Rapid Determination of Low Heavy Metal Concentrations in Grassland Soils around Mining Using Vis-NIR Spectroscopy: A Case Study of Inner Mongolia, China

Aru Han, Xiaoling Lu, Song Qing, Yongbin Bao, Yuhai Bao, Qing Ma, Xingpeng Liu, Jiquan Zhang2021 · DOI

TL;DR

This study presents a rapid and cost-effective method for determining low concentrations of heavy metals (Cr and Ni) in grassland soils using Vis-NIR spectroscopy. By exploring various spectral preprocessing techniques and statistical modeling approaches, the researchers developed an optimized method that significantly improves accuracy and stability compared to traditional lab analysis, offering a valuable tool for environmental monitoring.

What Problem Does This Paper Address?

Coal mining, while economically beneficial, poses a significant environmental threat through heavy metal contamination of surrounding soils. Traditional methods for monitoring these contaminants are laborious, expensive, and time-consuming, making large-scale, real-time assessment challenging. This limits our ability to understand and mitigate the risks associated with heavy metal accumulation in soil, which can impact food safety and human health. The paper addresses the need for a faster, more efficient, and cost-effective approach to quantify heavy metal concentrations in soil, particularly at low levels where detection is more difficult.

Methods

The study leverages Visible and Near-Infrared (Vis-NIR) spectroscopy, a non-destructive technique that measures how soil reflects light across a spectrum. The core of their methodology involves:

  • Continuum Removal (CR): This technique normalizes spectral reflectance to highlight absorption features, effectively enhancing subtle spectral signals that might otherwise be masked by background noise.
  • Spectral Preprocessing: A variety of methods were applied to clean and enhance the spectral data, including:
  • Savitzky-Golay (SG) smoothing: A common filter for reducing noise.
  • Normalization (NOR): Adjusts spectra to a common baseline.
  • Multiplicative Scatter Correction (MSC): Corrects for additive and multiplicative effects of light scattering.
  • Standard Normal Variate (SNV): Similar to MSC, it corrects for scattering effects.
  • Spectral Transformations: These methods further process the spectra to emphasize specific features:
  • First Derivative (FD): Highlights subtle spectral features and reduces baseline drift.
  • Second Derivative (SD): Further refines spectral features and can help resolve overlapping absorption bands.
  • Reciprocal Logarithm (log(1/R)): Can enhance absorption features.
  • Statistical Modeling: Three regression techniques were employed to build predictive models:
  • Partial Least Squares Regression (PLSR): A widely used method for handling multicollinearity in spectral data.
  • Principal Component Regression (PCR): Reduces dimensionality by extracting principal components.
  • Support Vector Machine Regression (SVMR): A powerful method for handling nonlinear relationships.

The researchers systematically tested 16 combinations of preprocessing and transformation methods with each statistical model to identify the optimal approach. They also explored using the entire Vis-NIR spectrum versus a specific band range (600-800 nm) associated with soil organic matter (SOM).

Datasets

The study utilized soil samples collected from grasslands around two open-cast coal mines in Inner Mongolia, China: the Huolinhe coal mine and the Baiyinhua coal mine. A total of 37 soil samples were collected from the top 0-10 cm layer. These samples were analyzed for their heavy metal concentrations (Cr and Ni), soil organic matter (SOM), and water content. The collected spectral data and corresponding chemical analysis formed the dataset for model development and validation.

Key Insights and Ideas

  • CR is Crucial for Low-Level Detection: Continuum Removal proved essential for enhancing subtle spectral signals, making it possible to detect and quantify low concentrations of heavy metals that would otherwise be obscured.
  • MSC and SNV Shine: Among the preprocessing techniques, MSC and SNV transformations, especially when combined with Second Derivative (SD) spectral transformation, showed the most significant improvement in model accuracy and stability.
  • Vis-NIR is a Powerful Tool: The study confirms that Vis-NIR spectroscopy, when combined with appropriate data processing and modeling, can accurately predict heavy metal concentrations in soil.
  • SOM as a Proxy: While direct spectral signatures of low heavy metal concentrations are weak, their correlation with spectrally active soil components like SOM provides an indirect but effective means of estimation.
  • PLSR Dominates: PLSR consistently outperformed PCR and SVMR in terms of both model stability (R²) and accuracy (lower RMSE), highlighting its suitability for this type of spectral analysis.

Key Results

The study achieved high accuracy in determining Cr and Ni concentrations:

  • Optimal Preprocessing: The combination of MSC-SD and SNV-SD preprocessing methods were found to be the most effective for determining Cr and Ni concentrations.
  • Model Performance (Vis-NIR):
    • For Cr, the best models (using MSC-SD or SNV-SD) achieved R² values above 0.98 and RMSE values as low as 0.51.
  • For Ni, the best models achieved R² values up to 0.99 and RMSE values as low as 0.07.
  • Model Stability and Accuracy: The Vis-NIR-based PLSR model demonstrated high stability with R² values above 0.99 and RMSE values between 0.07 and 0.34, indicating strong predictive power.
  • Comparison of Methods: PLSR consistently showed better performance than SVMR and PCR, with PCR exhibiting the lowest accuracy.
  • Validation: Independent validation using data from nine sampling points confirmed the PLSR model's good quantitative prediction ability, with R² values of 0.54 for Cr and 0.57 for Ni.

Limitations and Future Work

  • Study Area Specificity: The developed models are specific to the grassland soils of Inner Mongolia. Further research is needed to validate these methods in different soil types and geographical regions.
  • Low Concentration Focus: While the study excels at determining low concentrations, its performance at very high concentrations might differ.
  • Future Work: The authors suggest promoting and verifying the SNV-SD-PLSR method in other study areas, including field spectral analysis and the use of UAV and satellite remote sensing data. They also propose exploring other factors like Fe-Mn oxides and developing multi-spectral transformation indices for more comprehensive heavy metal monitoring.

Summary updated

BibTeX
@misc{han2021rapid,
  author = {Aru Han and Xiaoling Lu and Song Qing and Yongbin Bao and Yuhai Bao and Qing Ma and Xingpeng Liu and Jiquan Zhang},
  title = {{Rapid Determination of Low Heavy Metal Concentrations in Grassland Soils around Mining Using Vis-NIR Spectroscopy: A Case Study of Inner Mongolia, China}},
  year = {2021},
  doi = {10.3390/s21093220},
}
Figure