Agent Based Modelling

March 17th 2015 Land Surface Process Modelling (GEO4-4406) Judith Verstegen (J.A.Verstegen@uu.nl)

Topics

• ABM theory • Tools • Examples

ABM theory

Definition

• In an Agent Based Model (ABM), a system is represented as a set of agents, interacting with each other and their environment

• Also called: – Multi Agent System – Multi Agent Simulation – Agent Based Simulation – Individual Based Model

Differential equations vs. agents

Changes in the densities of prey population (N1) and the predator population (N2): • dN1/dt = b1N1 - k1N1N2 (1) • dN2/dt = k2N1N2 - d2N2 (2) In these equations b1 is the birth rate of the prey, d2 is the death rate of the predators, and k1 and k2 are constants.

http://ccl.northwestern.edu/papers/bio/long/

Why do we use ABMs?

• To try to reproduce observed system scale patterns potentially caused by individuals

• And to try to understand the processes causing these patterns

• To predict the future system state for a certain scenario. However, data availability is usually a huge problem in ABM

What is an agent?

• An agent is an autonomous, uniquely identifiable individual

What is an agent?

• An agent is an autonomous, uniquely identifiable individual

• Situated in an environment

bare soil

grass

What is an agent?

• An agent is an autonomous, uniquely identifiable individual

• Situated in an environment • Having behaviour, connecting

percepts (of the environment and other agents) to actions

bare soil

grass

grass Æ eat

What is an agent?

• An agent is an autonomous, uniquely identifiable individual

• Situated in an environment • Having behaviour, connecting

percepts (of the environment and other agents) to actions

• Having a state, consisting of one or a set of attributes, which changes over time

bare soil

grass

grass Æ eat

energy = energy + 10

Case 1

• We want to model decision making in the parliament

1. What should be the agents? 2. What is the environment? 3. What could be the behaviour? 4. What is the state (atrributes) of the agents?

Conceptual model ABM

(Potential) properties of ABMs

• emergence

“simple,  local,  individual  behaviour leads to complex and  ‘surprising’  global  patterns”

(Potential) properties of ABMs

• emergence • stochasticity and feedback loops Æ path-dependence

“Current  choices  determine  future  possibilities”

(Potential) properties of ABMs

• emergence • stochasticity Æ path-dependence • learning / adaptivity

“Agents  use  the  results  of  current  actions to adapt their future behaviour”

Educated agent example

When do we use field models and when ABMs? • Agents vs. fields (vector vs. raster)

field agent spatial variation continuous discrete objects attribute has value at all locations is linked to agent neighborhood adjacent cells /

Euclidean distance adjacent polygons / networks

processes behaviour of space as a whole

behaviour of a single agent

time environment changes agent’s  attributes  and/or location changes Æ macro-level patterns ‘emerge’

Coupling ABMs and fields

Coupling ABMs and fields

bare soil

grass

grass Æ eat

energy = energy + 10

growth  =  … grass = grass – grass_eaten + growth

Case 2

• When you want to model sediment transport in a river system, would you use an agent based model, field based model, or coupled model?

Tools

General or dedicated tools

• General tools: – General programming languages (Python, Java, C++) – Mathematics packages (MATLAB, Mathematica) – Spreadsheets with macros

• Dedicated ABM tools:

– NetLogo – Repast (Java or C++) – Mason (Java library) – Agent Analyst extension for ArcGIS

Model complexity vs. ease development

Examples

Learned before the break

• ABMs to reproduce, understand, predict

• Agent properties: uniquely identifiable, has an environment, has behaviour, has a state

• Potential ABM properties: emergence, path-dependence and learning

• General and dedicated tools

Usage

Agent Based models are used in, e.g.,: • Biology, ecology Æ animal migration, dispersion of plants • Sociology Æ crowd behaviour, segregation • Social geography Æ spatial planning, land use change • Economics Æ market models

Examples highlighted here: • traffic jams • ants • spatial planning

Traffic jams

Case 3

• Agent Based Model are not yet used widely in earth sciences. Discuss with your neighbour and find at least two potential earth science related systems/processes that can be represented with an ABM (5 mins)

Spatial planning

* LIGTENBERG, A. (2006) Exploring the use of Multi-Agent Systems for Interactive Multi-Actor Spatial Planning. Geo-Information and Remote Sensing. Wageningen, Wageningen University.

Conceptual Framework Agent Based Spatial Planning Model*

Reality

Spatial planning

• Three different groups have to agree on where to expand the city

• They have different initial desires considering the locations of expansion

Implemented learning: • Input = actions of the others • Output = changed desires

proposal

decisions

ideas about others updates

desire updates

Spatial planning With learning Without learning Average nr of proposals (-) 140 164 Average agreed area (ha) 52 45 Effectiveness (ha/proposal) 0.37 0.28

How often (% of time) a cell is selected for new urbanization with (left) and without (right) learning.

GIS

Summary

• In an Agent Based Model (ABM), a system is represented as a set of agents

• An agent is an autonomous entity, situated in an environment, having a state that can change by executing behaviour

• ABMs can have advantages over field based model (understanding, variability) but also disadvantages (data requirements)

• ABMs can be created in different general or dedicated toolsets

• ABMs can be used to simulate processes from very different domains at different scales

Practical

Install NetLogo from

http://ccl.northwestern.edu/netlogo/download.shtml at

C:\\temp

Practical

Carry out tutorial #1 and #3 on the NetLogo website: http://ccl.northwestern.edu/netlogo/faq.html If you have a Blackboard account: • Log in to Blackboard • Go to 2012 3 Land surface process modelling (GEO4-4406) • Go to Assessment Agent-based modeling • Answer the questions for tutorial #1 and #3

Assignment

Assignment 3: Agent-based modelling The tutorial on agent-based modelling by Macal & North (2010) in your reader explains what Agent-based models (ABM) are. It provides the structure of an ABM (agents, relationships & interactions and environment). Read the paper of Bennet and Tang (2006) (reading material for this topic). Write an essay that explains the relationships & interactions steering agent behaviour used in the Elk model, providing examples of rules used in the model. In addition, try to identify and explain some key ABM properties in the Elk model, like emergence and path-dependence. Write an approximately 2 page short paper or essay on this topic. Hand in before March 20th 17.00h by emailing Judith Verstegen, j.a.verstegen@uu.nl.

Questions?