Rapid analysis using Diffuse Reflectance Spectroscopy can enhance Soil Carbon Monitoring

To gain a better understanding of changes in soil carbon stocks and composition it is essential to have data on relevant soil properties with high temporal and spatial resolution. However, traditional laboratory analysis can be both time-consuming and expensive, ultimately limiting data availability. Mid-DRIFTS (Diffuse Reflectance Infrared Fourier Transformed Spectroscopy in the mid-IR range) is a cost-effective alternative to conventional analytical methods. It enables the simultaneous inference of multiple soil properties, such as soil organic carbon, nitrogen and phosphorus content, soil texture, and cation exchange capacity, from measured spectra. This makes it a valuable analytical tool for soil monitoring.

To successfully integrate mid-DRIFTS into a soil monitoring concept, a spectral library representative of the target is required as the basis for multivariate or machine-learning-based calibration models. Here, we want to present the initial results obtained using the BDF-SSL, a soil spectral library we created as the foundation for integrating mid-DRIFTS in agricultural soil monitoring in Saxony, Germany. The library's core consists of nearly 300 spectra obtained from retention samples from agricultural soil monitoring sites in Saxony, which were collected over the past 20 years.
We focused on the inference of soil carbon content, including three thermally derived carbon fractions (TOC₄₀₀, ROC, TIC₉₀₀) we measured according to DIN19539 by combustion. Additionally, we calibrated models for a wide range of other soil properties, such as soil texture, nitrogen and phosphorus content, elemental concentrations and cation exchange capacity. We used both Partial Least Squares Regression (PLSR), Cubist, and the Memory Based Learner (MBL) to calibrate the models.

Both Cubist and MBL consistently outperformed PLSR. Our models show high predictive accuracy for the carbon fractions, total nitrogen and phosphorous contents, cation exchange capacity and texture. In addition, we are also able to predict several elemental concentrations, such as Fe, Al, or Ni contents with high accuracy. Our results show that mid-DRIFT can be used to enhance spatial and temporal coverage of soil monitoring, allowing not only for more accurate estimations of soil carbon stocks and sequestration rates, but also for the rapid estimation of several other soil properties.