# GlobalMask

The released GlobalMask dataset is divided into fixed training and testing
splits. Both files contain the same arrays and dimensions, but differ in the
number of global grid-cell samples.

| Split | Samples | File |
|---|---:|---|
| Training | 3373 | `train/GlobalMask_train.npz` |
| Testing | 852 | `test/GlobalMask_test.npz` |

## Variables

- `ed_simulation_x` — monthly ED input features
- `ed_simulation_y` — age-specific ED simulation targets
- `ed_simulation_pft_bl` — annual age-specific ED broadleaf PFT fractions
- `ed_simulation_pft_nl` — annual age-specific ED needleleaf PFT fractions
- `ed_simulation_pft_gs` — annual age-specific ED grass-and-shrub PFT fractions
- `lidar_age_weight_fraction` — LiDAR-derived forest-age fractions
- `esa_cci_bl_fraction` — annual ESA CCI broadleaf PFT fractions
- `esa_cci_nl_fraction` — annual ESA CCI needleleaf PFT fractions
- `esa_cci_gs_fraction` — annual ESA CCI grass-and-shrub PFT fractions

All released arrays use sample-first orientation.

Detailed dimensions and normalization statistics are provided in:

- `../metadata/dimension_definitions.md`
- `../metadata/normalization_statistics.npz`