×
![]( "Buchcover ISBN 9783906467030")
Spatial Raster Data Management for Geo-Information Systems
A Database Perspective
von Stephan Nebiker, Vorwort von Alessandro CarosioIn numerous geomatics domains, vast amounts of spatial raster data are being acquired by high-performance scanning devices, digital cameras and high-resolution remote sensors. The efficient storage, handling, querying and retrieval of this data is one of the main challenges facing producers and users of spatial information. Existing raster management solutions are generally restricted to the file-based handling of limited-size raster objects with limited support for spatial aspects. In GIS, spatial raster data, such as raster maps, raster imagery, digital terrain models and thematic raster data are traditionally not integrated within the overall data management concept. Thus, they cannot benefit from standard database services such as querying, integrity and concurrency control, features which are usually available for the other GIS data types.
This dissertation documents the investigations and developed concepts for the management of large spatial raster objects and raster mosaics in a DBMS framework. Furthermore, it discusses the design and implementation of a prototype spatial raster management system, which was developed for the evaluation of these concepts.
In the first part, the theoretical foundation for the subsequent investigations is established. This includes the examination and documentation of the fundamental principles and characteristics of spatial raster data. It also incorporates the evaluation of raster modelling concepts, database models and technologies as well as storage management concepts and mass storage technologies.
In the second and main part, a number of new or modified spatial raster management concepts are described. Most of them are data model-independent and can be implemented on any modern DBMS technology. The first concept is a tile-based spatial partitioning concept for raster mosaics with an indexing scheme based on the Morton ordering. This space ordering scheme provides efficient spatial access to mosaic tiles through linear indexing techniques available on all DBMSs. In combination with the presented query optimisation concept, this indexing scheme also supports efficient spatial range queries with a low oversearch ratio. Secondly, a novel approach is presented with the tile-based multi-resolution concept for raster mosaics. This concept uses a resolution pyramid of constant-dimension tiles covering different spatial extents. With this approach, the number of database objects to be retrieved can be kept roughly constant over the entire zoom range. By combining the resolution access with the spatial access, the concept allows the reduced-resolution tiles to be spatially indexed and accessed via the same indexing scheme. It also allows the resolution access to be implemented on the DBMS server as a relatively simple filtering operation. Thirdly, a multi-level georeferencing concept for raster objects is presented. This concept allows the spatial extents of raster objects to be represented and queried at different levels of approximation. It therefore supports coordinate system-independent global and system specific local search scenarios.
The design and development of the prototype spatial raster management system GrIdS allowed a number of these concepts to be implemented and evaluated. The system architecture supports DBMS-internal and external raster data storage with a modular compression support mechanism DBMS-internal storage currently uses the Binary Large Object (BLOB) concept, whereas DBMS-external uses a file-based solution. The prototype system was implemented on the basis of a relational DBMS (Oracle 7 3).
The system tests were primarily focused on the functionality and performance of the mosaic management solution, which incorporated the majority of new concepts. The tests were performed on a series of mosaics which had been generated from raster maps and orthoimages. The tests showed excellent and consistent tile access times. The spatial range query performance was generally very good but occasionally it was affected by an outlier with a large tile oversearch. It was shown that these outliers can be completely eliminated if the proposed query optimisation is applied. Extensive window retrieval tests were performed, which yielded good retrieval times. The import and export tests with large raster objects revealed a relatively moderate system performance. The main limiting factor was identified as the low performance of the available test platform. Typical database features such as concurrent mosaic retrieval and update were successfully verified.
On the whole, the system implementation and tests demonstrated that the developed concepts can be implemented relatively easily and rapidly. They are suitable for building robust, efficient and scaleable raster management solutions on the basis of standard DBMS technology. The concepts could also be integrated with spatial database extensions used in numerous GISs. The shortcomings of the test platform with respect to I/O performance are discussed and recommendations for future system configurations together with high-performance and high-capacity storage solutions are provided.
This dissertation documents the investigations and developed concepts for the management of large spatial raster objects and raster mosaics in a DBMS framework. Furthermore, it discusses the design and implementation of a prototype spatial raster management system, which was developed for the evaluation of these concepts.
In the first part, the theoretical foundation for the subsequent investigations is established. This includes the examination and documentation of the fundamental principles and characteristics of spatial raster data. It also incorporates the evaluation of raster modelling concepts, database models and technologies as well as storage management concepts and mass storage technologies.
In the second and main part, a number of new or modified spatial raster management concepts are described. Most of them are data model-independent and can be implemented on any modern DBMS technology. The first concept is a tile-based spatial partitioning concept for raster mosaics with an indexing scheme based on the Morton ordering. This space ordering scheme provides efficient spatial access to mosaic tiles through linear indexing techniques available on all DBMSs. In combination with the presented query optimisation concept, this indexing scheme also supports efficient spatial range queries with a low oversearch ratio. Secondly, a novel approach is presented with the tile-based multi-resolution concept for raster mosaics. This concept uses a resolution pyramid of constant-dimension tiles covering different spatial extents. With this approach, the number of database objects to be retrieved can be kept roughly constant over the entire zoom range. By combining the resolution access with the spatial access, the concept allows the reduced-resolution tiles to be spatially indexed and accessed via the same indexing scheme. It also allows the resolution access to be implemented on the DBMS server as a relatively simple filtering operation. Thirdly, a multi-level georeferencing concept for raster objects is presented. This concept allows the spatial extents of raster objects to be represented and queried at different levels of approximation. It therefore supports coordinate system-independent global and system specific local search scenarios.
The design and development of the prototype spatial raster management system GrIdS allowed a number of these concepts to be implemented and evaluated. The system architecture supports DBMS-internal and external raster data storage with a modular compression support mechanism DBMS-internal storage currently uses the Binary Large Object (BLOB) concept, whereas DBMS-external uses a file-based solution. The prototype system was implemented on the basis of a relational DBMS (Oracle 7 3).
The system tests were primarily focused on the functionality and performance of the mosaic management solution, which incorporated the majority of new concepts. The tests were performed on a series of mosaics which had been generated from raster maps and orthoimages. The tests showed excellent and consistent tile access times. The spatial range query performance was generally very good but occasionally it was affected by an outlier with a large tile oversearch. It was shown that these outliers can be completely eliminated if the proposed query optimisation is applied. Extensive window retrieval tests were performed, which yielded good retrieval times. The import and export tests with large raster objects revealed a relatively moderate system performance. The main limiting factor was identified as the low performance of the available test platform. Typical database features such as concurrent mosaic retrieval and update were successfully verified.
On the whole, the system implementation and tests demonstrated that the developed concepts can be implemented relatively easily and rapidly. They are suitable for building robust, efficient and scaleable raster management solutions on the basis of standard DBMS technology. The concepts could also be integrated with spatial database extensions used in numerous GISs. The shortcomings of the test platform with respect to I/O performance are discussed and recommendations for future system configurations together with high-performance and high-capacity storage solutions are provided.