ul summary 
owing data 
howing the 
wowing data 
is our belief 
the success 
large spatial 
ODEL 
odel, let us 
mage object. 
XAD such as 
ation of the 
el precision, 
its used for 
meters used 
ity, etc.. The 
may be its 
of polygons 
sional array 
of rules to 
a rule like: 
less than 20, 
vater".) It is 
tional data 
e data (e.g. 
ta operation 
iandle by an 
| Wilkinson, 
Id define an 
th various 
methods to 
h class rules 
les[Hughes, 
iandle large 
'e: first, the 
'stract Data 
instructured 
3]. It allows 
| manipulate 
ystem to be 
its users. In 
unformatted 
formation is 
nd method 
classes, and 
r to define 
sses[Deux et 
re raw data 
and provide 
em through 
d as a whole 
  
and no abstraction methods are provided to the 
users. In addition most commercial systems 
generally do not allow the query language to be 
extended to capture the semantics of large objects. 
Therefore new functions and operators defined by 
users may not be able to use internal data indexing 
and clustering techniques provided by the system to 
manage BLOBs and make the user defined methods 
or operators inefficient. 
To make large object manageable within a GIS 
environment, some new methods such as data 
compression, data ordering, controlled data 
redundancy, indexing and data browsing needs to be 
investigated, evaluated and implemented based on 
our data model and the possible GIS query 
patterns. This paper will mainly discuss these 
methods and present our implementation 
strategies. 
Using the object oriented data model, we have 
implemented a visual based schema browser to 
help users to define and modify the data base 
schema[Zhou and Wilkinson, 1993]. An example of 
the data base schema browser is shown in Figure 1. 
5. LARGE SPATIAL OBJECT HANDLING 
METHODS 
Many current generation GISs have been designed 
to effectively manage point, line, surface and their 
attribute data. However, large spatial objects such 
as remotely sensed image, scanned document and 
DTM are series of large multidimensional arrays 
With various temporal, spatial and spectral 
characteristics. They are unformatted and quite 
often require long delayed processing in order to 
answer user's query or provide other useful 
information. Interactive, network based GIS 
queries may be very slow when large objects are 
concerned. Therefore efficient support of large 
objects must be designed into GISs. 
In order to manage large spatial objects effectively 
and efficiently, special techniques that allow 
applications to quickly store and retrieve large 
spatial object onto and from secondary memory 
devices such as a disk need to be developed. The 
traditional methods of storing large objects, such as 
imagery is linear allocation where the data is laid 
out linearly by a nested traversal of very 
dimension in a predefined order. The method may 
be optimized when the whole data of the large 
object is needed. Due to the limitation of the 
Current viewing system such as computer screen(e.g. 
1000 by 1000 screen pixels vs. 20 by 20 K scanned 
Map) and the nature of data usage(e.g. quite often 
only a small part of data is used at a time), the 
linear type data storage can be very inefficient in 
many GIS applications. This situation will get 
worse when the dimension and size of data 
increase. 
5.1 Data Compression 
One area that needs to be investigated is how we 
can reduce the amount of large object data while 
keeping most GIS queries satisfied. One obvious 
answer is data compression techniques such as 
reducing data quantization level or resolution, or 
using Run Length Encoding and JPEG compression 
[Wallace, 1991]. These techniques can significantly 
reduce the size of large objects and make large 
objects more manageable in data bases. 
The compression is one of the techniques for data 
reduction. It takes advantage of patterns in image 
data by removing redundancies and in effect, 
squeezing the data to maximize the information 
contained in each byte. Scanned maps, digital air 
photos and satellite images are considered good 
candidates for compression because of high 
correlation in the image data. The compression can 
be lossless(e.g. the decompressed data is identical 
to the original, Run Length compression is an 
example) or lossy (some information is lost. JPEG is 
an example for this type of compression). Lossy 
compression generally yields higher compression 
ratios than lossless. For some GIS applications, 
such as image display, this is a reasonable trade- 
off. However, some applications will not allow the 
lossy compression because of the precision 
requirement. 
Depending on the nature of image data, either 
lossless or lossy compression may significantly 
reduce the size of the image data to be managed. 
For scanned maps, lossless compression typically 
reduces the size by a factor of 2-10. This allows the 
database system to transfer more information with 
fewer bytes, thus making effective use of available 
band width. This can be very significant for 
network based GIS systems. 
Another method to perform data reduction is to 
reduce the data resolution and quantization level. 
This technique can significantly reduce the amount 
of data while satisfying most GIS queries. Table 1 
gives the example of various data reduction 
methods and their data sizes before and after the 
data reduction. Since most GIS queries are "Gold 
Mining” or “Landscape” type that do not require 
detailed information in the first several queries, 
this technique can be significant for GIS 
applications. 
215