Measuring in situ soil carbon stocks: A study using a novel handheld VisNIR probe

To be commercially viable, soil carbon project developers need to be able to measure soil carbon stocks across large scales (e.g., 100,000 to 1,000,000 ha). These measurements need to be accurate, unbiased, inexpensive, and fast. One potential measurement modality for carbon markets is visible and near-infrared diffuse reflectance spectroscopy (VisNIR). VisNIR has been widely used to predict soil properties including soil organic carbon (SOC) concentration and stock under both lab settings and in situ soil conditions. Recent developments in low-cost spectrometers have enabled the creation of easy to operate, rapidly deployed, handheld VisNIR-equipped devices for in situ soil measurement. Our objective for this study is to 1) test one such handheld in situ VisNIR probe (handheld probe) to measure SOC stocks to 30 cm depth in Midwest US Mollisols, 2) to quantify the role of bulk density and SOC concentration in VisNIR probe calibration for probe-based estimation on SOC stock in Midwest US Mollisols, and 3) to quantify the effect of indirect (SOC + BD) vs direct calibration modeling (SOC stock directly) of SOC stocks using VisNIR data. We collected handheld probe measurements and soil core samples from six non-contiguous farms across the state of Illinois, USA. A one-farm hold out PLSR modeling approach was taken for SOC concentration, bulk density, 5-cm incremented SOC stocks down to 45 cm; and 0 to 30 cm SOC stocks using the in situ VisNIR spectra from the handheld probe. Models accurately predicted SOC concentration (R2 = 0.72, RMSE = 0.33 %, RPIQ = 2.39, bias = 0.0005 %), 5-cm increment SOC stocks (R2 = 0.68, RPIQ = 2.41 Mg/ha, bias = 0.05 Mg/ha) and 0 to 30 cm SOC stocks (R2 = 0.88, RMSEP = 7.8, bias = -0.49 Mg/ha, RPIQ = 4.19 Mg/ha). Models were not able to accurately predict bulk density (R2 = 0.28). Direct SOC stock modeling resulted in lower bias compared to indirect computation of SOC stock (bias = 0.05 and 0.15 Mg/ha for direct and indirect methods, respectively) and results demonstrated that, in this loess landscape, SOC stock prediction accuracy was driven by accurate prediction of SOC concentration, rather than accurate prediction of bulk density. The handheld probe shows promise as a rapid, low-cost tool for measuring SOC stocks in the midwestern Mollisols and can provide the data necessary to support large spatial scale soil carbon market development. These results justify continued investment in in situ spectral libraries for the handheld probes and eventually posit a modeling framework for measurement-based soil carbon accounting.