Liu et al. (2025) Reconstruction of SMAP Soil Moisture Data Based on Residual Autoencoder Network with Convolutional Feature Extraction

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-11-16
• Authors: Yaojie Liu, Haoyu Fan, Yan Jin, Shaonan Zhu
• DOI: 10.3390/rs17223729

Research Groups

Not specified in the provided text.

Short Summary

This study introduces TsSMNet, a residual autoencoder model that combines multi-source remote sensing inputs with statistical time-series features to reconstruct gap-free surface soil moisture (SSM) estimates, demonstrating superior performance and improved spatial coverage compared to existing models.

Objective

• To develop and evaluate a novel residual autoencoder model (TsSMNet) for reconstructing seamless, gap-free surface soil moisture (SSM) estimates from satellite products, addressing issues of spatial gaps and reduced reliability.

Study Configuration

• Spatial Scale: Regional (China), 9 km resolution.
• Temporal Scale: 7 years (2016 to 2022).

Methodology and Data

• Models used: TsSMNet (residual autoencoder with 1D convolutional layers), AutoResNet, Transformer, Random Forest, XGBoost (for comparison).
• Data sources: SMAP product (satellite-based surface soil moisture), multi-source remote sensing inputs, statistical features derived from SSM time series (central tendency, dispersion, variability, extremes, distribution, temporal dynamics, magnitude, energy, count-based features), in situ observations from six networks in the International Soil Moisture Network (for evaluation).

Main Results

• TsSMNet significantly outperforms AutoResNet, Transformer, Random Forest, and XGBoost models.
• It reduces the root mean square error (RMSE) by an average of 17.1 percent.
• Achieves a mean RMSE of 0.09 cm³/cm³.
• Feature importance analysis reveals a strong contribution of temporal predictors to model accuracy.
• Compared to its variant without time-series features, TsSMNet provides better spatial representation and improved consistency with in situ temporal observations.
• The reconstructed product offers improved spatial coverage and continuity relative to the original SMAP data.

Contributions

• Introduction of TsSMNet, a novel residual autoencoder model that effectively integrates multi-source remote sensing inputs with comprehensive statistical time-series features for SSM reconstruction.
• Demonstrated superior performance of TsSMNet in generating seamless SSM data, outperforming several state-of-the-art machine learning and deep learning models.
• Generation of a high-resolution (9 km) and continuous SSM product over China for 2016-2022, enhancing data availability for regional hydrological and climate studies.
• Highlighted the critical role of temporal predictors in improving the accuracy and spatial representation of reconstructed SSM.

Funding

Not specified in the provided text.

Citation

@article{Liu2025Reconstruction,
  author = {Liu, Yaojie and Fan, Haoyu and Jin, Yan and Zhu, Shaonan},
  title = {Reconstruction of SMAP Soil Moisture Data Based on Residual Autoencoder Network with Convolutional Feature Extraction},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17223729},
  url = {https://doi.org/10.3390/rs17223729}
}