Soil charcoal prediction using attenuated total reflectance mid-infrared spectroscopy

Abstract

Despite strong evidence for the importance of charcoal as a long-term carbon sink in soils, simple methods to quantify charcoal in soil are still lacking. In this study, we tested the application of attenuated total reflectance mid-infrared spectroscopy (ATR-MIR) for quantification of charcoal in soil. To do this, we created calibration samples from defined quantities of pulverised rock, charcoal and litter sampled from a forest floor in Guy Fawkes National Park, New South Wales, Australia, and analysed them via ATR-MIR and dry combustion. The organic carbon concentration (mass proportion) of the samples ranged from 0.1 to 15% and the charcoal mass proportion from 0.02 - 11% (10 - 50% of soil organic matter). We then trained randomForest models to the spectral data and assessed the predictive performance of the models for both the quantity of charcoal and litter in the samples. The models were excellent at predicting both charcoal and litter contents of the samples, explaining 94% of variance in the mass proportion of charcoal and 93% of the variance in the litter content of the validation dataset (i.e. out-of-bag estimates of the models). Extracting the variable importance from the models showed that the spectral regions important to charcoal prediction differed from those important to litter prediction, highlighting the capacity of the models to distinguish between charcoal and litter components based upon ATR-MIR spectra. Our method enables a simple, cheap and efficient prediction of litter and charcoal without the need for complex chemical extraction or analyses.