多变量空间自相关分析：Moran指数与二元变量应用
需积分: 38 200 浏览量
更新于2024-07-20
3
收藏 505KB PDF 举报
"这篇文章探讨了如何在地理信息系统（GIS）中实现二元变量的空间自相关系数计算，特别是通过使用Moran指数。它介绍了一种替代方法，即利用专用函数库构建功能，而不是依赖现有的完整软件套件。这种方法是模块化的，并且独立于其他系统。在动态链接窗口的框架下，它结合了地图上的数据 cartographic 表示、传统的统计图形，如直方图、箱线图和散点图，特别扩展到了多变量的空间自相关性可视化，引入了Moran散点图矩阵和多元LISA地图。"
在地理数据分析中，空间自相关是一种重要的概念，它涉及到地理位置相邻的观测值之间的相似性。Moran指数是衡量这种空间自相关性的常用统计量，由PatrickJ. Moran在1950年提出。这个指数的值介于-1和1之间，其中1表示完全正相关，-1表示完全负相关，0表示没有空间自相关。如果Moran指数接近1，那么意味着空间数据在空间上呈现出聚集特征；如果接近-1，则表明数据呈现分散或反向聚集。
文章提出的动态链接窗口框架允许用户在多个视图之间交互，使得分析更为直观。例如，Moran散点图矩阵是一种展示各变量间空间关系的工具，每一行和列代表一个变量，点的位置反映了变量对之间的空间自相关性。如果点集中在对角线上，表明变量之间存在空间一致性；若远离对角线，可能揭示了空间异质性或负相关。
此外，多元LISA（Local Indicators of Spatial Association，局部空间关联指标）地图则用于识别空间中的热点和冷点，这些热点和冷点是指那些具有显著高或低值的区域，且这些值与周围邻居的值相比异常。多元LISA可以识别出在多变量环境中哪些特定区域的变量组合表现出强烈的空间聚集。
这种集成GIS和统计分析的新方法对理解和解释复杂的地理现象非常有价值，尤其在环境科学、城市规划、农业经济学等领域，帮助研究人员发现潜在的空间模式和趋势，从而做出更准确的决策。通过定制化的小型函数库，这种方法提供了更大的灵活性，可以根据具体研究需求进行调整和扩展。
Visualizing Multivariate Spatial Correlation
4
“map” is but one of several linked views.
4
Similar visions underlie several other recent
efforts to develop open and modular software frameworks for the visualization of high
dimensional (spatial) data.
5
In addition to being freestanding, DynESDA2 also includes a number of other advances
over its predecessors, such as the capability to handle both point and polygon coverages,
“true” brushing of maps, simultaneous linking of multiple maps with multiple statistical
graphics, and interactive LISA maps. It also extends the visualization of spatial correlation
to a multivariate setting. We turn to this first.
3 Multivariate Spatial Correlation
The visualization and exploration of multivariate association is a core functionality of cur-
rent exploratory data analysis (EDA), knowledge discovery and data mining tools (Buja
et al. 1996, Han and Kamber 2001, Gahegan et al. 2002). The incorporation of “spatial”
association in this framework is still in its infancy, however. Most suggested approaches
pertain to geostatistical analysis, where data are represented as points and the measure of
spatial correlation is derived from the variogram (see, e.g. Cook et al. 1996, Majure and
Cressie 1997). Similar progress has not been made for the analysis of multivariate spatial
correlation for lattice data, i.e., spatial objects represented as discrete points or polygons.
6
We develop a visualization device for multivariate spatial correlation in lattice data by
building on some of the ideas originally advanced in Wartenberg (1985). There, a multi-
variate coefficient of spatial autocorrelation between two standardized random variables z
k
and z
l
is defined as:
m
kl
= z
0
k
W
s
z
l
, (1)
where z
k
= [x
k
− ¯x
k
]/σ
k
and z
l
= [x
l
− ¯x
l
]/σ
l
have been standardized such that the mean
is zero and standard deviation equals one, and W
s
is a doubly standardized (or, stochastic)
spatial weights matrix. The weights matrix defines the “neighbor set” for each observation
(with non-zero elements for neighbors, zero for others) and has zero on the diagonal by
convention.
This concept of multivariate spatial correlation thus centers on the extent to which val-
ues for one variable (z
k
) observed at a given location show a systematic (more than likely
under spatial randomness) assocation with another variable (z
l
) observed at the “neighbor-
ing” locations. Note that this multivariate spatial correlation can be considered in addition
to or instead of the usual (non-spatial) correlation between the two variables at the same
location. Wartenberg (1985) used this statistic to develop a notion of spatial principal
components, for which the double standardization of the weights matrix (and the implied
symmetry) was necessary.
For the purposes of visualization, our focus is on the linear association between a vari-
able z
k
at a location i, z
i
k
and the corresponding “spatial lag” for the other variable, [Wz
l
]
i
.
7
In this context, the usual singly-standardized (row-standardized) form of the spatial weights
matrix can be used, which yields an interpretation of the spatial lag as an “average” of
neighboring values. Also, the cross-product statistic can be re-scaled by dividing by the
4
See Unwin (1996) and Wilhelm and Steck (1998) for recent examples. Similar ideas are behind the Tcl/Tk
based cdv toolkit of Dykes (Dykes 1997, 1998) as well as Brundson’s exploration of local spatial association
using a dynamically linked “map” constructed with tools available in Xlispstat (Brundson 1998).
5
See, for example, MacEachren et al. (1999), Sutherland et al. (2000) and Gahegan et al. (2002).
6
Note that the points used in geostatistical analysis are sample points from a continuous surface. In contrast,
for lattice data the points are not a “sample,” but fixed locations at which a spatial pattern for a random variable
can be observed.
7
The notation indicates that the spatial lag for location i is the i-th element of the vector Wz
l
. See Anselin
(1988), for an extensive treatment of the notion of a spatial lag.
剩余19页未读，继续阅读
2020-09-16 上传
2021-05-26 上传
2021-10-03 上传
2021-07-03 上传
2009-12-29 上传
2021-10-03 上传
2021-10-02 上传
ljlovejay
- 粉丝: 1
- 资源: 9
我的内容管理
展开
最新资源
- IEEE 14总线系统Simulink模型开发指南与案例研究
- STLinkV2.J16.S4固件更新与应用指南
- Java并发处理的实用示例分析
- Linux下简化部署与日志查看的Shell脚本工具
- Maven增量编译技术详解及应用示例
- MyEclipse 2021.5.24a最新版本发布
- Indore探索前端代码库使用指南与开发环境搭建
- 电子技术基础数字部分PPT课件第六版康华光
- MySQL 8.0.25版本可视化安装包详细介绍
- 易语言实现主流搜索引擎快速集成
- 使用asyncio-sse包装器实现服务器事件推送简易指南
- Java高级开发工程师面试要点总结
- R语言项目ClearningData-Proj1的数据处理
- VFP成本费用计算系统源码及论文全面解析
- Qt5与C++打造书籍管理系统教程
- React 应用入门：开发、测试及生产部署教程
