县域尺度表层土壤有机碳密度空间格局及影响因素定量刻画

A better understanding of the spatial variability and key influence factors of soil organic carbon density (SOCD) in topsoil at small scales may lead to more accurate soil organic carbon storage estimation, scientific land use management and effective environmental protection policy. However, the existing spatial analysis models are inadequate in explaining the spatial pattern and influence factors of soil properties. In this study, a county is selected to represent the small scale areas, meanwhile the existing theories and techniques are optimized and extended. Bayesian maximum entropy is firstly used to combine the multisource and heterogeneous environmental data, based on which the influence factors set of SOCD is built. Then, the mixed geographically weighted regression model is extended to be capable of handling the global linear, global nonlinear and local linear relationship between soil properties and its influence factors simultaneously, so as to refine the spatial quantitative relationship between SOCD and its influence factors in topsoil at small scale. Finally, the obtained spatial prediction results are used as the input parameters of the stochastic simulation, thus optimizing the simulation’s ability of dealing with the multisource and heterogeneous environmental data, aiming to accurately analyze the characteristics of spatial variability and uncertainty of spatial pattern of SOCD. The expected results would further our understanding of the formation mechanism of spatial characteristics of SOCD and its influence factors, as well as enriching the related theories and methods of spatial analysis.

精确掌握小尺度表层土壤有机碳密度空间变异特征及其关键影响因素，对多尺度土壤有机碳储量精确估算、土地利用科学管理和环保政策有效制定有重要的指导意义。既有空间分析模型对土壤属性空间格局及其影响因素的解释力度仍然有限。本研究拟以县域为小尺度代表，对相关理论和技术的优化扩展进行探索。本项目实施时，首先基于贝叶斯最大熵等方法融合多源异构环境数据，构建土壤有机碳密度影响因子集；然后扩展混合地理加权回归模型，使其具备土壤属性-影响因素之间全局线性、全局非线性和局部线性3种不同关系的同步分析能力，由此细化小尺度表层土壤有机碳密度和影响因素的空间定量关系，得到其空间分布预测结果；最后将预测结果作为随机模拟的输入参数，优化随机模拟处理多源异构环境数据的能力，并精确分析土壤有机碳密度空间变异特征和空间格局不确定性。本研究结果可加深土壤有机碳密度空间格局形成机制及其影响因素的认知，并可充实空间分析相关理论和方法。

精确衡量土壤属性与相关影响因子之间的空间定量关系是数字土壤制图模型成功应用的关键。然而，目前常用的环境数据在小尺度上的空间分辨率和精度有待提高；另一方面，既有空间分析模型往往只能体现特定类型的空间定量关系，因而对土壤属性影响因素的解释力度有限。针对此情况，本项目首先利用贝叶斯最大熵（Bayesian maximum entropy，BME）方法融合多源异构环境数据，以提高环境数据的可用性。然后尝试融合地理加权回归方法（Geographically Weighted Regression, GWR）和广义可加模型（Generalized Additive Model, GAM），使模型具备同步分析土壤属性和环境因子之间全局线性、全局非线性和局部线性3种不同关系的能力。本项目的研究内容主要包括两个方面：一是BME框架下构造软数据的新方法，即基于土壤属性-环境因子空间局部定量关系的区间类型软数据，以及基于异常值集合构建的直方图类型的软数据；二是融合GWR与GAM的模型建立与诊断方法：先利用赤池信息准则（Akaike Information Criterion）将环境因子划分为参数项和非参数项，然后通过迭代方法，求解得到最优的模型回归系数和最稳定的平滑函数形式，前者用以表征土壤属性和环境因子之间的局部线性关系，后者用以表征全局线性和全局非线性关系。对多个区域多种土壤属性的研究结果表明，本项目所获取的空间预测结果精度更高、不确定性更低而信息量更丰富，同时所提出的建模方法对土壤属性和相关影响因素之间空间定量关系的捕捉能力也优于传统方法。本研究对数字土壤制图模型的有机融合具有重要的借鉴意义，并可充实空间分析相关理论和方法。