BLOCHER_Julia_ABM_for_Migration_17.3

Report
AGENT-BASED MODELING
FOR MIGRATION
Julia M. Blocher
Sciences Po, 17 March 2014
What is agent-based modeling (ABM)?
• A computational method to model and simulate
complex systems in the real world
• Modeling agents (individual entities) that interact
with each other within an environment
• Explore the dynamics that arise from the
characteristics and behaviors of agents making
up biological, social, and other complex systems
When do you use ABM?
• When it is unrealistic, impossible, or unethical
to do real-life experiments
• To capture phenomena that can be difficult to
predict – or are counterintuitive
• When it’s important to include individuals describe activities rather than structure and
processes
• When you want to understand behavior – not
averages
What does ABM do well?
• Captures emergent phenomenon
• Non-linear behavior
• Thresholds
• If-then rules
• Nonlinear coupling (fluctuations, perturbations)
• Provides a ‘natural’ description of a complex
system
• Is flexible and relatively easy to program
• Stakeholder engagement is non-negotiable
Examples when this type of simulation
would be useful?
• Flow simulations: e.g. crowds at concerts,
panicking populations, transport and traffic –
implications for urban planning and evacuation
policies
• Market simulations: e.g. neural networks for
stock markets and trader behavior
• Organizational behavior: e.g. risk-taking in
banks
• Diffusion behavior: e.g. ‘social contagion,’
transmissible disease infection
Practical example
Source: Wilensky, U. (2003). NetLogo Ethnocentrism model. http://ccl.northwestern.edu/netlogo/models/Ethnocentrism. Center for
Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL.
What do you need for ABM?
• One or more agents:
• Self-contained
• Autonomous and self-directed
• State variables
• Communicative (social ability)
• Have memory, learning, strategies (adaption and path
dependence)
• Heterogeneous
• A representation of the environment
• Agent interactions
• Bounded rationality
AGENT-BASED MODELING FOR
ENVIRONMENT AND CLIMATE
CHANGE INDUCED MIGRATION
Why use ABM for environment and
migration?
• A quantitative approach that doesn’t assume
all people respond to climate in the same way
• Emphasis on unique context and
circumstances in real world phenomenon
• Recognizes that individual attitudes and
perceptions affect migration decision making
• Uses rules of behavior from real-life situations
and people in computer simulation model
Concretely – how do you build an ABM?
• Purpose
• Entities, state variables, and scales
Overview • Process overview and scheduling
Design
• Basic principles
• Agent and agent rule design
• Environment design
Details
• Initialization
• Input data
• Verification and validation
Agent-based migration model (RABMM-T)
Ten year ensemble for rate of total migration under non-scaled normal (N-) and
sigmoid (S-) rainfall scenarios tested. Error bars for S-EXTRAWET and SEXTRADRY.
Source: Smith, C.D. (2013) Modeling migration futures: development and testing of the RABMM-T. Sussex: Univ.
Sussex, UK.
Combining with other research methods
• Large N data
• E.g. census and household surveys, data mining
• Statistical distributions and other stylized facts
• Case studies
• E.g. Ethnographic studies, interviews
• Participatory methods
• Role-playing games and companion modeling (e.g.
Barreteau et al. (2001) )
• Geographic data: import topographies and GIS,
remote sensing
• Lab experiments to test computational models
Combining ABM with other methods
Flow rates in Al Zaa’tri refugee camp, northern Jordan, Sept. 3 2012. Source: UNOSAT
PRACTICAL
APPLICATIONS
Influence of climate on political drivers of
migration as adaptation in mixed livelihood
zones in North-Eastern Ethiopia
Conceptualizing the model
Drivers of migration. Source: Black, R., S.R. Bennett, S.M. Thomas, J. Beddington. Nature (2011) Vol 478, 27
Oct. 2011
Adaptive migration in mixed livelihood
zones in N.E. Ethiopia
• Question: What are the mechanisms underlying the
processes of influence of environmental change on
political drivers of migration – the indirect effects of CC
on migration decision-making process?
• Hypothesis: In past patterns (‘events’ and changes),
‘coping’ migration response increases overall, for most
individuals - rate of migration is decreased for
participants in local level policy schemes
Methodology & specific aims
• Treatment of large N data
• 20 year retrospective study
• 5-10 expert interviews at multiple levels of governance
• Questionnaires, reconstruction of migration histories
for 100-150 households of migrants and non-migrants
• Focus group discussions
• Participatory methods
• ABM validated by past patterns to predict migration
patterns for ‘future worlds’ simulated by high-end
warming scenarios (2, 4, and 6 degrees C warming)
Haraghe zones, Ethiopia
Livelihoods zones in Ethiopia. Source: FEWSNET (USAID)
Conceptualizing the model
Individual attitudes
and perceptions
High vulnerability:
need to change
situation (‘stress’)
Low vulnerability:
invest in migration
Vulnerability
assessment: Need
to change v.
employing existing
strategies
High vulnerability:
in-situ coping
Low vulnerability:
in-situ adaptation
Drivers of migration. Source: Black, R., S.R. Bennett, S.M. Thomas, J. Beddington. Nature (2011) Vol 478, 27
Oct. 2011
Agent-based migration model (RABMM-T)
Ten year ensemble for rate of total migration under non-scaled normal (N-) and
sigmoid (S-) rainfall scenarios tested. Error bars for S-EXTRAWET and SEXTRADRY.
Source: Smith, C.D. (2013) Modeling migration futures: development and testing of the RABMM-T.
Agent-based migration model (RABMM-T)
Ten year ensemble rate of total migration under the range of nondemographic scenarios.
Source: Smith, C.D. (2013) Modeling migration futures: development and testing of the RABMM-T.
What’s the catch?
• Trade off between generalizability and goodness
of fit with values observed empirically
• Survey data and statistics can leave out the most
vulnerable and marginalized
• How do you scale up the processes of a few
agents into the interactions among many agents?
• As complexity increases, the more difficult it is to
link the model’s structure to its behavior
(outcomes)
• Not as transparent as other methods
Sources
• Black, R., S.R. Bennett, S.M. Thomas, J. Beddington. Migration as adaptation. In:
•
•
•
•
•
•
•
Nature (2011) Vol. 478, 27 Oct. 2011
Black, R., et al. (2011). The effect of environmental change on human migration.
In: Global Environmental Change 21, Supplement 1(0): S3–S11.
Kniveton, D., C.D. Smith and R. Black (2012). Emerging migration flows in a
changing climate in dryland Africa. In: Nature Vol 2, pp. 444-447.
Railsback, S.F. and V. Grimm (2012). Agent-Based and Individual-Based
Modeling. Princeton: Princeton University Press, Princeton University.
Smith, C.D. (2013) Modeling migration futures: development and testing of the
RABMM-Tanzania. In: Climate and Development Vol 1 2014 [Accepted Sept
2013].
Smith, C.D. (2012) Assessing the Impact of Climate Change upon Migration in
Burkina Faso: An Agent-Based Modelling Approach. [DPhil Thesis] University of
Sussex.
Tacoli, C. (2011). The links between environmental change and migration; a
livelihoods approach. London, International Institute for Environment and
Development.
Wilensky, U. (2003). NetLogo Ethnocentrism model.
http://ccl.northwestern.edu/netlogo/models/Ethnocentrism. Center for Connected
Learning and Computer-Based Modeling, Northwestern University, Evanston, IL.
Contact me!
[email protected]
[email protected]
www.twitter.com/juliablocher
www.internal-displacement.org/

similar documents