基于环境变量的中国土壤有机碳空间分布特征

Characterization of Spatial Distribution of Soil Organic Carbon in China Based on Environmental Variables

研究中国土壤有机碳(Soil Organic Carbon,SOC)的空间分布特征对SOC储量估算以及农业生产管理具有重要意义。以全国第二次土壤普查2473个土壤典型剖面的表层(A层)SOC含量为研究对象,探寻地形、气候和植被等环境因素对SOC空间异质性分布的影响;以普通克里格法为对照,利用地理加权回归、地理加权回归克里格、多元线性回归和回归克里格模型建立SOC空间预测模型;并分别绘制了中国SOC的空间分布预测图。结果表明:(1)SOC含量与年均降水量、年均温、归一化植被指数、高程以及地形粗糙指数呈极显著相关关系;(2)平均绝对估计误差、均方根误差、平均相对误差和皮尔逊相关系数等模型验证指标表明地理加权回归的预测精度优于其他模型,可以更好地绘制SOC在大尺度上的空间分布特征;(3)较高SOC含量主要分布在研究区东北部、西南部以及东南部,而西北部SOC含量普遍偏低。本文以期从大尺度上探讨土壤属性与环境变量之间的相关关系,为全国土壤属性的空间制图提供一定的解决方案和思路。

【Objective】Soil organic carbon content in China and its variation plays a very important role in the global carbon cycle and variation of atmospheric carbon dioxide concentration.The processes of soil formation,development and erosion are subject to impacts of the complex and volatile environmental factors in the surroundings of the soil,so soil organic carbon is of strong spatial variability and dependence.【Method】Based on the data of soil organic carbon contents in the surface layers of 2473 soil profiles collected during the Second National Soil Survey,this paper explored influences of factors,like topography vegetation and climate on spatial distribution of soil organic carbon;With the ordinary Kriging method as control,geographically weighted regression,geographically weighted regression Kriging,multiple linear regression and regression Kriging were used separately to modeling for spatial prediction of soil organic carbon;indices,like mean absolute estimation error(MAE),mean relative error(MRE),root mean square error(RMSE)and Pearson correlation coefficient(r)were used to evaluate performance of these models;and soil organic carbon spatial distribution prediction maps were drawn separately.【Result】Results show:(1)Soil organic carbon varied in the range from 1.62 g·kg–1 to 223.88 g·kg–1 in content and averaged to be 22.28 g·kg–1 in the country.Its variation coefficient reached 96.10%,which indicates that organic carbon in the soil varies very sharply in range,and is of strong spatial heterogeneity;(2)Soil organic carbon content was significantly related to annual mean precipitation,≥10℃annual accumulated temperature,elevation,slope,aspect,normalized difference vegetation index,annual average temperature,topographic wetness index,topographic position index and topographic roughness index.Among them,slope,elevation,aspect,topographic roughness index,annual average precipitation and normalized vegetation index were positively related,while topographic position index,topographic wetness index,annual average temperature and≥10℃accumulated temperature were negatively related;(3)Multiple linear regression coefficients might reflect influences of the environmental variables on soil organic carbon globally,whereas the geographically weighted regression coefficient map might do clearly those of different environmental variables on soil organic carbon in different geographical locations;(4)The mean absolute estimation error,root mean square error,mean relative error and Pearson correlation coefficient of the model were used as model validation indices and indicated that the geographically weighted regression is higher than the other models in prediction accuracy,and hence can be used to plot soil organic carbon spatial distribution characteristics maps of large scales areas;and(5)Areas relatively high in soil organic carbon content were mainly distributed in the northeast and southwest of the studied region,and patches in the southeast,while areas relatively low were in the northwest.【Conclusion】The geographically weighted regression is higher than the ordinary Kriging,multiple linear regression,regression Kriging and geographically weighted regression Kriging in prediction accuracy.In this paper,efforts have been made to explore correlations between soil properties and environmental variables on large scales in an attempt to provide certain solutions and ideas for soil properties spatial mapping.