The New Zealand Soil Carbon Monitoring System (Soil CMS) was designed, and has been used, to account for soil organic carbon change under land-use change, during New Zealand’s first Commitment Period (2008–2012) to the Kyoto Protocol. The efficacy of the Soil CMS model has been tested for assessing soil organic carbon stocks in a selected climate–land-use–soil grouping (cell). The cell selected for this test represents an area of 709 683 ha and contains soils with a high-activity clay mineralogy promoting long-term stabilisation of organic matter, and is under low-producing grassland in a dry temperate New Zealand climate. These soils have been sampled at randomly selected positions to assess total soil organic carbon stocks to 0.3 m, and to compare with the modelled value. Results show no significant difference between the field estimation (67 ± 30 Mg C/ha), the mean value of the model calibration dataset (79 ± 28 Mg C/ha), and the value predicted by the model (101 ± 41 Mg C/ha), although all estimates have large uncertainties associated with them. The model predicts national soil organic carbon stocks as a function of soil texture, clay mineralogy, land use, climate class, and a slope–rainfall erosivity product. Components of uncertainty within the model include the size and distribution of the calibration dataset, and lack of representativeness of the calibration soil samples, which were sampled for other reasons, e.g. soil survey and forest mensuration.
Our study has shown that major components of uncertainty in our field estimation of soil organic carbon stocks (investigated using the indices reproducibility, RP; and coefficient of variation, CV) are short-range (within-plot) and regional (between-sites) spatial variability. Soil organic carbon stocks vary within our selected climate–land-use–soil cell due to varying stoniness (stony soils RP 44%, CV 21%; non-stony soils RP 27%, CV 13%), soil depth, slope position, and climatic effects. When one outlier soil was removed from the model calibration dataset, and the three very stony sites were removed from the randomly selected field validation set, the model calibration dataset and the field dataset means agreed well (78 ± 26 and 78 ± 21 Mg C/ha, respectively). The higher modelled value, before removal of the outlier, is likely to reflect a bias in the model dataset towards conventionally selected modal profiles containing less stony soils than those encountered by the random sampling strategy of our field campaign. Therefore, our results indicate (1) that the Soil CMS provides an adequate estimation of soil organic carbon for the selected cell, and (2) ongoing refinements are required to reduce the uncertainty of prediction.
Effects of adjacent land-use types on the distribution of soil organic carbon stocks in the montane area of central Taiwan