A Study of a Framework of Integration of Heterogeneous Spatio-Temporal Data
2.2 Classification of the observational data CCM
in this study, existing map data is regarded as "observational " data. We have to find out reasons of the gaps between " two kinds of CCM boundaries . if two types of CCCM boundaries data defined are generated based on different product specifications, it is necessary to determine how much degree they are close with each other, i.e. which part of the boundary data (observational data ) should be matched, because they are obtained from identical objects in real world, and which part should not because they are from different objectives. In case CCM boundary of CMS, which is also called regional statistic boundary consists of census tracts including some base unit tracts, while SDI consists of streets blocks, this is why gaps of CCM objects exist ( figure 2.). but it can be said that both are based on the same product specification if we cab trace back to the base unit blocks and street tracts because based unit tracts is equivalent to streets blocks in CMS data.
Even if specifications are same we need to consider a gap by change of object through the time. In this case imagine that town boundary was changed in certain time ad that the name of town was changed . this means the change of location and attributes of boundary through the time. If we take a gap of objects by different observation time onto account, an error in determining the location only remains. The locational error can be classified as a systematic error and a random error.
Like this , we cab overlay base unit tracts and street blocks as observational data from the commonly defined objects if we identify objects as CCM and regional statistic boundary referring each metadata.
Fig.2 Gaps between boundary data generated "partially" different product specification
2.3 Estimation of the CCM Boundaries
After the identification of observational data from the same objects we cab estimate the location , shape and an attribute of the object from base unit tracts data and street blocks data. We have only to take into consideration an error in measuring location and spatial extent of objects . taking an error and reliability into consideration by referring by referring to the quality of data described in metadata, we cab determine a probabilistic distribution for locatinal uncertainties of the boundary lines. Using the probabilistic function we can merge the boundary observational data to estimate the location boundary from by applying maximum likelihood method or some others . estimation accuracy cab be improved by considering constraints , which come form the shapes and nature of object itself, in addition to the observed data. It is necessary to prepare feature catalogue in advance including general information about feature as constraints.
3. framework of data integration
This section describes a framework of data integration. We define object as a thing , the state and changes of which we want to now by data integration, and define observational data as data to be integrated. According to the definition existing maps are also observational data.
3.1 Preparations
we summarize conditions used in classification before integration because process of data integration changes according to the nature of objects and observational data.
-
Type of object
(a) discrete distribution object (B) continuous distribution object
-
Types of observational data
(A) Classified observational data (B) Unclassified observational data
-
Dynamic nature of object
(A) immobile object (B) Mobile object
-
Data form
(A) Vector data (B) Raster data (C) Mixed data
3.1.1 types of object
There are two types of object ; discrete distribution object has a finite spatial extent or crisp boundaries,. While continuous distribution object has an infinite spatially continuous distribution such s temperature distribution . the essentials difference between these two is a definition of spatial extent. Spatial extent of a discrete distribution object is determined naturally when the object itself is defined, whereas the definition of continuous distribution object cab be made ind4pendently form the definition of aggregation and sampling unit, which is indispensable to generate data.
3.1.2 Types of observational data
Types of observational data depend on whether observational data is clearly classified to some object. For instance map data is observational data which acre already classified or attached to each object =. On the other hand each pixel in image data of remote sensing is not clearly classified to each object
3.1.3 Dynamic nature of objects.
It is necessary to classify in advance whether object changes through the time r not if we consider the temporal changes in object identification. This means that mobile object, whose shape and location changes every moment , is not applicable for identification of object as it is, though immobile object unchangeable through the time cab be used as the key of data integration object is known as unchangeable
3.1.4 Representation form of data
Representation form of data, which has no essential difference between raster and vector data , affects a design for algorithm of data integration .
