Network Spatial Analysis Supported By Network Tessellation Approach

As the basic skeleton and event context of geographic space, traffic network has always been treated as the main study object in geographic modeling and spatial analysis. Especially, with the general use of new handheld portable devices, location system and wireless sensor network, the information communication technologies currently are highly developed, and network spatial database becomes to accumulate huge amounts of human-activities spatiotemporal data. The characteristic of the ’big data’brings new challenges and opportunities to geographic network research, including data model transformation and spatial analysis technology evoluation. The central element of the model redefinition is dynamic network replacing static network as the basic unit of spatial modeling. Besides, regarding the network spatial analysis applications, network distance instead of traditional Euclidean distance should be used as the metric of spatial neighborhood.In a simple geographic environment, the node-segment model supported by the graph theory can be utilized to express the spatial linear characteristic, topological characteristic and connectivity characteristic of geographic network. However, since the graph model mainly focuses on the structural description of network from discrete object view, the detail of the geographic characteristic of spatial network would be disregarded, and thus that leads a large gap between the data model and the real cognitive result. Given this problem, this paper studies the network data model and the network spatial analysis algorithms based on network tessellation from the spatial continuum field view, and its main content includes the five following aspects:1. The application status and propects of geographic network model and spatial analysis technology are explored. We summarize the theoretical basis in spatial network research including the topological characteristic between network elements, the conceptual model of spatial network and the relevant spatial modeling technologies (e.g. computational geometry and graph theory). The spatial network model is divided into three levels:graph model, simple spatial network model and node-segment model, and also gives the major obstacles in representing microscopic geographic features of spatial network with object models.2. The application of continuum field view in the representation of geographic features is expounded. After analyzing the characteristics (i.e. data organization, structure definition and contruct rule) of general raster data model and real street network, we summarize the organizational form of 2-D raster model of spatial network, following some problems in representing network elements with grid pixel, e.g. the loss of topological relationship between node and segement, the difficulty in representing non-planar structure, the exaggerating of the estimated length of segment. To solve the above problems, we propose a network tessellation method based on the densing points, and establish a 1-D raster data model (namely LDM, or Lixel Data Model) to support network spatial analysis applications. The specific categories of the raster in LDM model and its relevant properties are introduced. On this basis, a proper index scheme for different spatial objects (i.e. point, line and area) is established and also a new definition of network distance is proposed. Then following the definition in mathematical morphology, we formalize the expansion operator for LDM model in order to support the regional search and the distance calculating in network spatial analysis. Specifically, according to the general static traffic environment and the complex dynamic traffic environment, the LDM model and expansion operator are further categoried as follows:● The conventional network LDM model and general expansion operator are defined, following the discussion of the spatial alignment, topological relationship and properties representation of raster lixel unit.● The constrained network LDM model and constrained expansion operator considering traffic constraints and generator difference are defined. The traffic constraints include dynamic traffic, directional road and street turn prohibition.3. Based on the LDM model description and expansion operator definition, the water flow extension idea in natural world is formalized, which lets streams spread on the network paths until meeting the other streams or arriving at the end of edge, with treating the event sources as the headwaters, raster unit length as expanded step length. And we also make a discussion about the algorithm complexity of flow extension idea, theoretically evaluate its validity in computing shortest path and extend the general algorithm suited to the applications in dynamic traffic condition. Considering the problem of distance calculating error due to unequal lixel lengths, a new parameter named lag is proposed to record error of distance, and also the corresponding expansion operator formalization presented.4. Using urban facility Point-of-Interests data, the research proposes several LDM-model-based spatial analysis and pattern recognization methods including network Voronoi model, network kernel density estimation and urban CBD (Central Business District) detection. The impact from traffic constraints and facility difference on these model and algorithms are evaluated. in addition, a 3-D density visualization approach based on two visual variables, color and height, is proposed in order to better reflect the linear nature of network space. The actual data experiment in Shenzhen and Guangzhou cities shows that the proposed algorithm is efficient and effective.5. The proposed models and algorithms are tested from physical level technologies. We present a practical strategy in developing network spatial analysis system, and also the system environment, data sturucture, operator library and key parameters in the developed LDM-model-based software.