A machine learning case study in estimating total organic carbon contents in soil and sediments of Tainan science and technology park, Taiwan
Journal
IET Conference Proceedings
Journal Volume
2025
Journal Issue
15
Start Page
267
End Page
272
ISSN
2732-4494
Date Issued
2025-08
Author(s)
Abstract
Total Organic Carbon (TOC) is an essential parameter in natural carbon cycle of soil and sediments and therefore is vital for global carbon storage estimation. Measurement, monitoring, and reporting TOC are essential in carbon reduction as a nature-based solution measure. Conventional chemical measurements are time-consuming, costly, and could produce further waste and cause environmental pollution. Non-destructive methods such as Fourier-transform infrared spectroscopy (FTIR) provides an alternative and much cleaner, faster, and safer way to estimate the TOC contents (relative weight percentage, %) in soils and sediments. However, converting spectral information into an accurate estimation of the TOC requires sophisticated chemometric efforts. This study aims to establish linkage between chemical and FTIR data by using modern and Holocene soil and sediment samples retrieved from southwest Taiwan where a variety of sediments deposited from various environments could provide a valuable spectrum of sediment samples. We collected FTIR spectral data and measured TOC from duplicates of the same samples. Data filtering and dimension reduction have been done by using Kernel Density Estimation (KDE) and Principal Component Analysis (PCA). Multiple machine learning models have been utilized, including Support Vector Regression (SVR), eXtreme Gradient Boosting (XGBoost), and Adaptive Boosting (AdaBoost) to predict TOC and their performances were evaluated. The XGBoost model is the best module with a mean standard error (MSE) of 0.015 and an R2 of 0.706. This study demonstrates that FTIR spectra is suitable for fast and accurate estimation of TOC content stored in soil and sediments when proper machine learning methods and procedures are conducted.
Publisher
Institution of Engineering and Technology (IET)
Type
journal article