Wei Zhang1, Kelin Wang, Hongsong Chen, Xunyang He, Jiguang Zhang. 1. Key Laboratory of Agro-ecological Processes in Subtropical Region, Institute of Subtropical Agriculture, Chinese Academy of Sciences, Changsha, Hunan, China.
Abstract
BACKGROUND: Soil carbon management at landscape scale requires reliable information on the spatial distribution of soil organic carbon (SOC). However, how to improve the accuracy of spatial prediction is not well addressed in the karst region of southwestern China. This study evaluates the performance of univariate kriging (ordinary kriging (OK)) and hybrid kriging (co-kriging (CK), regression kriging (RK) and residual maximum likelihood (REML)) in mapping the spatial distribution of SOC at a depth of 0-15 cm. Terrain attributes and the normalised difference vegetation index (NDVI) were used as ancillary variables. RESULTS: The distribution of SOC was significantly related to NDVI and terrain attributes. Furthermore, geostatistical analyses reflected a moderately structured spatial correlation of SOC. Regression analyses identified the NDVI and slope as the best predictors for describing the spatial pattern of SOC. Combined with NDVI and slope gradient, REML and RK performed better in increasing map prediction accuracy and decreasing the soothing effect of kriging. CONCLUSION: The spatial pattern of SOC was controlled by topography and cultivation activity. The predictive abilities of OK and CK were limited. Combined with the auxiliary variables, REML and RK can improve the prediction accuracy. This study is beneficial for the further research of precise SOC management in the typical karst landscape.
BACKGROUND: Soil carbon management at landscape scale requires reliable information on the spatial distribution of soil organic carbon (SOC). However, how to improve the accuracy of spatial prediction is not well addressed in the karst region of southwestern China. This study evaluates the performance of univariate kriging (ordinary kriging (OK)) and hybrid kriging (co-kriging (CK), regression kriging (RK) and residual maximum likelihood (REML)) in mapping the spatial distribution of SOC at a depth of 0-15 cm. Terrain attributes and the normalised difference vegetation index (NDVI) were used as ancillary variables. RESULTS: The distribution of SOC was significantly related to NDVI and terrain attributes. Furthermore, geostatistical analyses reflected a moderately structured spatial correlation of SOC. Regression analyses identified the NDVI and slope as the best predictors for describing the spatial pattern of SOC. Combined with NDVI and slope gradient, REML and RK performed better in increasing map prediction accuracy and decreasing the soothing effect of kriging. CONCLUSION: The spatial pattern of SOC was controlled by topography and cultivation activity. The predictive abilities of OK and CK were limited. Combined with the auxiliary variables, REML and RK can improve the prediction accuracy. This study is beneficial for the further research of precise SOC management in the typical karst landscape.