1. An Introduction to Agent-Based Modeling for Geographic Processes
Anthony Jjumba, PhD
GeoPath Research and Consulting
Prince George, BC, Canada

2. Outline Complex Systems Agent Based Modeling Prototype ABM
Discussion and Questions

3. Complex Systems Geography: people, places and environments.
Spatial and temporal dimensions
Relationships and interactions
Patterns and processes
Complex Systems: the collective behaviors of the parts gives rise to properties that can hardly be inferred from the properties of the constituent parts.
Complexity Science as a field of study
“The whole is more than the sum of its parts” – Aristotle

4. Complex Systems Examples of complex systems A colony of ants
A transportation system
The economy
The human body
A 1980 Chevy truck? A school of fish? Space shuttling? Raising a family? A rolling ball?

5. Complex Systems Modeling Complex Systems
Reductionism
Holism
The properties of the system could be explained in terms of its parts.
The whole system cannot be determined or explained by its component parts alone.
The understanding of the parts leads to the understanding of the whole
To understand the whole, we must understand the relationships between the parts in the system.
Parts
Relationships

6. Complex Systems
Modeling approaches take on combination of both the Parts and the Relationships
Consideration of feedback loops
Positive feed-back (Reinforcing)
Negative feed-back (Regulative)
Examples: A rolling ball, predator-prey dynamics
Feed-back loops can make systems counter-intuitive or unclear
New highways into cities tend to lead to more traffic jams. Initially it decreases the waiting times but this in turn increases the desirability of driving a car.

7. Agent Based Modeling Modeling Complex Systems
Cause and effect are not always clear, e.g. ecosystem
Non-linear relationships, emergence, self-organization, bifurcations
Global coherent patterns emerge out of local interactions.
Examples: flocking behavior of birds, swarming behavior of fish
Simple local rules
'separation' (avoidance of crowding, or short-range repulsion)
'alignment' (steering towards the average heading of neighbors)
'cohesion' (steering towards the average position of neighbors)

8. Agent Based Modeling Modeling Complex Systems
Bottom-up modeling approaches
Cellular Automata
Grid of cells
Cell states
Neighborhood
Transition rules
Time steps
Agent-based modeling
Neighborhoods of influence
Individual actions and interactions
Differ from spatial interaction models (gravity, location allocation)
Static, zone aggregate

9. Agent Based Modeling Cellular Automata Active Cell Neighborhood Cell
Inactive Cell
K. Clarke

10. Agent Based Modeling

11. Agent Based Modeling Complex Systems Modeling Agent-based Modeling
Focused on decisions and actions of individuals
Adds behavioral realism
Solves the immobility of the cells
Environment
Sensors
Effectors
set
get
Context Analysis
Processing
Data
Agent
listener objects

12. Agent Based Modeling
Crooks, A. et al

13. Agent Based Modeling
Perez, L. et al

14. Agent Based Modeling
Heppenstall, A. et al

15. Agent Based Modeling
Schelling’s Model

16. Agent Based Modeling
Riots in Kibera Slum
Crooks, A. et al

17. ABM – Urban Systems Modeling land use patterns in the city
Von Thunen’s regional land use model (1826)
The Burgess concentric model (1925)
Urban Dynamics (Forrester, 1969) is used in reference to the multiplicity of interacting systems in a city: land use, transport system, employment, housing

18. ABM – Urban Systems
“Cities [are] problems in organized complexity, like the life sciences.  They present situations in which half a dozen or several dozen quantities are all varying simultaneously and subtly interconnected ways …The variables are many but they are not helter skelter; they are interrelated into an organic whole.”
– Jane Jacobs (1961)
The city is a complex system
Relationships and interactions between different components
Continuous change makes the city a dynamic environment

19. ABM – Urban Systems
Urban growth is characterised by population increase, redevelopment, expansion, densification
One concern for urban planners is to forecast future population growth and plan for possible changes in land use patterns
Often urban planning is top-down and sector based housing, transportation, land-use

20. ABM – Urban Systems Significant components in a city Urban Population
Housing development
Infrastructure for services
Planning Regulations
Expression of the city’s vision and priorities
Land use, Sustainability
Transportation
Accessibility
Mobility
Urban Economy
Employment and Income

21. ABM – Urban Systems Representing the city Conceptualization
Geographic Space
Key Actors/Stakeholders
Fusion of data
Physical (L1, L2),
Social, Perceptual (L3) Spaces to
Derive Place Abstractions (L4)
for Different Locations (N1, N2).
Crooks, A. et al

22. ABM – Urban Systems
Urban landscape is composed of irregular spatial units

23. ABM – Urban Systems
A regular grid is inappropriate for high resolution modeling
Building on work by Stevens and Dragicevic (2007)
Agent-based modeling using irregular spatial tessellations

24. ABM – Urban Systems Understanding the city for Transportation modeling
Household mobility
Forecast of future urban land use patterns
Tools that support planning

25. Prototype ABM
Develop an agent-based model for the simulation of urban growth
Show results from testing the model on data from one of the neighbourhoods the City of Chilliwack, BC
Results of sensitivity analysis to see:
how the changes in social and economic conditions of households affect the land use patterns
if changes in the geometry of the cadastral parcels influence land use patterns

26. Prototype ABM Agent iCity an Agent-based Model
Drivers of urban land-use change
Planning Agent
Implementation of planning policies for urban growth management
Policy based zoning restrictions

27. Prototype ABM Agent iCity an Agent-based Model
Drivers of urban land-use change
Households
size, income
proximity to preferable land use classes
neighborhood characteristics
Move to empty units that are located in good neighborhoods

28. Prototype ABM Agent iCity an Agent-based Model
Drivers of urban land-use change
Developer
Identifies most profitable locations using the proximity score
Industrial and Commercial Firms
Employment in the City

29. Prototype ABM JUMP Plug-in Agents' Module Subdivision Module
Repast Executor Module
Neighbourhood Analyser
Agent iCity
Planner
Industrial
Households
Developer
Retailer
Subdivision to cadastral parcels
Subdivision to city blocks

30. Prototype ABM

31. Prototype ABM
Geometrical transformation for the parcels

32. Prototype ABM Agent iCity Settings Monthly time intervals
Developer 6 months
Planner yearly
Random distribution for household income
Uniform distribution for number of people per household
Random distribution for property value
Uniform distribution for number of rooms in a dwelling
Planning Policy to be implemented

33. Prototype ABM Different Scenarios for urban growth Scenario 1
No Urban Planning Policy
Scenario 2
Agricultural Land Preservation Policy
Scenario 3
Urban Containment Policy
Other parameters
Household Income: $80,000
Property Value: $250,000

34. SCENARIO 1

35. 0.5
1.0 Km
t=0

36. 0.5
1.0 Km
t=12

37. 0.5
1.0 Km
t=24

38. 0.5
1.0 Km
t=36

39. 0.5
1.0 Km
t=48

40. 0.5
1.0 Km
t=60

41. Scenario 2

42. 0.5
1.0 Km
t=0

43. 0.5
1.0 Km
t=12

44. 0.5
1.0 Km
t=24

45. 0.5
1.0 Km
t=36

46. 0.5
1.0 Km
t=48

47. 0.5
1.0 Km
t=60

48. Scenario 3

49. 0.5
1.0 Km
t=0

50. 0.5
1.0 Km
t=12

51. 0.5
1.0 Km
t=24

52. 0.5
1.0 Km
t=36

53. 0.5
1.0 Km
t=48

54. 0.5
1.0 Km
t=60

55. Sensitivity Analysis Exploration of the variations in:
Household income
Property value

56. Sensitivity Analysis Variation in Household Income Low Income
Property value: $250,000
High Income
Income: $110,000
Property value: $250,000
t=24
t=24

57. Sensitivity Analysis Variation in Property Value Low Value High Value
Income: $80,000
Property value: $250,000
High Value
Income: $80,000
Property value: $800,000
t=24
t=24

58. Sensitivity Analysis Sensitivity to Multiple Runs + = Run 1 Run 2
Binary
Overlay . …

59. t=0

60. t=12

61. t=24

62. t=36

63. t=48

64. t=60

65. average

66. Prototype ABM
The modular nature of the model provides the ability to extend and add improvements in the future
Model can be incorporated into spatial decision support system
Cadastral parcels are easier to explain VS raster data
Subdivision algorithm is a new contribution with respect to existing modeling algorithms
Further work is need to incorporate a complete set of actors.

67. Outlook
Performing model testing, particularly calibration & validation
Interdisciplinary/Institutional collaboration
Increased computation power, proliferation of data collection sensors, cheaper storage point to increase application

68. Thank You