Inversion Estimation of Soil Organic Matter in Songnen Plain Based on Multispectral Analysis

Siyu Tang, Chong Du, Tangzhe NieLand 2022 · DOI

TL;DR

This study leverages Sentinel-2A multispectral satellite imagery and advanced spectral processing techniques to accurately estimate soil organic matter (SOM) content in the Songnen Plain. By applying differential algorithms and comparing Multiple Linear Stepwise Regression (MLSR) and Partial Least Squares Regression (PLSR) models, the research demonstrates that MLSR, particularly with reciprocal logarithm first-order differential processing, offers a stable and reliable method for mapping SOM, outperforming traditional lab-based analyses for large-scale monitoring.

What Problem Does This Paper Address?

Accurately mapping soil organic matter (SOM) is crucial for understanding soil fertility, carbon sequestration, and sustainable land management. Traditional laboratory methods for determining SOM are time-consuming, labor-intensive, and impractical for large-scale assessments. While remote sensing offers a promising alternative, challenges remain in extracting reliable SOM information from spectral data, especially considering factors like mixed pixels, water content, and varying spectral resolutions. Existing studies have explored various spectral transformations and modeling techniques, but there's a need to systematically evaluate the effectiveness of advanced differential processing methods combined with robust regression models for SOM estimation using readily available satellite data like Sentinel-2A.

Methods

The researchers employed a multi-step approach to estimate SOM content:

  • Data Acquisition: Sentinel-2A multispectral satellite images from July 2021 were used, focusing on ten visible and near-infrared bands with 10m spatial resolution.
  • Ground Truth Data: 19 soil samples were collected from the study area (Daqing, Songnen Plain), and their SOM content was determined using the potassium dichromate method. These samples were divided into calibration (14 sites) and validation (5 sites) sets.
  • Image Pre-processing: Sentinel-2A images underwent radiometric and atmospheric correction, geometric correction, mosaicking, and clipping to isolate the study area.
  • Spectral Transformations: Raw spectral reflectance data was subjected to various mathematical transformations, including reciprocal, reciprocal logarithm, square root, square, and cubic transformations. Crucially, first-order (R') and second-order (SDR) differential processing was applied to these transformed spectra.
  • Model Development:
    • Single-Band Models: Correlation analysis was performed between individual spectral indices (derived from transformations) and SOM content to identify sensitive bands.
  • Multi-Band Models: * Multiple Linear Stepwise Regression (MLSR): This method iteratively selects the most significant spectral bands to build a predictive model, aiming to avoid multicollinearity.
  • Partial Least Squares Regression (PLSR): This technique extracts principal components from both spectral data and SOM content to establish a predictive relationship.
  • Model Evaluation: Models were assessed using calibration and validation metrics, primarily the coefficient of determination (R²) and Root Mean Square Error (RMSE).

Datasets

  • Sentinel-2A Multispectral Imagery: Six Sentinel-2A images acquired in July 2021 covering the Songnen Plain study area.
  • Soil Samples: 19 surface soil samples collected from the Daqing region, with SOM content determined in the laboratory. These were split into 14 calibration and 5 validation samples.

Key Insights and Ideas

  • Differential Processing Enhances SOM Correlation: Mathematical transformations, particularly square-root and reciprocal logarithm, significantly improved the correlation between spectral data and SOM content compared to raw reflectance. First and second-order differential processing further refined these correlations by reducing noise and enhancing spectral features.
  • Multi-band Models Outperform Single-band: Combining information from multiple spectral bands through MLSR and PLSR models consistently yielded better prediction accuracy and stability than single-band approaches.
  • MLSR with Reciprocal Logarithm First-Order Differential is Optimal: The study identified the MLSR model utilizing the reciprocal logarithm first-order differential spectral index as the most effective for SOM estimation, demonstrating the highest prediction accuracy and stability.
  • Soil Salinity Impacts Spectral Signatures: The discussion highlights that high soil salinity in the southwest of the study area likely masked spectral signatures of SOM, leading to lower inversion values, suggesting that soil properties can influence the effectiveness of remote sensing.
  • Sentinel-2A's Suitability: The research confirms Sentinel-2A's capability for monitoring and mapping soil organic matter, especially when combined with advanced spectral processing.

Key Results

  • Improved Correlation: Mathematical transformations, especially square-root, significantly improved the correlation between spectral values and SOM content.
  • MLSR Superiority: MLSR models generally showed better predictive performance and stability compared to PLSR models in this study.
  • Optimal Model Performance: The reciprocal logarithm first-order differential MLSR regression model achieved the best prediction and inversion results, closely matching real-world data.
  • Validation Accuracy: The optimal MLSR model demonstrated good prediction accuracy and stability in validation, indicating its reliability for mapping SOM.
  • Spatial Distribution: The study successfully generated a spatial distribution map of SOM content for the Songnen Plain, revealing generally low and uneven distribution, with higher concentrations in the northeast.

Limitations and Future Work

  • Limited Sample Size and Diversity: The study used a relatively small number of soil samples (19) and focused on a specific region. Future work should expand the sample size, diversify soil types, and incorporate soil moisture and microorganism data.
  • Influence of Soil Salinity: The high soil salinity in the study area was identified as a factor that can mask SOM spectral signatures. Future research could explore integrating salinity indices to improve SOM estimation in such environments.
  • Weather Impact: The study acknowledges that weather conditions can affect spectral acquisition. Future work could involve using indoor light sources to generate controlled reflectance spectra, eliminating weather impacts.
  • Further Model Refinement: While MLSR showed strong performance, further exploration of advanced spectral indices and integration with other remote sensing data (e.g., SAR) could potentially enhance SOM estimation. The authors also suggest exploring the integration of salinity indices for more precise spectral parameter screening.

Summary updated

Abstract

Sentinel-2A multi-spectral remote sensing image data underwent high-efficiency differential processing to extract spectral information, which was then matched to soil organic matter (SOM) laboratory test values from field samples. From this, multiple-linear stepwise regression (MLSR) and partial least square (PLSR) models were established based on a differential algorithm for surface SOM modeling. The original spectra were subjected to basic transformations with first- and second-derivative processing. MLSR and PLSR models were established based on these methods and the measured values, respectively. The results show that Sentinel-2A remote sensing imagery and SOM content correlated in some bands. The correlation between the spectral value and SOM content was significantly improved after mathematical transformation, especially square-root transformation. After differential processing, the multi-band model had better predictive ability (based on fitting accuracy) than single-band and unprocessed multi-band models. The MLSR and PLSR models of SOM had good prediction functionality. The reciprocal logarithm first-order differential MLSR regression model had the best prediction and inversion results (i.e., most consistent with the real-world data). The MLSR model is more stable and reliable for monitoring SOM content, and provides a feasible method and reference for SOM content-mapping of the study area.
BibTeX
@inproceedings{tang2022inversion,
  author = {Siyu Tang and Chong Du and Tangzhe Nie},
  title = {{Inversion Estimation of Soil Organic Matter in Songnen Plain Based on Multispectral Analysis}},
  booktitle = {Land},
  year = {2022},
  doi = {10.3390/land11050608},
}
Figure