Inversion of soil organic carbon content based on the two-point machine learning method.

Area of Science:

• Earth and Environmental Sciences
• Remote Sensing
• Soil Science

Background:

• Soil organic carbon (SOC) is crucial for global carbon cycling and sustainable development.
• Remote sensing offers a convenient method for monitoring SOC, but faces challenges with spatial heterogeneity and limited ground data.
• Existing methods struggle with complex spatial relationships between SOC and spectral data due to environmental influences.

Purpose of the Study:

• To introduce and evaluate the two-point machine learning (TPML) method for high-precision SOC inversion.
• To address limitations in current SOC monitoring, particularly complex spatial relationships and scarce ground samples.
• To provide a robust method for large-scale SOC estimation with uncertainty assessment.

Main Methods:

• Applied the two-point machine learning (TPML) method to invert SOC content in Hailun County.
• Utilized derived variables from Sentinel-1, Sentinel-2, topography, and environmental data.
• Employed 10-fold cross-validation and t-tests to assess accuracy and compare with other machine learning models.

Main Results:

• TPML demonstrated superior inversion accuracy (average r=0.854, MAE=0.384%, RMSE=0.558%, RPD=1.918) compared to random forest, gradient boosting, PLS, and SVM.
• TPML effectively evaluated inversion uncertainty by comparing actual and theoretical errors.
• Generated a 10m resolution SOC map with smoother details and better land-use consistency than other models.

Conclusions:

• The TPML method is highly effective for accurate and large-scale SOC inversion using remote sensing data.
• TPML offers significant advantages in handling spatial heterogeneity and providing uncertainty estimates.
• This study provides guidance for utilizing TPML for soil attribute prediction and low-cost, high-precision SOC monitoring.