As the energy transition accelerates, improving wind energy efficiency and forecasting becomes increasingly critical. One key challenge lies in reconstructing high-fidelity atmospheric boundary layer (ABL) flow fields from sparse measurements, especially in regions influenced by wind turbines. This thesis explores the use of Latent Diffusion Models (LDMs) to reconstruct physically plausible ABL flow fields from limited spatial data. Where previous work focused on homogeneous, neutral ABL states, this research extends the methodology to more diverse and realistic conditions by including both stable and neutral boundary layers with embedded wind farms. A conditional diffusion model is trained in the latent space of an autoencoder (AE), using both local measurements and global atmospheric labels. The model is evaluated using statistical, physical, and spectral metrics, and its ability to generalize across multiple ABL regimes is analyzed. Results show that the model can generate realistic reconstructions from extremely sparse inputs, including turbine wake structures, and can potentially serve as a tool for initializing Large Eddy Simulations (LES). These findings mark an important step toward integrating generative models into operational wind forecasting and control systems.

Wind energy becomes more and more popular since it is environmentally friendly. Wind farm control is one of the most popular topics and it works on steering the wind farm to extract energy from wind as much as possible. Generally, the model capturing wake effects between turbines in the wind farm plays a role in wind farm control. The existing FLORIS model is considered suitable for wind farm control due to the fact that it has the ability or potential to capture wake features with reasonably computational costs. A drawback of the FLORIS model is the lack of dynamics, which is improved by developing the FLORIDyn model.

This thesis focuses on a Gaussian FLORIDyn model. The objective is to explore the possibility of improving the model accuracy by quantifying the associated uncertainty in the model parameters. Uncertainty quantification consisting of sensitivity analysis and Bayesian calibration is conducted based on a 3-Turbine case simulation using the UQLab software. Since a MCMC algorithm associated with Bayesian calibration requires to evaluate the FLORIDyn model multiple times, it can result in massive computational expenses when directly applying the computational model to the simulation. To deal with this, a surrogate model is first constructed to replace the original model. This thesis assesses two types of approaches for surrogate model construction which are the Kriging-based approach and the PCE-based approach. One approach is chosen after the comprehensive comparison in terms of accuracy and efficiency. The constructed surrogate model is then applied to the sensitivity analysis using Sobol' indices to investigate how each model parameter of interest affects the model output. Last, the high-fidelity SOWFA data are used as experimental data for Bayesian calibration. Compared to non-calibrated model outputs, calibrated model outputs are closer to the SOWFA data, which means that the accuracy of the FLORIDyn model is improved.