【作者】 秦建成；

【导师】 魏朝富； 谢德体；

【作者基本信息】 西南大学 ， 农业资源利用， 2007， 博士

【摘要】 土壤适宜性评价是针对具体作物在特定地域种植的土壤适宜度所做出的定性、定量和定位的结论性评价，具有实践性、经验性、客观性及应用性的特点。随着现代数学理论的发展和计算机技术的不断更新，土壤适宜性评价方法逐渐从简单的单准则、单目标评价方法向多准则、多目标综合评价方法的发展和应用。在评价理论和方法上，目前具体土壤适宜性评价方法的选择通常是评价人员根据需要和研究目的，依据自身以往所积累的经验及个人偏好等条件，在众多方法中选取一种或多种方法进行评价，这在一定程度上忽略了各种评价方法的特性，造成了对具体评价方法的不适当运用，从而严重影响了评价结果的可靠性和准确性。特别是地理环境相对比较复杂的山地丘陵地区，土壤属性空间变异强度大的条件下，如何因地制宜地选取评价方法进行土壤适宜性评价才能最大限度地确保评价精度及科学指导实践等问题是摆在我们面前亟需解决的问题。同时，在评价尺度转换上，多数研究者都是把研究区土壤看作是一个均质单元，应用样本单元的评价等级结果代表整个区域的评价结果，实现评价由变量空间向评语空间的转换，但这种尺度转换缺乏一定的理论基础，存在主观性，背离了土壤是时空变异连续体的实际，很大程度造成了研究结果与土壤实际的差异。因此，本论文在采集、分析重庆市彭水县植烟区235个土壤样品理化性质的基础上，基于GIS技术及多种数学模型，采用检验指数和、地统计学、模糊数学理论及人工神经网络等方法对复杂地理环境条件下的土壤适宜性评价进行对比研究，并通过烟叶等级指标对评价结果进行精度检验，进一步完善了复杂地理环境区域土壤适宜评价的理论、方法；在此基础上，结合影响优质烤烟种植的气象要素及烟农种烟投资行为等因素综合评价彭水县烤烟种植适宜性并进行了种植区划，得到了以下结论：1基于经验指数和法的土壤适宜性评价基于GIS技术，采用频度统计、系统分析等方法选取土壤适宜性评价指标，并应用层次分析法(AHP)与特尔菲法对指标权重进行赋值并计算土壤样点经验指数；依据土壤样点经验指数分别采用行政村区划法和克立格插值法进行土壤适宜性评价并进行精度检验。(1)不同处理方式评价结果差异较大，且在面尺度(类型尺度和区域尺度)的研究中，两种处理方式所得的评价结果均随研究尺度的增大，适宜等级(中度适宜与高度适宜之和)和不适宜等级(勉强适宜与不适宜之和)比例差异性增强。行政村区划法在类型尺度(旱地)及区域尺度上适宜区与不适宜区比值分别为1.24：1和2：1；而克立格插值法在两种尺度上比值的为2：1和3：1；(2)基于经验指数和法的行政村区划法和克立格插值处理两种方式所得评价结果检验符合率分别为20.37％和46.30％，均未超过一半，但克立格插值法处理所得评价结果要相对较为准确。2基于地统计学方法的土壤适宜性评价在认真分析土壤属性空间分布特征的基础上，充分考虑地统计学在面尺度、山地条件下模拟评价的精度，提出并构建了不同类型土壤属性空间预测模拟模型，将传统统计学和地统计学相结合，在考虑各向异性和趋势效应的情况下，研究土壤属性的空间变异特征并进行不同尺度土壤的适宜性评价及精度检验。(1)土壤适宜性评价指标的样本数据受不同因素的影响而呈现不同的分布状况、趋势效应；土壤适宜性评价指标近似为几何异向性；球状模型、指数模型及高斯模型分别为土壤适宜性评价指标的半方差函数的最佳拟合模型。原始数据样本由于受地形、样点数量及实验分析误差的影响，多数呈非正态分布，指标数据以一阶、二阶趋势效应为主；不同土壤类型样本受样本数量及模拟方程因素的影响，各项评价指标多呈现正态分布，且趋势效应不显著；而混和数据样本指标同样受样本数量及模拟方程因素的影响多为正态分布，指标数据多以二阶趋势效应为主。(2)在不同研究尺度下，不同数据处理方式所得到的土壤适宜性评价结果存在较显著差异。原始数据处理在样本尺度、类型尺度及区域尺度下的适宜区与不适宜区比例分别为4.34：1、2.86：1和2.44：1；不同土壤类型处理在三种研究尺度比例分别为6.70：1、7.22：1和4.97：1；而混和数据处理在三种研究尺度比例分别为6.70：1、6.62：1和5.40：1。(3)将地理环境因子与土壤类型考虑到区域土壤属性的空间分布，利用原始数据通过评价指标的空间预测模拟模型来增加样本点的处理方式可以有效地提高评价精度，使评价结果更加接近现实，且不同土壤类型数据处理方式从理论方面要相对更加科学，评价结果更加精确，更符合实际。3基于模糊数学方法的土壤适宜性评价在Matlab7.0环境下，基于GIS技术，分别应用模糊ISODATA聚类评价和模糊综合评价方法进行不同尺度下土壤适宜性评价研究及精度检验。(1)模糊综合评价结果的适宜等级比例明显高于ISODATA聚类评价结果，这可能与模糊综合评价在因子加权时，过分突出较大值指标的影响相关。(2)不同研究尺度，土壤适宜范围差异较显著。土壤适宜性评价的适宜类与不适宜类的比值在样本尺度分别为1.78：1和1.16：1，类型尺度分别为1.86：1和1.10：1，而区域尺度分别为1.66：1和0.67：1。(3)模糊ISODATA聚类评价在面尺度的模拟精度高于模糊综合评价的精度。4基于人工神经网络(ANN)的土壤适宜性评价在Matlab7.0环境下，应用人工神经网络及GIS技术进行不同尺度下土壤适宜性评价及精度检验。(1)通过RBF网络和BP网络的学习测试，按不同研究尺度将土壤适宜性划分成4个等级，表明了人工神经网络模型应用于土壤适宜性分区的可行性。(2)不同研究尺度，RBF网络和BP网络适宜性评价的土壤适宜范围差异较显著。土壤适宜性评价的适宜等级与不适宜等级比值在样本尺度分别为3.35：1和2.35：1，类型尺度分别为1.01：1和2.09：1，而区域尺度分别为1.10：1和1.74：1。(3) RBF网络在面尺度的模拟精度高于BP网络的精度。5土壤适宜性评价方法比较(1)各种评价方法的理论基础不同，其评价过程存在较大差异，都有各自的优缺点。经验指数和法、模糊聚类及神经网络方法主要是针对样本在较小尺度上进行评价并划分等级，然后根据评价单元进行尺度的扩展，实现评价由变量空间向评语空间的转换，在对点尺度的评价上具有较高的精度，但在由点到面的尺度转换过程中缺乏一定的理论基础，存在主观性；地统计学法主要是针对连续变量在区域尺度上进行空间模拟和区划评价，通过变量图层叠加实现研究目标的综合评价，在区域评价过程中具备一定的数学理论基础，整个评价过程多通过计算机上实现，降低了人为误差，评价结果非常直观。(2)通过利用双变量简单相关分析和偏相关分析比较各种评价方法所得到的评价结果与评价指标数据的相关性认为：双变量简单相关分析与偏相关分析的结果有较大的相似性。海拔、pH及速效钾等指标都与不同适宜性评价方法所得评价结果呈现显著性相关，说明这些指标数据强烈影响评价结果的变化，是影响烟叶生长的主导因子；OM、碱解氮、有效磷、水溶性氯、坡度及物理性粘粒等指标在不同适宜性评价方法所得评价结果中相关性强度有所变化，其中以土壤养分元素的有效性变幅较大，其他土壤环境要素的变幅相对较小，说明这些指标是依据所选择的评价方法的不同，指标数据处理方式不同，其对评价结果的影响也有所不同；CEC指标在不同适宜性评价结果中均没有显著的相关关系，这说明CEC指标数值对评价结果的影响不大，如果评价体系中缺乏CEC指标，所得的评价结果不会产生大的变化。(3)地统计学方法适宜用于复杂地理环境条件下的土壤适宜性评价。通过比较不同方法适宜性评价精度检验表明：采用地统计学方法所得的评价结果与实际烟叶品质之间的对应关系的比例最高，符合率达到74.07％，RBF—ANN的符合率为61.11％，模糊ISODATA聚类法的符合率为57.40％，而经验指数和法的符合率为46.30％。如果单从符合率的角度，研究认为地统计学方法的适宜性评价结果与实际烟叶品质之间符合率最高，精度最好，该方法适宜用于复杂地理环境下的土壤适宜性评价。6彭水县烟叶种植适宜性综合评价及区划基于GIS技术，应用地统计学与数学模型相结合的方法，在借鉴国内外优质烟叶生产的自然基础环境条件上，充分考虑地统计学在区域尺度、山地条件下模拟评价的精度，依据彭水县30a(1975-2005)及近4a(2003-2006)气象要素资料，构建彭水县部分气候要素的空间分布预测模拟模型，并进行了单指标适宜性分区；通过参与性农户评估法对彭水县烟农种烟投资行为机理及主要影响因素进行了分析和评价；根据彭水县各气象观测站的分布状况及实际种植烟草的状况，按照自然地理条件及行政区划将彭水县植烟土壤划分为沿江河谷区、北部中山丘陵区及东南丘陵区三个区域，并应用综合指数评价法分别计算三个区域的适应性指数及进行适应性评价。(1)彭水县土壤基础环境信息的空间分布存在差异性，气候和地形地貌条件都具备大规模种植烟草的条件，自然地理条件整体上能够满足优质烟叶生长的需要。(2)烟农兼业经济行为、烟农土地经营规模以及烟农家庭劳动力状况是影响农户烟叶种植投入行为的主要因素。兼业化程度越高，烟农非农收入比例越高，烟农对烟叶种植进行生产性投资的可能性和规模就越小；土地细碎化经营会导致投资收益率降低，从而使烟农缺乏土地追加投资的意愿；而烟农劳动力状况常常成为烟叶生产顺利实施的制约因素。(3)按自然地理条件及行政区划将彭水县植烟区划分为沿江河谷区、北部中山丘陵区及东南丘陵区。沿江河谷区植烟土壤的适宜指数为0.634，属于勉强适宜级，不提倡种植烟叶；北部中山丘陵区植烟土壤的适宜指数为0.864，属于高度适宜级，适宜种植高产、优质烟草；东南丘陵区植烟土壤的适宜指数为0.723，属于中度适宜级，适宜烟草规模化种植。通过对复杂地理环境条件下土壤属性的空间变异及不同尺度下土壤适宜性评价方法的比较，提出的基于分布函数连续性数学理论的不同类型土壤属性空间模拟估算及尺度效应定量评价方法，有效解决了样本稀疏地区土壤环境因子在时空分布上的不均匀性和变异性所带来的复杂问题，一定程度上避免了将区域土壤作为均质体进行研究的状况，而基于GIS的区域尺度土壤评价方法能有效解决复杂环境条件下土地评价方法的选取，为复杂地理环境条件下土壤适宜性评价精度改善提供一种可以选择的方法。但土壤是一时空连续的变异体，其空间变异及其尺度转换研究仍需要多门类、多学科的专家、学者协同。更多还原

【Abstract】 Soil suitability evaluation aims to optimally allocate crop planting through measuring the coupling of designating crop to given land, and considers simultaneously soil physical features and current/future land use patterns. Nowadays, with the combination of mathematical theory and computer technique, multi-objective method based assessment for soil suitability has been increasingly observed. However, recognizing the excessive complexity of the interactions among factors influencing soil quality, some papers often confine these processes of soil quality to either a single process, or a single discipline. Certainly, assessment methods also are only limited to one or several in terms of the study objective, even individual experience and preference. Under such circumstances, the characteristics of different assessment methods can’t be presented; furthermore, the acourate and reliable assessment results for soil suitability can’t be obtained. In hilly and mountainous areas, complex topography features reduce the greater variability of soil properties at spatial scale. In these areas, currently, the greatest challenge has been what assessment methods, at some degree, can improve estimate accuracy of soil suitability thus being implemented to practices? Moreover, some researchers confine given soil to homogeneous units, and often the results of soil suitability from sampling units to the whole area. Seemingly, the results of these studies carry out spatial scaling, but fail to present soil being a continuous body at spatial-temporal scale, due to be short of theory and application. Undoubtedly, the pronounced differences between research results and soil practical value at some certain can occur.In this study, a GIS-based decision support system was built for assessing soil suitability in tobacco-planted areas, Pengshui of Chongqing, China. This was done by integrating an exponential test, geiostatistics, fuzzy mathematics and artificial neural network into geographical information system software to assess soil suitability. In this system, complex problems are solved within program: this paper, firstly, compared the results and their accuracy obtained by different methods, and then planned land use patterns in tobacco-planted areas through combining climate factors and farmers’ household behaviors. These results indicated:1. Assessment of soil suitability based on exponential testGIS-based assessment indicators of soil suitability were selected through frequent statistic and system analysis methods. Designing weight for every soil quality indicators and evaluating experiential index of every soil sampling were obtained by integrating AHP and expert decision support system. And assessment of soil suitability was carried out applying subareas of an administrative village and Kriging spatial interpolation, in terms of experiential index of every soil sampling, and tested by coupling classification index of tobacco sampling and these results from different methods. The difference between different treatments was significant. At slope scale, assessment results from dryland and regional treatments increased following research scale upscale. The variation proportions of suitable areas (e.g., moderate/highly suitability) to limited areas (marginal/non suitability) were enhanced. Based on subareas of an administrative village, at dryland scale, the proportions of suitable areas to limited areas was 1.24:1, while at regional scale, this number was 2:1. Applying Kriging spatial interpolation, they were 2:1 and 3:1, at dryland or regional scale, respectively. The coincidence rate of assessment results from subareas of an administrative village and Kriging spatial interpolation was 20.37% and 46.30%, respectively. They both were excessive 50%. But, the assessment results from Kriging spatial interpolation were more accurate.2. Assessment of soil suitability based on geostatisticsSpatial model of soil properties was built to predict the regional suitability of different soil types, analyzing the characteristics of spatial distributions of soil properties in tobacco-planted areas, and the accuracy of geostatistics method used in regional scale and mountainous conditions. The characteristics of spatial variation of soil properties were measured by integrating classical statistics and geostatistics, and considering anisotropic and trend effect. Thus soil suitability in tobacco-planted areas was assessed, and accuracy test was presented using the same methods described in first context. Sampling data of assessment indicators of soil suitability, associating with effects of factors, presented different distributions and trend effects. Assessment indicators of soil suitability reached geometric anisotropy. Spherical model, exponential model and Gaussian model were the best fitting model of semivariance function of soil suitable indicators. Most of raw data were non-normal distribution, due to the effects of terrain, soil sample size and analytical error. Indices data were trend effects of the first/second order. Every indicator was normal distribution, due to soil sample size and simulation equation, and trend effects were not pronounced. Similarly, indices of mixture samples were normal distribution, and presented trend effects of the second order as well. At different scale, significant differences between different data treatments occurred. The proportions of suitable areas to limited areas were 4.34:1, 2.86:1 and 2.44:1, when the raw data treated at sampling, dryland and regional scale, respectively. Under different soil types, the proportions of corresponding scale were 6.70:1, 7.22:1 and 4.97:1, respectively. However, mixture data treatments resulted in the proportions accounting to 6.70:1, 6.62:1 and 5.40:1, under above-mentioned corresponding scale, respectively. Increasing samplings of different soil types can effectively improve accuracy by adopting spatial simulation. This method integrated geographical environmental factors and spatial distributions of regional soil properties, and can obtained more closed results to the reality.3. Assessment of soil suitability based on fuzzy mathematicsUnder Matlab7.0, fuzzy ISODATA algorithm and comprehensive evaluation were adopted to assess soil suitability at different scales, and the same method of accuracy test was used. The proportions of suitable areas, including moderate/highly suitability, obtained by fuzzy comprehensive evaluation, were obviously greater than the results of clustered by fuzzy ISODATA. The reasons for these results were that fuzzy comprehensive evaluation emphasized excessively the effects of higher value indicators, when multifactor summation with different weight done. Significant differences between ranges of soil suitability can observed at different scales. The proportions of suitable areas to limited areas were (1.78:1 and 1.16:1), (1.86:1 and 1.10:1) and (1.66:1 and 0.67:1), at sampling, dryland and regional scale, respectively. The accuracy of fuzzy ISODATA algorithm was higher at regional scale than that of fuzzy comprehensive evaluation.4. Assessment of soil suitability based on artificial neural networkArtificial neural network and GIS were used to assess soil suitability in tobacco-planted areas, associating with Matlab7.0. In this part, the method of accuracy test was similar to that described in above context. Soil suitability of tobacco-planted areas was divided into four classifications, using the learning algorithm of RBF and BP neural network. This result indicated that artificial neural network is possible to assess soil suitability. Obviously differences between these results from RBF and BP soil suitability assessment can be observed at different scales. The proportions of suitable areas to limited areas were (3.35:1 and 2.35:1), (1.01:1 and 2.09:1) and (1.10:1 and 1.74:1), at sampling, dryland and regional scale, respectively. The accuracy of RBF neural network was greater than that of BP neural network, compared the results of soil suitability assessment.5. GIS-based comparison of soil suitability assessmentThe theory foundation of different evaluation methods was various, and the evaluation processes were clearly different. Each one had its own advantages and disadvantages. Exponential test method, fuzzy clustering and neural network were mainly for the evaluation and gradation of soil sampling at small scales. Then did upscale according to evaluation units, furthermore soil suitability assessment transformed from geological variable space to comment space. But, this method had the higher accuracy at the point scale and was lack of theory foundation, when transformed from point to space. The geostatistics spatially simulated and evaluated the continuous variables at regional scale, overlaying the variable layers to get the comprehensive research objectives. There was some mathematics knowledge, and the whole evaluation processes were executed through computers. So the manmade errors were reduced and the evaluation result was very objective. (2) Comparing the evaluation results obtained by above-mentioned methods and indicators through double variables and partial correlation analysis, showed: the simple correlation analysis of the double variables and partial correlation analysis were similar. Significant positive relation among elevation, pH, available K, etc. grained by different methods occurred. Thus those indicators strongly affected the evaluation results, and also were the major factors that affect the tobacco growth. The relations of organic matter, available N and P, water-soluble Chloride, slope and physical clay obtained by different methods were different. Availability of soil nutrients measured by different methods was clearly different, while the other indicators were lower different. Certainly, different methods applied, for the same indictor, could present different results. CEC was not obviously relation among different methods thus CEC possessing lowly effects on the result. (3) Geostatistics was suitable for the soil suitability evaluation in study area. Comparing the accuracy of different evaluation methods, and showed: the evaluation results through geostatistics and the actual quality of tobacco had a higher corresponding ratio accounting to 74.07%. Whilst, the corresponding ratio of RBF-ANN, fuzzy ISODATA and exponential test were 61.11%, 57.40% and 46.30%, respectively. The research believed that the corresponding ratio of geostatistics was the highest and the most accuracy from the point of corresponding views. This method was suitable for the soil suitability evaluation.6. Comprehensive suitability evaluation of tobacco-planted areas in Pengshui CountyIntegrating GIS, Geostatistics and mathematics, spatial distribution model was built. The geostatistics accuracy at regional scale and mountain simulation evaluation, and the meteorological data (1975-2005) of Pengshui County were applied to do the single factor suitability regionalization. The major factors and behaviors of tobacco farmer investment through participant rural appraisal were analyzed. According to meteorological observation distributions, factual tobacco planting situations, physical geography conditions and district regionalization of Pengshui County, this paper divided tobacco-planted soils into the region by along the river valley zone, Northern middle mountainous and hilly zone, Southeastern hilly zone. Applying comprehensive indices calculated the suitability of every subarea. (1) Soil spatial variability of Pengshui County was pronounced significant. Climatic and topographic conditions were suitable to tobacco growth. Physical geographical conditions could reach the tobacco growth demands. (2) Tobacco fanner part-farm behavior, land size and labors were the main factors that affected the tobacco planting investment. The higher tobacco farmer part-farm percentage and the tobacco non-agricultural income ratio were, and the smaller the possibility of tobacco farmer investment behavior and planting size. Land fragment leaded to investment benefit decreasing and make the farmers unwilling to addition input. Labor conditions of tobacco farmer often became a restrict factor for the tobacco planting. (3) The suitable indices of tobacco-planted soils along river valley zone were 0.634 thus belonging to the lowest suitable gradation, and don’t fit planting tobacco. In Northern middle mountainous and hilly zone, where was suitable to plant higher productive and quality tobacco, with the more suitable gradation 0.864. Moderate gradation located in Southeastern hilly zone with 0.723 was good to large-scale planting.This paper compared spatial variability of soil properties under different geographical conditions and methods of soil suitability assessment at different scales. Based on continuous distribution function and mathematics, spatial simulation model and scaling assessment methods of different soil properties. This research solved effectively these complex issues, which produced by spatial non-uniform and variability of soil properties due to sparse sampling, and at some certain, declined the effects of uniform soil viewpoint on study results. Moreover, GIS-based regional soil evaluation methods favored to soil quality measuring under complex environments, and improved evaluation accuracy. However, soil is a continuous body at spatial-temporal scale, and its spatial variability and scaling have been paid more attention to by multidisciplinary.更多还原