Cellular Automata Models

overwhelmedblueearthAI and Robotics

Dec 1, 2013 (3 years and 8 months ago)

82 views

INTERNATIONAL INSTITUTE FOR GEO
-
INFORMATION SCIENCE AND EARTH OBSERVATION

Transition Rule Elicitation
Methods for Urban

Cellular Automata Models

Junfeng Jiao¹ and Luc Boerboom
2



¹ Texas A&M University, USA, hkujjf@gmail.com

2

ITC, Enschede, the Netherlands, boerboom@itc.nl

Distance Education Course on Spatial Decision Support Systems

2


Several PhD and MSc projects on CA modeling


Always theoretical, not empirical


Expansion oriented rather than land use change and
land use conflict


Academic studies


Similar problems in MAS?


Junfeng Jiao looked at


What are different approaches of rule formulation


How knowledge rich?


How to elicit knowledge?


How can we empirically enrich future research


Distance Education Course on Spatial Decision Support Systems

3

Content


CA models


Transition rules


Data vs. knowledge driven elicitation of
transition rules


Knowledge elicitation methods to gain
understanding


Distance Education Course on Spatial Decision Support Systems

4

Cellular automata


Complex dynamic
behaviors based on a
relatively simple set
of rules


Applied to lattice of
cells (i.e. spatial)
interacting with
their environment


Cells interact over
time through rules



Distance Education Course on Spatial Decision Support Systems

5

Simulations done for different purposes
or with different aspirations


Land use dynamics (White and Engelen,1993,
White, Engelen, et al., 1997),


Regional scale urbanization (Semboloni,1997;
White and Engelen, 1997),


Poly centricity (Wu, 1998; Cheng, 2003)),


Urban spatial development (Wu and Webster,
1998),


Urban growth and sprawl (Batty, Xie, et al.,
1999; Clarke, Hoppen, et al., 1997).


Distance Education Course on Spatial Decision Support Systems

6

Content


CA models


Transition rules


Data vs. knowledge driven elicitation of
transition rules


Rule elicitation methods to gain
understanding


Distance Education Course on Spatial Decision Support Systems

7

Transition rules


The control component


Determines the future cell state as a
function


Current state


States of surrounding cells.


TP
T+1

T = f(S
T
, NB
T
)


TP
T+1
Transition Potential of tested cell in time T + 1


S cell state at time T


NB Neighborhood states at time T

Distance Education Course on Spatial Decision Support Systems

8

Land use CA


More considerations than neighborhood such
as


Access,


Suitability





TP
T+1

T = f(S
T
, NB
T
, AC, SU …)


TP
T+1
Transition Potential of tested cell in time T + 1


S cell state at time T


NB Neighborhood states at time T


AC Accessibility effect


SU Suitability effect

Distance Education Course on Spatial Decision Support Systems

9

Transition to what?

Distance Education Course on Spatial Decision Support Systems

10


we see the need to explicitly
differentiate transition rules and
consider transition potential and
conflict resolution rules

Distance Education Course on Spatial Decision Support Systems

11

Classification of transition rules

Distance Education Course on Spatial Decision Support Systems

12

Advantage of classification of transition
rules


Suppose:

Poor residential state convertible to


institutional or to


high quality residential


As function of current concentration of each of these three
states.



Is this a neighborhood effect?


Or is it a conflict resolution effect?



Could be modeled as both, but semantics are different.


We seem to treat cells as agents, although we are certainly not
talking about agent
-
based systems.

Distance Education Course on Spatial Decision Support Systems

13

Content


CA models


Transition rules


Data vs. knowledge driven elicitation of
transition rules


Rule elicitation methods to gain
understanding


Distance Education Course on Spatial Decision Support Systems

14

Regression analysis


Modeler
identifies the possible influence factors

of
land use change (neighborhood effect, suitability
effect, and accessibility effect)


Modeler uses some methods to
measure
these
different effects.


Modeler overlays different land use maps and
identifies change areas

and selects random
samples
.


Modeler uses
regression analysis

to calculate future
land demand based on past urban development.





Examples: Wu (2000) and Sui and Zeng’s (2001).





n
i
i
i
R
C
TP
1
Distance Education Course on Spatial Decision Support Systems

15

artificial neural network


Modeler identifies the
possible influence factors

of land use
change (neighborhood effect, suitability effect, and accessibility
effect)


Modeler uses some methods to
measure
these different effects.


Modeler forms a
neural network
.


Modeler selects
functions
to link the neurons.


Modeler
trains ANN
with historic land use change


Poor insight of how influence factors relate to land use change





Examples: Li and Yeh (2001, 2002)


Predicted land use

Influence factors

Distance Education Course on Spatial Decision Support Systems

16

Visual observation (trial
-
error)


Modeler
identifies the possible influence factors

of land use change
(neighborhood effect, suitability effect, and accessibility effect)


Modeler uses some methods to
measure
these different effects.


Unlike previously, trial
-
error to calibrate distance functions for
predictive modeling.

Or assumptions for scenario development (difficult to assess)


Uncertainty as to interaction of effects i.e. attraction as source of
change or repulsion by others?


Can be knowledge driven




Examples:
www.riks.nl



Distance Education Course on Spatial Decision Support Systems

17

Analytical Hierarchy Process and Multi
-
Criteria Evaluation (AHP
-
MCE)


Modeler
identifies the possible factors determing
land use
change (neighborhood effect, suitability effect, and accessibility
effect)


Modeler defines hierarchy to represent relationship between
these factors the simulation objective


Importance of factors is expressed by decision makers (i.e.
normative/prescriptive or descriptive)





Example: Wu and Webster (1998): simplified this step and
determined the factors’ weights according to possible planning
policies and their own understanding of the urban development



Unlike evaluation practice, where focus is on decision maker.

Distance Education Course on Spatial Decision Support Systems

18


Most CA models are data driven, few are
knowledge driven


Little empirical basis for many assumed
spatial relations


Distance Education Course on Spatial Decision Support Systems

19

Content


CA models


Transition rules


Data vs. knowledge driven elicitation of
transition rules


Rule elicitation methods to gain
understanding


Distance Education Course on Spatial Decision Support Systems

20

Brainstorming on examples of empirical
enrichment with knowledge elicitation


Interview to define stakeholders


Document analysis to understand actual
transitions and competition and possibly
derive an idea of dominance of land
uses


Free
-
listing and sorting to identify
factors and arrive at transition rules for
different stakeholders


Distance Education Course on Spatial Decision Support Systems

21

Distance Education Course on Spatial Decision Support Systems

22

(personal)

Conclusions


Could have more complicated conflict
resolution rules than a weighted summation,
e.g. as function to degree of demand


supply
gap.


Interesting to look at knowledge
-
driven
methods to come closer to understanding of
land use changes.


Focus on empirical support for transition
rules.


Focus on knowledge elicitation


Getting closer to MAS (but of course not
really)