ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001 
t i 
* 
t i+1 
♦■■■ 
ti+2 
> 
Phase 1 
Phase 2 
Phase 3 
Changent QbangeTiiL. ChangeTwt ChangeTi+i2... ChangeTt+zt ChangeTi+jZ.. 
Figure 1. A Method For Mining Sequential Pattern From Geospatial Data. 
To extract sequential pattern from geographical data, we have 
to first describe the change of spatial objects. Predicates are 
used in this paper to describe changes, then WINEPI is used 
to mining frequent episodes. In our method, the process of 
mining sequential consists of three steps. Generally geospatial 
data is stored in commercial GIS systems. The geometry of 
the objects is represented in terms of a set coordinates or pixel 
in some Cartesian reference grid. We cannot mine sequential 
pattern from it directly. Therefore, we have to first infer 
necessary qualitative representation of geospatial information 
of relevant spatial objects. Figure 1 illustrates that in phase 1 
spatio-temporal date sets can be described at selected time 
points e.g. ti, ti+1, that yield snapshot predicates sets or 
qualitative spatial representation which are then used to detect 
changes. In phase two, the successive snapshot predicates 
sets are compared to obtain the changes between them. Since 
all the changes bear time stamp, we may take these changes 
as a event sequence then WINEPI could be use to find 
frequent episodes which means certain event is often following 
by anther event or certain event often occurs together with 
another event. Rules can also be generated from frequent 
episodes. This pattern should be meaningful. For example, in 
meteorology study, the changing meteorological fields, such at 
temperature, may be lead to the movement of specific 
meteorological sysem. We are now collaborating with 
meteorologist to test our method. 
So far, the framework of our method is presented. In the 
following next three sections, three phases of our method are 
discussed one by one in detail. 
3. QUALITATIVE REPRESENTATIONS 
OF SPATIAL INFORMATION 
Today’s GISs capture geographic information at the geometric 
level in quantitative terms. However, to apply sequential 
pattern seeking algorithm and other Al methods to find pattern 
from geospatial data, the spatial information have to be 
represented symbolically. Therefore, in our method, we have 
to first represent spatial information qualitatively. 
Qualitative spatial relations have been the subjects of 
extensive research over the last years and numerous 
formalisms and prototype systems exit. With the advance of 
spatial reasoning, in particular qualitative spatial reasoning, 
there is a surprisingly rich diversity of qualitative spatial 
representations addressing may different aspects of space 
including topology, orientation, shape, size and distance. In 
this paper we mainly discuss topological relations, qualitative 
directions and qualitative distances. 
3.1 Topological Relations 
Topology is perhaps the most fundamental aspect of space 
and certainly one of the extensively studied fields. The relation 
algebra for topological relations is based on the 4-intersection, 
which analyzes the intersections of the two objects’ boundaries 
and interiors. We illustrate the use of this algebra in Figure 2 
(Egenhofer and Herring 1990; Jayant Sharma, Flewelling et al. 
1994). It is interesting these relations is quite similar to initial 
set of eight pairwise disjoint and jointly exhaustive spatial 
relationship which is proposed independently in the filed of 
qualitative reasoning (Egenhofer and Golledge 1998). A 
comprehensive formal categorization of such binary 
topological relations between regions, lines and points has 
been developed that is based upon the comparison of the nine 
intersections between the interiors, boundaries and exteriors of 
the two objects (Egenhofer and Herring 1990). To keep this 
paper brief, we do not consider this complex situation. 
o> 
’■“** B 
disjoint 
t§> 
a O>b 
contains 
inside 
equal 
B ® A 
A <2>* 
meet 
covers 
covered By 
overlap 
Figure 2. Topological Relations 
It should be noticed that deriving the topological relation 
between geographic objects is very computationally 
expensive. To accelerate the process, the topological 
relationship could be compute at a relatively coarse resolution 
level using less expensive spatial algorithm such MBR data 
structure, then only promising candidates are selected for 
further accurate computing, (koperski 1999) 
3.2 Qualitative Directions 
Qualitative directions could be in several forms. The choice 
depends on the application. One method is based on 
projections, where the directions are defined by half-plane. 
Four directions can be defined, such that they are pare-wise 
opposites and each pair divided that plane into two half 
planes. The direction operation assigns for each pair of points 
a composition of two directions, e.g. south and east, for a total 
of four different directions. (Frank 1992) 
We use an equally useful construction as illustrated in Figure 
3. The compass is usually divided into four cardinal directions, 
often with subdivisions for a total of eight or more directions. 
This results in cone-shaped areas for which a symbolic