复杂地理环境下土壤有机碳含量的精细空间预测方法

The accurate and fine knowledge of spatial distributions of organic carbon contents will help soil fertility management and development of soil carbon sequestration and emission reduction measures. Fine prediction of soil organic carbon contents is subject to data acquisition and effective prediction methods in the context of complex geographical environments. Six typical landscape-scale units in red-earth hilly region of the subtropical zone, ten typical medium-scale counties and large-scale four provinces in central south regions, are selected as study areas, due to their characteristics of typical geographical enrionments with subtropical terrestrial ecosystem.. Large amounts of soil organic carbon and geographic environmental data were early collected in the selected study areas. It is discussed that the three main issues of boundary constraint of model input variables,stability of model structure and reduction of the sensitivity of model parameters to sample sizes.To sovle above-mentioned questions, the following researches will be undertaken:.The factors of geographical environments are extracted, which affect the spatial variability of organic carbon contents, and taken as input variables of the model. Classification of the input variables of the model results in their boundary constraints, which lets unknown points be spatial ly similar to samples in the precess of model prediction..A modeling approach,which is based on low-dimensional feature vectors mapping to high-dimensional inner product space,is used to construct a linear model of structural stability.In such a model,its structure is not sensitive to the changes of the input variables..A method to flexibly adjust model parameters is presented to reduce their sensitivity to sample sizes. Finnally, Based on above-mentioned results, the accurate and fine prediction model of soil organic carbon contents will be created..The innovation of this proposal is the creation of a new method to make model structure stable and limit input variables and parameters, and it will provide new ideas for the expansion of existing digital soil mapping methods and their applications in the prediction of soil properties in medium or large scale.

土壤有机碳含量的精细分布信息有助于土壤肥力管理及制定合适的土壤固碳和减排措施。复杂地理环境下有机碳含量的精细空间预测受制于大量数据的获取及有效的预测方法。选择亚热带红壤丘陵区6个典型小尺度景观单元、中尺度上中南区10个典型县及大尺度上中南四省为研究区域。依托前期在研究区域积累的大量有机碳和地理环境数据,围绕约束模型输入变量的边界、稳定模型结构及降低模型参数对样本数量的敏感性三个关键问题，开展如下工作：提取与有机碳含量变异有关的地理环境因子，作为模型输入变量；构建自收敛分类算法，对模型输入变量进行分类，以约束输入变量的边界，保障模型预测时未知点与监测样点的空间相似性；提出结构稳定的线性模型的建模方法，降低模型结构对输入变量的敏感性；建立模型参数的弹性调节方法，降低模型参数对样本量的敏感性；以此构建有机碳含量的精细空间预测模型。研究成果可为现有土壤制图方法在中/大尺度上的拓展和应用提供新思路。

土壤有机碳含量的精细空间分布信息有助于土壤肥力管理及制定合适的土壤固碳和减排措施。复杂地理环境下影响土壤有机碳含量的地理环境因子复杂多变。准确估算土壤有机碳含量，不仅需要明确土壤有机碳含量变异与环境变量的关系，还需要构建有效的空间估值模型。 . 本研究以亚热带丘陵区为研究区域，充分利用了课题组前期积累的景观尺度（湖南省桃源县盘塘）、流域尺度（湖南省长沙县金井）以及中南4省（湖北、湖南、广西、广东）土壤有机碳数据以及相关地理环境数据，开展了土壤有机碳含量变异与地理环境变量的关系研究，建立了土壤有机碳的空间估算模型，主要研究成果如下：. （1）揭示了影响亚热带丘陵区土壤有机碳含量变化的主控因子。在亚热带丘陵区景观和流域尺度上，地形因子和土地利用变化是驱动土壤有机碳含量变化的主要因素；但是省域尺度上土壤有机碳含量变化依赖外源有机物质的投入。这可能是由于省域尺度上土壤样点主要采集于稻田生态系统所致。. （2）建立了基于环境协变量相似的空间同质区划分方法。该方法以土壤景观模型理论为指导，用与土壤有机碳含量变化相关联的环境变量为空间同质性度量指标，通过空间聚类实现与待估值点（单元）空间同质的同质邻域构建和划分。. (3）提出了基于环境协变量相似的空间估值模型。该模型整合了2种自主研发的空间估值算法（基于环境协变量相似的反距离加权算法以及基于环境协变量相似的局部加权回归算法），可有效改善复杂地理环境土壤有机碳含量的空间估算。进一步将以上集中式空间插值算法拓展到分布式插值算法，建立基于云计算平台Hadoop的空间插值系统，为大尺度海量数据的插值计算提供了技术支撑。