Variation in the precision of soil organic carbon maps due to different laboratory and spatial prediction methods

Digital mapping of soil organic carbon (SOC) is important for site-specific crop management and for environmental modeling and planning. Our objective was to evaluate how the precision of SOC maps is affected by the laboratory method used for measuring SOC and by the spatial prediction method used for mapping. In two irrigated maize fields in Nebraska, soil samples were collected, and SOC was either determined directly by automated CN analyzer (reference method, SOC c ) or estimated from weight loss-on-ignition (LOI) as a cheaper alternative. The latter involved conversion of LOI to soil organic matter (SOM) content using a standard laboratory calibration, followed by converting SOM to SOC values by assuming a constant C mass fraction in SOM (estimates denoted as SOC L ) or by regressing SOC on LOI (denoted as SOC R ). Interpolation methods evaluated were ordinary kriging (OK) and regression kriging (RK). Exhaustive ancillary variables used in RK included relative elevation, slope, soil electrical conductivity, and remotely sensed soil surface reflectance. Soil organic C was correlated with most of these ancillary variables, but the magnitudes of correlation varied among locations. Direct measurement of SOC C in combination with RK as spatial prediction method resulted in the most precise SOC maps. The relative improvement in map precision was 15% in Field 1 and 6% in Field 2 over OK of SOC C . Maps of SOC derived from LOI estimates were biased and less precise than maps that were based on direct measurement of SOC, but utilizing secondary information for spatial prediction alleviated some of the loss in precision. Using SOC L or SOC R estimates of SOC decreased map precision by 10% to 14% in OK or by 7% to 10% with RK as compared to the SOC C method. Regardless of the laboratory method chosen, secondary information should be used in SOC mapping to reduce sampling cost and/or increase map precision. However, the relative improvement of hybrid geostatistical techniques over OK largely depends on the strength of the correlation between SOC and ancillary variables.