COS 200-4 - What can we learn from predictive modeling of soil quality by using mid-infrared spectroscopy?

Abstract: Increasing our understanding of soil quality improvement through C sequestration may influence agroecosystem resiliency in a changing climate. In this context, it has been demonstrated that the application of mid-infrared spectroscopy (MIR) is a viable alternative to conventional laboratory methods and could help us deal with the problem of analyzing large quantities of samples. Moreover, the most significant advantage of MIR is its ability to take rapid and non-destructive measurements of different soil properties. In this way, MIR has an enormous potential to identify large-scale variations and drivers of microbial communities, improving soil monitoring at a lower cost. Our findings in long-term plots at Pendleton, Oregon, show that strong drivers of soil microbial communities such as soil organic carbon (SOC), mineral-associated organic carbon (MAOM), total N (TN), and pH can be successfully predicted by using MIR (R-square >0.8). However, soil microbial properties such as C mineralization are less accurate than soil physical and chemical properties. Recently, machine learning algorithms have received increasing attention for improving prediction accuracy using spectroscopy data. However, there are no standard criteria for choosing the ideal approach since prediction accuracy depends significantly on each dataset's sample distribution, sample size, training algorithm, sampling depth, and soil spectral diversity. This study offers insight into MIR methodology as an efficient, scalable, and cost-effective tool that provides reliable estimation for some parameters. Still, we also examine the most critical challenges that affect the model performance.