Sampling designs for soil organic carbon stock assessment of soil profiles

•Two or three vertical transects are required to capture the short range horizontal variations of SOC.•Depth-wise stratified random sampling reduces the sample size and is suitable when horizontal variation is high.•Fuzzy c-means sampling is useful to determine the minimum sample size for profile SOC stock assessment.•Horizon-based sampling is effective to assess profile SOC stock provided horizon depths are accurately measured.

We investigated four sampling designs for soil organic carbon (SOC) stock assessment of soil profiles: (i) sampling by horizons, (ii) vertical transect sampling, (iii) depth-based stratified random sampling, (iv) fuzzy c-means sampling in which we explored the use of vis-NIR spectroscopy, image analysis and color models. An Alfisol and Mollisol profile wall (1 × 1 m) was divided into a 10 × 10 cm raster and 100 samples (about 200 g each) were collected at the centers of grid cells for SOC analysis. Bulk density samples were collected from each 10-cm depth interval along a single vertical transect and the SOC stock was calculated using 100 points in the profile wall. Horizon-based sampling for the Mollisol (5 horizons) ranged from 231 to 262 Mg C ha− 1, whereas it ranged from 69 to 99 Mg C ha− 1 in the Alfisol (3 horizons). The SOC stocks obtained by 1 to 7 vertical transects ranged from 68 to 81 Mg C ha− 1 in the Alfisol, and 239 to 246 Mg C ha− 1 in the Mollisol. Depth-based stratified random sampling resulted in the SOC stocks ranging from 77 to 82 Mg C ha− 1 in the Alfisol and 234 to 257 Mg C ha− 1 in the Mollisol, and the standard errors decreased with increasing sample size from 10 to 70. Fuzzy c-means clustering created clusters similar to the field delineated horizons. A sample size of 7 in both profiles was sufficient to estimate the mean profile SOC stock by fuzzy c-means sampling. The CIE L*a*b* color model resulted in more accurate estimation in the Alfisol, but the vis-NIR spectra resulted in more accurate estimation in the Mollisol. Soil depth improved the performance of vis-NIR spectra. It is concluded that in these soils, at least two or three vertical transects are required to capture the horizontal variation for estimating profile SOC stock. Depth-wise stratified random sampling reduces the number of samples and is suitable when horizontal variation is high. Fuzzy c-means sampling is useful to determine the minimum sample size for profile SOC stock assessment but requires ancillary data and processing before sampling the soil profile.