Contents lists available at ScienceDirect Soil & Tillage Research journal homepage: www.elsevier.com/locate/still Estimating soil organic carbon density in plains using landscape metric- based regression Kriging model Zihao Wu a , Bozhi Wang a , Junlong Huang a,f , Zihao An a , Ping Jiang a , Yiyun Chen a,b,c,d,e, ⁎ , Yanfang Liu a,b,c, ⁎ a School of Resource and Environmental Science, Wuhan University, Wuhan 430079, China b Key Laboratory of Geographic Information System of the Ministry of Education, Wuhan University, Wuhan 430079, China c Key Laboratory of Digital Mapping and Land Information Application Engineering, National Administration of Surveying, Mapping, and Geo-information, Wuhan 430079, China d Collaborative Innovation Center of Geospatial Technology, Wuhan University, Wuhan 430079, China e State Key Laboratory of Soil and Sustainable Agriculture, Chinese Academy of Sciences, Nanjing 210008, China f Department of Ecology and Evolutionary Biology, University of Toronto, Toronto M5S 3B2, Canada ARTICLE INFO Keywords: Soil organic carbon density Spatial estimation Landscape pattern Landscape metric-based regression Kriging Scale effects ABSTRACT The spatial distribution of soil organic carbon density (SOCD) is crucial for understanding land use impact on carbon budget. The spatial estimation and accurate mapping of SOCD in plains remain challenging, partly due to the relatively invariant topography and the lack of consideration of landscape patterns. Here, we propose a novel landscape metric-based regression Kriging (LMRK) for the spatial estimation of SOCD in plains. Using 242 topsoil samples collected in the Jianghan Plain, China, we (i) investigate the scale-dependent relationship between SOCD and 24 landscape metrics and (ii) develop LMRK models with multi-scale buffers (100–1000 m) for SOCD estimation and compare their performance with ordinary Kriging (OK) and regression Kriging (RK) that in- tegrates land use types. Results showed that LMRK outperformed other models. The relationships between SOCD and landscape metrics were found to be scale-dependent, and the buffer of 300 m exhibited the optimal scale in our case. The LMRK also revealed that a highly connected and water-sufficient landscape was conducive to the accumulation of soil organic carbon in farmlands. These results indicated that landscape metrics serve as good predictors, and the proposed LMRK method is effective for SOCD mapping in plains. Our findings highlight the scale-dependent relationship between landscape metrics and SOCD and provide a new perspective for soil mapping in plains. 1. Introduction Soil is the largest and most active terrestrial carbon pool that maintains more than thrice the carbon as that of the atmosphere (Adhikari et al., 2014; Zomer et al., 2017). Soil can act as a carbon source or sink depending largely on the land use patterns and their changes (Bhattacharya et al., 2016; Zomer et al., 2017). Soil organic carbon density (SOCD) is an important measurement of soil carbon stock, the accurate estimate of the spatial variation of SOCD is therefore crucial for understanding the carbon budget and guiding land use management, which helps achieve the target of the Paris Climate Agreement to ensure that global temperature increase should be well below 2 °C (Wollenberg et al., 2016). Current digital soil mapping (DSM) methods based on the soil–- landscape model concept (Jenny, 1941; McBratney et al., 2003) mostly use environmental variables that are easily obtained, such as topo- graphical factors, to estimate soil properties that are difficult to mea- sure (Wang et al., 2018; Zhao et al., 2014; Zhu et al., 2010). Numerous interpolation techniques that integrate environmental variables have been used for DSM; these techniques include multiple linear regression (MLR), co-Kriging (coK), and regression Kriging (RK) (Bilgili et al., 2011; Dong et al., 2019; Mirzaee et al., 2016). The RK technique is a hybrid interpolation method that utilizes auxiliary information via re- gression and interpolates the residuals from the regression using or- dinary Kriging (OK) (Odeh et al., 1995). In many cases, RK has been confirmed to be superior to inverse distance weighted (IDW), MLR, OK, and coK when environmental factors can explain a partial variation of the target variable (Hengl et al., 2007; Knotters et al., 1995; Mirzaee et al., 2016; Xiang et al., 2007). Although RK has the advantages of easy implementation, detailed results, and high prediction accuracy (Hengl https://doi.org/10.1016/j.still.2019.104381 Received 1 February 2019; Received in revised form 14 August 2019; Accepted 16 August 2019 ⁎ Corresponding authors at: School of Resource and Environmental Science, Wuhan University, Wuhan 430079, China. E-mail addresses: chenyy@whu.edu.cn, kellypcyy@126.com (Y. Chen), yfliu610@163.com (Y. Liu). Soil & Tillage Research 195 (2019) 104381 0167-1987/ © 2019 Published by Elsevier B.V. T