Soil Carbon Maps

The quality of soil property maps may be improved and spatial sampling intensities reduced by incorporating secondary data to enhance spatial estimates. The purpose of this study was to evaluate how scale of sampling and secondary spatial information (terrain attributes) affected the quality of spatial estimates of soil C. A field in Central Michigan was sampled using 30.5‐ and 100‐m regular grids and the samples were analyzed for total C. Extracting a 61‐m grid from the 30.5‐m regular grid (G 30 ) data set created an additional data set. Total C maps were created at each scale using ordinary kriging, kriging with a trend model, cokriging, kriging with an external drift, and multiple regression. Each resulting map was compared with an independent validation data set ( n = 24) to evaluate map quality. At the 30‐m grid scale, there were modest differences between maps created with ordinary kriging (root mean squared error RMSE = 2.9 g kg −1 , isotropic model; RMSE = 3.0 g kg −1 , anisotropic model), kriging with a trend model (RMSE= 2.8 g kg −1 ), cokriging (RMSE = 2.9 g kg −1 ), kriging with an external drift (RMSE = 2.6 g kg −1 ), and multivariate stepwise regression (RMSE = 3.2 g kg −1 ). Prediction errors were generally larger at the 61‐m grid scale and procedures that utilized secondary the terrain data (cokriging, RMSE = 3.8 g kg −1 ; kriging with an external drift, RMSE = 3.8 g kg −1 ; stepwise multiple regression, RMSE = 3.0 g kg −1 ) outperformed procedures that did not (isotropic ordinary kriging, RMSE = 4.3 g kg −1 ; kriging with a trend model, RMSE = 4.3 g kg −1 ). At the 100‐m grid (G 100 ) scale, geostatistical procedures were not appropriate because of the small sample size ( n = 12) yet multiple regression performed well (RMSE = 3.8 g kg −1 ). Maps of soil C created with regression most resembled the soil color patterns evident in an aerial photograph of the field.