Research On Cartographic Visualization Methods For Mining Spatial Distribution Patterns

Spatial distribution is the basic research issue for geography.Mathematical,statistical,and computational methods and techniques are mainly used to mine spatial distribution of geographic phenomenon,such as regression,cluster analysis,Bayesian classification,support vector machine,and Markov Model.These methods and techniques belong to abstract thinking,using formalized expression,model,reasoning and logical thinking to abtain abstract results.However,geographic phenomenon is spatial referenced,and not all the knowledge can be obtained through methods and techniques based on abstract thinking.Based on concrete thinking,visulization methods can be combined with methods and techniques based on abstract thinking.Recently,researchers have recognized the importance of visualizaiton and proposed"visual data mining".However,it simply uses visualization to represent original data or the results obtained through mathematical,statistical,and computational methods and techniques."Visual data mining" is quite differnet from visualization that bases on visual thinking.In visual data mining,visualization merely acts as a media to conveny information or knowledge rather than a tool to mine geo-data and discovery knowledge.This paper studies cartographic visualization methods for mining spatial distribution charcteristics,and its main content includes the three aspects:(1)Mining spatial static distribution characteristics,especially geographic phenomenon with sematic or temporal hierarchy.We propose a visualization method to mine distribution pattern of geographic phenomenon with sematic or temporal hierarchy.Firstly,we describe sematic and temporal hierarchy based on ontology.Then we introduce a hierarchical information visualization method from information visualization domain-treemap to visualize sematic hierarchy,and iintroduce a calendar view to visualize temporal hierarchy.As a treemap or a calendar view cannot represent geographic reference,we design spatial layout algorithms to represent geographic reference of geographic phenomenon.As to geographic phenonmenon associated with geographic areas,we combined a treemap or a calendar view with a cartogram to represent geographic reference.As to geographic phenonmenon associated with geographic points,we proposed a new kind of map-a point grid map to represent geographic reference.A point grid map transforms an input point data set into a grid in which each point is represented by a square grid cell of equal size while preserving the relative position of each point,which leads to a clear and uncluttered appearance.We use our method to mine the spatial distribution pattern of Chinese farmers' income structure and spatial-temporal pattern of air quality in the three economic zones of China.(2)Mining spatial dynamic distribution characteristics.We adjest OD(origin-destination)map and propose its adjusted DO(destination-origin)map to mine spatial dynamic distribution characteristics.The OD or DO map can be regarded as a two-level spatial treemap representing flows recorded by pairs of locations.In accordance with the different hierarchy of the two-level spatial treemap,the OD and DO are distinguished.The first level spatial treemap represents the origins(destinations),while the second level spatial treemap represents the destinations(origins)in the OD(OD)map.The two different views can visualize respective spatial patterns correctly.They can provide an overview at the national scale and detail at the provincial scale,offer the ability to distinguish origins from destinations,preserve spatial configuration,and distinguish in-flows and out-flows,while avoiding overlaps and occlusions of flows effectively.We use an OD maps and its adjusted DO map to mine the interprovincial floating population of China.The migration indicators including migration volume,migration effectiveness,migration preference indexes,and sex ratios are represented by the OD and DO maps,which assist in to reviewing and studying the deep patterns of floating population.(3)Mining spatial distribution association characteristics.Spatial distribution association could indicate certain association relationship among a set of geographic phenomenon,and co-location rule mining is an important content of spatial distribution association mining.We propose a visualization method for co-locaiton mining in Eculidean space and along network.Considering spatial heterogeneity,this method can mine regional co-location pattern.Different from existing co-locaiton methods,this method makes use of visual thinking,and it is intuitive and can be easily understood.The visual language is used to represent mutual influence between two geographic phenomena along networks in this method.Firstly,taking Tobler's first law of geography into consideration,we use a kernel density estimation method(network kernel density estimation method)to express distribution pattern of geographic phenomena in Euclidean space(along networks),and construct a mapping between the distribution pattern of geographic phenomenon and color.Secondly,based on the law of additive color mixing,two colors representing two geographic phenomena are mixed to get cognition of the interaction between the two geographic phenomena.Finally,Pearson correlation coefficient is used to evaluate the association relationship between geographic phenomenon quantitively.