The National Center for Atmospheric Research (NCAR) – Sage Management Source Term Estimation (STE) algorithm consists of a combination of modeling systems, including the Second order Closure Integrated PUFF model (SCIPUFF), a hybrid Lagrangian-Eulerian Plume Model (LEPM), and its formal adjoint. This STE algorithm has been demonstrated in a prototype version of DTRA's HPAC system. The implementation in HPAC and use of this algorithm in more operationally relevant conditions uncovered new technical challenges to addressing the STE and downwind hazard refinement requirements of the chemical and biological defense community. While the algorithm meets the initial requirements, the current challenge is to make this algorithm perform more robustly in an operational environment. These challenges are associated with inaccuracies/uncertainties in the observational data used by the algorithm. In this presentation we will highlight the challenges these inaccuracies create for the STE algorithm and approaches for minimizing their effects on the algorithm accuracy. We illustrate the challenges posed by inconsistencies in the observations available to solve the STE problem, in particular, scenarios where the meteorological observations are not representative of the atmospheric conditions where the CB agent was released and dispersed. A method is required to estimate the uncertainty in the observations and combine this uncertainty with other information to bound the STE problem. In the work presented here we explore approaches to link the uncertainty in wind direction measurements with uncertainty estimates of the source location via the background error covariance matrix utilized by the adjoint minimization. Results from this evaluation that use data from FFT07 dispersion trial and an ensemble of “single realization” plume simulations created in Virtual THreat Response Emulation and Analysis Testbed (VTHREAT) that closely replicate this FFT07 trial are shown.