A major challenge of novel scientific visualization using Augmented Reality is the accuracy of the user/camera position tracking. Many alternative techniques have been proposed, but still there is no general solution. Therefore, this paper presents a system that copes with different conditions and makes use of context information, e.g. available tracking quality, to select adequate Augmented Reality visualization methods. This way, users will automatically benefit from highquality visualizations if the system can estimate the pose of the realworld camera accurately enough. Otherwise, specially-designed alternative visualization techniques which require a less accurate positioning are used for the augmentation of real-world views. The proposed system makes use of multiple tracking systems and a simple estimation of the currently available overall accuracy of the pose estimation, used as context information to control the resulting visualization. Results of a prototypical implementation for visualization of 3D scientific flow data are presented to show the practicality.