Prediction of Soil Organic Carbon under Varying Moisture Levels using Reflectance Spectroscopy

摘要

Assessment of soil organic C (OC) spatial variability with proximal and remote sensing is complicated by interactions with soil constituents and moisture content. The objectives of this study were to (i) assess how well C could be predicted across a wide range of soils, (ii) determine how varying soil moisture impacted OC predictions, and (iii) determine the spectral wavelengths useful for assessing OC. Soil samples from the North American Proficiency Testing (NAPT) Program soil library were utilized in this study. Spectral reflectance (800-2200 nm) was measured with a spectrometer for air-dried soil and with 15, 20, and 25% soil moisture. Several pretreatment spectral reflectances were analyzed with partial least square (PLS) regression. The best pretreatment was the first derivative, explaining 88% of OC variability with air-dried samples and 70% at 15% soil moisture. Predictions for the samples of 20% (r(2) = 0.64) and 25% (r(2) = 0.63) soil moisture were as good as the combined datasets. These datasets included the 15 and 20% (r(2) = 0.56), 15 and 25% (r(2) = 0.64), 20 and 25% (r(2) = 0.59), and 15, 20, and 25% (r(2) = 0.64) soil moistures. The variable importance for prediction identified wavelengths associated with organic components including aromatics, aliphatics, and amides. Clustering the latent vectors suggested that PLS was able to distinguish samples with different clay and Fe content despite that they were not included as predictors. This study suggests that spectral OC prediction with varying soil moisture content (i. e., between 15 and 25% moisture) is of acceptable quality (i. e., r(2) >= 0.56) even across a range of soils from the United States. These findings have important implications for estimating OC with proximal and remote sensing techniques