Fine-resolution mapping of soil organic carbon based on multivariate secondary data

Our objective was to quantify the improvement in fine-resolution maps of soil organic carbon stock (CS, Mg C ha(-1)) resulting from utilizing multivariate sources of secondary information. Different geostatistical techniques for mapping CS in the top 0.3 m of soil with or without secondary information were assessed in three large no-till fields (49 to 65 ha) in Nebraska, which were sampled at a density of 3.9 to 4.2 samples ha(-1). Geostatistical methods evaluated were ordinary kriging (OK), co-kriging (COK), kriging with external drift (KED), and regression kriging (RK). Ancillary variables assumed to provide indirect information on spatial patterns of CS included maps of soil series, surface reflectance derived from satellite images (REF), relative elevation (EL), and soil electrical conductivity (EC). Root mean square error (RMSE) of CS predicted by OK ranged from 10.8 to 12.5 Mg C ha(-1) for the three sites. Methods that utilized secondary information reduced the RMSE by 5% to 38% compared to OK. Relative improvements in map accuracy were highest (16% to 38%) in multivariate regression kriging approaches, which also performed better than COK, KED, or RK methods that utilized only one ancillary variable. The relative gain from incorporating secondary information increased with decreasing sampling density. Reducing sampling intensity of CS to one half of the original samples increased the bias and RMSE of maps of CS produced by OK, whereas no or only little loss of map accuracy occurred in RK. Among the hybrid methods tested, RK performed best in terms of consistently increasing map accuracy and flexible modeling of the multivariate relationships between CS and secondary information. Among the ancillary variables, EC was most useful for CS mapping. Terrain attributes had limited value at the three study sites, while the value of REF depended on the surface conditions at the time of image acquisition. To reduce uncertainties, we recommend using independently measured, multivariate secondary information in RK approaches for mapping of soil organic carbon. (c) 2005 Elsevier B.V. All rights reserved.