Physically based models have been widely used around the world to study landslide occurrence. The accuracy in a physically based landslide susceptibility/hazard assessment depends mostly on the input parameters. In this research study, three physically based models' applicability and performance were assessed using deterministic and probabilistic approaches. It was carried out in a data-scarce environment: a tropical mountain basin of the Colombian Andes. TRIGRS, SLIP and Iverson models were applied with back analysis of the landslide events in the La Liboriana basin on May 18, 2015. The performance of the models was evaluated using ROC (receiver operating characteristic) analysis and %LR_class index. The results showed that the back analysis using landslide events could be a good alternative to define the input parameters for physically based models in data-scarce tropical mountain areas. The ROC analysis and %LR_class are considered useful techniques for assessing landslide modeling performance. The metric indices calculated should be seen as complementary information, and the drawbacks of these indices should be identified, as elucidated in this paper.

基于物理的模型已在世界范围内广泛用于研究滑坡的发生。基于物理的滑坡敏感性/危害评估的准确性主要取决于输入参数。在这项研究中，使用确定性和概率方法评估了三个基于物理模型的适用性和性能。它是在缺乏数据的环境中进行的：哥伦比亚安第斯山脉的热带山区盆地。将TRIGRS，SLIP和Iverson模型应用到2015年5月18日拉里博里亚纳盆地滑坡事件的反分析中。使用ROC（接收器工作特性）分析和％LR_类对模型的性能进行了评估指数。结果表明，使用滑坡事件进行的反分析可以很好地替代为数据稀少的热带山区基于物理模型定义输入参数的方法。ROC分析和％LR_类被认为是评估滑坡建模性能的有用技术。如本文所阐明的，应将计算出的度量指标视为补充信息，并应识别这些指标的缺点。