Soil organic carbon measurements influence FT‐NIR model training in calcareous soils of Saskatchewan DOI Creative Commons
Gbenga Daniel ADEJUMO, David Bulmer, Preston Sorenson

et al.

Soil Science Society of America Journal, Journal Year: 2025, Volume and Issue: 89(2)

Published: March 1, 2025

Abstract This study compares acid digestion and temperature ramping methods for obtaining soil organic carbon (SOC) reference data to train Fourier transform near infrared (FT‐NIR) models in carbonate‐rich Saskatchewan agricultural soils. FT‐NIR spectra were measured on samples ( n = 431) from Dark Brown Chernozem soil, with quantification of inorganic carbon. Spectra transformed using continuous wavelet analyzed cubist regression tree models. Models built a 70:30 test split validation approach. Spectral feature selection, scale, model hyperparameter optimization conducted fivefold cross‐validation analysis the training dataset. All metrics calculated testing The method identified outliers (SIC) greater than 1.5%, which not detected method. SOC SIC prediction accuracy was higher (coefficient determination: R 2 0.66 0.63, Lin's concordance: ccc 0.78 0.77) compared 0.44 0.42, 0.64 0.62). Total (TC) similar both 0.58, 0.71). Removing high carbonate (SIC > 1.5%) improved TC 0.70, 0.81 SOC; 0.64, 0.75 TC) but when suggests that content may negatively affects accuracy, especially relying upon data.

Language: Английский

Soil organic carbon measurements influence FT‐NIR model training in calcareous soils of Saskatchewan DOI Creative Commons
Gbenga Daniel ADEJUMO, David Bulmer, Preston Sorenson

et al.

Soil Science Society of America Journal, Journal Year: 2025, Volume and Issue: 89(2)

Published: March 1, 2025

Abstract This study compares acid digestion and temperature ramping methods for obtaining soil organic carbon (SOC) reference data to train Fourier transform near infrared (FT‐NIR) models in carbonate‐rich Saskatchewan agricultural soils. FT‐NIR spectra were measured on samples ( n = 431) from Dark Brown Chernozem soil, with quantification of inorganic carbon. Spectra transformed using continuous wavelet analyzed cubist regression tree models. Models built a 70:30 test split validation approach. Spectral feature selection, scale, model hyperparameter optimization conducted fivefold cross‐validation analysis the training dataset. All metrics calculated testing The method identified outliers (SIC) greater than 1.5%, which not detected method. SOC SIC prediction accuracy was higher (coefficient determination: R 2 0.66 0.63, Lin's concordance: ccc 0.78 0.77) compared 0.44 0.42, 0.64 0.62). Total (TC) similar both 0.58, 0.71). Removing high carbonate (SIC > 1.5%) improved TC 0.70, 0.81 SOC; 0.64, 0.75 TC) but when suggests that content may negatively affects accuracy, especially relying upon data.

Language: Английский

Citations

0