Agent Based Modelling and Scenario Analysis in the Policy Process Lessons from the FIRMA project informed by the first morning of the European Water Scenarios Workshop Scott Moss, Centre for Policy Modelling
Andrea Tilche’s objectives Model supported consistent scenarios Integrating qualitative and quantitative approaches Richly expressive analysis Story lines and modelling approaches Transparent and well documented Stakeholder involvement Scenarios support risk based a risk based approach to decision making Scott Moss, Centre for Policy Modelling
Joe Alcamo’s scenario vision Qualitative and quantitative scenarios different but complementary Iterative process between experts, modellers and stakeholders/policy makers to develop scenarios “Final scenario” necessary for wider use “We have to be visionary but also practical” Scott Moss, Centre for Policy Modelling
Scenario Development Process Scott Moss, Centre for Policy Modelling
Models from the FIRMA Project Validation Integration of qualitative and quantitative Integration of natural science and social behaviour in single model Scott Moss, Centre for Policy Modelling
Centre for Policy Modelling Validation Agents designed to describe real human or organisational actors Individual behaviour Social interaction how with whom Rules can be validated by domain experts (eg stakeholders) Scott Moss, Centre for Policy Modelling
Model Structure - Overall Structure Policy Agent Ownership Frequency Volume Households Ground Temperature Rainfall Sunshine Aggregate Demand Scott Moss, Centre for Policy Modelling
An Example Social Structure - Global Biased - Locally Biased - Self Biased Scott Moss, Centre for Policy Modelling
Statistical validation – model version 2 Scott Moss, Centre for Policy Modelling
Social Behaviour and Climate Change Reference runs MH climate change Individual Social Social influence: individual=33%, social=80%. All runs: 1973=100. Scenarios broadly correspond to EA reference scenarios: individual (alpha and beta); social (gamma and delta). Scott Moss, Centre for Policy Modelling
Model of UK Foresight Scenario OFV specified by Environment Agency Water saving devices specified by Agency Increased volatility turned out to be a bug Scott Moss, Centre for Policy Modelling
Some simulation general results The more agents influence one another, the less variability across scenarios in aggregate water for demand Confirmed that agent interaction and threshold behaviour generates heavy-tailed distributions Any population distribution has undefined variance (and maybe undefined mean). Conforms to observed domestic water demand Scott Moss, Centre for Policy Modelling
Two approaches Compared Agent based: Volatile Unpredictable Dynamic simulation: Smooth scenarios Predictable through convergence Scott Moss, Centre for Policy Modelling
Centre for Policy Modelling Risk Clustered volatility results from inter alia Threshold behaviour by individuals Dense patterns of social interaction with neighbours/acquaintances Implies heavy tailed population distributions or (more plausibly?) no population distribution Actuarial science based on finite-variance distributions If distributions change endogenously through social interaction, does this affect prevailing risk/uncertainty assessment techniques? Scott Moss, Centre for Policy Modelling
Centre for Policy Modelling Environmental Scenario Objectives (paraphrased from Alcamo and Henrichs) To raise awareness about environmental problems To take into account large time and spatial scales To illustrate alternative environmental futures To explore alternative policy pathways To assess policy robustness to different conditions To investigate connections among future problems To combine qualitative and quantitative information Scott Moss, Centre for Policy Modelling