Numerical weather prediction - currently the most widely-used tool for weather forecasting - relies on observations and computer models to produce forecasts. To produce accurate forecasts, computer models require an accurate initial state. The initial state is comprised of the previous model run's outcome, or the background information, and the observations. The most challenging aspect of the initial state to calculate is the background error covariance matrix, which is a measure of the uncertainty associated with the background information. The matrix can be calculated directly from the model, but that is computationally impractical. Current operational methods involve a less expensive broad estimation. This project examines the use of ensemble forecasting to produce an estimate for the background error covariance matrix, using an idealized supercell thunderstorm as a test case. This estimate is a more accurate representation than what is currently used in operational weather forecasting, which should produce a better estimate of the initial state and improve the model forecast.

Knowing the uncertainty associated with the model forecast can help forecasters determine the likelihood of severe weather that is currently not well predicted. This paper studies portions of the background error covariance matrix for implications regarding the predictability of the modeled supercell thunderstorm, as well as examining the interdependencies between variables that are missed by simpler operational matrices. This project found that the areas of highest uncertainty were associated with the strongest part of the storm, and that improving the areas of highest uncertainty would improve estimates throughout the storm.