Maps of regional spatial distribution of soil organic carbon (SOC) in detail are needed to guide sustainable soil uses and management decisions. Spatial interpolation methods include deterministic methods and geo-statistical methods. Deterministic method is based on mathematical functions, whichsearch for the spatial similarities within a certain geographical area, such as inverse distance weighting method(IDW), or create prediction surface by the known samples based on smoothness such as radial-based function method(RBF); Geo-statistical method is one of the important means for soil mapping supported by geographical and ecological parameters. It can achieve unbiased optimal estimation of regionalized variables by using the statistical property of the known samples, and on the basis of theoretical analysis of semi-variogram, such as ordinary kriging method(OK). Many of these methods perform well in areas with gentle terrains. However it is uncertain how these methods perform to capture SOC variations in complex terrains, especially in those areas where land uses are highly influenced by human activities, such as in the Loess Plateau of China. Therefore, four methods were applied to predict SOC content spatial explicitly at typical small watershed in the Loess Plateau of China including Ordinary Kriging method(OK), Inverse Distance Weighted (IDW) method, Radial-Based Function method(RBF), and Ordinary kriging modified with land-use type method(OK_LU) in the present research to evaluate their usability.
The purpose of this study is to find appropriate methods which are suitable to the complex terrain in Loess Plateau region of China. The study area was a typical watershed in Loess Plateau with complex hilly gully terrain and various land-use types. A field sampling dataset of 188 points was divided into two parts randomly:75% for model building and 25% for accuracy validation. The accuracy validation points were used to compare the predicted value with measured value to check the similarity at each point. Prediction results were validated using Pearson's Correlation Coefficient(R), Mean Absolute Error (MAE), Root Mean Square Error(RMSE) and Accuracy (AC).The model with higher R and lower MAE and RMSE, while the AC value is closer to 1 was regarded as more relevant, less biased and more accurate.
Results showed that: (1) The values of MAE and RMSE produced by OK method were the lowest, while the values of R and AC were the largest, which indicated OK's highest effectiveness in SOC interpolation to yield the most accurate results among the three ordinary methods. IDW method ranked behind OK and followed by RBF method in interpolation effectiveness. (2) OK_LU was the best method for predicting SOC in this study. The prediction by OK_LU had more details of depicting the local variation under land-use types. MAE by OK_LU decreased from 0.900% to 0.567%, RMSE decreased from 1.101% to 0.777%, R increased from 0.4026 to 0.5589, and AC increased from 0.9081 to 0.9505 comparing with OK. In conclusion, OK_LU gave more precise evaluation which was critical for obtaining accurate spatial distribution of SOC in Loess Hilly regions. This will be of interests for related research in similar environments.