The increasing frequency of extreme and clear-air turbulence (CAT) in aviation poses a significant challenge, as many of these events are invisible to current onboard sensing and forecasting tools. Effective mitigation strategies, such as gust load alleviation or buffet suppression, require anticipatory control systems. These systems are critically dependent on robust, real-time estimation of the high-dimensional flow state. This estimation, however, is operationally constrained to using sparse, non-intrusive, wall-mounted sensors. Recent advances in data-driven reduced-order modelling (ROM) offer a path to solving this problem. Latent-space models, such as observableaugmented autoencoders, can create physically interpretable, low-dimensional representations of complex flow phenomena. AI-driven estimators, including recurrent neural networks, can then map sparse sensor histories to these latent states. This presentation will explore how these components—nonlinear ROMs, AI-driven estimators, and interpretable sparse sensing—converge to enable the robust state estimation required for next-generation control systems, including Model Predictive Control and Model-Based Reinforcement Learning.