Agent Analyst: Integration of ABM and GIS

1
Agent AnalystIntegration of ABM and GIS

• Hamid Ekbia
• hekbia_at_indiana.edu
• SLIS, IU
• March 2008

2
Motivation

• ABM and GIS
• On the rise, but separate
• research communities
• intellectual origins
• technologies
• Potentials
• For GIS change and movement
• Land cover/land use, predator/prey, network flows
  and congestions
• For ABM real-world environmental data
• Spatial analysis, resource management, disaster
  management

3
Three Birds, One Stone

• Science Models
• To explain a macroscopic regularity by providing
  micro-specifications
• Generative explanation
• Policy Simulations
• Scenarios in decision processes
• Inexpensive computation
• Manageable complexity
• Practice Hybrid
• Real-time measurements
• Dynamical data-driven analysis and simulation

4
Initiatives

• Funding
• NSF/E-science
• Venues
• Publications
• Blogspot http//gisagents.blogspot.com/
• Education
• Courses
• Software
• Agent Analyst

5
Outline

• Desert Tortoise Project
• Agent Based Modeling and Simulation
• Integration of ABM and GIS
• How does Agent Analyst Work?
• Urban Growth Model
• Bird Migration Sample Model

6
Desert Tortoise Project (DTP)

• Improve access to DT scientific information
• Facilitate understanding of DT science threats,
  population, and habitat
• Evaluate new and emerging technologies is
  assessing the status of DT
• Evaluate and develop tools and methods for
  knowledge management, discovery, modeling, and
  decision support

7
Data Collection in DTP

• GPS data from dog movement
• Distribution of location
• Distribution of speed

8
The Random-Walk Model

• Assumption
• Maximum speed of 4 m/sec
• ? Surface similarity with data

9
The Newtonian Model

• Physical constraints
• e.g., friction
• Speed and heading direction are related

10
The Behavioral Model

• Behavioral constraints
• Dogs on sight move rapidly on a straight line
• Dogs on scent move at moderate speed with
  differential headings
• Dogs on bay-up have low-velocity and high
  differential heading

11
Information in Movement A Comprehensive Model
12
Particles as Agents
13
Challenges

• How best to represent data?
• To capture change and movement
• How best to visualize data
• To fit and enhance human cognition
• How best to communicate data?
• To facilitate decision-making

14
Conceptual Shifts

• From
• Snapshots
• S1 - S2 ? ?s
• Objects/location
• Features and layers
• Queries
• What? Where?

• To
• Transitions
• S1 ?s ? S2
• Agents
• Actions and rules
• Scenarios
• What if?

15
Agent-Based Modeling Simulation

• Agents actions lead to system and agent
  evolution
• System-level patterns emerge from agent-level
  interactions
• Possible Scenarios can be developed

16
Agent Analyst ABM and GIS

• ArcGIS
• Modeling, analysis, and visualization
• Repast
• Toolkit developed by Argonne National Laboratory
• Free/Open Source
• Python, Java
• Repast Simphony

17
AA Architecture
Geoprocessing Tool
Calls
ArcMap/ArcGlobe
Calls
Refreshes Display
Refresh.exe
Inputs
Vector and/or Raster
18
AA How it works

• Create, edit, and run models within ArcGIS
  geoprocessing environment
• Setup, initialize, and schedule agents in Repast.
• Display agent movement and state change in ArcMap
  or ArcGlobe

19
AA Features

• GUI allows for the rapid creation of the modeling
  environment, agents, actions, and action
  scheduling
• Actions and rules are programmed using
  Not-Quite-Python (NQPy) for rapid model
  development
• NQPy models may be exported to Java to allow for
  more advanced modeling using RepastJ

20
Urban Growth Model
21
Data

• Study area 1 medium sized city and 2 small towns
• 38,000 parcels, 12,000 undeveloped parcels
• Parcels shapefile, raster layers of various
  constraints and suitability factors, dbf table
  for parcel neighborhood information

22
Physical constraints

• Deep slopes (gt 20 percent)
• Rivers and streams (with buffers)
• Fault lines (with buffers)
• Flood zones
• Combined into Hard constraints layer

23
Suitability Factors
Ownership
Access to Local Roads
Distance to Town
Distance to Major Roads
24
Neighborhood Factors

• Percentage of developed neighbors
• Local road access through developed neighbors (up
  to 2nd order neighbors)

25
Policy Constraints

• Conservation-oriented policy
• Agricultural land
• Rural living
• Parks
• Resource preservation
• Wild life corridor
• Soft constraints layer

26
Scenario Building

• Suitability parameters with user adjustable
  weights
• Distance to town
• Distance to major roads
• Land use policy constraints

27
Agent Rules

• Parcel agent initialization
• Parcel agent rules applied at each simulation
  cycle
• Random factors

28
Putting It All Together
29
Conservation Conscious Development
Non Conservation Conscious Development
30
Bird Migration Model
31
Salton Sea Bird-Banding

• Bird locations are tracked when banded,
  encountered, and recovered
• Analysis of banding data to study the importance
  of the Salton Sea to migratory birds
• Birds recovered from as far away as Russia and
  Peru
• 2/3 of all migratory birds in the continental US
  can be found at the Salton Sea

32
Data

• Targets
• Pintails
• Winter Gulf Coast or Salton Sea
• Summer Prairie Pothole Region
• Geese
• Winter Salton Sea or Gulf Coast
• Summer Banks Island, ANWR, Arctic North Slope
• Migration Trigger
• Photoperiod (rate of increase of hours of
  daylight. Usually around equinoxes.)

Snow Goose (1119 records)
Northern Pintail (9328 records)
33
Navigation Rules

• Generally follow North/South pattern
• Follow foothills of mountain ranges
• Follow coastlines
• Follow major rivers
• Barometric Pressure
• Solar, stellar, lunar, magnetic compass

34
Migration Rules

• Avoid flying over ocean and other large bodies of
  water
• Stop at large lakes to feed
• Avoid flying over mountain ranges

35
Constraints Data
Global Elevation (30m)
Distance from land
Distance from major rivers
Distance from Summer Targets
Distance from Winter Targets
Distance from Major Water Bodies
36
AGENT DECISION-MAKING

• Cellular Automata
• Evaluate neighboring cells for most suitable
• Evaluate probability scores for each neighboring
  cell of each constraint layer
• Weigh all probabilities to assign a master cell
  probability
• Local cost/opportunity surface
• Apply a random factor to select 1 cell among
  highest probability cells
• Moves point agent to the centroid of the selected
  cell

3 2 5 3 6 5 8 8 7
9 7 4 5 8 3 4 7 5
8 8 5 3 5 7 3 2 6
5 1 1 8 5 5 6 9 1
5 7 8 4 6 6 4 2 7
1 9 7 3 4 9 8 4 5
3 5 4 7 5 3 8 4 5
3 6 9 6 4 1 2 1 6
5 5 7 4 6 4 9 7 4
37
Agent Decision-Making
3 2 5 3 6 5 8 8 7
9 7 4 5 8 3 4 7 5
8 8 5 3 5 7 3 2 6
5 1 1 8 5 5 6 9 1
5 7 8 4 6 6 4 2 7
1 9 7 3 4 9 8 4 5
3 5 4 7 5 3 8 4 5
3 6 9 6 4 1 2 1 6
5 5 7 4 6 4 9 7 4

• Cellular Automata
• Evaluates neighboring cells for most suitable
• Evaluates probability scores for each neighboring
  cell of each constraint layer
• Weights all probabilities to assign a master cell
  probability
• Apply a random factor to select 1 cell among
  highest probability cells
• Moves point agent to the centroid of the selected
  cell

38
Bird Migration Model Demo
39
AA WEBSITE
http//www.institute.redlands.edu/AgentAnalyst
40
Courses at IU

• Summer Session I, 2008
• S603 Modeling in GIS
• S603 Agent-based Modeling
• Fall 2008
• S604 Modeling and Simulation

41
Thanks

• Tony Turner
• Naicong Li
• Nathan Strout
• Aditya Agrawal
• Paul Burgess
• Tim Krantz
• Jordan Henk