采样尺度对土壤养分空间变异分析的影响

以高密度土壤养分采样数据为数据源,通过随机抽取生成不同采样尺度的样点数据,分析采样尺度对土壤养分空间变异特征分析的影响。研究结果表明:区域土壤养分预测均值随采样尺度减小呈下降趋势,而变异系数增加;养分空间分布的全局趋势随采样尺度增大而增强,但不影响半方差模型;当采样尺度较大,样点间自相关较弱时,相对较少的样点也能满足区域统计参数估测分析需要,但不能用于空间变异特征和插值分析;当样点数大于最佳采样数时,养分统计参数、空间变异特征和插值分析随着采样尺度减小而精度提高,当采样尺度达到0.2左右时,能够满足中等空间变异的土壤养分空间插值分析需要;样点空间布局对相关距和空间插值分析精度的影响比采样尺度本身更为显著。     

土壤制图中多等级代表性采样与分层随机采样的对比研究

采样设计是土壤地理研究中备受关注的重要问题。本文以区域尺度土壤属性制图为例,将多等级代表性采样与经典采样中的分层随机采样进行对比研究。以安徽宣城研究区的表层砂粒含量为目标要素,采集数量均为59个的两套样点,设计不同数量(46、58和59)的样点分组,采用两种制图方法进行制图并利用独立验证点进行评价。结果表明:1)无论是采用多元线性回归方法还是基于环境相似度的制图方法,在同等样点数量下,利用代表性样点所得土壤图精度均高于利用随机样点所得精度,并且利用少量代表性样点(46个)所得土壤图精度也高于利用多量随机样点(59个)所得精度;2)随着代表性较低样点的增加,土壤制图精度基本有一个提高的趋势,而采用随机样点所得土壤图的精度波动较大。因此,可认为多等级代表性采样方法是一种可用于区域尺度土壤调查的有效采样方法,且比分层随机采样高效、稳定。     

一种基于样点代表性等级的土壤采样设计方法

采样设计是获取土壤空间分布信息的关键环节,直接影响到土壤制图的精度。目前常用的采样设计方法大多存在着设计样本量大、采样效率不高的问题。当可投入资源难以完成一次性大量采样时,采样往往需要多次、分批进行。然而现有分批采样方法多考虑各批采样点在地理空间的互补性,可能造成样本点在属性空间的重叠,影响采样资源的高效利用。鉴于此,本研究通过对与土壤在空间分布具有协同变化的环境因子进行聚类分析,寻找可代表土壤性状空间分布的不同等级类型的代表性样点,建立一套基于代表性等级的采样设计方法。将该采样方法应用于位于黑龙江省嫩江县鹤山农场的研究区,利用所采集的不同代表性等级的样点进行数字土壤制图并进行验证,探讨采样方案与数字土壤制图精度的关系,以评价本文所提出的采样方法。结果表明,通过代表性等级最高的少量样点可获取研究区的大部分主要土壤类型(中国土壤系统分类的亚类级别),且制图精度较高;随着代表性等级较低样点的加入,土壤图精度提高;但当样点增加到一定数量时,土壤图的精度变化不大。因此,与样点数相比,样点的代表性高低对制图精度的影响更大。该方法所提出的代表性等级可以为样点采集顺序提供参考,有助于设计高效的逐步采样方案。     

县域土壤有机质空间变异特征及合理采样数的确定

以有机质为例,以高密度土壤养分采样数据为数据源,通过随机抽取生成不同采样密度的样点数据,分析了不同采样密度下土壤有机质的空间变异特征及县域合理采样数。研究结果表明,在一定研究尺度下采样密度对土壤养分的模型拟合、变程和空间相关性没有显著影响,即适当减少样点数可以满足插值分析的需要,充分考虑土壤养分空间变异评价的精度分析,确定县域土壤有机质合理采样数应控制在400个以上。     

基于空间模拟退火算法的最优土壤采样尺度选择研究

以研究区0.5 km×0.5 km(尺度a)网格的7050个样点为基础,分别得到1 km×1 km网格的1757个样点(尺度b),2 km×2 km网格的444个样点(尺度c),4 km×4 km网格的110个样点(尺度d),以土壤有机质(SOM)为目标属性,运用模拟退火算法对4种采样尺度的土壤样点进行优化选择,确定区域土壤调查的最优采样尺度。研究发现,通过模拟退火算法优化选择后,尺度a、b、c、d的最优样点数量分别为956、751、283和95个,优选的样点在空间上均匀分布。随着采样尺度的减小,采样点数量呈倍数增长,但对土壤属性的预测精度并没有相应比例的增加,且随着样点数量的增加,土壤属性预测精度的增加量逐渐减小。从样点数量与土壤属性预测精度综合来看,2 km×2 km的采样尺度是最优的土壤采样尺度。     

基于地统计的土壤养分采样布局优化

传统的土壤养分采样布置方法都是基于采样区土壤特征状态空间随机变异的假设。而地统计学研究表明,土壤特征状态在空间上有关联性,因此利用传统方法来制定采样方案并不是最优的,因为它没有考虑土壤特性的空间相关性,不能反映其局部的变化特征。该文在分析土壤肥力空间变异的基础上,研究利用经典统计学方法确定合理的采样点数目,并基于地统计学的半方差函数拟合与Kriging方法确定合理的采样点布局的方法,选择典型地区的土壤肥力进行空间变异分析和采样点布置的优化设计。研究结果表明：在合理的位置布置14个采样点就可以满足典型基地种植区绘制施肥处方图进行变量施肥决策的要求;利用经典统计学与地统计分析相结合的方法进行农田尺度的土壤肥力采样布点优化分析具有良好的可行性。     

精确农业田间土壤空间变异与采样方式研究

以英国Hillsborough农业研究所附近的一块7.9 hm²的牧草地为研究区，采用地统计的半方差分析和克立格方法研究其空间变异性和空间插值。同时对研究田块的样点根据不同间距、不同形状进行删选，对不同布局状况下的结果进行统计比较，以获取满足一定精度下的最少采样个数和采样形状。研究结果表明，单纯利用样方统计，土壤有效钾需要65个采样点，大致为原始采样点的一半。而在考虑空间采样形状和空间插值效果，再采用最小显著性差异（LSD）进行比较，该田块土壤有效钾采样最好使用规则三角网布点（样点数为62个）。     

样点规模与采样方法对表层土壤pH空间预测精度的影响

土壤空间预测与数字化制图的精度与质量受土壤样点规模、采样策略、预测模型选择、目标区域地貌与成土环境复杂程度、协变量数据质量等多种因素共同制约。选择河南省为研究区，基于9种土壤样点规模、5种采样方法，应用5种最具代表性的机器学习（Machine learning，ML）算法对耕地表层土壤pH实施空间预测与数字化制图，用以对比分析不同样点规模与采样方法对ML模型的性能表现及土壤pH预测精度的影响。结果表明：（1）当研究区土壤样点规模从200个经由400个、800个、1 200个、1 600上升到2 000个时，无论使用何种采样方法，所有ML模型的性能表现与预测精度均呈快速上升的总体趋势；当样点规模达到并超过2 000个时，大多数ML性能表现趋于稳定，预测精度上升快速趋缓，表明2 000个土壤样点可能是这些ML模型预测研究区耕地表层土壤pH的样点规模阈值。（2）5种ML模型性能表现及其土壤pH预测精度存在明显差距，基于树结构的随机森林（Random forests, RF）和Cubist表现最好，无论使用哪种采样方法，这两种模型预测结果的决定系数（R2）均可稳定在0.75~0.80之间、RMSE保持在0.50以下。（3）当土壤样点规模足够大时，采样方法对ML模型性能和土壤pH预测精度的影响很小，五种采样方法的效果相差不大。当土壤样点规模小于2 000个时，采样方法的影响逐渐凸显。比较而言，条件拉丁超立方采样在样点规模较小时具备优势。当样点规模为1000个时，条件拉丁超立方采样仍可使随机森林和Cubist预测的R2维持在0.80左右；在样点规模小至200个时，条件拉丁超立方采样方法下5种ML模型预测的R2均在0.55以上。（4）不确定性分析结果显示，平均73.9%的验证样点表层土壤pH观测值落入随机森林模型90%预测区间，表明该模型的可靠性被轻微高估，但处于可接受范畴。此外，数据显示模型预测的不确定性与样点规模无明显关联。     

基于模拟退火算法的土壤样点布局和土壤景观模型优化

本研究利用多重线性回归方程,以地形因子为预测变量,构建关于土壤有机质的土壤景观模型,并以西南山地丘陵区的一块面积为2 km2的汇水盆地为研究区,对该区域的土壤有机质空间分布进行预测。在此基础之上,探讨最少可用多少个点来预测土壤有机质的空间分布,并使之预测精度不低于原始集合的精度;同时,找出最优土壤样点布局,确定不同地形部位的取样单元,使之预测精度最高。研究结果表明:在预测误差最小化的情况下,最少可用7个优化的样点就可以代替原始200个采样点,且优化的样点数为124时,模型预测土壤有机质空间分布的精度最高。优化后的土壤景观模型的拟合度比原始模型提高了3.28%,MAE降低了5.3%,RMSE降低了3.94%。     

土壤优化采样策略研究进展

[目的]对现有的采样方式进行系统地总结归纳,并探寻一种优化的采样策略,对采样强度、分析成本及其研究精度三方面进行均衡,即以最小的经济投入换取最大化的试验精度。[方法]广泛查阅近几年国内外的相关文献,对土壤优化采样策略的设定进行了系统的总结。将优化采样策略的理论分为合理采样数和样点布设两方面,就此分别介绍确定采样数的3种方法和确定样点布设的4种方式,详细介绍其发展现状,并对该领域今后的研究进行展望。[结果]目前的优化采样法大多基于模型来优化采样设计的研究,其初步采样的方式忽略了空间相关性所导致的信息损失,必然在一定程度上造成试验结果的偏差和人力物力的浪费。而且,研究尺度多集中在县级以下,没有对采样方案的设计确立统一的评价体系和标准。[结论]未来优化采样设计在样点优先级方面还应进行更加深入的研究。     

基于影像分析的黄土丘陵沟壑区土壤水分采样研究

[目的]以山西省寿阳县三眼井小流域土壤水分采样为例,对采样点进行了均匀性和代表性分析,以此验证土壤采样设计的合理性。[方法]利用统计分析SPSS 13.0软件和ArcGIS软件,以规则网格采样点作为初步布设图,然后结合遥感影像,把自动生成的规则网格采样点在兼顾土地利用类型和可操作性的前提下进行微调,即采用大均匀小随机布点法进行土壤采样点布设。[结果]最近邻点指数和变异系数均显示采样点分布较均匀;从采样点的坡度代表性来看,由于大均匀小随机采样考虑到实际可操作性,采样点布设多集中在0°~15°,坡度的代表性较差;从采样点的坡向代表性来看,采样点代表性较好;从土地利用类型代表性来看,大均匀小随机法得到的采样点对草地与耕地的代表性较好,对林地的代表性较差,主要因为林地多分布在陡坡上难以采样。[结论]总体来看,大均匀小随机采样法得到的样点均匀性和代表性都比较好,且由于黄土丘陵沟壑区沟壑纵横,地形复杂,利用此方法可以兼顾均匀性、代表性和可操作性。     

Sampling Designs for Validating Digital Soil Maps: A Review

Sampling design (SD) plays a crucial role in providing reliable input for digital soil mapping (DSM) and increasing its efficiency. Sampling design, with a predetermined sample size and consideration of budget and spatial variability, is a selection procedure for identifying a set of sample locations spread over a geographical space or with a good feature space coverage. A good feature space coverage ensures accurate estimation of regression parameters, while spatial coverage contributes to effective spatial interpolation. First, we review several statistical and geometric SDs that mainly optimize the sampling pattern in a geographical space and illustrate the strengths and weaknesses of these SDs by considering spatial coverage, simplicity, accuracy, and efficiency. Furthermore, Latin hypercube sampling, which obtains a full representation of multivariate distribution in geographical space, is described in detail for its development, improvement, and application. In addition, we discuss the fuzzy k-means sampling, response surface sampling, and Kennard-Stone sampling, which optimize sampling patterns in a feature space. We then discuss some practical applications that are mainly addressed by the conditioned Latin hypercube sampling with the flexibility and feasibility of adding multiple optimization criteria. We also discuss different methods of validation, an important stage of DSM, and conclude that an independent dataset selected from the probability sampling is superior for its free model assumptions. For future work, we recommend:1) exploring SDs with both good spatial coverage and feature space coverage; 2) uncovering the real impacts of an SD on the integral DSM procedure; and 3) testing the feasibility and contribution of SDs in three-dimensional (3D) DSM with variability for multiple layers.     

Spatial analysis of categorical soil variables with the wavelet transform

This paper describes a wavelet transform for the analysis of categorical (multistate) soil variables, i.e. ones (such as profile classes) that have two or more discrete states. The states are transformed to a continuous variable by a mapping which is optimized by scale and location to highlight local variation. The method is illustrated with data from a transect across a gilgai landscape in Australia. A categorical variable on relief, with three states, was recorded from the sample sites, from which soil cores had also been collected and analysed. The wavelet analysis showed a transient feature of the variation at scales up to 32 m. There was an interval where the characteristic alternation of depressions with the level plain was interrupted. The variation at scale 64 m appeared to be non-stationary. The relief was more variable on one side of a change point than it was on the other. This complex variation of relief was matched by that of the electrical conductivity of the soil, most strongly at the 64-m scale. The periodicity of conductivity, and the strength of its correlation with relief, were also different either side of the change point identified in the analysis of relief alone. Conductivity also showed similar transient features to relief. Evidently the wavelet transform can be used to elucidate the variation of categorical soil variables. The information from such an analysis is likely to be useful for planning surveys of the soil to measure continuous variables by sampling and laboratory analysis.     

结合多等级代表性采样和模糊隶属度的区域土壤制图方法

High-resolution and detailed regional soil spatial distribution information is increasingly needed for ecological modeling and land resource management. For areas with no point data, regional soil mapping includes two steps: soil sampling and soil mapping. Because sampling over a large area is costly, efficient sampling strategies are required. A multi-grade representative sampling strategy, which designs a small number of representative samples with different representative grades to depict soil spatial variations at different scales,could be a potentially efficient sampling strategy for regional soil mapping. Additionally, a suitable soil mapping approach is needed to map regional soil variations based on a small number of samples. In this study, the multi-grade representative sampling strategy was applied and a fuzzy membership-weighted soil mapping approach was developed to map soil sand percentage and soil organic carbon(SOC) at 0–20 and 20–40 cm depths in a study area of 5 900 km2 in Anhui Province of China. First, geographical sub-areas were delineated using a parent lithology data layer. Next, fuzzy c-means clustering was applied to two climate and four terrain variables in each stratum. The clustering results(environmental cluster chains) were used to locate representative samples. Evaluations based on an independent validation sample set showed that the addition of samples with lower representativeness generally led to a decrease of root mean square error(RMSE). The declining rates of RMSE with the addition of samples slowed down for 20–40 cm depth, but fluctuated for 0–20 cm depth. The predicted SOC maps based on the representative samples exhibited higher accuracy, especially for soil depth 20–40 cm, as compared to those based on legacy soil data. Multi-grade representative sampling could be an effective sampling strategy at a regional scale. This sampling strategy, combined with the fuzzy membership-based mapping approach, could be an optional effective framework for regional soil property mapping. A more detailed and accurate soil parent material map and the addition of environmental variables representing human activities would improve mapping accuracy.     

辅助时序数据用于土壤盐分空间预测及采样研究

以普通克立格法作为参考，利用辅助数据的两种预测方法，即协同克立格法和回归克立格法对海涂区土壤盐分进行空间内插计算，并在目标变量的采样数目不断减少的情况下，利用80个检验样本，对比了这3种方法的预测精度。结果表明，不论目标变量的样品数目如何减少，利用了辅助变量的协同克立格法和回归克立格法的预测精度较普通克立格法都有了较大提高，而且回归克立格法的预测精度总体上要好于协同克立格法。对不同样品数目下3种方法的预测误差进行T检验发现，回归克立格法对普通克立格法和协同克立格法预测误差的减少在不同的样本数目下都达到了极显著水平。研究结果表明，利用连续几个时段上辅助的时序数据，来对同样点位上下一个时段的变量进行估值，可以较大地提高估值精度，节省采样成本。尤其是回归克立格法的回归部分可以是一般的线性模型，也可以是非线性模型,在预测时无疑更具灵活性。     

How geostatistics can help you

Abstract. Geostatistics is basically a technology for estimating the local values of properties that vary in space from sample data. Research and development in the last 15 years has shown it to be eminently suited for soil and ripe for application in soil survey and land management. The basic technique, ordinary kriging, provides unbiased estimates with minimum and known variance. Data for related variables can be incorporated to improve estimates using cokriging. By more elaborate analysis using disjunctive kriging the probabilities of deficiency and excess can be estimated to aid decision.
The variogram is crucial in all geostatistics; it must be estimated reliably from sufficient data at a sensible scale and modelled properly. Once obtained it can be used not only in the estimation itself but also to choose additional sampling sites, improve a monitoring network or design an optimal sampling scheme for a survey. It may also be used to control a multivariate classification so that the resulting classes are not too fragmented spatially to manage.