Temporal GIS for Meteorological Applications
Visualization, Representation, Analysis, Visualization, and Understanding
May Yuan
Department of Geography
College of Atmospheric and Geographic Sciences
The University of Oklahoma
Outline
• A brief history of time in GIS
• Temporal GIS for Meteorological Applications: a
new approach: visualization > representation >
analysis > visualization > understanding
• A case study
A brief history of time in GIS
GIS development: no room for time
•
•
•
•
Mapping tradition
Static views of the world
Space-centered
Using location to get information
Adding Time in RDBM
Time-stamped tables
1993
County
Population Avg. Income
1994
20,000
Nixon
17,000
County Population Avg. Income .
1995
Time-stamped values
19,800
Nixon
20,000
32,000
Cleveland 35,000
County Population Avg. Income
Nixon
20,900
21,000
Cleveland
35,000
32,000
Oklahoma
86,000
28,000
[0,20] U [41,51]
Tom
Gadia and Vaishnav (1985):
From
Department
[11, 49] 15K [11,44] Toys
[50, 54] 20K [45, 60] Shoes
[55, 60] 25K
[0, 20] 20K [0, 20] Hardware
[41, 51] 30K [41, 51] Clothing
[0,44] U [50, Now] [0,44] U [50, [0,44] U [50, Now]
Mary
Now] 25K
Credit
Gadia and Yeung (1988):
Time-stamped records
Stock Price
Salary
Name
[11,60] John
To
IBM
IBM
IBM
16
19
16
10-7-91 10:07am 10-15-91 4:35pm
10-15-91 4:35pm 10-30-91 4:57pm
10-30-91 4:57pm 11-2-91 12:53pm
IBM
25
11-2-91 12:53pm
11-5-91
Snodgrass and Ahn (1985):
2:02pm
Adding Time in GIS
• Snapshot: Time-stamping layers
• Space-time composite: Time-stamping spatial
objects (records)
• Spatiotemporal object: Time-stamping attributes
Snapshots
• Temporal time sets
Metro Denver Temporal GIS Project by Temporal GIS, Inc.
http://www.rrcc-online.com/~gey235/bpop.html
Space-Time Composite
• Spatial change over time
– History at location
– Cadastral mapping
3
5
2
1
Poly id
1
2
3
4
5
T1
4
T2
T3
Rural Rural Rural
Rural Urban Urban
Rural Rural Urban
Rural Rural Urban
Rural Rural Rural
Langran and Chrisman (1988)
T4
Rural
Urban
Urban
Urban
Urban
Spatiotempoal Object Model
• Spatial objects with beginning time and ending
time
U
t 3 S2 T U
t
S2
t2
S1
t3
T
T2
T1 t 1
S
t1
t3 U
T2
t 2 T3
t3
t2
T1
t1 S1
Agriculture
Urban
t 2 S2
Industry
Worboys (1992)
Change at Location
• History at a location
• Nothing moves
Space
Time
Semantics
Geographic semantics = something
(concrete or abstract) meaningful in
geographic worlds, including objects,
fields, ideas, authority, etc.
Commercial TGIS
• 4Datalink (2002)
• STEMgis (2003?)
• TerraSeer (2004)
4Datalink (2002)
• Time Travel Through Data
• Spatiotemporal objects with initial time (ti)
and finishing time (tf)
• AM/FM applications
No considerations on changes in geometry or attributes.
STEMgis (2003)
• Time-stamp spatial objects
• Hierarchical database
TerraSeer (2004)
• Object chains
• Public health and surveillance
Current Temporal GIS Technology
•
•
•
•
Mostly point data
Change-based information
Uniform change
Do not consider:
–
–
–
–
Change with spatial variation
Split
Merge
Development (temporal lineage)
TGIS based on “event” and “change”
Peuquet and Duan (1995)
Event-based
SpatioTemporal Data
Model (ESTDM)
Changes
from ti-1 to ti
Temporal GIS for Meteorological Applications
New temporal GIS approach
Shift our emphasis
• From “storage” : how data are ingested
– Observation based
– Organize data accordingly how data were collected by sensors or
observers
• To “analysis” : what we want to get from the data
– Process based
– Organize data according to how data were resulted from geographic
processes
Let’s start with a scenario
May 3, 1999 Oklahoma City
Tornado outbreaks
Visualization
• An entry point for
–
–
–
–
Investigation
Exploitation
Hypothesis generation
Understanding
• A means to
– Communicate results
– Discern correlation and relationship
– Reveal patterns and dynamics
SEE
what
THINK
why
how
CONCEPTUALIZE
SEE
UNDERSTANDING
Temporal GIS: “reversed engineering”
TEMPORAL GIS
HUMAN
UNDERSTANDING
SEE
SEE
CONCEPTUALIZE
what
what
REPRESENTATION
why
THINK
why
how
how
ANALYSIS
CONCEPTUALIZE
UNDERSTANDING
SEE
UNDERSTANDING
See ST Data > Think Geog. Processes
Digital Precipitation Arrays
7Z
8Z
9Z
10Z
11Z
12Z
13Z
14Z
15Z
What do we have?
• Observations from sensor networks: satellites, radars, or
ground-based stations at discrete points in time.
• Model simulation data
• Raster or point-based data
• Point-based data: may be transformed to raster data
through spatial interpolation.
What do we want?
Information beyond pixels and points:
•
•
•
•
•
How does “something” vary in space?
How does “something” change over time?
How does “something” progress in space?
How does “something” develop over time?
How often do similar “things” occur in space and time?
Want to know about “something”
What is “something”?
Event, Process and State
trigger
event
process
drive
state
spatiotemporal data
measured by
Events and Processes
Energy or mass
• An event introduces additional energy or mass into
a system
• Triggers processes to adjust the system
Time
State
•
•
•
•
Fields
Objects
Fields of objects
Objects of fields
Temporal GIS for understanding and
discovery
• Representation
– identify constructs for geographic processes
– organize ST data based on geographic
processes that generate the data
• Analysis
– elicit process signatures and their implications
– diagnose how a geographic process evolves
– examine how a geographic process relates to
its environment
– categorize and relate processes in space and
time
• Visualize
Issues
• Scale
• Granularity
• Uncertainty
Considerations
• Integration of fields and objects
• Hierarchies of events, processes, and states
Koestler (1967): holons
• Duality of a holon:
– Self-assertive tendency: preserve and assert its
individuality as a quasi autonomous whole;
– Integrative tendency: function as an integrated part of
an existing or evolving larger whole.
• Field of objects: rainfield of storms
• Object of fields: storm of rainfields
Weinberg (1975): General Systems Theory
• Small-number simple systems
– Individuals’ behaviors
– Mathematical
• Large-number simple systems
– Collective behaviors
– Statistics
• Middle-number complex systems
– Too large for math
– Too small for stats
– Both individually and collectively
Hierarchy Theory Is For …
Middle-number complex systems in which elements are…
• Few enough to be self-assertive and noticeably unique in
their behavior.
• Too numerous to be modeled one at a time with any
economy and understanding.
A hierarchy is necessary to understand middle-number
complex systems (Simon 1962).
Hierarchy Theory (HT)
• Reality may or may not be hierarchical.
• Hierarchy structures facilitate observations and
understanding.
• Processes at higher levels constrain processes at lower
levels.
• Fine details are related to large outcomes across levels.
• Scale is the function that relates holons and behavior
interconnections across levels.
Key HT Elements
•
•
•
•
•
•
Grain (resolution)
Scale (extent)
Identification of entities
Hierarchy of levels
Dynamics across levels
Incorporation of disturbances
Event
Zone
ts
in
Cell/point
ra
nt
co
a
at
D
n
io
t
Process
ga
e
gr
g
a
Squence
n
io
at
rm
fo
In
Temporal scale and granularity
Levels of fields and objects
Spatial scale and granularity
Levels of Organization
Extratropical Cycle
Squall-lines
Supercell
Hail
Tornado
Data, States, Processes, and Events
7Z
8Z
9Z
10Z
11Z
12Z
13Z
14Z
15Z
Objects formed through spatial aggregation
A process is formed…
Temporal aggregation of
state sequences
Extent 1
Process 1
Extent 2
Extent 3
State 2
State 3
Spatial aggregation of
observatory data
State 1
Levels of Observations
Occurrence of a rain event
Movement of a rainstorm
States of precipitation
Objects formed by temporal aggregation
Zones
Zone B
Levels of Spatiotemporal Aggregation
Zone C
Zone
T3
Zone
Zone A
T 1 T2
Sequence B
Sequence A
T5
T4
Sequence
T1
T2 T 3
T5
Sequence C
T4
Process A
Process
T5
T
T 1 T 2 3T 4 T 5
T5 T 6 T7 T8
Process B
T4
T5
Event B
T 22 T
23
Event
T
T1 T 2 3 T4 T 5
Event A
T5 T 6 T 7 T 8
204 mm/hr threshold
Sequences
Zone B
Levels of Spatiotemporal Aggregation
Zone C
Zone
Zone A
T3
T 1 T2
Sequence B
Sequence A
Sequence
Sequence
T5
T4
T1
T2 T 3
T5
Sequence C
T4
Process A
Process
T5
T
T 1 T 2 3T 4 T 5
T5 T 6 T7 T8
Process B
T4
T5
Event B
T 22 T
23
Event
T
T1 T 2 3 T4 T 5
Event A
T5 T 6 T 7 T 8
Process
Zone B
Levels of Spatiotemporal Aggregation
Zone C
Zone
Zone A
T3
T 1 T2
Sequence B
Sequence A
Sequence
T5
T4
T
T1 T2 3
T5
Sequence C
T4
Process A
Process
Process
T5
T
T 1 T 2 3T 4 T 5
T5 T 6 T7 T8
Process B
T4
T5
Event B
T 22 T
23
Event
T
T1 T 2 3 T4 T 5
Event A
T5 T 6 T 7 T 8
Event
Zone B
Levels of Spatiotemporal Aggregation
Zone C
Zone
Zone A
T3
T 1 T2
Sequence B
Sequence A
T5
T4
Sequence
T1
T2 T 3
T5
Sequence C
T4
Process A
Process
T5
T
T 1 T 2 3T 4 T 5
T5 T 6 T7 T8
Process B
T4
T5
Event B
T 22 T
23
Event
Event
T
T1 T 2 3 T4 T 5
Event A
T5 T 6 T 7 T 8
Data Structures
objects
fields
Time Series of Gridded Snapshots
A Case Study
Collaborator: Dr. John McIntosh
Data for Our Case Study
The Arkansas Red River Basin
Forecast Center generates
hourly radar derived digital
precipitation arrays
– 8760 raster layers per year
– Organized as temporal
snapshots and available online
Storm paths and velocity
Interactions with a geographic feature
How long did a storm last and
how much rainfall was received in this watershed?
4/15/98/00
116,670 m3
4/15/98/03
491,908 m3
4/15/98/01
2,193,379 m3
4/15/98/02
697,902 m3
Duration: 4 hours
Cumulative volume: 3,499,857 m3
4/15/98/04
0 m3
Find storms occurring at certain time and
duration
A query builder dialog to support queries
based on the modeled relationships and
object attribute values
Characterization indices
Index Type
Static
Object
Index Name
Elongation
Orientation
Object Relationship Distribution
Dynamic
Object
Growth
Granularity of Change
Object Relationship Relative Movement
Elongation
Orientation
Growth
Granularity of Change
Distribution
Relative Movement
Elongation Orientation
1.000
-0.184
1.000
-0.085
0.031
-0.123
0.033
0.217
-0.156
-0.126
0.052
Growth
Granularity
of Change Distribution
Relative
Movement
Cross Correlation Matrices
1.000
0.009
0.151
0.103
1.000
-0.321
-0.128
1.000
0.233
1.000
Granularity
of Change Distribution
Relative
Movement
Table 2. Index Cross Correlation Matrix for Events
process
Elongation Orientation
Elongation
1.000
Orientation
-0.096
1.000
Growth
-0.088
0.001
Granularity of Change
-0.158
-0.023
Distribution
0.003
0.014
Relative
Movement
0.167
-0.033
Table 3. Index Cross Correlation Matrix for Processes
event
Elongation
Orientation
Growth
Granularity of Change
Distribution
Relative Movement
Elongation Orientation
1.000
-0.184
1.000
-0.085
0.031
-0.123
0.033
0.217
-0.156
-0.126
0.052
Growth
1.000
0.241
-0.106
-0.022
Growth
1.000
0.009
0.151
0.103
1.000
-0.001
-0.251
1.000
-0.009
Granularity
of Change Distribution
1.000
-0.321
-0.128
1.000
0.233
1.000
Relative
Movement
1.000
Little shared information among the indices for both process and
event objects.
Granularity
Relative
Table 2. Index Cross Correlation Matrix for Events
Elongation
Orientation
Growth
of Change Distribution
Movement
Find storms with rotations
Similar change from T1 to T2
Cases from a cluster determined
by the six indices
Similar change from T1 to T2
Cases from a cluster determined by the six indices
a.
b.
c.
d.
Compare two processes
Dynamic time warping: the sequences are stretched
so that Imperfectly aligned common features align
14
12
10
8
6
4
2
0
11 33 5 5 7 79 11
13
19 21
29 25
31 33
9 Time
11 151317 15
17 23
19 252127 23
27 352937
Time
Find matching storms …
Return:
Query
storm systems with
similar behaviors
Categorize processes
hour 5
hour 4
hour 3
hour 2
hour 1
Figure 8. Test set of processes with durations between 3 and 5 hours.
4
Hierarchical Clustering
3
8
hour 5
hour 4
hour 3
hour 2
hour 1
hour 5
hour 4
hour 3
hour 2
hour 1
processes with durations between 3 and 5 hours.
13
17
2
Figure 10. Dendogram of test set clusters.
Figure 8.
20
9
10
7
1
14
hour 5
hour 4
hour 3 between 3 and 5 hours.
Figure 8. Test set of processes with durations
hour 2
1
Test set of processes with durations between
5 hours.
hourh5ou3r and
15
12
18
16
Figure 8.
hour 5
hour 4
hour 3
hour 2
hour 1
19
3
1
6
5
4
11
2
hour 5
hour 4
hour 3
hour 2
hour 1
hour 4
hour 3
hour 2
hour 1
hour 5
hour 4
hour 3
Test
set of8.processes
with
durations
between
3 and
5 hours.
Figure
Test set of
processes
with
durations
between
3 and 5 hours.hhoouurr 21
Events and Processes (Features) for
Data Retrieval
• As a catalog to identify what is of interest
• As a filter to specify time and area of interest
Features
specify
access
Radar data
Spatial bounds
Temporal bounds
Satellite data
In-situ observations
Events and Processes to Identify
Correlates
• Spatiotemporal relationships among features (e.g.
NDVI, ENSO, and LULC)
• Spatial lags
• Temporal lags
Features for impact analysis
• Use features to retrieve environmental
and socio-economic data
• Case based evaluation
• Case comparison
• Impacts along the evolution of a process
Temporal GIS for Meteorology
•
•
•
•
•
•
•
•
•
Ingest meteorological data
Analyze patterns and behaviors
Identify anomalies
Find spatiotemporal relationships among weather
events (e.g. teleconnections)
Incorporate model output and observations with
environmental data; Model validation
Elicit environmental correlates
Evaluate environmental consequences
Assess socio-economic impacts
Facilitate emergency planning, rescue, and decision
making
Questions, Comments?