Estimation and mapping of soil carbon concentration with in situ vis-NIR at the field scale

Yuanhong Song

Soil organic carbon (SOC) serves a series of soil physical, biochemical and ecological functions. Knowledge of SOC spatial distribution at the field scale is important for precision farm management and scientific soil monitoring. Such knowledge can be obtained by digital soil mapping with the support of high spatial resolution soil data. Visible near-infrared (visNIR, 350nm-2500nm) diffuse reflectance spectroscopy provides an efficient tool for rapid, cost-effective, and non-destructive soil surveying. The objectives of the present study were to 1) test the practicality of estimating soil carbon concentration from a newly designed visNIR penetrometer in situ, 2) develop 2D mapping models to investigate soil carbon spatial distribution at the field scales, and 3) improve the mapping accuracy with the in situ vis-NIR. 404 in situ soil vis-NIR spectral profiles and 407 soil reference core profiles of the top 30 cm at four study sites across the Palouse region were taken during 2014 to 2016. Lab-based vis-NIR spectral scan were obtained with air-dried, sieved sample. Partial least square regression models were calibrated for the in situ and lab-based spectra. The lab-based regional calibration yielded a standard error of prediction (SEP) of 0.17 g kg-1 and a residual product deviation (RPD) of 3.10; the in situ regional calibration resulted a SEP of 0.29 g kg-1 and a RPD of 1.64. Regression analysis and random forest techniques showed that the combination of terrain attributes, remote sensing (RapidEye) imagery, and apparent electrical conductivity can explain more than 40% of the variability in carbon concentration. Leave-group-out cross-validation for the regression kriging mapping models suggested no mapping accuracy improvement was achieved by including in situ vis-NIR predicted carbon concentration. Finally, a three-dimensional (3D) interpolation of the carbon concentration was created by regression kriging with 3D variogram. Results from this study suggest that the in situ vis-NIR is useful for estimating soil carbon concentration. More sophisticated modeling techniques and strategies should be considered to improve the mapping accuracy by using in situ vis-NIR.

Estimation and mapping of soil carbon concentration with in situ vis-NIR at the field scale

Files and links (1)

Abstract

Metrics

Details