Developing a near-infrared spectroscopy calibration algorithm for soil organic carbon content in South Africa

Near-infrared (NIR) spectroscopy has emerged as an easy, rapid and cost-effective alternative for soil organic carbon (SOC) analysis and the accounting of carbon credits. South Africa currently lacks a calibration algorithm for predicting SOC content from NIR spectroscopy. This study aimed to develo...

Full description

Saved in:
Bibliographic Details
Main Authors: Willie Herman Cloete, Gerhard du Preez, George Munnik Van Zijl
Format: Article
Language:English
Published: Elsevier 2025-06-01
Series:Soil Advances
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2950289625000077
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Near-infrared (NIR) spectroscopy has emerged as an easy, rapid and cost-effective alternative for soil organic carbon (SOC) analysis and the accounting of carbon credits. South Africa currently lacks a calibration algorithm for predicting SOC content from NIR spectroscopy. This study aimed to develop a NIR spectroscopy calibration algorithm for SOC content, specific to South Africa. Soil samples were collected from 2 fields and 3 catchments across South Africa. These samples were analysed using the total dry combustion (TDC) method and scanned with a NIR spectrometer. Sixty NIR calibration algorithms were developed on a regional scale. The impact of methodological parameters, such as sample state, sampling design, processing and machine learning models, on the root mean square error (RMSE) of the validation statistics was also assessed. Although 60 regional-scale calibration algorithms were developed, none were suitable (RMSE = 0.39 % and RPIQ > 2) for SOC content prediction, which was attributed to the small sample size (n = 238). However, local calibration models for the Tsitsa catchment and Ottosdal fields presented great accuracy (RMSE < 0.1 and RPIQ > 1.5) that can be used for future SOC content prediction. The study found that the open spectral library global prediction model poorly predicted SOC content using local data (RMSE = 1.23 % and R² = - 0.83). This was attributed to South African samples being underrepresented in the global dataset. Sample state and sampling design were the most influential parameters affecting RMSE. To develop a national calibration algorithm, effort should be placed on developing accurate calibration algorithms for smaller areas that could be added to the national spectral library.
ISSN:2950-2896