【作者】 林清火；

【导师】 石元春； 李保国； 李虹；

【作者基本信息】 中国农业大学 ， 土壤学， 2013， 博士

【摘要】 天然橡胶是一种典型的资源约束型产业,在我国国防建设和国民经济中都具有重要地位。实施合理的养分综合管理是提高橡胶树生产力的重要手段。养分精准管理的前提是必须详细了解橡胶园土壤属性空间分布特征。目前野外调查取样是获取土壤基本属性和空间变化的基本手段,正确科学的土壤取样方法是橡胶园养分精准管理是否成功的关键。因此,在进行橡胶树养分精准管理之前,有必要明确橡胶园土壤养分的取样方法,这是进一步开展橡胶树养分精准管理基础。现行采样过程中橡胶园土壤样品采集区域主要位于萌生带位置,但其是否合理仍是个未知数。本文针对橡胶树生产管理上的特点,结合生产实际,首先研究植株尺度下土壤属性的取样位置,在此基础上利用地质统计学和空间模拟退火算法探讨田块尺度下橡胶园土壤样本取样数目和布局,主要研究结果如下：(1)基于植株尺度围绕着9株橡胶树,通过在84m2范围内按1mx0.5m规则网格采集168个橡胶园表层土壤样品,利用普通克里格对全氮、有机质、速效磷、速效钾和pH五种土壤属性按照0.5m网格进行插值,而后描述各自空间分布特征,结果表明：全氮、速效磷、有机质呈现出明显的“肥岛”效应。其中全氮“肥岛”面积最大,影响距离最远；速效磷的“肥岛”面积最小,与周围浓度成倍数递减,成环状同心圆分布；有机质的空间分布介于两者之间。速效钾的变程最大,斑块面积大且过渡均匀。橡胶园中管理措施(例如挖施肥穴以及修筑环山行等)导致植株尺度范围内土壤表层空间变异大,同时这些土壤属性的高变异系数特征也揭示了橡胶园土壤表层适合于取样分区管理。(2)通过主成分分析将植株尺度下全氮、有机质、速效磷、速效钾和pH五种土壤属性转化为具有高度代表性的协同变量主成分,而后利用模糊聚类方法将前几主成分进行划分管理分区,研究结果表明：植株尺度下橡胶园表层土壤可划分为萌生带、树头以及施肥穴位置三个分区。萌生带区域土壤全氮和有机质与整个研究区域的均值相近,而速效磷和速效钾略低于均值。由于萌生带区域保持植被自然生产且少受外界干扰,具有较高的稳定性,因此,从该处获取的土壤样本分析结果可代表氮肥和有机质的养分水平。(3)根据描述性统计结果,将植株尺度下橡胶园表层土壤属性划分为正态和偏态两组数据。针对土壤全氮和速效钾这两个正态或对数正态分布的土壤属性,本文应用普通克里格方法描述其合理取样位置,结果表明橡胶园土壤全氮和速效钾综合养分取样位置主要受土壤速效钾含量分布决定。在空间分布上,综合养分的合理取样位置主要位于远离施肥穴的橡胶树行间萌生带和中间树头附近区域。综合外界因素,其最优取样位置位于远离施肥穴的橡胶树行间萌生带位置。(4)针对土壤有机质具有偏态分布的数据特点,应用指示克立格法描述了土壤全氮和有机质的合理取样位置分布概率。研究结果表明橡胶园植株间土壤全氮和有机质含量均值±10%相对标准偏差范围的概率分布具有相似性。土壤全氮和有机质含量同时处于均值±10%相对标准偏差范围的高概率区域主要分布在东边地势较高的橡胶树行间萌生带和树头株间位置,其最佳取样位置即综合指标最大概率集中于地势高的橡胶树行间正中央。研究结果表明,应用指示克立格法可为南方丘陵地区多年生经济作物复杂表层土壤样品的采集提供一种可行的解决途径。(5)在田块尺度围绕231株橡胶树,总面积为4200m2(60m×70m)范围,按照6m×7m网格采集100个土壤样本。结果表明田块尺度橡胶园土壤养分属性均属于中等变异程度,但速效磷和速效钾的变异程度明显大于全氮和有机质,前者的变异系数约为后者的两倍。从土壤属性的空间分布特征来看,全氮、有机质和速效磷整体上表现出明显西高东低分布,而速效钾呈现出四周高中间低的“岛屿”分布。应用克立格方法计算田块尺度橡胶园土壤属性合理取样数目,其中全氮、有机质、速效磷和速效钾合理取样数目分别为9、9、29和32,利用克里格指导的采样效率比传统统计学方法提高约2-5倍。(6)将空间模拟退火算法引入到橡胶土壤取样布局中并探讨了不同约束条件下田块尺度橡胶园土壤取样方案,研究结果表明：在橡胶园土壤取样过程中,如果研究区域既无先验方差也无早期观测样点情况下,在给定一定取样数量情况下可以基于最小均值距离(MMSD)准则进行优化布局；如果具有早期观测样点或者具有类似区域的先验方差,则可通过MMSD和最小克里格方差(MMKEV)准则相结合或者先验方差知识和MMKEV准则相结合进行指导取样布局。空间模拟退火算法在处理橡胶园土壤取样区域障碍以及充分利用前期先验知识方面具有现实指导意义。更多还原

【Abstract】 Production of natural rubber is a typical resource-constrained industry. Natural rubber is of strategic importance in national economy and defense in China. The best management practice (BMP) for natural rubber’s nutrient management is a major pathway to improve its yield and quality. Prior to BMP in nutrient management, the spatial distribution of key soil physical and chemical properties in rubber plantation have to be identified and mapped out. In-field soil sampling is a widely-accepted method to acquire key soil chemical properties and their spatial variations. Hence, a sampling method based on a sound-developed spatial sampling theory is crucial for precision nutrient management. Therefore, we must study the sampling design in the rubber plantation to obtain essential, prior knowledge on BMP for rubber tree management. At present, the conventional soil sampling sites in rubber plantation are in the shrub and ruderal zones, which are usually located in a specific free land between the adjacent rubber planting strips with vegetative growth of controlling nature. Whether it could reliable represent the nutrient level in the rubber plantation is unknown. In this dissertation, we first studied the sampling locations of soil properties in an individual rubber scale. Second, we investigated the optimal soil sampling numbers in the rubber field scale based on the geostatistics methods. Third, we introduced the spatial simulation annealing to optimize the soil sampling design in the rubber field scale. The main results were as follows:(1) The study was conducted around nine selected rubber trees in a typical hilling area of84m2at Yangjiang State-owned Farm, Hainan Province, China. The experimental plot was divided into168equivalent rectangles. The dimension of each rectangle grid for sampling was1m x0.5m. Ordinary Kriging (OK) was employed to interpolate five soil variables (i.e. Total Nitrogen or TN, Organic Matter or OM, Available Phosphorous of AP, Available Potassium or AK and pH) into a0.5m grid cell in the non-sampling locations and delineated spatial distribution for the five soil chemical properties. Results showed that the distributions of soil chemical properties were obviously different. Contour maps of TN, AP and OM variables showed as an island of fertility. Area of the island of TN variable was the biggest, OM was second, and AP came the last and revealed a sharp decrease in the nearby locations. The range for the AK variable was3.36m and its contour map was relatively uniform. Management practices, such as digging fertilization caves and building contour ledges, resulted in high spatial variability of soil chemical properties in rubber tree plantation. The coefficients of variation for the soil chemical properties revealed considerable spatial variability and soil nutrients could be classified into several zones for management purposes.(2) Principal Component Analysis (PCA) was applied to transform original five soil variables into new, uncorrelated variables (axes) called the principal components (PCs), which retain as much as possible information present in the original data. The PCs with eigenvalues larger than1.0were selected for the development of management zone classes. Results showed that fuzzy cluster algorithms could classify the chemical properties in the soil into three zones such as rubber rhizome neck areas, shrub and ruderal zone and the areas around fertilization caves. The conventional soil-sampling sites were in the shrub and ruderal zones, where the soil TN and OM variables were approximate equal to the mean values and AP and AK concentrations were slightly lower than the mean values of interpolation estimations. Soil samples in the shrub and ruderal zones were not disturbed by human activity, therefore the contents of TN and OM variables in this zone could be more reliable for corresponding levels.(3) Based on descriptive statistics, we divided the five soil properties in the rubber plantation into two groups:normality group and abnormality group. We used the OK method to analyze the spatial distribution of TN and AK variables and the sampling sites of the combined nutrients was determined. Results indicated that the optimum sampling site of combined soil TN and AK was determined by distribution of the AK variable. The sampling site of combined soil TN and AK variables were located between the rubber rhizome neck and in the shrub and ruderal zone which was far away the fertilization cave. As the location between the rubber rhizome neck is often affected by some uncertain factors, so the best sampling sites are in the shrub and ruderal zone which was far away the fertilization cave.(4) As soil organic matter content in the rubber plantation was showed as abnormality,we employed nonparametric indicator kriging to analyze the spatial probability distribution of soil TN and OM between the mean±10%relative standard deviation (RSD) respectively and the probability maps of the combined index was presented. The results indicated that the high probability location of the soils,where total nitrogen concentration and organic matter content were between the individual mean±10%RSD, were between the rubber rhizome neck and in the shrub and ruderal zone in the high terrain. As the location between the rubber rhizome neck is often influenced by some uncertain factors, so the best sampling sites should be in the shrub and ruderal zone with the high terrain. The highest probability of combined index of TN and OM variables are located in the middle of the line spacing of rubber tree. The results present a new method to evaluate the reasonable soil sampling site of Latosol for multi-years of growth period of crops planted in hilly region of South China.(5) In the field scale, the study was conducted around231rubber trees covering an area of4200m2. The plot was divided into100equivalent rectangles. The dimension of each rectangle grid for sampling was16m×7m. Results indicated that all the soil properties were with medium spatial variability in the field scale. The coefficient of variability of AP and AK concentrations were double that of TN and OM. Spatial distribution of the soil properties revealed that concentrations of TN, OM and AP were higher in the west than in the eastern areas. Concentrations of AK were displayed as a fertilizer island, which was low in the center and high in the periphery. The optimal sampling numbers of TN, OM, AP and AK variables were9,9,29and32, respectively. Sampling efficiency based on geostatistics was2to5times than the estimated by classical statistics in this study.(6) In this study, we also introduced spatial stimulation annealing to optimize soil sampling design with different constrained conditions in the field scale. If there were no priori variance, no priori sampling points and sample numbers was available, soil sampling design should be optimized based on minimizations of the mean of shortest distrances (MMSD) criterion in the study area. If there had priori variance or priori sampling points in the study area, soil sampling design should be optimized based on minimizations of the mean of kriging estimation variance (MMKEV) criterion or MMSD plus MMKEV criterion. Spatial stimulated annealing could be well applied to optimize soil sampling strategy in those study area, where had constrained area or priori knowledge.更多还原