Spatiotemporal Change Pattern Mining: A Multi-disciplinary Perspective - Space-Time Integration in Geography and GIScience - page 302

each pixel represents an area of 0.5 by 0.5 degrees (Jones and Harris 2008 ). This
data type is commonly seen in climate and earth science. Finding spatial areas with
value changes may help identify patterns of environmental change.

17.2.3 Spatiotemporal Data

17.2.3.1 ST Event Reports

Discrete events may be represented as points in space and time. Each event has a
time stamp, a pair of coordinates, and possibly other information. For example, in
public health report data, each point rep- resents one disease case, along with its
location, time, and other information such as patients information. In public safety
data, each crime report is also a ST point with information such as crime type,
loss, etc.
Furthermore, ST event reports can also be represented in an aggregated manner
given a spatial zoning (e.g., jurisdictions) schema. The number of reports in each
zone can be totaled so that each zone (represented as a polygon in space) will
have a series of report totals at different time. These datasets may be used for
the detection of disease or crime outbreak patterns. Spatiotemporal raster data:
Similar to spatial raster data, STraster data are gridded space-time framework with
a number of functions defined over it. It is usually used to represent the distribution
of continuous variables over space and time. For example, the daily precipitation
data of the world shown in Fig. 17.4 a is available for a period of 106 years. The
entire dataset thus can be represented as ST raster dataset where each spatial grid
represent a fixed area (e.g., 0.5 by 0.5 degrees in latitude and longitude), and each
grid in time represent a day. Finding change pattern on such data (e.g., decline in
rainfall) can be used to confirm theories of climate changes.

17.2.3.2 Image Sequences

An image sequence contain images or snapshots of a spatial region at two time
instances (e.g., satellite image). Such datasets are commonly used in the area of
remote sensing to identify the changes that occur between the two time points.
Figure 17.4 b shows an example of image sequences where the two snapshots of
Washington D.C. were taken in 1984 and 2011 respectively (NASA 2011 ).

17.2.4 Statistical Models of Data

17.2.4.1 Temporal Statistical Model

Time series are usually modeled as i.i.d. samples drawn from a statistical

distribution. However, it is always the case that time-referenced attributes