Several industrial sectors (energy, naval, agroecology) show an increasing need for fluid flows monitoring and control/command innovations. Wind turbines structures are routinely monitored but their instantaneous aerodynamics are barely followed. Control solutions are also getting more and more popular. Accurate active control of wind turbines (e.g. farm control or pitch control) decrease maintenance costs, increase their life cycle and performances. Farm control will prevent the wakes induced by upstream wind turbines to reduce the efficiency of downstream wind turbines. Pitch control properly orientates blades to increase performances and alleviate wind turbine mechanical fatigue exposed to wind variability. All these control loops generally rely on real-time estimation and prediction of the airflow around wind turbines. Nacelle-mounted LIDAR provide valuable information at the farm and wind turbine scales whereas eTellTale sensors provide valuable information at the blade scale. However, this useful information are far from complete. Monitoring, predicting and preventing local freeze event is another big challenge in wine-growing and arboriculture. Industrial and academic actors (e.g., Weather Measure and INRAE) do provide solutions of sensors networks and weather downscaling at the plot scale. Besides, antifreeze towers can often locally protect part of the field from freezing. However, efficient solutions for monitoring and for driving the freezing prevention would necessitate portable computational fluid dynamic (CFD) predictions coupled with the sensors networks. To improve these airflow estimations and predictions further, we propose an innovative solution [1,2], which combines data, physical models and measurements in this purpose. Thanks to data and learning methods, this solution can run in real time even using limited computational resources. Besides, physical models provide the additional information, which are missing in databases, and enable aerodynamic and micro-weather predictions in the close future. Using a single noisy local sensor, we have obtained impressive results on wake flows (see Fig. 1). To scale up and address high-Reynolds-number flows, our solution now relies on reduced order models of turbulence models and on ITHACA-FV [3] – a C++ OpenFOAM-based library. REFERENCES [1]Resseguier, V., Picard, A., Mémin, E., & Chapron, B. (2021). Quantifying truncation-related uncertainties in unsteady fluid dynamics reduced order models. SIAM/ASA Journal of Uncertainty Quantification, 9(3), 1152-1183. [2]Resseguier, V., Ladvig, M., & Heitz, D. Real-time algorithm to estimate and predict unsteady flows from few measurements using reduced-order models. (2022). Journal of Computational Physics, 471, 111631. [3]Stabile, G., & Rozza, G. (2018). Finite volume POD-Galerkin stabilised reduced order methods for the parametrised incompressible Navier–Stokes equations. Computers & Fluids, 173, 273-284.