A multi-resolution dense geospatial dataset comprising spatially-aligned 4-band RGBI imagery (0.2m GSD), LiDAR(ALS)-derived planar forest metrics (1m resolution: Canopy Height (CHM), Plant Area Index (PAI), Foliage Height Diversity (FHD), height percentiles (5, 50, 95)), and 3D vertical Plant Area Density profiles (1m resolution, 12 height bins) for 87 tiles covering forested regions in Saxony, Germany.

Data Structure

The daatsets are following the 2x2km grid from the original GeoSN data source: https://www.geodaten.sachsen.de/downloadbereich-dop-4826.html

Each 2x2 km tile is stored as 3 dense numpy binary files, in format:

{grid_coords}_sn_rgb.npy(4, 10000, 10000) --> band order: ['r', 'g', 'b', 'i']
{grid_coords}_sn_planar.npy (6, 10000, 10000) --> band order: ['chm', 'p05', 'p50', 'p95', 'fhd', 'pai']
{grid_coords}_sn_vertical.npy (12, 2500, 2500) --> band order index '0'= 0 cm 0-1m and index '11' = 11-12m

Metrics Definitions

CHM (Canopy Height Model): Maximum vegetation height [meters]
P05, P50, P95: 5th, 50th, 95th percentile heights [meters]
FHD (Foliage Height Diversity): Vertical structural complexity [Shannon entropy]
PAI (Plant Area Index): Total leaf area per ground area
PAD (Plant Area Index): The amount of plant material in a vertical slice of the forest

Usage Example

import numpy as np

# Load tile
data = np.load('33282_5586_2_sn_rgb', allow_pickle=True) # (4, 10000, 10000)

metadata = data['metadata'].item()

print(f"Tile: {metadata['origin']}")
print(f"Bounds: {metadata['bounds']}")

If you use ForestStack, please cite:

@misc{taimur_khan_2026,
    author       = { Taimur Khan },
    title        = { ForestPatch (Revision c05a74a) },
    year         = 2026,
    url          = { https://huggingface.co/datasets/thisistaimur/ForestPatch },
    doi          = { 10.57967/hf/7462 },
    publisher    = { Hugging Face }
}

Downloads last month: 17