This study addresses some critical issues in the selection of input data for probabilistic seismic risk assessment at a local scale, considering available open-source information. It focuses on the derivation of vulnerability curves, which should be consistent with the local hazard and building practices, analysing the effect of (1) the record selection methodology, (2) the regression procedure followed for the analytical vulnerability derivation and (3) the consideration or not of local structural characteristics in the modelling process. It illustrates the significant differences in the results when distinct assumptions and sources of information—global, regional, or local—are used for the analysis. Based on a case study in Medellin, Colombia for assets representing the most vulnerable building classes in the city (unreinforced masonry houses and low code reinforced concrete buildings), accounting for more than 60% of its building stock, the effects of the previously mentioned parameters are studied. It is shown that hazard-consistent record selection is extremely important in the derivation of vulnerability models to use at a local scale, for sites with contributions from different tectonic regimes. Considerable variability is found in risk metrics such as Probable Maximum Loss curves and Average Annual Loss Ratios, rendering crucial the communication to decision-makers of these assumptions and the bias they could generate. Given the hazard characteristics of the site (with common low intensity events) it was seen that the lower tails of vulnerability curves have a large impact in loss results and should be given special attention.

考虑到可用的开源信息，本研究解决了在选择用于局部规模概率地震风险评估的输入数据时的一些关键问题。它侧重于脆弱性曲线的推导，应与当地的灾害和建筑实践相一致，分析 (1) 记录选择方法，(2) 分析脆弱性推导所遵循的回归程序和 (3)在建模过程中考虑或不考虑局部结构特征。它说明了在分析中使用不同的假设和信息来源（全球、区域或本地）时结果的显着差异。根据麦德林的案例研究，哥伦比亚对于代表该市最脆弱建筑类别的资产（无钢筋砌体房屋和低代码钢筋混凝土建筑），占其建筑存量的 60% 以上，研究了上述参数的影响。结果表明，对于具有不同构造制度贡献的地点，在推导脆弱性模型以在局部范围内使用时，灾害一致记录选择极为重要。在风险指标中发现了相当大的可变性，例如可能的最大损失曲线和平均年度损失率，这使得与决策者就这些假设及其可能产生的偏差进行沟通变得至关重要。