Large-scale multi Multidimensional coverage databases
Peter Baumann
Active Knowledge GmbH
Kirchenstr. 88, D-81675 Munich, Germany
Email: baumann@active-knowledge.com
Erhard Diedrich
German Aerospace Center
Oberpfaffenhofen, D-82234 Wessling, Germany
Email: erhard.diedrich@dlr.de
Clemens Glock
Bayerisches Landesvemessungsamt
Alexandrastr. 4, D-80538 München, Germany
Email: clemens.glock@blva.bayern.de
Michael Lautenschlager, Frank Toussaint
Max-Planck Institut for Meteorology
Bundesstr. 55, D-20146 Hamburg, Germany
Email: lautenschlager@dkrz.de,
toussaint@dkrz.de
Abstract
With GIS technology for Web-enabled vector and meta data access becoming mature, the next
quest is towards fast, flexible services on extremely large raster data assets. Currently 2-D images
(satellite imagery, ortho photos) prevail, but at the horizon higher-dimensional objects appear, such as
3-D time satellite image time series and seismic data, and 4-D climate simulations.
At the same time, database researchers are getting aware of the specific challenges such
multidimensional coverages pose. Some promising recent work shows that database systems with
optimising query languages actually are superior to file-based systems in performance, flexibility, and
scalability.
We demonstrate three applications: a 2-D ortho photo maintained by the Bavarian State Survey, a
3-D AVHRR time series deployed as experimental Web service by the German Aerospace
Association, and a 4-D climate simulation database implemented by Max-Planck-Institute for
Meteorology. All of them use the identical underlying database technology, namely RasDaMan /
Oracle with the RasDaMan raster query language.
Why Raster databases?
Increasingly raster data are complementing vectorial data in geo applications. On the one hand,
storage facilities allow to keep substantial raster archives online, on the other hand acquisition of
raster data is rapidly getting easier and less expensive. Additionally, timeliness of raster imagery is
significantly higher as compared to vector data, due to the easier acquisition.
This advance in technology is paired by a rapidly increasing user demand for flexible, ubiquitous
access to large archives of raster images and maps of various types. A broad range of raster imagery
has to be considered in geo data management: grayscale and colour aerial images, multi- and
hyperspectral satellite images, topographical raster maps, RADAR and laser scan, and Digital
elevation Models (DEMs). Sometimes it is desirable to keep older versions, resulting in 3-D time
series data cubes.
Retrieval technology for large raster repositories, however, is lagging behind. Raster archives
today commonly are implemented in a file-based manner; databases serve only for meta data search,
but not for image retrieval itself. Hence, storage usually is driven by the data acquisition process
rather than by user access patterns. Moreover, versatile retrieval as known from database query
languages for alphanumeric data is impossible, not to speak of other services like transaction support,
The following example we consider as representative for an important, although still basic class of
queries: “Overlay selected channels of a multi-band satellite image with cadastral maps; colour all
areas in blue that would be flooded if water rose to level L, based on the DEM; do this for the geographic
area selected; zoom the result into my browser window”. Notably, each of the map items
usually has a distinct pixel resolution.
Aside from flexibility in task definition, there are several more arguments which advocate the use
of database systems. Query languages allow to define complex tasks to the server rather than a small
set of atomic steps as in procedural APIs; the consequence is that the query optimiser gains a lot of
freedom to rephrase the query optimally for the particular situation1. Further, application integration is
much higher because there is one central instance in charge of data integration and consistency. All in
all, file-based solutions frequently re-invent all the features which have been developed by database
technology over decades; using existing mature technology obviously is preferable.
A system offering all these classical database features on raster maps is RasDaMan2 (Baumann
1999, Ritsch 1999, Widmann 1999). It is implemented as middleware running on top of a relational
database system, offering a query language streamlined to semantically adequate raster retrieval.
In this paper, we present a series of cross-dimensional raster database case studies: a 2-D seamless
aerial image map, a 3-D seamless map extended in the time dimension, and a 4-D database of climate
simulation results. All use the same database platform, RasDaMan.
The remainder of this paper is organized as follows. In Section 2, a brief overview of the RasDa-
Man model, query language, and architecture is presented. Sections 3, 4, and 5 describe the 2-D, 3-D,
and 4-D databases. Section 6 reviews the state of the art, and Section 7 concludes the paper.
