[Inversion of Soil Organic Matter Content Using Hyperspectral Data Based on Continuous Wavelet Transformation]

Area of Science:

• Soil Science
• Remote Sensing
• Agricultural Engineering

Background:

• Soil organic matter content (SOMC) is a key indicator of soil fertility and essential for crop production.
• Dynamic monitoring of SOMC is vital for advancing precision agriculture practices.
• Proximal hyperspectral sensing is increasingly used for soil analysis, necessitating robust predictive algorithms.

Purpose of the Study:

• To investigate the effectiveness of hyperspectral techniques combined with wavelet analysis for predicting soil organic matter content (SOMC).
• To compare different machine learning models for SOMC estimation using spectral and wavelet features.
• To identify the optimal model for stable and accurate SOMC prediction in soil samples.

Main Methods:

• Hyperspectral reflectance data were collected using an ASD FieldSpec3 spectrum analyzer.
• Soil organic matter content (SOMC) was determined using the potassium dichromate external heating method.
• Continuous Wavelet Transform (CWT) was applied to raw spectra (R) and continuum removal curves (CR) to extract wavelet coefficients. Partial Least Squares Regression (PLSR), BP Neural Network (BPNN), and Support Vector Machine Regression (SVMR) were used for model building.

Main Results:

• Correlation analyses identified sensitive wavebands and wavelet coefficients for SOMC prediction, with R2 values exceeding 0.15 for raw reflectance and 0.3 for wavelet coefficients.
• Models utilizing wavelet coefficients (R-CWT, CR-CWT) showed improved prediction accuracy (increase in R2 by 0.15-0.2) compared to raw spectral reflectance.
• The CR-CWT-SVMR model achieved the best performance on the validation set with R2=0.83, RMSE=4.02, and RPD=2.48, demonstrating comprehensive and stable SOMC estimation. The CR-CWT-PLSR model offered computational efficiency.

Conclusions:

• Hyperspectral sensing combined with continuous wavelet transform significantly enhances the accuracy of soil organic matter content (SOMC) prediction.
• The CR-CWT-SVMR model provides a robust and accurate method for estimating SOMC, suitable for precision agriculture applications.
• Wavelet analysis offers a valuable approach for extracting meaningful features from hyperspectral data, improving soil property estimation and potentially enabling efficient field-based soil sensing.