Estimation of soil free Iron content using spectral reflectance and machine learning algorithms

Area of Science:

• Soil Science
• Remote Sensing
• Data Science

Background:

• Free iron is a key soil property influencing soil evolution and characteristics.
• Traditional methods for determining soil free iron are often destructive and time-consuming.
• Spectral reflectance offers a rapid, non-destructive, and cost-effective alternative for soil property estimation.

Purpose of the Study:

• To evaluate the feasibility of using spectral reflectance and machine learning (ML) to estimate soil free iron content.
• To compare different spectral transforms, variable selection techniques, and ML algorithms for optimal model performance.

Main Methods:

• Collected spectral reflectance data from 540 soil samples across 135 locations.
• Applied spectral transforms: original, first derivative (FD), standard normal variate (SNV), and continuum removed (CR).
• Utilized principal component analysis (PCA) for variable selection and employed ML algorithms: partial least squares (PLS), support vector machine (SVM), random forest (RF), and deep neural network (DNN).

Main Results:

• The first derivative (FD) transform showed efficient training performance (R²=0.797).
• Principal component analysis (PCA) for variable selection improved model accuracy (training R²=0.821, testing R²=0.692).
• The optimal model combination (FD + PCA + SVM) achieved high accuracy in both training (R²=0.876) and testing (R²=0.803), with low RMSE and RRMSE.

Conclusions:

• Spectral reflectance coupled with machine learning provides a reliable method for estimating soil free iron content.
• This approach offers rapid, non-destructive, and economical soil analysis.
• The findings support applications in soil property mapping, crop nutrient management, and environmental monitoring.