Cross-Resolution Combat Model Calibration Using Bootstrap Sampling

The US Air Force uses many combat simulation models to assist them in performing combat analyses. BRAWLER is a high-resolution air-to-air combat simulation model used for engagement-level analyses of few-on-few air combat. THUNDER is a low-resolution combat simulation model used for campaign-level analyses of theater-level warfare. BRAWLER is frequently used to ensure that THUNDER air-to-air inputs are valid. This thesis describes the confederation of THUNDER and BRAWLER by clearly showing how one particular BRAWLER output, the effectiveness of a missile type, is transformed into THUNDER air-to-air input data. Since BRAWLER is a stochastic simulation model, it is necessary to replicate a number of BRAWLER simulation runs in order to obtain a sufficiently accurate estimate of the mean missile effectiveness, a number that varies for each different BRAWLER combat scenario. This thesis focuses on using two different sequential methods to determine when the minimum number of BRAWLER runs has been performed to obtain a specified relative precision. One method uses classical statistical analysis techniques, while the other uses the more modern technique of bootstrap resampling. The performance of these two methods is compared.