling based on transfer learning for enhancing crop mapping through synergistic integration of InSAR coherence and multispectral satellite data

Identification

• Journal: Computers and Electronics in Agriculture
• Year: 2025
• Date: 2025-12-23
• Authors: Niantang Liu, Qunshan Zhao, Richard Williams, Si-Bo Duan, Yingwei Sun, Brian Barrett
• DOI: 10.1016/j.compag.2025.111332

Research Groups

• State Key Laboratory of Efficient Utilization of Arid and Semi-arid Arable Land in Northern China, Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, Beijing, China
• Urban Big Data Centre, School of Social and Political Sciences, University of Glasgow, Glasgow, United Kingdom
• School of Geographical and Earth Sciences, University of Glasgow, Glasgow, United Kingdom
• China Institute of Water Resource and Hydropower Research, Beijing, China
• Earth Sciences New Zealand, Hamilton, New Zealand

Short Summary

This study proposes an innovative ensemble deep learning framework, Transformer-AtLSTM-RF, to enhance crop mapping in smallholder intercropping systems by synergistically integrating multi-temporal Sentinel-1 InSAR coherence with Sentinel-2 and RapidEye multispectral data, achieving high classification accuracies in Bei'an county, China.

Objective

• To propose and evaluate an innovative framework for enhancing crop mapping in smallholder croplands, particularly in complex intercropping patterns, by synergistically integrating multi-temporal Sentinel-1 InSAR coherence with Sentinel-2 and RapidEye multispectral satellite data using ensemble deep learning models and transfer learning.

Study Configuration

• Spatial Scale: Smallholder croplands in Bei’an county, China, focusing on two specific test sites (Site A and Site B).
• Temporal Scale: Multi-temporal (time series) data from Sentinel-1, Sentinel-2, and RapidEye satellites.

Methodology and Data

• Models used:
  • Deep Learning: 3-Dimensional U-Net (3D U-Net), Transformer, Attention-based Long Short-Term Memory (AtLSTM).
  • Machine Learning: Random Forest (RF).
  • Ensemble Model: Transformer-AtLSTM-RF (a new architecture using ensemble learning to fuse features from different classifiers with a rule-based strategy).
  • Techniques: Transfer learning, ensemble learning, feature importance assessment.
• Data sources:
  • Satellite: Multi-temporal Sentinel-1 Interferometric Synthetic Aperture Radar (InSAR) coherence.
  • Satellite: Sentinel-2 multispectral data.
  • Satellite: RapidEye multispectral data.

Main Results

• The proposed Transformer-AtLSTM-RF ensemble architecture, fine-tuned with region-specific data, achieved high crop mapping accuracies in complex intercropping patterns.
• For test site A, the model yielded an Overall Accuracy (OA) of 96.2%, a mean F1 score of 92.7%, and a mean Intersection over Union (mIoU) of 86.9%.
• For test site B, the model achieved an OA of 90.7%, a mean F1 score of 88.6%, and an mIoU of 79.7%.
• Feature importance assessment provided insights into critical temporal features during the model inference process, improving the understanding of underlying patterns in feature learning.

Contributions

• Proposes an innovative ensemble deep learning framework (Transformer-AtLSTM-RF) that effectively integrates multi-temporal InSAR coherence with multispectral optical data for enhanced crop mapping.
• Demonstrates the synergistic value of combining SAR-derived and optical time series data for improving crop classification, particularly in challenging smallholder intercropping systems.
• Evaluates the transfer learning capabilities of various deep learning models (3D U-Net, Transformer, AtLSTM) and a baseline Random Forest model in complex agricultural landscapes.
• Provides an in-depth understanding of underlying patterns in the feature learning process through feature importance visualization.

Funding

• Not specified in the provided text.

Citation

@article{Liu2025Ensemble,
  author = {Liu, Niantang and Zhao, Qunshan and Williams, Richard and Duan, Si-Bo and Sun, Yingwei and Barrett, Brian},
  title = {Ensemble modelling based on transfer learning for enhancing crop mapping through synergistic integration of InSAR coherence and multispectral satellite data},
  journal = {Computers and Electronics in Agriculture},
  year = {2025},
  doi = {10.1016/j.compag.2025.111332},
  url = {https://doi.org/10.1016/j.compag.2025.111332}
}