| Summary: | Quantitative assessment of soil functions requires the characterization of soil capability and condition. Mid-infrared (MIR) spectroscopy has been suggested as a viable alternative to the wet chemistry method. However, the extensive set of soil properties that can be well predicted have yet to be explored. The USDA MIR spectral library contains approximately 45,000 samples with more than 119 soil properties. This unique dataset allows us to establish which soil properties that can be accurately measured. Memory-based learning (MBL) algorithm achieved higher accuracy than the Cubist. The prediction accuracies of different properties were then categorized further into four classes (A, B, C, and D) based on four accuracy metrics, namely coefficient of determination (R2), Lin's concordance correlation coefficient (LCCC), ratio of performance to the interquartile range (RPIQ), and standardized bias (Stb). We found that a single MIR spectrum can infer 50 soil properties with high accuracy (A or B category), and 44 properties can be estimated approximately (C category). These properties can then be used to evaluate a range of soil functions, including food production, carbon storage, water storage, nutrient cycling, and habitat function.
|
|---|