• anemometers
• atmosphere
• cfd
• fluid dynamics
• graph
• graph neural network
• machine learning
• mesoscale
• microscale
• Navier-Stokes
• Rans equations
• spatial temporal forecasting
• Stode
• turbulence
• weather
• wind energy
• wind fields
• wind forecasting
• wind resources
• Windsim
• wrf

The production of wind energy is one of the main priorities in the present days. To estimate the wind resources of a specific site, a Computational Fluid Dynamics (CFD) software is needed. Such a tool solves numerically the equations describing the dynamics of the air, to predict the wind fields in a given domain. The main approach used in the wind industry is a microscale approach, given resolutions and length scales of respectively tens of meters and some kilometers. One aim of the present thesis is to improve the estimations of the WindSim CFD software, exploiting atmospheric data. Such data can be provided by the Weather Research and Forecasting (WRF) model. It is a complete physics model describing the evolution of the atmosphere with a resolution of some kilometers in domains of thousands of kilometers: for that reason, it is referred as a mesoscale model. The usage of mesoscale data in microscale computation is known as the mesoscale-microscale coupling. A second challenge faced in this thesis is wind forecasting. The machine learning model adopted is the Spatial-Temporal Ordinary Differential Equation (STODE) Graph Neural Network. It uses past anemometers wind speed and wind direction time series for the forecasting of future ones. The improved fluid dynamics computations and the new forecasting tool achieved in this thesis can be used together to estimate the energy production of a wind turbine.