Data of Super-Resolution Wind Mapping with Deep Learning for Scalable Renewable Energy Planning

Datasets High-resolution model data (HR; ICON)o Type and variables: Hourly gridded wind fields from the ICON numerical weather prediction model, stored as an xarray.Dataset with 10 m and 100 m wind speed: • wind_speed_10m (time, latitude, longitude) • wind_speed_100m (time, latitude, longitude)o Spatial domain and resolution: Regular latitude–longitude grid over the Taiwan region, with • latitude: 21.38°N–26.38°N (41 grid points, ≈0.12° spacing) • longitude: 118.10°E–123.10°E (41 grid points, ≈0.12° spacing)o Temporal coverage and resolution: • time: 2022-12-01 00:00 to 2025-08-29 23:00 (24 072 hourly time steps)o Implementation details: • Stored as an xarray.Dataset with Dask chunking ((720, 41, 41) along time × latitude × longitude) to enable out-of-core processing. Low-resolution reanalysis data (LR; ERA5)o Type and variables: Hourly ERA5 wind fields on the same 41×41 grid, stored as an xarray.Dataset including both wind components and derived quantities: • u10, v10: 10 m zonal and meridional wind components • u100, v100: 100 m zonal and meridional wind components • wind_speed_10m, wind_speed_100m: wind speed magnitude at 10 m and 100 m • wind_dir_10m, wind_dir_100m: wind direction at 10 m and 100 mo Spatial domain and resolution: • latitude: 21.38°N–26.38°N (41 points, ≈0.12°) • longitude: 118.10°E–123.10°E (41 points, ≈0.12°)o Temporal coverage and resolution: • time: 2022-12-01 00:00 to 2025-08-29 23:00 (24 072 hourly time steps)o Implementation details: • Stored as an xarray.Dataset with Dask chunking ((3009, 11, 11) along time × latitude × longitude), allowing efficient subsetting and parallel computation. Ground-station observations (GS)o Type and variables: Hourly in-situ wind observations aggregated as a 2-D xarray.DataArray of station-wise mean wind speed: • wind_speed_mean (time, station_id)o Network and metadata: • 345 surface stations (station_id), each with associated metadata: – latitude, longitude: station coordinates – elevation: station height above sea level – grid_id: identifier of the collocated model grid cell – group: integer label grouping stations (e.g. by region or data source)o Temporal coverage and resolution: • time: 2023-01-01 00:00 to 2025-07-01 23:00 (21 912 hourly time steps)o Implementation details: • Stored as an xarray.DataArray with Dask chunking ((720, 64) for time × station_id) to handle long time series across many stations. High-resolution land reanalysis (BM; ERA5-Land benchmark)o Type and variables: Hourly near-surface wind speed from the ERA5-Land reanalysis, used as a benchmark dataset: • wind_speed (time, latitude, longitude)o Spatial domain and resolution: • latitude: 21.30°N–26.40°N (52 grid points, 0.10° spacing) • longitude: 118.10°E–123.20°E (52 grid points, 0.10° spacing) • This domain fully covers Taiwan and surrounding coastal waters with ≈0.1° (~10–12 km) grid spacing.o Temporal coverage and resolution: • time: 2023-01-01 00:00 to 2025-07-01 23:00 (21 912 hourly time steps)o Implementation details: • Stored as an xarray.DataArray with Dask chunking ((720, 41, 41) along time × latitude × longitude), enabling scalable computation over the full 3-D field.