Exploratory Spatial Data

Analysis
Part III
Analysis of spatial time series

Contents
Introduction: types of temporal variance
Exploration of events
Exploring changes of spatial patterns
Detecting spatio-temporal clusters

Analysis of thematic changes

Exploring changes of spatial patterns
Analysing local changes and their spatial
distribution
Investigating local behaviours
2

Types of Temporal Variance

Existential changes: appearing and
disappearing of objects; events
e.g. earthquakes, traffic incidents, observations
of rare plants or animals

Changes of spatial properties: location, size,

shape, orientation, altitude, etc.
e.g. movement of vehicles, urban growth

Changes of thematic properties (attributes)

e.g. district population, daily temperature
3

Event Exploration: Major

Questions
How are the events distributed in space?
at a particular time moment or all events that occurred
over a time period

How are the event occurrences distributed over

time?
E.g. how does event frequency vary?

How does the pattern of spatial distribution of the

events change over time?
How are the events distributed in space-time? Are
there any spatio-temporal clusters or/and trends?
4

Spatial Distribution of Events

The small circles represent the earthquakes
that occurred in Western Turkey and the
neighbourhood between 01.01.1976 and
30.12.1999

Here we see only the earthquakes

that occurred during 30 days from
15.05.1977 to 13.06.1977
By applying the temporal filter, we can
investigate the spatial distribution on any
time interval

Spatial Distribution of Event

Characteristics (1)
Earthquake
characteristics
(e.g. magnitudes)
can be represented
by circle sizes or
colors

Spatial Distribution of Event

Characteristics (2)
To explore the spatial
distribution of earthquake
characteristics, we can also
apply the thematic filter
(dynamic query)
Now we see only the
earthquakes with
magnitudes 4 and more that
occurred at depths not less
than 20 meters

Spatial Distribution of Event

Characteristics (3)
The temporal and thematic
filters may be combined

Now we see only the

earthquakes with
magnitudes 4 and more that
occurred at depths not less
than 20 meters during the
year 1999
8

Progress of Spatial Patterns over

Time
Map animation allows us to see how the spatial distribution of
events and their characteristics evolve over time
15.05.1977 - 13.06.1977

14.06.1977 - 13.07.1977

25.05.1977 - 23.06.1977

24.06.1977 - 23.07.1977

04.06.1977 - 03.07.1977

04.07.1977 - 02.08.1977

Each animation frame in this example covers 30-days time interval. The step between the
frames is 10 days. Hence, there is 20 days overlap between the adjacent frames.

Controls for Time Filtering and

Animation
The start of the currently
shown time interval

The slider can be dragged

(manual animation)

The end of the currently

shown time interval
We may fix either
interval start, or end,
or interval length

Interval length
Starts automatic
animation
Stops animation

Delay (decreases
animation speed)

Makes one step

forth or back

Step length

Fix interval start: on each step, step length is added to the interval end
Fix interval end: on each step, step length is added to the interval start
Fix interval length: on each step, step length is added to both interval start and end
If the step length is less than the fixed interval length, we will have a time overlap
between adjacent map states (animation frames)

10

Event Distribution in Space-Time

time

The perspective view (space-time

cube) uses the vertical dimension to
represent time while two other
dimensions represent geographical
coordinates. Events are placed in
the cube according to their spatial
locations and time of occurrence.

This is how a spatio-temporal

cluster looks like, i.e. sequence of
events close in space and time
11

Using Display Links for Exploring

Spatio-Temporal Distribution (1)
Let us select this
snake in the
cube and look
where these events
are on the map

We see that all but one events really

occurred very close to each other.
We can conclude that this is indeed a
spatio-temporal cluster and, hence,
there may be a relationship between
these events
12

Using Display Links for Exploring

Spatio-Temporal Distribution (2)
This is a spatial
cluster of events.
Let us see whether
they are close in
time

We see that the events seem to split

into two sequences with a certain
time lapse between them
13

Manipulating Perspective View

Time focuser: defines the time subinterval to show
Viewpoint position:
from what side and
what height we look
into the cube

Stretches the chosen time

interval to the whole
height of the cube
Dynamically updates the
view in the process of
time focusing and
viewpoint changing
14

Exploring Thematic Changes:

Major Questions
How are attribute values distributed over
the territory at a given time moment?
How do the attribute values at a given place
vary over time? (local behaviour pattern)
How does the overall spatial pattern of
value distribution evolve over time?
How are different behaviour patterns
distributed over the territory? Are there
spatial clusters of similar behaviours?
15

Exploring Spatial Distribution

Time-dependent data may be represented on a time map, which is
manipulated through time controls and, in particular, allows animation
For a time map, we
can use any
representation
method which is
suitable for timeirrelevant data.
Thus, a choropleth
map is good for
exploring spatial
patterns
Using the temporal controls, we can look at the spatial distribution at any time
moment. We can also run map animation and see how the pattern evolves
over time.
16

Time Map Properties

On a time map, the colour
scale or diagram sizes are
calculated taking into
account the minimum and
maximum attribute values
on the whole time period.
Therefore, for a particular
time moment, the darkest or
lightest shades (or largest/
smallest diagram sizes) may
not be present on the map.
Time maps allow all
interactive operations
available for usual maps,
e.g. outlier removal
17

Map Series
Map animation is good
for revealing general
trends in the development
of spatial patterns but
less suitable for a more
detailed investigation of
changes.
For comprehensive
comparison of states at
two or more time
moments, these states
need to be visible on the
screen simultaneously.
Best of all is to use
multiple maps displayed
in a common panel and
manipulated through a
common set of controls.
18

Exploring Distribution of Changes

Instead of original attribute
values, a time map can
represent changes, that is,
differences or ratios to the
previous moment or to any
selected moment
Here the maps correspond to
years 1990, 1991, and 1992
and represent differences to
the previous years. Positive
differences (i.e. increased
values) are shown in brown
and negative differences (i.e.
decreased values) in blue
19

How to Transform Data for

Change Exploration
1) Click here!

2) Push here!

3) Select one of
these options

4) Select either
differences or ratios
(5) Enter the
reference moment

20

10

Other Useful Data Transformations

Smoothing, value accumulation, residuals
For each time moment computes differences or ratios
to a particular object or value, to the mean or median
among all objects at this moment
Value scale transformation, e.g. logarithmic

The transformations can be applied in a sequence. Thus,

temporal aggregation may be applied to results of
temporal comparison, general comparison to the output
of temporal aggregation, and arithmetic transformation
to results of all previous transformations

21

Exploration of Behaviours (1)

The value flow symbols show us

the evolution of attribute values
(behaviour) at each location.
Unfortunately, symbol overlapping
creates significant inconveniences,
and zooming does not always help
22

11

Exploration of Behaviours (2)

Temporal smoothing allows us to disregard small

fluctuations and see development trends.
Here the values for each year have been
replaced by 5-year means. You can
compare to the previous variant and see the
effect of the smoothing.
23

Exploring Spatial Distribution of

Behaviours
This appears to be a spatial cluster
of similar behaviours

Around the Great Lakes, the

theft rates are high, but tend
to decrease in last years

There are also some

unusual behaviours
(behavioural outliers)
The theft rates along
the western coast
are, in general,
higher than inland

The theft rates are

relatively moderate,
but tend to grow
24

12

Data Transformations for Behaviour

Exploration
As with time maps, various data transformations
can be applied to value flow maps.

Here we have applied the comparison to the mean: the values for each
moment are replaced by their differences to the countrys mean at the
same moment. Yellow colour corresponds to positive differences, and
blue to negative. We have received a rather clear spatial pattern.

25

Comparing Behaviours and

Detecting General Trends
Putting all behaviours together, as in
these time graphs, makes their
comparison more convenient. When we
point on any line, we may see the
corresponding objects name and value.
Besides lines for individual objects, a
time graph can also show us the mean
behaviour (the line obtained by
connecting each year means) or the
median behaviour (obtained from
each year medians). This helps us to
understand the general development
trends for the whole set of objects.
26

13

Comparing Variations and

Trends of Different Attributes (1)
Time graphs are suitable for comparing
temporal variations of two or more
attributes.
Here we observe that the attributes
Motor vehicle theft rate and
Burglary rate have quite different
trends of general development.

27

Comparing Variations and

Trends of Different Attributes (2)
Smoothing (value averaging
over intervals) mitigates small
fluctuations and exposed the
trends more clearly.

28

14

Varying Level of Detail in Trend

Analysis
mean

The mean or median behaviour line

gives only a very coarse picture of the
general properties of value variation.
For a finer analysis, we may prefer to
look also at the quartile lines or even
smaller percentiles

median

quartiles

10, 20, , 90 percentiles

29

Combining Generic and Specific

Information Display (Example)

1) We select the counties with burglary rates

in 2000 fitting in the bottom 10%

2) and look how these countries behaved

over the whole time period 1960-2000

3) We also see where these counties are on

the map

4) and how motor vehicle theft

rates progressed there

30

15

Explanation of Some Controls

Focusing on a
value subrange

Selection box
(appears when
the mouse is
dragged)
Switches on/off
drawing of lines
for individual not
selected objects

Switch on/off
drawing of mean
and median lines

Specifies the number

of quantiles
(percentiles) to display

31

Summary
This lection was supposed to
explain what types of temporal variation
may happen to spatial phenomena
demonstrate that different tools are needed
for analysing each type of variation
show some state-of-the-art tools for
exploratory analysis of spatio-temporal data
explain what tasks each tool is good for
32

16