以人流、物流或信息流等方式存在的空间O-D流将不同区域紧密联系在一起，形成具有一定结构和功能的复杂网络系统。基于频率学派的空间数据分析，将影响空间O-D流变化的诸因素中的不确定性归结为噪声，忽视了与模型设定有关的不确定性，难以对空间网络系统的结构和功能演变做出更为合理的解释。本项研究从空间O-D流所在的网络系统的复杂性出发，围绕节点属性、节点环境与模型结构表达中可能存在的不确定性，结合贝叶斯统计学，建立表征不确定性的空间O-D模型，通过贝叶斯模型平均方法优化模型，以期对空间O-D流的模式演变与内在机理做出较为科学的解释，并对未来发展趋势做出概率预测。本项研究通过对中国省际人口迁移流的建模与预测分析，以实例形式表明面向不确定性的空间O-D模型构建与优化在技术与方法上的可行性。本项研究将深化GIS和空间数据挖掘等领域的研究和应用，一定程度上推动地理学研究范式的转型，提升其定量化研究水平。

We are in a mobile society. Different types of spatial origin-destination flows, whatever tangible entities such as persons and commodities or intangible ones such as information, have weaved different regions in a complex network system with specific structures and functions. The spatial data analysis and modeling techniques based on frequentist statistics, cannot provide a sound explanation of the structural and functional evolution in a network system by ignoring the uncertainties in model specification. Starting from the complexities in a spatial.O-D network system, this project attempts to apply Bayesian statistics to build a series of spatial O-D models with uncertainties about nodes and their environments. These models are then compared and optimized by Bayesian model averaging methods in order to provide a much sounder interpretation of the evolution of the spatial O-D flows. The feasibility of these techniques and methods in spatial O-D model optimization will be demonstrated by a case study of Chinese interprovincial migration flows. This project in spatial O-D model comparison and uncertainty analysis will deepen the capability of GIS spatial analysis and the related spatial data mining and knowledge discovery fields. In addition, it will drive the paradigm transformation of geographic research from frequentist to Bayesian, and improve the development of quantitative geography to some extent due to the symbolic role of spatial interaction modeling in geography.