Soil Organic Carbon Prediction Based on Vis-NIR Spectral Classification Data Using GWPCA-FCM Algorithm.

Area of Science:

• Soil Science
• Spectroscopy
• Data Analysis

Background:

• Visible and near-infrared reflectance spectroscopy offers rapid soil organic carbon (SOC) estimation.
• Large spectral libraries are valuable for SOC research, but direct application faces challenges due to soil variability.
• Improving SOC prediction accuracy from spectral data is crucial for soil management.

Purpose of the Study:

• To enhance soil organic carbon (SOC) prediction accuracy using spectral classification techniques.
• To evaluate the effectiveness of geographically weighted principal component analysis (GWPCA) combined with fuzzy c-means (FCM) clustering for spectral classification.
• To compare the performance of spectral classification with land cover type classification for SOC prediction.

Main Methods:

• Utilized the European Land Use and Cover Area frame Survey (LUCAS) spectral library.
• Applied GWPCA and FCM clustering to classify soil spectra.
• Employed partial least squares regression (PLSR) and Cubist models for SOC prediction, comparing spectral vs. land cover classification.

Main Results:

• The GWPCA-FCM-Cubist model achieved the highest prediction accuracy (R² = 0.83, RPIQ = 2.95), improving upon unclassified models.
• Spectral classification using GWPCA-FCM significantly outperformed land cover type classification for both PLSR and Cubist models.
• Cubist models generally provided better SOC prediction accuracy than PLSR models.

Conclusions:

• Spectral classification, particularly using GWPCA and FCM, substantially enhances SOC prediction from large spectral libraries.
• The proposed GWPCA-FCM-Cubist approach offers a robust method for accurate SOC estimation.
• This methodology improves the direct application of spectral libraries in soil science research and management.