Effects of prediction methods for detecting the temporal evolution of soil organic carbon in the Hilly Red Soil Region, China

Though numerous studies have evaluated the effect of prediction method on soil organic carbon (SOC) spatial distribution, the influence of prediction method on detecting the temporal evolution of SOC has rarely been reported. This study was conducted in Yujiang County, Jiangxi Province, China using 174 and 257 samples collected in 1982 and 2007, respectively. The SOC spatial distributions for those 2 years and the associated temporal evolution were predicted via ordinary kriging (OK) and kriging combined with soil types and land-use (KSTLU) patterns. Results showed that KSTLU had better spatial prediction accuracy than OK for both years. The mean absolute errors (MAEs) of OK for 1982 and 2007 were 2.45 and 5.99 g kg⁻¹, and the root mean square errors (RMSEs) were 3.37 and 7.23 g kg⁻¹, respectively. Meanwhile, the MAEs of KSTLU for 1982 and 2007 were 1.99 and 3.23 g kg⁻¹ which are 19 and 46% lower than those of OK, respectively. The RMSEs were 2.76 and 3.97 g kg⁻¹ which were 18 and 45% lower than those of OK, respectively. Moreover, the result of temporal evolution from the KSTLU prediction showed that the area of increasing SOC (70%) is remarkably larger than the area of decreasing SOC (30%). The KSTLU prediction was more consistent with the statistical results than the OK prediction. Furthermore, the contours of SOC temporal evolution from KSTLU had a more detailed description for local change of SOC than OK. Thus, the prediction methods greatly affect the detection of SOC temporal evolution. Results of this study indicated that KSTLU is more efficient and rational than OK in studying the SOC temporal evolution in the Hilly Red Soil Region of South China.