Sensitivity-driven simulation development: a case study in forced migration

Area of Science:

• Computational modeling
• Simulation science
• Computational social science

Background:

• Millions are forcibly displaced globally, necessitating accurate migration prediction for humanitarian aid.
• Agent-based models (ABMs) are crucial for simulating forced migration but require robust validation.
• Direct calibration to validation data can be challenging and may not improve model reliability.

Purpose of the Study:

• To introduce and demonstrate a novel approach, sensitivity-driven simulation development (SDSD), for refining computational models.
• To guide simulation development using sensitivity analysis (SA) without direct calibration to validation data.
• To enhance the reliability and reproducibility of agent-based models for simulating forced displacement.

Main Methods:

• Implementing sensitivity analysis (SA) to identify pivotal assumptions and parameters in agent-based models (ABMs).
• Utilizing SA results to focus model ruleset refinement on critical assumptions.
• Iteratively applying SA to assess and balance parameter sensitivity for improved model robustness.

Main Results:

• The sensitivity-driven simulation development (SDSD) approach successfully guided the refinement of agent-based models.
• Initial SA identified several pivotal parameters influencing model validation outcomes.
• Subsequent SA iterations demonstrated an average reduction of 54% in the relative sensitivity of these parameters.

Conclusions:

• SDSD offers an effective alternative to direct calibration for improving simulation model reliability.
• By focusing refinement on sensitive parameters, SDSD enhances the trustworthiness of computational models for forced migration.
• This approach contributes to verification, validation, and uncertainty quantification in computational science.