Organic substrate, clay type, texture, and water influence on NIR carbon measurements

Near-infrared reflectance (NIR) spectroscopic detection of soil organic carbon (SOC) is an alternative to more resource-intensive methods. However, interrelated impacts from specific soil characteristics on NIR performance are not well understood. To address this, a population of artificial soils was created based on several primary soil characteristics, and a single optimized NIR model's predictive capability was compared by each soil characteristic subset. Objectives were to determine how NIR prediction of SOC is affected by: i) SOC substrate and primary soil constituents [quartz sand, leonardite, and manufactured humic acid (MHA)]; ii) clay mineral [i.e. two standards (kaolinite and smectite)] and percent composition [three textures per clay type (0, 50, and 98.6%)]; and, iii) hydroscopic soil water (hydrated and non-hydrated) content, with three replications. The RPD statistic (ratio of performance to deviation) further compared prediction accuracies within soil subsets, and was 2.04 for the model across the entire set. NIR predicted SOC in the MHA better (R²=0.86) than naturally derived leonardite (R²=0.74). The greatest factor was texture, with prediction improvement for lower clay content (from R²=0.79 to 0.92 for 98.6% to 0% clay, respectively). Clay type and hydration affected accuracy of SOC prediction for samples created with leonardite. Amongst leonardite samples, SOC prediction improved in kaolinitic-hydrated (hydroscopic water) compared to smectitic soils. Conversely, in oven-dry soils, SOC was almost equally accurate for smectitic (R²=0.77) and kaolinitic (R²=0.78). Consequently, NIR calibration sets developed per soil population textural characteristics should improve SOC prediction.