Published October 31, 2026 | Version v1
Dataset Restricted

Code and data for Adaptive Land Units from Earth Embeddings

Description

Adaptive land units of Denmark built from the 2024 AlphaEarth embeddings with parallel Bayesian hierarchical clustering at s = 1. There are 1,265,975 units and 3,491,995 edges between touching units.

Files

- dk_units_2024_s1.tif has the unit id of every 10 m pixel (EPSG:32632, int32, −1 outside land).
- dk_units_2024_s1_nodes.parquet has one row per unit. `unit_id` matches the raster. It also holds `n_pixels`, `area_m2`, the centroid `cx`, `cy`, the bounding box, the sums `s00`–`s63` and sums of squares `ss00`–`ss63` of the centered embeddings, and the BHC values `log_d`, `log_p`.
- dk_units_2024_s1_edges.parquet has one row per pair of touching units `u < v`, with the shared border (`border_px`, `border_m`), the BHC log-odds of merging them (`merge_logodds`) and the distance between their mean embeddings (`dist`).
- dk_units_2024_s1_settings.json holds the model settings and the national mean `mu`.
- code.zip is the code.

The mean embedding of a unit is `s00..s63 / n_pixels + mu`.

License

The data are CC BY 4.0 and the code is MIT. The data are derived from the AlphaEarth Foundations Satellite Embedding dataset (Google and Google DeepMind, CC BY 4.0). The land mask is from Basemap05 (DCE, Aarhus University).

Files

Restricted

The record is publicly accessible, but files are restricted. Log in to check if you have access.

Additional details

Software

Repository URL
https://github.com/alonfnt/pbhc
Programming language
Python