Simulation Analysis Of The Influence Of Spatial Weight Matrix On Moran's I

With the development of technology, GIS doesn't any more limit on storage, query, display, of data, main works now are to analysis in depth transformation rule and development of geographic phenomena happenings based on characteristics of dynamics. Spatial autocorrelation analysis becomes focus of geography, and many parameters of spatial autocorrelation have been brought out. In these parameters, Moran's I is the earliest and was used widely.In order to reveal spatial relationship among phenomena, it is necessary to define neighborhood relationship among phenomena first. Spatial weight matrix is the main express form of this relationship. With the deep research, different definitions of spatial weight were built.In the research of spatial neighborhood relationship now, almost all only use one kind of spatial weight matrix, such as matrix based on distance, binary join, K nearest. For the influence of spatial weight definitions on Moran's I, and spatial cluster, spatial autocorrelation analysis result, there still lack of research. Different definitions of spatial weight directly influence on Moran's I, analysis and view on the problems of spatial relationship. Further more, spatial attribute data exist some errors, they influence on statistic and analysis of data. However, how these influence on spatial autocorrelation analysis still lack of necessary evaluation.In this paper, simulation data and GDP data of main counties in Jiangsu province were analyzed and got the following conclusions:(1 )In spatial weight matrix, K nearest weight and threshold weight definitions can derived a stable global Moran's I, and left-right weight, up-down weight are worse, based on Monte Carlo simulation.(2) Data error can bring obvious influence on Moran's I. Global Moran's I may become small when error is bigger. Distance weight, threshold weight, up-down weight are sensitive to error; K nearest weight and binary join weight aren't sensitive to error.(3) A case analysis of Jiangsu province showed that the different definitions of spatial weight matrix can result in the different explanation, and result of Local Moran's I based on the K nearest weight are more reliable.