Predicting Profile Soil Properties with Reflectance Spectra via Bayesian Covariate-Assisted External Parameter...

Area of Science:

• Soil Science
• Spectroscopy
• Precision Agriculture

Background:

• In situ diffuse reflectance spectroscopy (DRS) offers rapid, high-resolution soil property prediction for agriculture and environmental monitoring.
• Environmental factors and data collection artifacts often limit the accuracy of in situ DRS data.
• Advanced processing (EPO) and Bayesian modeling (Bayesian Lasso) show promise in overcoming these limitations.

Purpose of the Study:

• To predict soil organic carbon (SOC), total nitrogen (TN), and texture fractions using in situ profile DRS spectra.
• To compare the performance of traditional Partial Least Squares (PLS) regression against enhanced models: PLS-EPO, PLS-EPO-BL, and covariate-assisted PLS-EPO-BL.

Main Methods:

• Collected in situ profile DRS spectra and soil cores (~1 m depth) from 708 samples across 22 fields in Missouri and Indiana, USA.
• Laboratory analysis of air-dried, sieved soil samples (<2 mm) for measured soil properties.
• Employed traditional PLS, PLS with external parameter orthogonalization (EPO) transformation, and Bayesian Lasso (BL) with and without categorical covariates (soil textural class).

Main Results:

• PLS-EPO significantly improved prediction accuracy over PLS alone, reducing root mean square error of prediction (RMSEP) by 53-76% for SOC, TN, and texture fractions.
• The addition of Bayesian Lasso further reduced RMSEP by 4-11%, and the categorical covariate provided an additional 2-9% reduction.
• The combined EPO transformation and Bayesian modeling approach effectively mitigated environmental factors and data collection artifacts.

Conclusions:

• The combination of EPO spectral transformation and Bayesian modeling techniques is highly effective for accurate in situ DRS data analysis.
• This approach overcomes challenges posed by environmental factors and data collection artifacts, enhancing soil property prediction.
• In-field DRS spectroscopy, when coupled with these advanced methods, holds significant potential for rapid, high-resolution soil property assessment.