Rapid, cost-effective and reliable methods are in need to assess the soil carbon (C) pools and carbon sequestration potential at landscape scales. Visible/near-infrared diffuse reflectance spectroscopy is a rapid and cost-effective method that provides inferences on multiple soil properties. The objective of our study was to relate six different soil carbon attributes (total, recalcitrant, hydrolysable, mineralizable, and two hot water extractable size fractions) to soil spectral signatures derived using VNIRS. Our approach targets multiple aspects of soil carbon and provides a comprehensive assessment of biogeochemically active carbon pools of a wide variety of soils.

Soil samples were collected from 0-30 cm at 141 sites located in the Santa Fe River Watershed in north-central Florida representing typical soils (Ultisols, 37%; Spodosols, 26%; Entisols, 15%; Histosols, 2%; Inceptisols, 1%; and Alfisols, 1%) and land uses (upland forest and pine plantation, 43%; agriculture and rangeland, 31%; wetland, 14%; and urban, 7%). Soil samples were analyzed for total C, hydrolysable C, mineralizable C, and dissolved organic C. Recalcitrant C was calculated as the difference between total C and hydrolysable C. Soil reflectance was measured in the visible/near-infrared range and chemometric modeling was applied to relate the spectral data with laboratory measurements using five different parametric and non-parametric regression methods: stepwise multiple linear regression, principal components regression, partial least squares regression, and regression and committee trees. Accurate models of all soil carbon properties, except hydrolysable C, were developed using VNIRS, with overall better models obtained using partial least squares regression. These results indicate the suitability of VNIRS associated with chemometric modeling to more rapidly assess total, labile and stable carbon pools at landscape scales.