【作者】 王帅；

【导师】 王秋兵； 张甘霖； Qianlai Zhuang；

【作者基本信息】 沈阳农业大学 ， 土地资源与信息技术， 2018， 博士

【摘要】 本研究以黑河流域中游地区为研究区。由于该区空间尺度大,景观条件复杂,致使区域内成土环境的空间异质性强、土壤的形成过程多种多样,土壤属性的精确模拟较为困难。此外,现有数字土壤制图的研究普遍基于大批量密集性的采样,往往需要大量的时间和经费投入,因此,基于有限采样点位的预测性土壤制图研究具有重要意义。目前的数字土壤制图研究中,不管空间预测模型还是深度函数模型的建立都缺少对土壤发生学知识的运用。运用土壤发生学知识不仅可以了解土壤的形成因素、发生过程、类型及其性质,而且可以提高土壤属性的空间预测真实性和准确性。本论文首先假设土壤发生学知识可以用于提高土壤属性空间预测的合理性和准确性。为了验证该假设,在黑河流域中游地区运用目的性采样方法,采集了80个土壤剖面数据和43个土壤表层饱和导水率数据,并结合12个环境变量,将土壤发生学知识与指数函数、增强回归树模型和环境相似度度量模型有机的结合起来,探讨了区域内土壤有机碳、饱和导水率和田间持水量的空间分布特征,其主要结果如下:（1）运用等面积样条函数和分类指数函数,对土壤有机碳和田间持水量进行了深度分布模拟。等面积样条函数可以很好的模拟出黑河流域中游地区土壤有机碳的深度分布特点,基于区域内土壤田间持水量采样数据所建立的深度分布函数,可以反映黑河流域中游地区1 m深土体内的3种田间持水量的分布模式,即随深度递减型、随深度递增型和随深度先递减再递增型。这两种深度函数的模拟精度均较高,模拟R²的均值分别为0.93和0.86。（2）将土壤形成过程知识与景观划分相结合,利用高斯混合模型将研究区内复杂的景观环境划分为6个环境条件相对一致的―气候-植被‖景观单元。每个景观单元表达为区域内土壤饱和导水率分布类型与特征要素的组合集,基于景观单元的数字土壤制图方法可以提高模型预测的灵活性与真实性。（3）利用深度分布函数对土壤有机碳和田间持水量的三维分布进行模拟,并结合增强回归树模型对其三维空间进行预测制图。这两种方法可以较好地描述黑河流域中游地区的土壤有机碳和土壤田间持水量进行三维空间分布特征,土壤有机碳的MAE（mean absolute error）、RMSE（root mean of squared error）、R²（Coefficient of Determination）和LCCC（lin’sconcordance correlation coefficient）范围为0.40-0.85 kg m^-3,0.55-1.21 kgm^-3,0.39-0.65和0.56-0.80,而土壤田间持水量为1.57%、7.89%、0.63和0.81。在大尺度复杂景观条件下,基于土壤发生学知识的数字土壤制图方法可以对土壤属性的空间分布特征进行较准确的描述,符合土壤属性预测制图的需求,为区域土壤有机碳库估算和农田灌溉等提供数据支撑。更多还原

【Abstract】 In this study,the middle reaches of the Heihe River Basin was selected as a research area,due to its large spatial scale and complex landscape conditions,a strong spatial heterogeneity of the soil-forming environment,and the formation of a variety of soil processes.Given these complexisities,it is difficult to accurately simulate the two-dimensional（2D）and three-dimensional（3D）soil properties in the area.In addition,the current study of digital soil mapping（DSM）is generally based on sampling in large quantities and often requires significant time and expense.The 2D and 3D DSM has low accuracy in predicting soil properties at large scales.Therefore,using finite sampling points is of great significance in the research with DSM.However,neither the spatial prediction model nor the depth function model has used the knowledge of soil genesis.Using the knowledge of soil genesis can not only understand the formation of soil factors,occurrence process,type and nature,but also improve the accuracy of spatial prediction of soil properties.In this dissertation,we first assume that the knowledge of soil genesis can be used to improve the spatial prediction of soil properties.To test this hypothesis,we collected 80 soil profile datasets and 43 soil surface saturated hydraulic conductivity（SHC）datasets using the purposeful sampling method in the middle reaches of the Heihe River Basin.In combination with 12 environmental variables,soil knowledge,exponential function,boosted regression tree model and similarity-based model were combined to explore the spatial prediction of SOC,SHC and soil field capacity（SFC）.The main results are as follows:（1）Combining the equal-area spline function and classification exponential function and the knowledge of soil genesis,the depth distribution of SOC and SFC was simulated.The equal-area spline function can well simulate the depth distribution of SOC in the middle reaches of the Heihe River basin.The classification exponential function was established on the basis of the depth distribution pattern of SFC capacity in the region.The depth distribution function can reflect the distribution pattern of three kinds of SFC in the depth of 1m in the middle reaches of the Heihe River basin,including a decrease with depth,an increase with depth,and a decrease initially and an increase afterwards.The results show that the depth functions provided high fitting accuracy,with a mean R² of 0.93 and 0.86,respectively.（2）According to the Gauss mixture model,the complicated landscape environment in the study area is divided into six"climate-vegetation"landscape units under relatively uniform environmental conditions.Each landscape unit is expressed as a set of features of the type and distribution of SFC in the region.Compared with other DSM methods,this method can provide the flexibility and authenticity of model predictions.（3）The depth distribution function was used to define the 3D distribution of SOC and SFC.The 3D spatial prediction maps were constructed based on the boosted regression tree model.These two methods can describe the characteristics of 3D spatial distribution of SOC and SFC in the middle reaches of the Heihe River basin.The ranges of MAE（mean absolute error）,RMSE（root mean of squared error）,R²（Coefficient of Determination）,and LCCC（lin’s concordance correlation coefficient）of SOC are 0.40-0.85 kg m^-3,0.55-1.21 kgm^-3,0.39-0.65和0.56-0.80,respectively.The MAE,RMSE,R²,LUCC of SFC is 1.57%,7.89%,0.63 and 0.81,respectively.Under the conditions of large-scale and complex landscapes,the DSM method based on the knowledge of soil genesis can accurately describe the spatial distribution of soil properties,which satisfies the relevant administrative and research departmental need,such as estimating SOC pool and farmland irrigation level.更多还原