A MULTISCALE APPROACH TO DETECT SPATIAL-TEMPORAL OUTLIERS 
Tao Cheng 
Zhilin Li 
Department of Land Surveying and Geo-Informatics 
The Hong Kong Polytechnic University 
Hung Hom, Kowloon, 
Hong Kong 
Email: {lste; Iszlli}(@polyu.edu.hk 
WG: THS9 Uncertainty, Consistency and Accuracy of Data and Imagery 
Keywords: Detection, Dynamic, Multitemporal, Multiresolution, Thematic, Quality 
ABSTRACT 
A spatial outlier is a spatial referenced object whose non-spatial attribute values are significantly different from those of other 
spatially referenced objects in its spatial neighborhood. It represents locations that are significantly different from their 
neighborhoods even though they may not be significantly different from the entire population. Here we adopt this definition to 
spatio-temporal domain and define a spatial-temporal outlier (STO) to be a spatial-temporal referenced object whose thematic 
attribute values are significantly different from those of other spatially and temporally referenced objects in its spatial or/and 
temporal neighborhood. Identification of STOs can lead to the discovery of unexpected, interesting, and implicit knowledge, such 
as local instability. Many methods have been recently proposed to detect spatial outliers, but how to detect the temporal outliers 
or spatial-temporal outliers has been seldom discussed. In this paper we propose a multiscale approach to detect the STOs by 
evaluating the change between consecutive spatial and temporal scales. 
1. INTRODUCTION 
Outliers are data that appear inconsistent with respect to the 
remainder of the database (Barnett and Lewis, 1994). While in 
many cases these can be anomalies or noise, sometimes these 
represent rare or unusual events to be investigated further. In 
general, there are three direct approaches for outlier detection: 
distribution-based,  depth-based and distance-based. 
Distribution-based approaches use standard statistical 
distribution, depth-based technique map data objects into an 
m-dimensional information space (where m is the number of 
attribute) and distance-based approaches calculate the 
proportion of database objects that are a specified distance 
from a target object (Ng 2001). 
A spatial outlier is a spatial referenced object whose non- 
spatial attribute values are significantly different from those 
of other spatially referenced objects in its spatial 
neighborhood. It represents locations that are significantly 
different from their neighborhoods even though they may not 
be significantly different from the entire population (Shekhar, 
et al, 2003). Identification of spatial outliers can lead to the 
discovery of unexpected, interesting, and implicit knowledge, 
such as local instability. 
Many methods have been recently proposed to detect spatial 
outliers by the distribution-based approach. These methods 
can be broadly classified into two categories, namely 1-D 
(linear) outlier detection methods and multi-dimensional 
outlier detection methods (Shekhar, et al, 2003). The 1-D 
outlier detection algorithms consider the statistical 
distribution of non-spatial attribute values, ignoring the spatial 
relationships between items. The main idea is to fit the data 
set to a known standard distribution, and develop a test based 
on distribution properties (Barnett and Lewis, 1994; Johnson, 
1992). Multi-dimensional outlier methods can be further 
1008 
grouped into two categories, namely homogeneous multi- 
dimensional metric based methods and spatial methods. The 
homogeneous multi-dimensional metric based methods do not 
distinguish between attribute dimensions and geo-spatial 
dimensions, and use all dimensions for defining neighborhood 
as well as for comparison. In the spatial methods, spatial 
attributes are used to characterize location, neighborhood, and 
distance, and non-spatial attribute dimensions are used to 
compare a spatially referenced object to its neighbors. Among 
others, Shekhar et al (2003) developed a unified modeling 
framework and identify efficient computational structure and 
strategies for detecting spatial outliers based on a single non- 
spatial attribute from a data set. 
Depth-based techniques are also applied extensively as 
clustering for spatial outlier detection, i.e. identifying the 
neighborhood of an object based on spatial relationship, and 
considering the proximity factor as the main basis for 
deciding if an object is an outlier with respect to neighboring 
objects or to a cluster. The limitation of these approaches is 
ignoring the influence of some of the underlying spatial 
objects that might be different at different spatial locations 
despite the close proximity, i.e., the semantic relationship is 
not considered in the clustering. An exception is that, Adam et 
al. (2004) identified spatial outliers by taking into account the 
spatial and semantic relationships among the objects. 
Ng (2001) used distance-based measures to detect unusual 
paths in two-dimensional space traced by individuals through 
a monitored environment. These measures allow the 
identification of unusual trajectories based on entry/exit 
points, speed and geometry; these trajectories may correspond 
to unwanted behaviors such as theft. Other methods used in 
data mining such as classification and aggregation, are also 
applied in spatial outlier detection (Miller, 2003). 
Intern 
ve 
2.] 
Th 
ba 
clt 
kn 
ba: 
cla 
(se 
ne 
dai 
reg 
2.2 
Th 
in 
mi 
2.3 
Th 
wi