Soil Moisture Retrieval from the Chinese GF-3 Satellite and Optical Data over Agricultural Fields

Linlin Zhang, Qingyan Meng, Shun Yao, Qiao Wang, Jiangyuan Zeng, Shaohua Zhao, Jianwei Ma2018 · DOI

TL;DR

This paper introduces a novel algorithm for retrieving soil moisture using China's GF-3 C-band SAR satellite data, combined with optical imagery. By developing a backscattering coefficient simulation database and effectively accounting for vegetation effects, the method achieves accurate soil moisture estimates, particularly when using both HH and VV polarizations.

What Problem Does This Paper Address?

Accurate and timely soil moisture information is crucial for agricultural monitoring, impacting everything from irrigation scheduling to yield prediction. While satellite remote sensing offers a way to monitor soil moisture over large areas, several challenges remain. Active microwave sensors like SAR are valuable for their high spatial resolution and all-weather capabilities, but their measurements are significantly influenced by vegetation cover. Vegetation scatters and attenuates the radar signal, making it difficult to isolate the soil's contribution. Existing methods often struggle to accurately correct for these vegetation effects, especially over diverse agricultural landscapes. This paper aims to address this by developing a robust soil moisture retrieval algorithm for the newly launched Chinese GF-3 satellite, which offers multi-polarization capabilities, by effectively integrating optical data to mitigate vegetation interference.

Methods

The core of the proposed method involves a multi-step process:

  1. Vegetation Water Content (VWC) Estimation: The study first identifies the most suitable optical vegetation index (VI) for estimating VWC. Eight different VIs derived from Landsat-8 data were evaluated against field measurements. The Normalized Difference Infrared Index (NDII) showed the best correlation (R=0.885) and was used to derive VWC for agricultural fields. This VWC is a critical input for the subsequent vegetation correction step.
  2. Bare Soil Backscatter Simulation: An Advanced Integral Equation Model (AIEM) was employed to generate a comprehensive simulation database of bare soil backscattering coefficients. This database covers a wide range of soil moisture, surface roughness, incidence angles, and polarizations (HH and VV), tailored to the GF-3 satellite's specifications.
  3. Vegetation Correction using Water Cloud Model (WCM): The Water Cloud Model (WCM) was used to account for the scattering and attenuation effects of vegetation. By combining the simulated bare soil backscatter, the estimated VWC, and GF-3's observed total backscatter, the WCM was used to derive the bare soil backscattering coefficient. This step effectively "removes" the vegetation's influence from the SAR signal.
  4. Soil Moisture Retrieval: The retrieved bare soil backscattering coefficient was then matched against the simulated bare soil backscattering coefficient database. The soil moisture value that minimized the difference between the observed and simulated values was selected as the retrieved soil moisture. This minimization was performed using a cost function, and the accuracy was evaluated for HH, VV, and HH+VV polarization combinations.

Datasets

  • GF-3 Satellite Data: Quad-polarization strip I (QPSI) mode images acquired on May 27, 2017 (wheat fields) and July 12, 2017 (corn fields). These were preprocessed to Level-1A Single-Look Complex (SLC) imagery.
  • Landsat-8 Operational Land Imager (OLI) Data: Land surface reflectance products (Collection 1 Level-2) with 30m spatial resolution were used to derive vegetation indices.
  • In-situ Field Measurements: Ground measurements of soil moisture (0-5 cm depth) and vegetation water content (VWC) were collected from 38 wheat fields and 28 corn fields in the Luancheng study area. These were used for both VWC estimation and algorithm validation.

Key Insights and Ideas

  • NDII as a Strong VWC Proxy: The study highlights the effectiveness of the Normalized Difference Infrared Index (NDII) derived from optical data as a reliable proxy for vegetation water content in agricultural settings, outperforming other common vegetation indices. This is crucial for accurate vegetation correction in SAR-based soil moisture retrieval.
  • WCM's Effectiveness for Vegetation Correction: The Water Cloud Model (WCM) proved effective in separating the vegetation's contribution from the total radar backscatter, allowing for the extraction of a cleaner bare soil signal. This is a critical step for improving soil moisture retrieval accuracy.
  • Synergy of GF-3 and Optical Data: The combination of GF-3's C-band SAR data and optical data (Landsat-8) offers a powerful approach for soil moisture monitoring, leveraging the strengths of both sensor types.
  • HH+VV Polarization Advantage: Combining HH and VV polarizations for soil moisture retrieval yielded the highest accuracy, suggesting that utilizing multiple co-polarization channels provides more robust information than single channels alone.

Key Results

  • VWC Estimation: NDII provided the best fit for VWC estimation with R = 0.885 and RMSE = 0.484 kg m⁻².
  • WCM Validation: Simulated total backscatter from the WCM showed good agreement with GF-3 measured backscatter (R = 0.839 for HH, R = 0.797 for VV).
  • Bare Soil Backscatter Correlation: Bare soil backscatter (σsoil) showed a stronger correlation with in-situ soil moisture (R > 0.650) compared to total backscatter (σtotal) (R < 0.383).
  • Soil Moisture Retrieval Accuracy: The proposed algorithm achieved satisfactory soil moisture retrieval accuracy:
  • HH+VV polarization: RMSE = 0.044 m³m⁻³, R = 0.717
    • HH polarization: RMSE = 0.049 m³m⁻³, R = 0.652
    • VV polarization: RMSE = 0.053 m³m⁻³, R = 0.636

Limitations and Future Work

  • Co-polarization Focus: The study primarily utilized HH and VV polarizations. The authors acknowledge that GF-3's quad-polarization capability offers potential for further improvement, particularly by incorporating cross-polarization (HV or VH) data, which could account for multiple scattering effects more effectively.
  • Vegetation Parameter Dependency: The WCM vegetation parameters (A and B) are crop-dependent and were derived for the specific study area. Future work will focus on obtaining these parameters for broader applicability and for different crop types.
  • Limited Validation Area: The algorithm was validated in an agricultural region dominated by wheat and corn. Future research will involve validating the algorithm in more diverse regions with varying surface conditions to assess its generalizability.
  • AIEM Limitations: While AIEM is a powerful model, the authors note that it doesn't inherently account for multiple scattering in cross-polarization, which is a limitation for future cross-polarization analysis.

Summary updated

BibTeX
@misc{zhang2018soil,
  author = {Linlin Zhang and Qingyan Meng and Shun Yao and Qiao Wang and Jiangyuan Zeng and Shaohua Zhao and Jianwei Ma},
  title = {{Soil Moisture Retrieval from the Chinese GF-3 Satellite and Optical Data over Agricultural Fields}},
  year = {2018},
  doi = {10.3390/s18082675},
}
Figure