Biochar Stability Revealed by FTIR and Machine Learning.

Abstract

Biochar is a carbon-rich and environmentally recalcitrant material, with strong potential for climate change mitigation. There is a need for rapid and accessible estimations of biochar stability, the resistance to biotic and abiotic degradation in soil. This study builds on previous work by integrating Fourier-transform infrared spectroscopy (FTIR) data with predictive modeling to estimate standard stability indicators: H:C and O:C molar ratios. Lignocellulosic feedstocks were pyrolyzed at highest treatment temperatures (HTT) ranging from 150-700 °C, and all samples achieved H:C < 0.7 and O:C < 0.4 at HTT of 400 °C and above. Several statistical and machine learning models were developed using FTIR spectra. The random forest (RF) models, which incorporated full data preprocessing, yielded the highest accuracy (R ² = 0.96 for both ratios) when tested on an unseen feedstock. Variable importance analysis identified spectral regions linked to aromaticity and inversely correlated to C-O stretches in cellulose and lignin as key predictors. The findings of this study verify that FTIR data can serve as a rapid and accurate tool for estimating biochar stability.