This paper presents an approach named sensitivity-driven simulation development (SDSD), where the use of sensitivity analysis (SA) guides the focus of further simulation development and refinement efforts, avoiding direct calibration to validation data. SA identifies assumptions that are particularly pivotal to the validation result, and in response model ruleset refinement resolves those assumptions in greater detail, balancing the sensitivity more evenly across the different assumptions and parameters. We implement and demonstrate our approach to refine agent-based models of forcibly displaced people in neighbouring countries. Over 70.8 million people are forcibly displaced worldwide, of which 26 million are refugees fleeing from armed conflicts, violence, natural disaster or famine. Predicting forced migration movements is important today, as it can help governments and NGOs to effectively assist vulnerable migrants and efficiently allocate humanitarian resources. We use an initial SA iteration to steer the simulation development process and identify several pivotal parameters. We then show that we are able to reduce the relative sensitivity of these parameters in a secondary SA iteration by approximately 54% on average.

This article is part of the theme issue ‘Reliability and reproducibility in computational science: implementing verification, validation and uncertainty quantification in silico’.

本文提出了一种名为灵敏度驱动模拟开发 (SDSD) 的方法，其中灵敏度分析 (SA) 的使用指导进一步模拟开发和改进工作的重点，避免直接校准验证数据。SA 识别对验证结果特别关键的假设，并作为响应模型规则集细化更详细地解决这些假设，在不同的假设和参数之间更均匀地平衡敏感性。我们实施并展示了我们改进邻国被迫流离失所者的基于代理的模型的方法。全世界有超过 7080 万人被迫流离失所，其中 2600 万人是逃离武装冲突、暴力、自然灾害或饥荒的难民。在今天，预测被迫迁移很重要，因为它可以帮助政府和非政府组织有效地帮助弱势移民并有效地分配人道主义资源。我们使用初始 SA 迭代来引导模拟开发过程并确定几个关键参数。然后，我们证明我们能够在二次 SA 迭代中将这些参数的相对敏感度平均降低约 54%。

本文是主题问题“计算科学中的可靠性和再现性：在计算机中实现验证、确认和不确定性量化”的一部分。