Large-eddy simulations of a yawed and tilted actuator disc in a veering neutral atmospheric boundary layer: five yaw misalignments and two tilts (AMR-Wind, IEA 22 MW scale)

This dataset contains seven large-eddy simulations (LES) of a single wind turbine rotor in a neutral atmospheric boundary layer with Coriolis-driven wind veer. The rotor is a uniform-thrust actuator disc of the IEA Wind 22 MW reference turbine size (D = 284 m, hub height 170 m, C_T = 0.8, no rotation). In five cases the rotor has -20, -10, 0, +10 and +20 degrees of yaw misalignment relative to the measured hub-height flow direction. Two aligned cases have a shaft tilt of +6 and -6 degrees in the solver's convention. All seven cases restart from the same precursor checkpoint and use the same saved inflow planes. The differences between cases come from the rotor setting only. The simulations were made in September 2026 with AMR-Wind (now Kynema-SGF) on two GPUs. Hub-height inflow: 9.36 m/s, turbulence intensity 0.056, friction velocity 0.381 m/s, flow direction 8.78 degrees from the x axis, veer about 4.4 degrees across the rotor. The domain is 7168 x 3072 x 1536 m with 16 m base cells and 8 m cells from the ground to 400 m (35.5 cells per rotor diameter). Each case covers 3449 s. The output is five cross-stream planes at 3, 6.5, 10, 13.5 and 20.5 D across the full domain width every 20 steps, and a hub-height plane over the whole domain every 100 steps. The planes are given in the solver's native format and as ASCII, with the input decks, solver logs and boundary-layer statistics. Known flaws. The solver did not output rotor force or power. The rotor is a constant-C_T disc without rotation or controller. In this code a positive tilt is nose-down (the +6 case), and the -6 case is the conventional nose-up tilt. The precursor failed one of five acceptance items (boundary-layer depth drift 9.4 percent against a 5 percent limit) and has no capping inversion. The planes are normal to x, and the flow is 8.78 degrees off x. There is one inflow condition and one realisation. The summary averages cover about 1100 s and have no uncertainty estimate. Details are in README.md in the record.