Reflectance Spectroscopy Detects Management and Landscape Differences in Soil Carbon and Nitrogen

Summary statistics for laboratory dry combustion SOC and TN analysis of the plot data set and independent calibration data set are given in Table 2. Th e two data sets were relatively simi-lar in composition, although the means and standard deviations were somewhat higher in the plot data set. Th is was probably because approximately 50% of the plot data were from the grass systems, with generally higher SOC and TN, while only about 25% of the calibration data were from grass systems.Th e performance of the cross-validated calibration models for refl ectance estimation of SOC and TN using PLS regression was good for the oven-dry soil. Th e model performance with fi eld-moist samples, however, was worse in terms of both R 2 and RPD (Table 3). Refl ectance characteristics in VNIR wavelengths are primarily determined by chemical and physical characteris-tics of the sample surface. Th e physical structure of the oven-dry samples was much more consistent than that of the wet samples, where samples with higher clay content were somewhat smeared due to the manual manipulation necessary to obtain a fl at sur-face. Morgan et al. (2009) stated that one possible reason for reduced accuracy with fi eld-moist, intact, smeared samples was that cores with a high clay content smeared more than others, leading to greater variability. It seems likely that a similar eff ect occurred here with the fi eld-moist sieved samples. Additionally, Lobell and Asner (2002) reported that the water content in wet samples may decrease accuracy because increasing water content can reduce the strength of an important absorption feature of C and N. Malley et al. (2002) also reported less accurate prediction of soil properties including organic matter and NH