EmbeddingsOG/uk-farm-assets-embeddings-v1

🌾 UK Farm Assets Embeddings v1

Algorithmic attribution of UK agricultural field to farming assets. Generated with help of AlphaEarth embeddings, to ensure rich context capture with help of EO features.

---

🌍 Live sample map

Explore a synthetic 500-parcel demo in the interactive map below.
Each parcel is colored by confidence of attribution to its farm.

(Demo is synthetic for illustration; full dataset embeddings & farm IDs are included here in tabular form.)

---

📦 Dataset contents

Each row corresponds to one parcel centroid.
Main fields:

• parcel_id — unique parcel identifier
• farm_id — farm grouping (cluster of parcels operated as one asset)
• lat, lon — centroid coordinates (EPSG:4326)
• area_ha — parcel area in hectares
• vec — [256-D float32 embedding] (L2-normalized AlphaEarth vector)
• confidence — probability that the parcel attribution to this farm is correct
• provenance — JSON with model + parameters (e.g., embed_dim, months_back, scales)

Format: Parquet shards (for efficient loading).

---

🧑‍💻 Usage in Python

Load with 🤗 Datasets:

from datasets import load_dataset
import numpy as np

ds = load_dataset("EmbeddingsOG/uk-farm-assets-embeddings-v1", split="train")

row = ds[0]
print("Parcel:", row["parcel_id"], "Farm:", row["farm_id"])
print("Lat/Lon:", row["lat"], row["lon"], "Area:", row["area_ha"], "ha")

vec = np.array(row["vec"], dtype=np.float32)
print("Embedding shape:", vec.shape)  # (256,)