面向地理变量空间分布推测的知识集成方法研究

Predictive mapping is one of the most important approaches to mapping spatial variation of geographical variables. A key component in predictive mapping is the knowledge on the relationships between the target variable (geographic variable to be predicted) and the covariates (other geographic variables which co-vary with the target variable). The sources which contain the knowledge on the relationship often come in various forms and styles, such as samples, maps and texts. The knowledge in these different sources each has its own strengths and drawbacks and is often complement to each other. However, existing methods can only extract knowledge from single source, thus unable to effectively use the knowledge in other forms. To address this problem, this project intends to develop a method for effectively integrating knowledge from different forms to increase the quality of the obtained knowledge. The proposed research consists of four elements: design of a prototype-based scheme for representing knowledge, development of methods for transferring knowledge in different sources into the prototype-based scheme, integration of knowledge expressed in prototypes and quantification of trustworthiness of each prototype, and enhancement of existing predictive methods to be able to use the integrated knowledge in the form of prototypes. The proposed research aims to overcome the bottleneck of knowledge integration from different sources in predictive mapping, in turn to push forward the development of theory and techniques for predictive mapping. The findings of this research will have theoretical and practical implications for geographers and ecologists who use predictive mapping for mapping spatial variation.

在地理变量空间分布信息获取方法中，基于地理目标变量与环境协同变量相关关系的方法近年来发展迅速，是空间推测领域的重要发展方向。该方法的关键在于获取地理目标变量与环境协同变量之间的关系知识。关系知识的来源有多种，包括样点、专题图和文本知识等，所含知识往往优势互补。由于这些知识来源具有不同的结构和特征，现有知识获取方法往往只能利用单一知识来源，难以有效集成这些多源异构知识进行地理变量空间分布的推测。针对这一问题，本项目拟根据原型理论构建多源异构知识的统一表达模式，探索不同来源和结构的知识向原型的映射方法，利用信用互评方式集成所得原型并度量原型的可信度，进而发展基于多源知识集成的地理变量空间分布推测方法。本项目的研究有望突破现有空间推测方法难以充分利用多源知识的限制，为地理变量空间推测提供新的思路，推动地理变量空间分布推测方法体系的发展，对地理学、生态学等领域均具有重要理论意义和应用价值。

地理目标变量和环境协同变量相关关系是空间推测方法所需的关键内容。刻画相关关系的知识来源往往有多种，如样点、专题图和专家等，这些不同知识来源具有不同的结构和特征。项目针对现有知识获取方法往往只能利用单一知识来源，难以有效集成这些多源异构知识进行地理变量空间分布推测的问题，根据原型理论构建了多源异构知识的统一表达模式，基于这一表达模式，建立了将来源不同和结构各异的知识向原型的映射方法，利用信用互评方式集成所得原型并度量原型的可信度，进而研发了基于多源知识集成的地理变量空间分布推测方法。本项目的研究突破了现有空间推测方法难以充分利用多源知识的限制，为地理变量空间推测提供了新的思路，推动了地理变量空间分布推测方法体系的发展，对地理学、生态学等领域均具有重要理论意义和实际应用价值。