|
Abstract:
|
Soil organic carbon is vital to soil functions and ecosystem services. The increasing demand for soil organic carbon concentration data to support precision agriculture, spatial modeling and climate change mitigation efforts is being met by a growing interest in methods for its precise estimation. One such method is diffuse reflectance, mid-infrared soil spectroscopy. In soil spectroscopy, a model relates the spectral data to the soil property values for prediction. This study explores the effectiveness of subsetting a large, continental soil spectral library to construct smaller calibration models, as a resource-efficient method to improve the prediction accuracy of soil organic carbon concentration. The Functional Data Explorer platform of JMP Pro was used to derive functional principal components, which identified the most important dimensions in the spectral data. The functional principal components were subsequently used as covariates in a Generalized Regression to predict soil organic carbon concentration. These analysis techniques provide an effective alternative to the traditional Partial Least Squares regression for the accurate prediction of soil organic carbon.
|
