Ji et al. (2026) Robust hyperspectral reconstruction from satellite and airborne observations via a deep hierarchical fusion network across heterogeneous scenarios

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-03-25
• Authors: Fujiang Ji, Jiaqi Yang, Philip A. Townsend, Ting Zheng, Kyle R. Kovach, Tong Yu, Ruqi Yang, Ming Liu, Min Chen
• DOI: 10.1016/j.rse.2026.115385

Research Groups

• Department of Forest and Wildlife Ecology, University of Wisconsin-Madison, USA
• Department of Biological Systems Engineering, University of Wisconsin-Madison, USA
• Department of Computer Sciences, University of Wisconsin-Madison, USA

Short Summary

This study develops a deep learning framework for robust high spatial resolution hyperspectral imagery (HR-HSI) reconstruction by fusing low-resolution hyperspectral (EMIT) and high-resolution multispectral (PlanetScope) satellite data. The framework consistently outperforms state-of-the-art models, demonstrating high spectral fidelity and reconstruction accuracy across diverse landscapes.

Objective

• To develop and evaluate a deep learning fusion framework capable of reconstructing high spatial resolution hyperspectral imagery (HR-HSI) by integrating low-resolution hyperspectral data from NASA's EMIT with high-resolution multispectral imagery from PlanetScope, specifically addressing performance under real cross-sensor conditions.

Study Configuration

• Spatial Scale: Three ecologically distinct landscapes in the western United States, enabling fine-scale ecological and environmental monitoring.
• Temporal Scale: Single acquisition period for the satellite and airborne imagery used for reconstruction and validation.

Methodology and Data

• Models used: A custom deep hierarchical fusion network (deep learning framework).
• Data sources:
  • Low-resolution hyperspectral imagery: NASA's Earth Surface Mineral Dust Source Investigation (EMIT) satellite.
  • High-resolution multispectral imagery: PlanetScope satellite.
  • Reference data (airborne): AVIRIS-NG and AVIRIS-3 imagery.
  • Sensor inter-calibration pipeline applied to EMIT and PlanetScope data.

Main Results

• The proposed deep learning framework consistently outperformed seven peer state-of-the-art fusion models.
• Achieved a best spectral angle mapper (SAM) of 2.64 and a root mean squared error (RMSE) of 0.0267 when validated against AVIRIS observations.
• Additional metrics confirmed strong spectral fidelity and reconstruction accuracy: Mean Absolute Error (MAE) = 0.0174, Erreur Relative Globale Adimensionnelle de Synthèse (ERGAS) = 3.62, Peak Signal-to-Noise Ratio (PSNR) = 28.87.
• Stratified analyses across vegetation density gradients (NDVI classes), spectral subregions (Visible, Near-Infrared, and Shortwave Infrared), and pixel-level distributions demonstrated strong robustness in heterogeneous landscapes.
• A sensor inter-calibration pipeline successfully ensured radiometric consistency between EMIT and PlanetScope, enhancing the reliability of the fusion inputs.

Contributions

• Developed a novel deep learning fusion framework that robustly reconstructs HR-HSI from real, heterogeneous satellite observations (EMIT and PlanetScope), addressing the limitations of models trained on synthetic data.
• Demonstrated superior performance and robustness of the proposed framework compared to existing state-of-the-art fusion models across diverse and heterogeneous landscapes.
• Integrated a sensor inter-calibration pipeline to ensure radiometric consistency between disparate satellite sensors, improving the accuracy and reliability of the fused products.
• Provided a scalable and high-fidelity solution for generating HR-HSI from current satellite observations, supporting fine-scale applications in vegetation trait retrieval, ecosystem monitoring, and environmental change analysis.

Funding

Not specified in the provided text.

Citation

@article{Ji2026Robust,
  author = {Ji, Fujiang and Yang, Jiaqi and Townsend, Philip A. and Zheng, Ting and Kovach, Kyle R. and Yu, Tong and Yang, Ruqi and Liu, Ming and Chen, Min},
  title = {Robust hyperspectral reconstruction from satellite and airborne observations via a deep hierarchical fusion network across heterogeneous scenarios},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115385},
  url = {https://doi.org/10.1016/j.rse.2026.115385}
}