Spatial Prediction Methods For Soil Organic Carbon In Peak-cluster Depression Areas

Soil organic carbon is one of the key factors to soil fertility and crop yield.The unique karstification makes soil organic carbon play an important role in the carbon cycle of surface ecosystem in karst area.In peak-cluster depression areas,complex topography,discontinuous soil spatial distribution and extremely tough field survey bring great challenges to the estimation of soil organic carbon density and storage.Therefore,it is vital to explore new methods to improve the spatial prediction accuracy of soil organic carbon density.This paper selected the typical peak-cluster depression area in Guohua base,Guangxi as the study area,obtained 135 topsoil(0-10 cm)organic carbon data through soil survey,and utilized 10 m resolution DEM to extract 11 topographic factors such as topographic relief(AL),elevation(h)and distance from the sample point to the ridge line(DTR).On the basis of analyzing the relationship between soil organic carbon and topographic factors,five methods were used to predict soil organic carbon density including kriging,multiple linear regression,geographically weighted regression,multiple linear regression kriging and geographically weighted kriging.By contrasting the prediction accuracy,the best prediction model was ascertained.Finally,the study used the exposure rate of bedrock to modify the soil organic carbon density predicted by the best model,and then estimated the topsoil organic carbon storage.The main results are as follows:(1)Soil organic carbon showed significant positive correlations with AL(0.404),slope(0.379),and LS(0.233),and a significant negative correlation with DTR(-0.389).Multivariate stepwise regression analysis indicated that AL,h and DTR could reflect the spatial heterogeneity of soil organic carbon well and be used as the auxiliary variable to predict soil organic carbon density.(2)These five methods were in the descending order of the size of prediction accuracy geographically weighted regression kriging model,multiple linear regression kriging model,geographically weighted regression model,kriging model and multiple linear regression model.Hence,the geographically weighted regression kriging model was the best method for predicting the spatial distribution of soil organic carbon density in the peak-cluster depression area.The soil organic carbon density was modified by the exposure rate of bedrock and then the surface soil organic carbon storage in the study area was estimated to be approximately 3.19×10~7kg.(3)The surface soil organic carbon density in the study area has a strong spatial heterogeneity.The high value areas scatter like flowers across the western part of the study area,while the low value areas are blocky or patchy in the south and northeast of the study area.The soil organic carbon density and storage both decreased with the increase of rocky desertification levels,and were largest in the non-rocky desertification regions.