Predicting the proportion of centennially stable soil organic carbon using mid-infrared spectroscopy

Abstract

Recent work has shown that it is possible to quantify the proportion of centennially stable soil organic carbon (SOC) by using Rock-Eval® thermal analysis results as input variables for PARTYsoc, a learning model calibrated on long term experiments data. This method of quantifying SOC biogeochemical stability holds promise for improving projections of SOC stock evolutions. Here, we assessed the potential of mid-infrared spectroscopy (MIR) as a lower-cost, higher-throughput technique to facilitate its wide-scale deployment.

We compiled a spectral library of over 1,800 records obtained through the scanning of samples from the French Réseau de Mesure de la Qualité des Sols to calibrate a model using MIR data to predict the proportion of centennially stable SOC. The model gave accurate predictions (RMSE = 0.06, RPD = 2.21, RPIQ = 3.26), suggesting that MIR spectra contain relevant information on SOC biogeochemical stability. We then tried to transfer this model directly to datasets acquired using another MIR spectrometer on German and Finnish soil samples. The accuracy of the predictions was degraded, even when using the CORAL data alignment method to harmonise the different spectral datasets.

Our results show that it is possible to predict the proportion of centennially stable carbon determined by the PARTYsoc model using MIR spectroscopy. However, we found that the transfer of such models to different soils, scanned with different instruments or different protocols, is difficult. Large-scale deployment of such models will require careful calibration transfer, probably associated to local calibration within a similar spectral space.