Analysis of error propagation within hierarchical air combat models

Abstract

Operations research analysts often use a hierarchy of combat models to provide insight to military decision makers. Briefly, lower-level, higher-resolution models provide input to higher-level, lower-resolution models. This allows analysts to explore how engineering and tactics changes can affect campaign effectiveness. This thesis builds upon previous research and examines various methods for employing distributions of engagement-level model outputs as input to campaign-level models, instead of just using the average. We contrast methods for linking the engagement-level model to the campaign-level model. Previous research indicates that when expected values alone are propagated through layers of combat models, the final results will likely be biased, and risk underestimated. An air-to-air engagement model is developed to generate a data library that is used as input in a stochastic Lanchester campaign model. A variety of sampling methods are employed to sample from the engagement model's output data library to provide input to the campaign model. The results indicate that the manner in which the engagement and campaign models are linked has substantial impact on the estimates of operational effectiveness and risk. Additionally, our research illustrates how running a designed experiment on the engagement-level model, to generate a library of data that can be linked to the campaign-level model, can support robust decision making.