Back to list of ACASIAN papers Back to ACASIAN Home Page
DESIGN AND IMPLEMENTATION OF HISTORICAL DIMENSIONS IN THE SIIASA/CHINA GIS PROJECT DATABASES
Paul Yates and Dr. Lawrence W. Crissman Australian Centre of the Asian Spatial Information and Analysis Network Griffith University, Nathan, QLD 4111, Australia
The China Geographical Information System Project was initiated in early 1989 in the Faculty of Asian and International Studies, Griffith University. It became the pilot project for the Spatial Information Infrastructure for Asian Studies in Australia (SIIASA) Project, funded by the Australian Research Council for 1992 through 1994. The SIIASA Project now involves a consortium of nineteen Australian universities, and is establishing GIS databases containing local administrative boundaries for all of East, Southeast, and South Asian countries, while a companion project (SIIRCEASA) is doing the same for all of the Republics of the former Soviet Union. Because of the more advanced stage of the China GIS Project in 1992, it was the early major beneficiary of the SIIASA grant, and considerable progress has been made in digitizing information from the Land-use Map of China and the sixteen published sheets of the 1:1,000,000 Map of China. An early goal of the China GIS Project was the vectorisation of Xian and Shi boundaries matching the 1990 census results, which are now in their final form.
The China GIS Project spatial databases were designed from the outset to incorporate an historical dimension for the administrative system of the PRC. That was necessary because it is not a static system, as is more or less true of many western countries, but one which is continually undergoing local adjustments as well as systematic changes designed to support China's developmental policies. As a result, there are changes of one kind or another to approximately one percent of the PRC's county-level administrative units each year. Many of the changes during the past decade have involved the creation of shi from what had been xian or diqu. Such changes of administrative type or level can involve name changes, but always involve changes to the official Guobiao coding system used by the State Statistical Bureau for such things as census tabulations.
Although such changes do not usually involve any actual boundary modifications, county-level units are divided or merged with some frequency. Therefore, it is not possible to directly compare the 1982 census materials with their 1990 counterparts using a standard GIS because some of the units have changed spatially between those dates. Although it would be possible to establish continuities for statistical purposes by aggregating figures up to the level at which no spatial changes occurred between two dates, that proceedure obviates the basic advantage of using a GIS in the first place, which is that it allows data to remain disaggregated and spatially specific. Some other means that retained the locational desegregation of attribute data organised into a GIS was therefore required to deal with the temporal changes in China's administrative system.
As most of us working in the GIS field well know, commercially available software programs do not incorporate temporal dimensions or topologies. They can be used to construct 'slices in time', but because unchanged spatial data must be stored redundantly for all such times, the inefficiencies in dealing with many frequently occurring changes soon make such an approach unviable. For instance, the file containing the 1990 administrative system of the PRC, as digitised at 1:1m scale, comprises many Megabytes. Even if the administrative changes were only needed on a year by year basis between 1982 and 1992, it would require nine times the storage capacity to maintain those coverages than would be needed for a single year although the actual changes to the spatial data from year to year would be minimal.
The 'slices in time' approach is also deficient in allowing the relationships between earlier and later manifestations of the 'same' element to be maintained if names or codes have changed. It also cannot trace the relationships among antecedent and descendent manifestations of elements that have changed spatially, unless the cumbersome polygon overlay method is used. That of course soon becomes impractical when dealing with multiple time periods.
The problem, therefore, was to devise a means to maintain relationships among changing elements over time, and to store the temporal variations in the spatial data economically in the same, non-redundant database. As the SIIASA project supports multiple GIS software systems, it was also necessary to develop a general solution that could be implemented in a variety of different commercial programs, such as MGE, ARC/INFO, and even MapInfo.
The Spatio-Temporal Database Structure.
The solution developed for incorporating administrative changes in the China GIS Project databases involves two basic methodologies. Firstly, for temporal spatial data, a space-time composite as describe by Langran (1991) is maintained. In a space-time composite, areal features such as administrative units are decomposed into the minimum units that can be identified as being part of one and only one administrative unit at any one time period. This method essentially implements time as a third dimension, projected and stored in a two dimensional file. Secondly, for temporal aspatial (attribute) data, a versioning method is employed, where changes in the attributes pertaining to a spatial feature cause a new row to be created in the attribute table. Both the old version and new versions of the attributes are stored with an appropriate timestamp, or 'from-to' dates.
Four essential types of changes in China's administrative system were identified and incorporated into the database design. For each type of change there are distinctive storage requirements for both spatial and aspatial data. For example, a change in boundary requires that alternative data be stored in the spatial database, while changes in name alone require multiple entries in the attribute database. Table 1 shows the types of change and the storage requirements.
Table 1: Significant Types of Administrative Change
Change type Storeage Requirements ------------------------------------------------------------------------- Guobiao Code (C) The new code and as well as data establishing relationship between new and old codes. Status (S) The new status. Name (N) The new name. Boundary (B) The new boundary data with data establishing the space-time composite by relating the all new polygons to their Guobiao codes. -------------------------------------------------------------------------Clearly, the different change types can occur simultaneously, eg. changes in code and status, changes in code, name and boundary, etc. Note that change in aspatial data does not imply an alteration to the spatial data and similarly change in the spatial data does not imply a change to aspatial attributes.
Figure 1 shows the data structures we are using to accommodate the temporal requirements using the methodologies identified above.
The Administrative Feature Table contains the Guobiao codes with the appropriate name and status, and answers the storage requirements for change in name and status through the use of a row versioning method. When a given date is queried, the RDBMS simply selects out all rows that are 'current' for that time using the 'date from' and 'date to' columns, yielding a unique mapping of code to name and status. The Polygon Table allows us to uniquely identify each polygon. To store the relationship between the polygons and the administrative areas, a Composite Polygon Table is used, which stores the history of attribute changes to particular polygons. Again the rows in this table are duplicated and assigned appropriate temporal information when change occurs. At different times, a single administrative unit may be represented by multiple polygon elements.
The Spatio-Temporal Database Operation.
Sources for change in the administrative structure in china are often descriptive, and contain no spatial information. This descriptive information must be transformed so that it may be stored in a structured database to enable spatio-temporal operation. Table 2 shows such descriptive information as it is found, along with an indication of the change type in the form suggested in Table 1.
Table 2: Administrative Changes
Time Boundary Change Change Description of Change Event Type ----------------------------------------------------------------------- T0 E0 B Dongshan Xian (394222) and Beishan Xian (394223) established from part of Wushan Xian (394221). T1 - N Wushan Xian (394221) renamed to Shan Xian. T2 E1 B Dongshan Xian (394222) and Beishan Xian (394223) merged to form Wushan Xian (394223). T3 - C,S Shan Xian upgraded status to Shi and renamed to Wushan Shi (394201). -----------------------------------------------------------------------
Figure 2 shows the resulting spatial and aspatial database elements in our structure for the changes indicated above.
Retrieving Information from the Database
The database design described above allows the construction of "views" of the GIS for any given date, and also allows the historical lineage of an administrative unit to be traced. The approach used to implement "view" generation in currently available commercial software involves copying the base polygon table, retrieving the valid Guobiao code for each polygon, using utilities to merge adjacent polygons, and lastly retrieving the current name and status information for each element. This method can essentially be implemented in ARC/INFO, using the dropline program, and MapInfo for Windows using the Merge using Column facility. As MGE does not require polygon topology to be stored, the implementation is more straightforward.
The construction of a historical lineage may take two forms. Firstly, GIS software can be used to overlay constructed 'views', identify changes of the same area. The results of the overlay may be stored for future queries, or discarded. Secondly, map data can be generated showing the change of particular administrative units, by displaying their present or past manifestations.
Using a variety of documentary sources, the record of changes to the county- level administrative system of the PRC has been compiled and maps showing earlier boundaries have been assembled. All changes between the 1990 census and the 1982 census are now also incorporated into the China GIS Project databases, and work is well along to reconstruct all changes back to December, 1949. The next step will be to go back to 1911, from which it will not be difficult to reconstruct the Qing Dynasty system of field administration for circa 1820. Scholars with historical interests will then take up the task of pushing the coverage back to earlier dynasties.
Langran, Gail, Time in Geographical Information Systems. Technical Issues in Geographical Information Systems Series. London: Taylor and Francis, 1992.