【作者】 於家；

【导师】 吴健平； 陈芸；

【作者基本信息】 华东师范大学 ， 地图学与地理信息系统， 2010， 博士

【摘要】 土地是人类赖以生存的基本条件和物质基础。随着人口的增长和经济社会的发展,对土地的需求也在不断增加。在有限的土地资源条件下,如何合理配置人类生产、生活所需用地,保证土地资源的可持续利用,协调人地之间的矛盾,是摆在我们面前的重大课题。土地利用适宜性评价根据特定的用地类型,以土地合理利用为目标,对土地属性进行鉴定,并阐述土地适宜性程度。土地利用适宜性评价是土地规划与决策的重要依据,对土地利用方式的可持续发展与永续利用具有非常重要的意义。但是传统的土地利用适宜性评价工作中,仍然存在评价结果主观性较强,评价过程效率不高,难于做长期的潜在适宜性评价等问题。面对这样的情况,一些人工智能的方法被应用到该领域中了。但是当前使用的人工智能方法相对单一,很多优势性很强的新兴人工智能方法还未得到应有的利用,有必要我们做进一步深入的研究和探讨。本文以常用的层次分析法为起点,指出层次分析法中存在的缺点,并使用敏感性分析来考察其中的不确定性。研究中,还尝试使用新的适宜性规则分类方法来替代层次分析法,从而减少主观指定的因子权重对评价结果的影响。为了实现潜在适宜性评价,本文又尝试用地理模拟系统来揭示某种开发模式下土地利用适宜性的转换规律,为可持续性的土地规划提供更好的依据。本文的主要工作和研究成果包括：一、提出了土地利用适宜性模拟的概念。文中将土地利用适宜性模拟定义为运用地理模拟系统来实现土地利用潜在适宜性评价的方法。利用地理模拟系统能够模拟复杂系统的特点,来支持潜在的土地利用适宜性评价,揭示在特定土地利用方式下适宜性分布形态中的隐含内容,挖掘土地利用适宜性中潜在的规律。二、用元胞自动机机理模拟潜在土地利用适宜性。根据提出的土地利用适宜性模拟概念,设计了基于元胞自动机的适宜性模拟方法,这也是土地适宜性研究领域首次运用元胞自动机理论来实现评价工作。该工作是在三个假定：(a)土地利用适宜性领域效应(b)土地利用开发模式(c)土地适宜性限制性因子,都成立的情况下展开的。基于元胞自动机的潜在土地利用适宜性模拟在一定程度上使预测性土地评价工作更规范化和精确化,使土地利用适宜性评价工作更符合土地利用规划和决策人员的实际要求,为土地可持续利用提供更好的方法措施和技术支持。三、蚁群算法发掘土地利用适宜性分类规则。在获取土地利用适宜性分类规则的方法上,本文创新性的引入了最新的人工仿生学智能理论——蚁群算法。该方法避免了层次分析法中权重分配的主观因素,降低了评价过程中权重不确定性的干扰。本文借鉴了基于规则的分类法中对规则的定义,将适宜性规则表达为IF-THEN的条件关系的形式,同时把由样本获取的知识信息也通过该形式转换,并输入训练集,供蚁群算法发掘分类规则使用。由蚁群算法中优化路径的机制,抽象出训练数据集中发掘分类规则的数据结构,来发掘规则,进行土地利用适宜性分类,形成评价结果图。四、空间权重敏感性分析。本文的空间权重敏感性分析是运用改进的OAT (one-at-a-time)方法展开的,由此探究评价结果的稳定性、准则因子的相对权重敏感性,以及如何减低多准则决策方法的不确定性等内容。结果通过表格、图表和专题图的形式表达,能方便明确的找出敏感性高的地理区位。五、土地利用适宜性评价模型工具的开发。本文基于Microsoft C#.NET开发平台、运用ESRI ArcGIS Engine开发组件、Mathworks MATLAB嵌入式开发组件等设计开发了LSA-GIS模型工具,并给出了关键的设计流程与示例代码。设计中特别注重了以用户良好感受为中心的交互设计方法,提升用户使用的工作效率。交互设计中贯彻了UML统一建模的方法,使设计过程更规范化,为今后的模型工具的功能扩展打下基础。六、研究区灌溉农业用地实例分析。本文选取澳大利亚Macintyre Brook流域作为研究区,分别用层次分析法、土地利用适宜性分类规则发掘方法和基于元胞自动机的土地利用适宜性模拟方法做了灌溉农地适宜性分析与评价。这三种评价结果根据一定的规则分别进行空间分析对比,得出各种评价方法的可行性、合理性和存在的局限性。实验证明,LSA-GIS模型工具在研究区的评价工作中取得了良好的效果,同样可以在其他研究区的评价工作中推广使用。更多还原

【Abstract】 Lands provide basic materials for human life. The demand for land is constantly increasing with population growth and economic development. It is a major issue for us that how to rational allocate lands for human production and living with limited land resourses. It is also essential to ensure the sustainable use of land resources and harmonize the relationship between human and land. Land-use suitability assessment (LSA) aims to rational use lands according to specific type of lands, identifies land properties and describes land suitability extent. LSA is an important basis for land-use planning and decision-making. It is of great significance for sustainable development and usage of lands. But it remains problems in traditional methods of solving LSA, such as subjective assessment, inefficient evaluation and difficult to do long-term evaluation. To cope with these problems, some artificial intelligence (AI) methods have been applied to this research field. However, current usages of AI methods are relatively homogeneous. Many new advanced AI methods have not been used. It is necessary to make further study and discussion.The discussion in this paper starts from Analytical Hierarchy Procedure (AHP), which is a common method in LSA. The shortcomings of AHP have been pointed out and we use sensitivity analysis to examine the uncertainties in it. In this study, it is also attempt to use the suitability rule classification method to replace AHP to reduce the subjective effects of criteria weights to the assessment results. In order to conduct the potential suitability assessment, the paper also tries to make use of geographical simulation system to reveal the land-use suitability conversion rules under certain development pattern, which provides a better basis for sustainable land-use planning.The main scientific work and findings of this paper include:(1) Propose the conception of land-use suitability simulation.In this paper, the land-use suitability simulation is defined as the method of implementing potential LSA using geographical simulation system. It utilises the characteristic of geographical simulation system that it could simulate complex systems to support potential LSA. The method can reveal implicated content in suitability distribution form under specified land-use type, discover prospective rules.(2) Simulation of potential land-use suitability using the mechanism of cellular automata (CA).According to the proposed concept of land-use suitability simulation, a CA based suitability simulation method has been designed. This is the first use of CA theory in the field of land-use suitability study to achieve the evaluation. The work is on the basis of three assumptions:(a) neighbour effect of land-use suitability (b) land-use development mode (c) restricted land-use suitability factors. CA simulation of potential land use suitability makes land evaluation more standardized and accurate, so as to make the work better meet the actual requirements of planners and decision-makers, and provide better solutions and technical support for sustainable use of lands.(3) Land-use suitability classification rules discovery for LSA using Ant colony optimization (ACO).This paper introduces new artificial intelligence theory - ACO, to obtain classification rules for LSA. The method avoids the subjective factors of weighting in AHP, reduces the interference of weight uncertainty to the evaluation process.The paper uses the definition of rules in rule-based system as reference, expresses suitability rules as a conditional which is in IF-THEN form. Meanwhile, the knowledge information obtained from the samples is also converted to this form and imported to the training set, which supply ACO discover classification rules. The data structure of classification rule discovery based on training set is abstracted with the inspiration of optimal path mechanism of ACO. It is utilized to discover rules for land-use suitability classification and generate the evaluation result map.(4) Spatial weight sensitivity analysis.The spatial weight sensitivity analysis makes use of improved OAT (one-at-a-time) methods. It explores the stability of the evaluation results, the relative weight sensitivity of criteria, and the problem that how to reduce the uncertainty in multi-criteria decision-making methods, etc. The results are presented in different forms including tables, charts and thematic maps, which make it easy to identify the geographical locations of high sensitivity. (5) Development of LSA Tool (LSA-GIS).Based on Microsoft C #. NET development platform, a model tool named LSA-GIS has been developed using ESRI ArcGIS Engine components, Mathworks MATLAB embedded development components. Critical flow charts and sample codes are also presented in this paper.The implementation of the tool focuses on the interaction design, which specially pays attention to user experience, to improve the efficiency of using the tool. Interaction design carries out unified modeling (UML) approach, which makes the design process more standardized and builds a good basis for function extension of the modelling tool in the future.(6) Case study of irrigated agricultural land-use suitability.This paper selected the Macintyre Brook catchment, Australia, as the study area. Three methods:AHP, land-use suitability classification rule discovery and CA based land-use suitability simulation have been conducted to assess and analyse irrigated agricultural land-use suitability. These three evaluation results were separately compared in spatial context according to certain regulation. It proved that the methods are reasonable, feasible. But there still existed limitation. Experiments showed that LSA-GIS modelling tool has generated satisfied results in the evaluation of study area. The methods and the tool are able to be popularised to complete LSA work in other study areas.更多还原