基于决策树模型的土壤性质空间推断

许多水文、生态环境和土地管理规划的决策当中,都迫切需要空间连续的土壤性质信息来提高其建模和决策的精确性和可靠性。依靠传统土壤调查技术获得的土壤图则无法满足这种需求。本研究以浙江省龙游县研究区为例,利用第二次土壤普查的土壤性质数据,生成土壤性质-环境因子空间数据库。运用决策树建模方法将土壤性质含量与一些易于广泛观测的景观属性,包括地形、地质、土地利用和遥感影像建立联系,从而将有关土壤性质含量分布的知识转入一种清楚的、定量的、与环境因子相关联的规则系统中,并以此来预测研究区土壤性质的连续空间分布。研究结果表明,所建立的决策树模型可以解释75～81%的土壤性质空间变异。     

土壤有机质的空间变异性

本文根据田间采样分析资料，探讨了土壤有机质含量的空间变异性，在一定置信水平和估值精度条件下，确定了合理的取样数目，根据半方差分析结果，得出中等变异程度的有机质分布结构，并求出了取样点距之间的最大相关距离。最后讨论了土壤有机质含量的半方函数模型，提出了研究地区土壤有机质采样的间距应增加到２０米，以节省人力物力，提高样本分析结果的代表性。     

土壤有机质预测性制图方法研究进展

有机质是土壤重要组成部分,是衡量土壤肥力的重要指标。对土壤有机质空间制图,不仅可以了解土壤质量动态变化,而且对土壤区域变异性研究具有重要意义。对现今土壤有机质预测性制图的主要方法,包括地统计学、土壤-景观关系理论和遥感反演三种方法,分别进行了阐述,对其优缺点进行分析比较,以此明确各方法的使用特点和运用环境,并对土壤预测性空间制图中存在的问题进行分析,展望了土壤有机质预测性制图未来的研究趋势,以此为精准农业的实施和农业的可持续发展,提供基本的信息及建议。     

基于分类树方法的土壤有机质空间制图研究

以浙江省龙游县研究区为例 ,提供了一种推断和表达土壤有机质 (OM)含量空间分布信息的方法 ,通过一种数据挖掘方法———分类树建模方法将土壤OM含量与一些易于广泛观测的景观属性 ,包括地形、地质、土地利用和遥感影像建立联系 ,从而将有关土壤OM含量分布的知识转入一种清楚的、定量的、与景观因子相关联的规则系统中 ,并以此来预测研究区土壤OM水平的连续空间分布。树分析选取了高程、岩石类型、土属类型、PC4 、PC2 、土地利用类型、PC3、PC1、上坡贡献面积、坡度、坡向、平面曲率和剖面曲率来预测研究区土壤OM等级的分布。其中 ,高程、岩石类型、土属类型和反映植被覆盖度的PC4 、PC2 以及土地利用类型对于研究区土壤OM等级预测更为重要。从分析结果来看 ,依据分类树所划分出的景观类型与土壤OM含量有着较好的关联性     

基于遥感影像和决策树算法的土壤制图

传统土壤信息获取方法已经无法完全满足当前各领域对土壤数据的需求,如何结合新的技术提高土壤普查效率,获取高精度土壤图成为了现阶段的研究重点。本研究综合利用高分二号遥感影像提取的遥感光谱指数以及DEM数据提取的地形因子,通过决策树算法进行数据挖掘,获取各土壤类型的土壤—环境规则,然后利用SoLIM结合土壤—环境规则进行推理制图,获得研究区的土壤类型分布图。结果表明,预测土壤图总体精度为88%,高于传统土壤图的精度72%,且在三种不同的采样方式(均匀采样、横截面采样和主观采样)下土壤预测精度分别为89%、88%、86%,均高于传统土壤图。这说明,预测土壤图比传统土壤图更能反映土壤类型空间差异,且预测土壤图在表达土壤类型整体空间分布信息的同时也可捕捉到土壤类型与地貌类型的耦合关系。     

基于不同模型的区域尺度耕地表层土壤有机质空间分布预测

研究不同模型对土壤有机质空间预测的性能差异对制定更加科学合理的采样策略、提升采样效率和提高土壤空间预测精度有着重要的指导意义。本研究将6496个土壤样点按8∶2的比例分层随机分成训练集与验证集,应用普通克里格、随机森林以及随机森林－回归克里格三种有代表性的数字化土壤制图（Digital Soil Mapping,DSM）模型,对河南省许昌市耕地表层土壤有机质含量及空间分布进行预测,对三种模型性能表现进行综合评价。三种模型输出的预测结果显示:研究区耕地表层土壤有机质含量水平一般,均值为18.70 ~ 18.81 g kg?1,变异系数0.15 ~ 0.17,属中等强度变异;空间分布总体格局为西北与西南部分山地褐土区、东南部砂姜黑土区表层有机质含量高,中北部脱潮土、石灰性潮土区表层有机质含量低。验证结果表明:三种模型性能表现无明显差距,预测精度基本一致,输出结果对研究区耕地表层土壤有机质变异解释百分比在33% ~ 34%之间,在相同和相近尺度土壤有机质空间预测案例研究里属中等水平。在协变量有限且样点分布较为均匀的情况下,普通克里格模型便于快速获得研究区目标变量的空间分布;如果协变量比较丰富且易于收集利用,或是进行空间预测的同时还需要甄别不同因素对目标变量的影响大小,则建议采用随机森林模型;协变量有限,但样点密度较大时,随机森林－回归克里格模型可能是对目标变量进行空间预测的不错选择。     

土壤有机质的生物学稳定性及其转化模型

土壤有机质是土壤中最重要的一种物质，本文首先对土壤有机质各组分的生物学稳定性，以及非腐殖物质在转化为腐殖质的过程中进入腐殖质各组分的比例进行概述；然后分四种类型总结了土壤有机质的转化模型，提出各类模型的优缺点及建模的发展趋势。     

土壤有机质周转计算机模拟原理

土壤有机质变化与土壤肥力和大气CO2 变化有密切的关系 ,建立土壤有机质周转计算机模型并进行模拟是预测土壤有机质长期变化的重要手段。本文简要回顾了土壤有机质周转计算机模拟模型的发展和现有主要模型 (如RothC、CENTURY等 )的特点和应用范围 ,着重讨论了建模基本原理 ,包括模型结构和过程、主要因子选择、模拟和预测能力的检验和评估方法等 ,并分析了目前所存在的问题。     

基于神经网络优化算法的云南土壤有机质含量数字制图——以景洪市为例

  目的  利用自然成土作用变量,预测并制作栅格化的土壤有机质分布图,对发展热带数字化精细农业具有重要意义。  方法  使用2006年云南省景洪市测土配方样点数据,应用BP神经网络（BPNN）、基于强分类器算法的BP神经网络模型（BPNN-Ada）、基于粒子群算法优化的BP神经网络（PSO-BPNN）、基于遗传算法优化的BP神经网络（GA-BPNN）和多元线性回归（MLR）对土壤有机质的含量预测并进行空间化制图。  结果  ① 土壤样点X、Y坐标值能够有效提高算法精度且充分表现环境因子与土壤有机质空间分布上的协同关系。② 4种神经网络算法预测结果土壤有机质空间分布基本类似,均呈现南高北低的趋势。③ 研究区域内4种神经网络模型的在建模集拟合程度从高至低依此次为:BPNN-Ada > GA-BPNN > PSO-BPNN > BPNN,在建模集中PSO-BPNN和GA-BPNN与BPNN拟合精度一致,BPNN-Ada的拟合精度R2最高为0.98。在验证集的预测能力由高至低依次为:BPNN-Ada > GA-BPNN > PSO-BPNN > BPNN。BPNN-Ada有着最高的预测精度和算法稳定性:RMSE = 4.47、MAE = 3.3、MRE = 0.05、R2 = 0.976。  结论  在景洪地区进行土壤有机质神经网络建模时加入地理坐标能够有效提高模型精度,且基于学习规则的神经网络优化算法效果要优于优化初始权重和阈值的神经网络算法及传统的BPNN算法。     

基于神经网络与决策树的土壤粗糙度测量

利用野外拍摄照片判读土壤粗糙度是一种简单快速的测算方法。为了克服人工判读处理效率低、结果易受人为因素影响和目前的计算机自动读取结果易受野外杂草影响、自动化程度有待提高的缺点,提出了一种基于神经网络和决策树的土壤粗糙度测量方法。该方法在建立神经网络时充分考虑土壤边界线、白板边缘及参考方框的特征选取输入特征向量,消除噪声影响,实现图像边缘检测;并在此基础上建立决策树区分白板、土壤、参考方框等信息,完成土壤边界线和比例尺的获取。野外试验中不可避免有杂草和光照的影响,考虑到阴影、土壤和杂草色彩的差异性,决策树判决准则的选择不仅包含了纹理信息而且考虑了色彩信息。试验表明,基于神经网络与决策树的土壤粗糙度测量采用带有简单方框的参考白板能快速高效地从复杂的野外照片中获取土壤粗糙度信息,降低了拍摄要求。自动提取结果与人工判读相比,所得均方根高度的误差在5%以内,相关长度的计算误差在1%以内,具有较高的精度和可靠性。该研究方法为土壤粗糙度实时在线测算提供了有效的解决方案。     

运用岩石热解产生的S3和S4信号进行土壤有机质特征研究

Variations in the abundance of soil organic matter（SOM） constituents with different stability have a major impact on important environmental processes, e.g., carbon dioxide（CO2） fluxes between the soil and the atmosphere. Recently, besides the bulk Rock-Eval（RE） data, the mathematical deconvolution of the signals derived from hydrocarbon-like compounds released by thermal cracking of SOM during RE pyrolysis has been increasingly used to estimate the relative contribution of the major SOM classes differing in origin and preservation. This study applied the mathematical deconvolution of the S3 and S4 signals of carbon monoxide（CO） and CO2, produced both by the pyrolysis of the oxygen-containing moieties and by the oxidation of the residual highly resistant organic matter, to characterize the stability of these components. Our results suggested that the stability of the oxygen-containing moieties was controlled by the precursor material and was strongly affected by the land use and the presence of humic substances in the surface horizon of some main soil types in Hungary. In consistence with the bulk RE data, results of the mathematical deconvolution also proved to be diagnostic markers for discriminating the aquatic or terrigenous plants as the main sources of SOM. The mathematical deconvolution of S4 signals derived from the highly resistant SOM fraction allowed us to quantify the contribution of constituents with different stability. Furthermore, the results of this study displayed that the stability of this highly abundant SOM fraction in the surface soil samples depended on source biomass and intensity of leaching.     

Soil organic matter dynamics under grain farming in Northern Kazakhstan

Because of their ability to store a high amount of soil organic matter (SOM), Chernozem soils are one of the most important resources from both agricultural and environmental viewpoints. This study was carried out to determine the SOM budget under grain farming in the Chernozem soil of northern Kazakhstan through the analysis of in situ soil respiration and soil environmental factors such as soil temperature as well as moisture content. Five experimental plots including one fallow field were established at the experimental farm of Barayev Kazakh Research and Production Center of Grain Farming, Shortandy, northern Kazakhstan (mean annual precipitation and average year temperature are 323 mm and 1.6°C, respectively). Mean daily soil temperature increased to above O°C in early April, remaining at above 20°C from mid-June to mid-August, and then sharply decreased to below 5°C at the end of September. Most of the biological activities were considered to be limited from April to September. On the other hand, the soil moisture content remained high after thawing until mid-June and then continuously decreased in the cropped plots except during the rainfall events. The soil respiration rate recorded the highest values from late June to early July and overall fluctuations were similar to those of the soil temperature, unlike the fluctuations of soil microbial C and N contents, which exhibited similar patterns to those of the soil moisture content. In order to represent the daily soil respiration rates using the soil environmental factors, the following relationship was introduced as a model function: Cem = aM ^pbexp(-E/RT). The coefficients, a, b, and E (activation energy in Arrhenius equation), were determined by stepwise multiple regression after logarithm transformation using the measured data, Cem (daily soil respiration rate), M (volumetric soil moisture content), and T (absolute soil temperature). As a result, a significant relationship was always obtained between the soil respiration rate and the activation energy, E, while the contribution of the soil moisture content to the soil respiration rate was uncertain. Using the regression equations and monitored data of soil temperature and moisture content, cumulative soil respiration throughout the cropping period was calculated to be in the range of 2.5 to 3.2 Mg C ha^p-1 On the other hand, the amounts of crop residues in the cropped plots that were expected to be incorporated into the soils ranged from 1.6 to 4.4 Mg C ha^p-1 Except for the plot planted with oats (higher amounts of residues than for wheat), the SOIL budget was slightly negative in this year, that is, the soils lost their organic matter stock. Although it is difficult to generalize the C budget in different years because of the large variations in crop growth due to fluctuating water resources, the disadvantage of summer fallow (no residues) was obvious in terms of SOM budget. The net soil respiration rate in the fallow plot, 2.9 Mg C ha^p-1 was approximately equivalent to 4% of the total SOM stock in the plow layer (30 cm) (70 to 80 Mg C ha^p-1 To reduce further loss of SOM, at least evenly extensive use of summer fallow should be reconsidered.     

基于Boosting的决策树集成土地评价

传统的土地评价方法易受人为因素的限制,探索更科学合理的土地资源评价方法,对土地利用与规划具有重要意义.由于决策树具有分类精度高、分类器可解释性强的优点,特别是C5.0采用了提高决策树分类精度的Boosting技术,提出利用Boosting技术的决策树集成C5.0进行土地评价的方法.采用C5.0算法对广东省七地资源进行了评价,对不使用Boosting的决策树和使用Boosting决策树集成的评价结果进行了分析和比较.研究结果表明利用决策树进行土地质量评价能够得到较高的评价精度,且Boosting决策树集成的土地评价精度高于不使用Boosting的决策树的精度.     

聚类支持下决策树模型在耕地评价中的应用

为了挑选合理的学习样本,增强决策树模型在耕地评价应用的有效性,运用聚类方法挑选学习样本,用改进C5．0决策树算法建立耕地评价模型,提出一种新的评价思路。运用此方法以广东省龙川县耕地为研究对象,以试验法挑选出6种聚类结果的学习样本,确定4000个样本作为最终的学习样本;利用决策代价权重来改进决策树评价模型,最终建立的评价模型的预测精度达到94．92％,满足了实际情况的需要。试验结果表明综合运用聚类和决策树模型进行耕地评价是可行的,其建立的评价模型具有精度高、鲁棒性和易理解性等特点。     

土壤有机质对疏水性有机污染物的非线性吸附及其影响因素

土壤有机质(SOM)是土壤的重要组成部分,它对疏水性有机污染物(HOCs)的吸附作用对污染物在环境中的迁移转化有着至关重要的影响。近年来的一些研究表明,HOCs在SOM上的吸附行为并不严格地符合传统的线性分配模型。本文综述了近年来一些学者提出的SOM对HOCs的非线性吸附作用的机制及其影响因素,认为SOM的组成和结构均对HOCs的吸附有着重要的影响。SOM吸附在土壤矿物表面后,其结构或各组分的比例发生了改变,从而对其本身的吸附作用产生影响。溶液的化学环境(主要是阳离子)会对SOM的结构和吸附作用产生影响,特别是当溶液中存在一些过渡金属(如,铜或银)的离子时,共吸附的金属离子可能会与多环芳烃类分子形成阳离子-π键的作用,从而促进多环芳烃类物质在SOM上的吸附。     

基于改进决策树模型的矿区土地复垦适宜性评价

针对矿区传统的土地复垦适宜性评价方法,运用模糊数学的隶属函数法,对决策树的1D3算法加以改进,建立基于改进决策树的矿区土地复垦适宜性评价方法.以平禹二矿为研究区,将改进的评价模型运用到塌陷区土地复垦适宜性评价中,选取7个评价指标作为测试属性,提取100个已确定复垦方向的评价单元作为样本,运用改进的ID3算法,构建研究区土地复垦适宜性评价决策树,然后运用生成的决策树分类规则,对各评价单元按其属性进行分类,最终确定各评价单元的复垦方向,并将分类结果与模糊综合评价方法得到的评价结果进行对比分析.结果表明:基于改进决策树模型的矿区土地复垦适宜性评价方法,不依赖于经验知识,依靠样本的自主学习,制定分类规则,得到的分类结果较为客观合理,而且决策树分类规则确定后,可以直接应用于同一地区的其他待评价区,且分类过程简单明了,便于操作,避免传统评价方法中较为繁琐和复杂的计算过程,提高矿区土地复垦适宜性评价的效率,为矿区土地复垦适宜性评价工作提供一种新的思路.     

基于决策树和图层叠置的精准农业产量图分析方法

作物产量限制因子的提取是精准农业变量施肥的重要环节之一。决策树和GIS图层叠置方法研究结果表明,土壤有机质始终占据决策树全部知识规则的第一次分支,起着重要的决定性作用,其次为速效磷和碱解氮;产量空间分布与有机质、碱解氮、速效磷高度吻合,平均吻合度分别达到91%、83%和49%。土壤有机质、速效磷和碱解氮含量是宁夏暖泉农场小麦产量主要限制因子,提高土壤有机质含量是提高单产的重要措施。结论得到生产实际验证,说明运用决策树和GIS图层叠置分析方法挖掘产量限制因子在技术上可行。     

Soil organic matter maps by nuclear magnetic resonance

Preliminary studies are reported on a soil and a litter fraction which have the aim of exploring the potential of two-dimensional ¹³C nuclear magnetic resonance spectroscopy for the analysis of soil organic matter structure. It has been shown that contour plots can be obtained in which one co-ordinate gives chemical shift information and the other the magnitude of the dipolar interaction. For aromatic and oxygenated aliphatic carbon, the latter parameter is a measure of the degree of protonation. Hence the method is useful for distinguishing between soils containing highly substituted and condensed aromatic structures and those with protonated aromatic carbon. Likewise, it is possible to distinguish between soils containing dioxygenated carbon which is. protonated (e.g. ketal).     

利用烧失量方法精确测定土壤有机质

Wet oxidation procedure,i.e.,Walkley-Black (WB) method,is a routine,relatively accurate,and popular method for the determination of soil organic matter (SOM) but it is time-consuming,costly and also has a high potential to cause environmental pollution because of disposal of chromium and strong acids used in this analysis.Therefore,loss-on-ignition (LOI) procedure,a simple and cheap method for SOM estimation,which also avoids chromic acid wastes,deserves more attention.The aims of this research were to study the statistical relationships between SOM determined with the LOI (SOMLOI) and WB (SOMWB) methods to compare the spatial variability of SOM in two major plains,Shahrekord and Koohrang plains,of Chaharmahal-va-Bakhtiari Province,Iran.Fifty surface soil samples (0-25 cm) were randomly collected in each plain to determine SOM using the WB method and the LOI procedure at 300,360,400,500 and 550 ℃ for 2 h.The samples covered wide ranges of soil texture and calcium carbonate equivalent (CCE).The general linear form of the regression equation was calculated to estimate SOM LOI from SOM obtained by the WB method for both overall samples and individual plains.Forty soil samples were also randomly selected to compare the SOM and CCE before and after ignition at each temperature.Overall accuracy of the continuous maps generated for the LOI and WB methods was considered to determine the accordance of two procedures.Results showed a significant positive linear relationship between SOM LOI and SOM WB.Coefficients of determination (R2) of the equations for individual plains were higher than that of the overall equation.Coefficients of determination and line slopes decreased and root mean square error (RMSE) increased with increasing ignition temperature,which may be due to the mineral structural water loss and destruction of carbonates at higher temperatures.A temperature around 360 ℃ was identified as optimum as it burnt most organic carbon,destroyed less inorganic carbon,caused less clay structural water loss,and used less electrical energy.Although the trends of SOM in the kriged maps by the two procedures accorded well,low overall accuracy was observed for the maps obtained by the two methods.While not suitable for determination where high accuracy is required,determination of organic carbon through LOI is likely suitable for exploratory soil surveys where rough estimation of organic matter is required.