An understanding of the ramifications of an off-nominal event occurring during commercial space launches is becoming more necessary with the forecasted increase in frequency. Decision makers must be made aware of potential debris field size and location for the purpose of protecting nearby aircraft. Various debris modeling approaches have been developed to evaluate the effect of an off-nominal event, however current methods do not support the dynamic generation of Aircraft Hazard Areas (AHAs). A model to facilitate a real-time situational awareness of an off-nominal event can be proven to be an extremely useful tool. To support dynamic AHA generation, a debris field is created and analyzed through an accurate debris propagation model. During launch and reentry operations, the proposed trajectory does not always coincide with the real worlds during an off-nominal event resulting in the static model results being insufficient. Fast computation of the AHA is required to ensure online operations effectiveness. A parallel computing framework utilizing Graphic Processing Units (GPUs) is developed to aid in the computational speed up of the debris propagation model. The AHA is not shown as a contour of the debris, rather a bounding box of the contour. Due to the short comings of Kernel Density Estimation (KDE) in the form of computation time, clustering methods are explored to determine the feasibility and effectiveness of their use. The real-world component of this dynamic AHA generation model requires an AHA assessment metric to evaluate the parameter selection to balance the trade offs of accuracy and efficiency. Some models can be time consuming due to their large size so a learning-based parameter optimization model can be trained in an off-line format to combat this issue. The optimal policy generated from the learning model can support online operations to determine an accurate AHA. The learning-based models effectiveness is evaluated to support its use in dynamically generating AHAs for off-nominal events.