Wind conditions and output power characteristics of a wind farm in Japan are evaluated with highly resolved weather predictions from the so-called cloud resolving storm simulator. One year of 30-hour-ahead predictions with 2-km spatial resolution and 1-hour time resolution are evaluated against 10-minute averaged measurements (averaged to hourly data) from the wind farm. Also, extremely detailed shorter-term predictions with 200-m spatial resolution and 1-second time resolution are evaluated against 1-Hz measurements. For the hourly data, wind speeds are predicted with an RMSE of 3.0 to 3.5m/s, and wind power with about 0.3 per unit. Wind direction is predicted with a standard deviation of errors of 16 degrees to 28 degrees for hourly data, and generally below 10 degrees for the 1-Hz data. We show that wind power variabilityhere in terms of incrementscan be assessed on the timescale of several hours. The measured and predicted wind spectra are found similar on both short and long timescales.