GNSS-R Soil Moisture Retrieval Based on a XGboost Machine Learning Aided Method: Performance and Validation

Yan Jia, Shuanggen Jin, Patrizia Savi, Yun Gao, Jing Tang, Yixiang Chen, Wenmei LiRemote 2019 · DOI

TL;DR

This paper introduces an XGBoost-based machine learning approach to improve the accuracy and understanding of GNSS-Reflectometry (GNSS-R) for soil moisture retrieval. By analyzing the importance of various input variables, the study reveals that the signal-to-noise ratio (SNR) is the most critical factor, offering valuable insights for future GNSS-R model development and experimental design.

What Problem Does This Paper Address?

Accurately measuring soil moisture is crucial for agriculture, hydrology, and climate monitoring. GNSS-Reflectometry (GNSS-R) offers a promising, low-cost remote sensing technique for this purpose, utilizing reflected GNSS signals. However, existing GNSS-R soil moisture retrieval methods face challenges due to the complexity of modeling, the variability of experimental environments, and the inherent uncertainty in the data. Specifically, understanding which input parameters most significantly influence retrieval accuracy and how they interact remains a key challenge. This paper aims to address this by systematically evaluating the importance of different input variables in the GNSS-R soil moisture retrieval process.

Methods

The core of this research lies in the application of the XGBoost (Extreme Gradient Boosting) machine learning algorithm. XGBoost is an ensemble learning method known for its efficiency and ability to handle complex datasets. The authors leverage XGBoost's capability to perform variable importance assessment, which allows them to quantify the contribution of each input parameter to the soil moisture retrieval.

The study employs a two-pronged approach:

  1. Simulated Data Analysis: A simulated dataset was generated based on bistatic radar equations and established soil permittivity models. This allowed for an initial assessment of variable importance under controlled conditions.
  2. Ground-Truth Experiments: Real-world experiments were conducted at two distinct sites with different soil compositions (loamy sand and silty clay loam) and moisture conditions (dry and wet). Data from these experiments, collected using GNSS-R equipment and Time-Domain Reflectometry (TDR) for ground truth, were used to validate the findings from the simulated data.

The input variables considered for the XGBoost model included:

  • Elevation angle ($\theta$)
  • Receiver Gain ($G_r$)
  • Signal-to-Noise Ratio ($SNR_{rh}$)
  • Noise Power ($P_n$)

The output variable was primarily permittivity, which is directly related to soil moisture content.

Datasets

The study utilized two main types of datasets:

  • Simulated Dataset: Generated using bistatic radar equations and soil permittivity models to create a controlled environment for initial analysis. This dataset contained 1000 samples.
  • Ground-Truth Datasets: Collected from two experimental sites in Grugliasco and Agliano, Italy, during January-March 2016. These datasets involved measurements under both dry and wet soil conditions, with TDR providing the ground-truth soil moisture and permittivity values. The paper details four specific campaigns within these sites.

Key Insights and Ideas

  • SNR Reigns Supreme: The most striking finding is the overwhelming importance of the Signal-to-Noise Ratio (SNR) from the reflected GNSS signal. It consistently emerged as the most sensitive and influential variable for both permittivity and soil moisture retrieval, contributing significantly (40-70%) to the model's performance. This highlights the critical role of signal quality in GNSS-R.
  • Elevation Angle's Minor Role: Conversely, the elevation angle ($\theta$) of the satellite was found to have the least impact on retrieval accuracy, often contributing less than 10% and sometimes close to zero. This suggests that while crucial for signal reception, its direct influence on soil moisture estimation is less pronounced compared to signal strength.
  • Soil Type Matters, But SNR Dominates: While different soil compositions (loamy sand vs. silty clay loam) showed some variations in sensitivity, the dominance of SNR remained consistent across all soil types and conditions. This implies that robust signal acquisition is paramount, regardless of the specific soil characteristics.
  • XGBoost as a Powerful Tool: The study successfully demonstrates XGBoost's utility not just for prediction but also for interpreting the underlying physics of GNSS-R by quantifying variable importance. This offers a data-driven approach to understand complex remote sensing phenomena.
  • Validation of Bistatic Radar Principles: The ground-truth experiments and subsequent analysis (including linear fitting) provided strong validation for the relationships between SNR, elevation angle, permittivity, and soil moisture, aligning well with theoretical expectations.

Key Results

  • Variable Importance Ranking: The consistent order of variable importance was found to be: SNR > $P_n$ ≈ $G_r$ > $\theta$.
  • SNR Contribution: The received SNR was identified as the predominant variable, contributing between 40% and 70% to the retrieval accuracy.
  • Elevation Angle's Low Importance: The elevation angle ($\theta$) was the least important variable, with a contribution often below 10%.
  • Good Correlation: The GNSS-R retrieval results showed a good correlation with the TDR-measured soil moisture and permittivity, especially when the TDR measurements fell within the satellite's Fresnel zone.
  • Sensitivity Differences: The study noted that silty clay loam soils exhibited more permittivity variation with changes in SNR compared to loamy sand soils, indicating a higher sensitivity in wetter, clay-rich environments.

Limitations and Future Work

  • Limited Spatial and Temporal Scale: The ground-truth experiments were conducted at specific sites and over a limited time frame. The authors suggest future work should involve longer-term monitoring to account for seasonal effects and expand the analysis to a regional scale.
  • More Terrain Types: The study focused on two main soil types. Future research could benefit from including more diverse terrain types to further evaluate the robustness of the findings.
  • Experimental Environment Uncertainty: The authors acknowledge that factors like equipment interference and complex environmental conditions can affect received signals and introduce uncertainty. While XGBoost helps mitigate some of this, further investigation into these factors is warranted.
  • Focus on Input Variables: While the study excels at evaluating input variable importance, future work could delve deeper into optimizing the GNSS-R receiver construction and impact analysis based on these findings.
  • Exploration of Other ML Algorithms: While XGBoost proved effective, exploring other machine learning algorithms for comparison and further insight could be beneficial.

Summary updated

Abstract

Global navigation satellite system (GNSS)-reflectometry is a type of remote sensing technology and can be applied to soil moisture retrieval. Until now, various GNSS-R soil moisture retrieval methods have been reported. However, there still exist some problems due to the complexity of modeling and retrieval process, as well as the extreme uncertainty of the experimental environment and equipment. To investigate the behavior of bistatic GNSS-R soil moisture retrieval process, two ground-truth measurements with different soil conditions were carried out and the performance of the input variables was analyzed from the mathematical statistical aspect. Moreover, the feature of XGBoost method was utilized as well. As a recently developed ensemble machine learning method, the XGBoost method just emerged for the classification of remote sensing and geographic data, to investigate the characterization of the input variables in the GNSS-R soil moisture retrieval. It showed a good correlation with the statistical analysis of ground-truth measurements. The variable contributions for the input data can also be seen and evaluated. The study of the paper provides some experimental insights into the behavior of the GNSS-R soil moisture retrieval. It is worthwhile before establishing models and can also help with understanding the underlying GNSS-R phenomena and interpreting data.
BibTeX
@inproceedings{jia2019gnss,
  author = {Yan Jia and Shuanggen Jin and Patrizia Savi and Yun Gao and Jing Tang and Yixiang Chen and Wenmei Li},
  title = {{GNSS-R Soil Moisture Retrieval Based on a XGboost Machine Learning Aided Method: Performance and Validation}},
  booktitle = {Remote},
  year = {2019},
  doi = {10.3390/rs11141655},
}
Figure