Jiao et al. (2025) Multi-Layer Soil Moisture Profiling Based on BKA-CNN by Integrating Sentinel-1/2 SAR and Multispectral Data

Identification

• Journal: Agronomy
• Year: 2025
• Date: 2025-10-31
• Authors: Menglong Jiao, Xuqing Li, Xiao Sun, Jian Wu, Tianjie Zhao, Ruiyin Tang, Yu Bai
• DOI: 10.3390/agronomy15112542

Research Groups

• Innovation Base for Natural Resources Monitoring Technology in the Lower Reaches of Yongding River, Geological Society of China, Langfang 065000, China
• State Key Laboratory of Remote Sensing Science, Beijing Normal University, Beijing 100875, China
• College of Remote Sensing and Information Engineering, North China Institute of Aerospace Engineering, Langfang 065000, China
• State Key Laboratory of Remote Sensing and Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China

Short Summary

This study developed a BKA-CNN model integrating Sentinel-1 SAR and Sentinel-2 multispectral data to estimate multi-layer soil moisture (SM) in the Shandian River Basin, achieving high accuracy (R² up to 0.799) across depths from 3 cm to 50 cm, with superior performance compared to single-source data and traditional machine learning models, and demonstrating robust generalization.

Objective

• To explore the performance of different data combinations (multispectral, SAR, and multispectral + SAR) in estimating soil moisture at various depths using machine learning algorithms.
• To investigate the performance differences of three models (Random Forest, XGBoost, and Convolutional Neural Network) in estimating soil moisture at different depths.
• To validate the optimal model at sites with different vegetation cover and explore how vegetation types affect the model’s functionality.

Study Configuration

• Spatial Scale: Shandian River Basin, China (approximately 12,700 square kilometers), with high-resolution (10 m) soil moisture maps generated.
• Temporal Scale: Data collected and analyzed from 14 December 2018, through 31 December 2019. Sentinel-1 data has a temporal resolution of 6 days, and Sentinel-2 data has a temporal resolution of 3 days.

Methodology and Data

• Models used: Black-winged Kite Algorithm (BKA)-optimized Convolutional Neural Network (BKA-CNN), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). The BKA was used for hyperparameter optimization of all models.
• Data sources:
  • Satellite: Sentinel-1 Synthetic Aperture Radar (SAR) data (VV, VH polarization, incidence angle) and Sentinel-2 Multispectral (MS) data (13 bands used to calculate 13 vegetation indices).
  • Observation: In-situ multi-layer soil moisture measurements from the SMN-SDR network (34 sites) at 3 cm, 5 cm, 10 cm, 20 cm, and 50 cm depths, obtained from the International Soil Moisture Network (ISMN).
  • Platform: Google Earth Engine (GEE) was used for satellite data retrieval and preprocessing.

Main Results

• Data Fusion Superiority: The combined use of multispectral (MS) and SAR data significantly outperformed single-source data (MS or SAR alone) for multi-layer soil moisture (SM) estimation (p < 0.05). MS data showed better performance in root-zone estimation, while SAR data excelled in surface soil moisture (SSM) estimation.
• Model Performance: The BKA-CNN model achieved significantly higher accuracy than RF and XGBoost models across all depths.
  • Five-fold cross-validation R² values for BKA-CNN were: 0.768 ± 0.011 (3 cm), 0.777 ± 0.013 (5 cm), 0.799 ± 0.011 (10 cm), 0.792 ± 0.01 (20 cm), and 0.782 ± 0.011 (50 cm).
  • Corresponding RMSE values for BKA-CNN were: 0.041 ± 0.004 m³/m³ (3 cm), 0.038 ± 0.003 m³/m³ (5 cm), 0.034 ± 0.004 m³/m³ (10 cm), 0.036 ± 0.004 m³/m³ (20 cm), and 0.035 ± 0.003 m³/m³ (50 cm).
• Optimal Depth for Fusion: The highest accuracy for the fused MS + SAR data was achieved at a 10 cm depth (BKA-CNN R² = 0.812).
• Vegetation Type Impact: Temporal validation showed that the BKA-CNN model performed better in grassland (R² ranging from 0.732 to 0.785) than in farmland (R² ranging from 0.706 to 0.758).
• Spatial Mapping: High-resolution (10 m) multi-layer SM spatial distribution maps were generated for the Shandian River Basin, demonstrating the model's effectiveness in capturing detailed spatial variations and generally increasing SM with depth, with regional variations influenced by precipitation and vegetation.

Contributions

• Systematic evaluation of multispectral, SAR, and their fused data to quantify their respective performances in soil moisture estimation across varying depths.
• Initial exploration of the feasibility and applicability of Convolutional Neural Networks (CNNs) optimized by the Black-winged Kite Algorithm (BKA) for multi-layer soil moisture retrieval.
• Production of high-resolution maps of multi-layer soil moisture, offering technical support for precision irrigation management and drought monitoring.

Funding

• Innovation Base for Natural Resources Monitoring Technology in the Lower Reaches of Yongding River, Geological Society of China (KY02202501)
• Central government guide local science and technology development fund project Science and technology innovation base project (246Z7401G)
• Open Fund of State Key Laboratory of Remote Sensing Science (Grant No. OFSLRSS202303)
• National Key Research and Development Program of China (No. 2021YFB3900104)
• Graduate Innovation Funding Project of North China Institute of Aerospace Engineering (YKY-2024-91)

Citation

@article{Jiao2025MultiLayer,
  author = {Jiao, Menglong and Li, Xuqing and Sun, Xiao and Wu, Jian and Zhao, Tianjie and Tang, Ruiyin and Bai, Yu},
  title = {Multi-Layer Soil Moisture Profiling Based on BKA-CNN by Integrating Sentinel-1/2 SAR and Multispectral Data},
  journal = {Agronomy},
  year = {2025},
  doi = {10.3390/agronomy15112542},
  url = {https://doi.org/10.3390/agronomy15112542}
}

Generated by BiblioAssistant using gemini-2.5-flash (Google API)