【作者】 刘志鹏；

【导师】 邵明安；

【作者基本信息】 中国科学院研究生院（教育部水土保持与生态环境研究中心） ， 土壤学， 2013， 博士

【摘要】 严重的水土流失、土壤肥力和生产力下降、土地退化和荒漠化等一系列生态环境问题威胁着黄土高原地区自然、社会和经济的可持续发展。土地利用调整、植被恢复以及水土保持工程的大力开展是实现该区生态恢复和重建的根本途径。在土壤贫瘠的黄土高原，土壤养分是制约植被生长和土地生产力的重要因素，同土壤水分一起影响着该区土地利用和植被类型的空间分布。从区域尺度上，系统而科学地认识整个黄土高原地区土壤养分资源空间分布特征及其影响因素，是各项生态重建工程有效实施的有力保障，对区域宏观决策具有重要的指导意义。本论文以整个黄土高原地区为研究区域，旨在摸清现阶段主要土壤养分的含量及储量水平，明晰其空间变异特征及空间分布格局，并揭示不同尺度下土壤养分与相关环境因素之间的关系。研究团队于2008年在黄土高原地区开展了大规模、高密度的野外试验，在382个代表性采样点上共采集表层0-20cm和20-40cm以及深层0-100cm和100-200cm土壤混合样品共1528个；表层0-20cm和20-40cm原装土壤样品764个；同时获取了各采样点经纬度、海拔高度、坡向、坡位、坡度、土地利用、植被类型等相关信息。室内统一测定的指标有土壤有机碳、土壤全氮、土壤全磷、土壤全钾、土壤pH值、土壤机械组成（美国制分级）、土壤容重等。使用经典统计学（相关分析、线性回归、方差分析、多重比较等）、地质统计学（半方差函数、克里格插值、因子克里格分析）、状态空间模拟等方法，对各土壤养分指标的空间变异特征、空间插值及预测、不同空间尺度下的主要影响因素进行了分析，主要研究结果如下：（1）黄土高原地区表层土壤有机碳浓度变化在0.38-54.03g kg^-1之间，0-20cm和20-40cm土层中的均值分别为10.34g kg^-1和6.78g kg^-1；表层0-20cm和0-40cm土壤中有机碳密度分别为2.64kg m^-2和4.57kg m^-2，相应深度的有机碳总储量分别为1.64Pg和2.86Pg (1Pg=10¹⁵g)。深层0-100cm和0-200cm土壤有机碳密度分别为7.70kg m^-2和12.45kg m^-2，相应深度的有机碳总储量分别为4.78Pg和5.85Pg。黄土高原地区0-20cm和0-100cm深度土壤有机碳总储量分别占世界总储量的0.36%和0.31%，占我国土壤有机碳总储量的8.21%和5.32%。表层土壤有机碳浓度和密度的变异系数表明，它们在区域范围内均表现为中等程度变异。皮尔逊相关分析结果表明表层土壤有机碳浓度与相关环境因素之间存在显著（p<0.05）相关关系，其中与土壤全氮、土壤pH值以及粘粒含量关系最为密切。多元逐步线性回归（p<0.05）结果表明，土壤全氮、粘粒含量、土壤pH值、海拔高度以及气温对土壤有机碳浓度具有显著影响。方差分析（p<0.05）结果表明，气温、降雨、海拔高度、土壤细颗粒（<20μm）以及土地利用类型对土壤有机碳储量影响显著。区域尺度上，农地中土壤有机碳含量显著高于林地和草地。地质统计学结果表明，表层土壤有机碳浓度（0-20cm和20-40cm）和密度（0-40cm）的最大空间相关距离分别为384km、393km和339km。块金基台比分别为0.52,0.50和0.45，表明土壤有机碳在区域尺度上表现为中等程度空间依赖性。使用克里格插值方法绘制了整个黄土高原地区表层土壤有机碳浓度和密度的空间分布图。总体而言，整个区域中心有一个土壤有机碳含量相当低的区域，它被由中心向区域边界发散的含量递增的同心圆包围着。这种分布特点与区域尺度上大地形区的分布相对应，同时也影响着土地利用类型的分布：黄土高原的四周边界上分布有山地；河谷平原区主要分布在中部地区；鄂尔多斯风沙台地位于整个区域的中部。（2）不同土地利用类型下，表层0-20cm土层中土壤全氮和全磷浓度均值变化在0.58-0.81g kg^-1和0.50-0.73g kg^-1之间；20-40cm深度土层中，土壤全氮和全磷浓度均值的变化范围分别为0.46-0.60g kg^-1和0.48-0.61g kg^-1；0-40cm深度土壤全氮和全磷密度均值变化范围分别为0.27-0.39kg m-2和0.27-0.38kg m-2。黄土高原地区0-40cm深度土壤全氮和全磷的总储量为0.217Pg和0.205Pg，占到我国总储量的5.4%和7.3%。变异系数表明，不同土地利用类型下土壤全氮和全磷含量均表现为中等程度变异。方差分析（p<0.05）结果表明，气温、降雨以及土地利用类型对表层土壤全氮和全磷含量具有显著影响。降雨和气温的影响在不同土地利用类型下不同，土地利用类型的影响在不同降雨和气温区内也表现出不同的特点。总体而言，农地中土壤全氮和全磷含量高于林地和草地；具有较多降雨和较高气温的地区土壤全氮和全磷含量更高。相关和线性回归结果表明，土壤全氮和全磷与相关环境因素，如土壤有机碳、降雨、气温、海拔高度、经纬度、坡度、粘粒和粉粒含量以及土壤pH值之间存在显著的相关关系，而它们之间的关系随土地利用类型的变化而不同。针对不同土地利用类型，建立了土壤全氮和全磷的线性预测方程。区域尺度上，土壤全氮和全磷浓度和密度的块金基台比表明，它们具有中等程度空间依赖性，最大空间相关距离分别在374-461km以及546-664km之间。通过克里格插值绘制了黄土高原地区表层0-40cm深度土壤全氮和全磷密度的空间分布图。（3）土壤全钾浓度在表层0-20cm和20-40cm土层中变化范围分别为10.07-30.97gkg^-1和12.82-32.39g kg^-1，均值分别为19.25g kg^-1和19.10g kg^-1；变异系数分别为31.7%和26.9%，表现为中等程度变异；块金基台比分别为31.7%和26.9%，在最大相关距离546km和564km范围内表现出中等程度空间依赖性。使用经典线性回归和状态空间模拟方法分析了表层0-20cm土壤全钾含量与土壤容重、粘粒和粉粒含量、土壤pH值、降雨、气温以及海拔高度之间的相互关系。最优状态空间方程能够解释97%的土壤全钾总变异，而最优线性回归模型仅能解释26%的总变异。使用相同变量，所有的状态空间方程在预测土壤全钾含量时均优于对应的线性回归方程。气温、容重以及粘粒含量被认为是影响土壤全钾局地变异的最重要因素，因为它们均出现在最优状态空间方程中。结果表明，空间状态方程可以作为有效的工具很好地模拟区域尺度上土壤养分的局地空间变异。（4）土壤pH值在表层0-20cm土层中变化范围为6.06-10.76，均值为8.49，中位数为8.48。土地利用类型对土壤pH值具有显著（p<0.05）影响，草地土壤pH值显著高于农地和林地。区域尺度上，土壤pH值表现出较弱程度的变异和较强的空间依赖性，其变异系数为5%，块金基台比为0.243。以土壤pH值为研究对象，分析了四种空间插值方法，即反距离法、样条函数法、普通克里格法和协克里格法及其相关参数对空间插值精度的影响。交叉检验结果表明，克里格方法相对于反距离法和样条函数法具有更高的精度，而使用土壤有机碳作为协变量的协克里格法能够进一步提高插值精度。四种空间插值方法绘制的土壤pH值空间分布图，在整体上具有相似性，在细节上存在不同。整体而言，土壤pH值的相对低值区分布在黄土高原的东南部，可能与该区较多降雨、土壤淋溶以及较高的土壤有机质含量有关。土壤pH值的相对高值区分布在黄土高原的中北部，与该区干旱的环境和不合理灌满以及严重的土壤盐渍化有关。（5）经典统计学中的相关分析并没有考虑各变量自身以及相互间的空间位置关系以及它们的区域化特征。使用结合了多元统计和地质统计学的因子克里格方法研究了各土壤性质（土壤有机碳、全氮、全磷、全钾、土壤pH值、容重、粘粒和粉粒含量）与相关环境因素（气温、降雨、海拔高度、土壤类型和土地利用类型）之间的尺度依赖性相关关系。使用含有块金效应和两个球状结构的协同区域化线性模型拟合单变量以及双变量交互的半方差函数，并据此分别研究了块金尺度（<30-50km），短变程尺度（最大相关距离200km）以及长变程尺度（最大相关距离400km）上各变量之间存在的尺度依赖性相关关系。使用了主成分分析以及单位圆投影的方法表达了多变量之间复杂的相关关系。结果表明，各变量之间的相关关系随尺度的变化表现出不同特征。总体而言，土壤有机碳和全氮在块金尺度和长变程尺度上紧密相关，而在短变程尺度相关性不明显。降雨和土壤粘粒含量在块金尺度和长变程尺度上与土壤全磷含量关系密切。土壤全钾在各尺度上与其它变量之间的相关关系均不明显，但其与土壤类型在长变程尺度上关系密切。土壤pH值在块金尺度、短变程尺度以及长变程尺度上分别与土壤容重、土壤类型以及海拔高度密切相关。土壤容重与土地利用类型在各尺度上都具有紧密关系。土壤利用类型和土壤类型被认为是控制短变程尺度上土壤性质空间变异的主要影响因素，而影响长变程尺度土壤变异的主要因素为降雨、气温和海拔高度。本论文以大量的野外试验数据为基础，从整体上认识了黄土高原地区土壤有机碳、全氮、全磷、全钾以及土壤pH值的空间变异特征及其与相关环境因素之间的关系。可靠的土壤养分空间数据丰富了黄土高原土壤数据库，为该区土壤养分空间变异及相关研究的深入开展提供了整体框架指导，也为今后该区大尺度上数字土壤制图、碳氮循环模拟、面源污染评估等提供了可靠的数据支持，相关研究结果也将为黄土高原地区各项生态重建工程的宏观决策提供理论和实践指导。更多还原

【Abstract】 A series of eco-environmental problems, such as severe soil erosion, decreases in soilfertility and productivity, land degradation and desertification, have threatened thesustainable development of the ecosystems and social economy on the Loess Plateau ofChina. Regional and local projects aiming on ecological restoration and reconstructionhave been launched to combat these problems on this area, practically through land useoptimization, vegetation recovery and soil and water conservation. Soils on the LoessPlateau are poverty due to the scarcity of both soil nutrients and water resources. Thesetwo factors together greatly limit plant growth and agricultural production, controlling thespatial variations of land use and vegetation. From regional perspective, systematic andaccurate information on the spatial variations of soil nutrients and the impact factors atdifferent scales is basic and essential for effective applications of these ecological projects,and would be helpful in related macro policy makings.The main purposes of this dissertation were to （1） explore the current level and stocksof several main soil nutrients across the entire Loess Plateau region;（2） to reveal thespatial variability of these soil nutrients and illustrate their distribution patterns;（3） toinvestigate the relationships between these soil nutrients and pertinent environmentalfactors at different scales;（4） and to generate accurate prediction models usingeasy-to-measure variables. An intensive soil survey with a sampling interval of about30-50km was accomplished within one year during2008, by investigating382representative sampling sites across the entire Loess Plateau region （62.4×104km2）. Atotal of1528composite soil samples were corrected using a handy soil auger （5cm indiameter） from0-20cm and20-40cm topsoil layers, and0-100cm and100-200cm deepsoil layers. Additionally,764undisturbed soil cores were collected with cutting rings （100 cm³in volume）. The environmental conditions of each sampling site were recorded, suchas latitude, longitude, elevation, aspect, slope gradient, slope position, land use type andvegetation type. All the soil samples were taken to the laboratory for measurements of soilorganic carbon （SOC）, soil total nitrogen （STN）, soil total phosphorus （STP）, soil totalpotassium （STK）, soil pH, soil mechanical composition and bulk density （BD）. Traditionalstatistics methods （correlation analysis, linear regression, ANOVA and post-hoc）,geostatistics methods （semivariogram, kriging interpolation, factorial kriging analysis） andstate-space modeling approach were used for spatial analysis and generation of predictionmodels. The main resultes are listed as follows:（1） Soil organic carbon concentrations （SOCC） varied within a wide range throughout theregion from0.38g kg^-1to54.03g kg^-1, with mean values of10.34g kg^-1and6.78gkg^-1for the topsoil （0-20cm） and subsoil （20-40cm）, respectively. The mean SOCdensity （SOCD） was2.64kg m^-2in the0-20cm soil layer and4.57kg m^-2in the0-40cm soil layer, and it was estimated that1.64and2.86Pg (1Pg=10¹⁵g) of organiccarbon were stored in these soil layers, respectively. Estimates for deeper soil layersindicated that mean SOCD in the0-100cm and0-200cm layers was7.70and12.45kgm^-2, respectively, while the total organic carbon stocks amount to4.78Pg （0-100cm）and5.85Pg （0-200cm）, respectively. The SOC stocks in the0-20cm and0-100cm ofsoils in the Loess Plateau region contribute0.36%and0.31%to the global SOC storedin these respective layers. In addition, our results indicated that the0-20and0-100cmsoil layers of the Loess Plateau, which covers nearly6.5%of the area of China,currently holds about8.21%and5.32%of the total SOC stocks in these layers in China,respectively. Coefficient of variation values showed moderate variation for both SOCconcentration and density values in both0-20cm and20-40cm soil layers. Significantcorrelations were detected between SOCC and these environmental variables, notablywith soil total nitrogen （STN）, soil pH and clay content. Multiple linear regressionanalysis indicated that STN, clay content, soil pH, elevation and temperature hadgreater effects on regional SOCC variability, among all the selected soil and sitevariables. The results of ANOVA showed that precipitation, temperature, elevation,clay plus silt contents and land use showed significant regional impacts on SOCD. Theresults also show that human activities have heavily affected SOC accumulation. Measured SOCD under cropland was relatively higher than under grassland andforestland.Geostatistics analysis showed that the maximum autocorrelation ranges were384km,393km and339km for SOCC （0-20cm and20-40cm） and SOCD （0-40km）,respectively. Nugget-to-sill ratios were0.52,0.50and0.45, which also indicatedmoderate spatial dependence. The distribution maps of SOCC in both topsoil layers andSOCD in0-40cm soil layers were produced by geostatistical method showed that theoverall spatial pattern was characterized by an area of low SOC content surrounded bybands with higher values that generally increased towards the region’s boundaries. Thedistribution pattern corresponded to that of the major regional landforms, which alsoinfluenced land use, whereby the sandy Ordos Plateau is surrounded by relativelyfertile plains and valleys, where the human population density is highest, and theregional boundary is mountainous.（2） In0-20cm soil layers, mean STN concentrations （STNC） and STP concentrations（STPC） ranged from0.58g kg^-1to0.81g kg^-1and from0.50g kg^-1to0.73g kg^-1,respectively, under different land types. In20-40cm soil layers, mean STNC and STPCranged from0.46g kg^-1to0.60g kg^-1and from0.48g kg^-1to0.61g kg^-1, respectively.Mean STN and STP densities in0-40cm soil layers ranged from0.27kg m^-2to0.39kgm^-2and from0.27kg m^-2to0.38kg m^-2, respectively, under different land use types.All the concentrations and densities of STN and STP under different land use typesshowed moderate variations, which was indicated by the values of coefficient ofvariation. We detected significant （p<0.05） effects of land use, precipitation andtemperature on both STN and STP. But the results varied among different precipitationand temperature regions and different land use types. Generally, croplands had higherconcentrations and densities of STN and STP than forestlands and grasslands, andregions with more precipitation and higher temperature had higher STN and STPdensities. Significant correlations were found between STN, or STP, with selectedfactors, i.e. soil organic carbon, precipitation, temperature, elevation, latitude, longitude,slope gradient, clay content, silt content and soil pH. The results were not consistentwithin either the variable or the land use types. We generated land-use specific linearmodels to predict STN and STP using these related variables. Geostatistical analysis showed moderate spatial dependence of both STN and STP, indicated by the values ofnugget to sill ratio. The spatial range of STN and STP ranged from374km to461kmand546km to664km, respectively. This range was much larger than our samplingintervals （30-50km）. The distribution maps of STN and STP densities were made withkriging interpolation. Finally, the stock of STN and STP was estimated to be0.217Pgand0.205Pg in the upper0-40cm soil layers, which was about5.4%and7.3%of thetotal nitrogen and phosphorus stocks in China. Our study suggests that it is important totake land use into account when considering variation of STN and STP at regionalscale.（3） In0-20cm and20-40cm soil layers, soil total potassium （STK） concentration variedfrom10.07-30.97g kg^-1and12.82-32.39g kg^-1, with mean values of19.25g kg^-1and19.10g kg^-1, respectively. The coefficients of variation for STK were13.4%and13.3%,defined as moderate variation. The spatial ranges of STK were546km and564km.The nugget-to-sill ratios were31.7%and26.9%, showing moderate spatial dependence.Two methods, state-space modeling and classical linear regression, were used toquantify the relationships between STK （0-20cm） and bulk density, clay and siltcontent, soil pH, precipitation, temperature, and elevation. The best state-space modelsexplained more than97%of the STK variation, while the best linear regression modelexplained less than26%of the STK variation. The results showed that all thestate-space models described the spatial variation of STK much better than thecorresponding linear regression models. Temperature, bulk density and clay contentwere identified as important factors that affected localized variation of STK, since theywere connected to the better performance of the state-space models. State-spacemodeling is recommended as a useful tool for quantifying spatial relationships betweensoil properties and other environmental factors in large-scale regions.（4） In0-20cm soil layers, soil pH values ranged from6.06to10.76, with a mean of8.49and a median of8.48. Land use type had a significant effect （p <0.01） on soil pH;grassland soils had higher pHs than cropland and forestland soils. From a regionalperspective, soil pH showed weak variation and strong spatial dependence, indicatedby the low values of the coefficient of variation （5%） and the nugget-to-sill ratios（<0.25）. Indices of cross-validation, i.e. average error （AE）, mean absolute error （MAE）, root mean square error （RMSE） and model efficiency coefficient （MEC） wereused to compare the performance of the four different interpolation methods, i.e.inverse distance weighting （IDW）, splines, ordinary kriging and cokriging. The resultsshowed that kriging methods interpolated more accurately than IDW and splines.Cokriging performed better than ordinary kriging and the accuracy was improved byusing soil organic carbon as an auxiliary variable. Regional distribution maps of soilpH were produced. The southeastern part of the region had relatively low soil pHvalues, probably due to higher precipitation, leaching, and higher soil organic mattercontents. Areas of high soil pH were located in the north of the central part of theregion, possibly associated with the salinization of sandy soils under inappropriateirrigation practices in an arid climate.（5） Traditional statistical analysis of the correlations between spatially distributed variablestakes no account of their regionalized nature. Factorial kriging analysis （FKA） wasapplied to investigate scale-dependent correlations between selected soil properties （i.e.soil organic carbon （SOC）, soil total nitrogen （STN）, soil total phosphorus （STP）, soiltotal potassium （STK）, soil pH, bulk density （BD）, and clay and silt contents） andenvironmental factors （i.e. elevation, precipitation, temperature, land use type and soiltype）. A linear model of coregionalization, including a nugget effect and two sphericalstructures （effective ranges of200and400km）, was fitted to the experimental auto andcross-variograms of the variables. Scale-dependent correlations were calculated fornugget effect scale （<30-50km）, short-range scale with a range of200km andlong-range scale with a range of400km. Principal component analysis was conductedto clearly illustrate the correlations at each spatial scale. The scale-dependentcorrelations were different from the general correlations and varied at different scales.Generally, SOC and STN were strongly correlated at the nugget effect scale and thelong-range scale, but not at the short-range scale. Precipitation and clay content showedclose correlations with STP at the nugget effect scale and long-range scale. The STKwas weakly correlated with the other variables at each spatial scale, but closelycorrelated with soil type at the long-range scale. Soil pH was closely correlated withBD, soil type and elevation at the nugget effect, short and long spatial scales,respectively. Close correlations were found between BD and land use type at each spatial scale. Land use and soil type were considered to be the important factorscontrolling spatial variation of soil properties at the short-range scale while at thelong-range scale the likely factors were identified as precipitation, temperature andelevation.Based on intensive filed sampling and uniform laboratory measurements, our studyprovided an overview on the spatial variation and impact factors of soil organic carbon,soil total nitrogen, soil total phosphorus, soil total potassium and soil pH across the entireLoess Plateau region of China. The reliable spatial data updated the soil database for thestudy region, and can be used as important input layers in regional digital soil mapping,carbon and nitrogen cycle modeling and evaluation of the potential non-point sourcepollution associated with soil erosion. Moreover, the results presented in this dissertationcan serve as an important background for the future studies in related fields, and can beuseful in macro decision making for regional eco-environment restoration on the LoessPlateau.更多还原