Basin2Vec Embeddings v1.0.0

Basin2Vec Embeddings v1.0.0

This release provides the pretrained Basin2Vec embedding archive generated using the full Basin2Vec model with monthly hydroclimatic inputs, static and annual physiographic variables, basin metadata, relation-aware contrastive learning, and drainage-area regularization.

Released asset

basin2vec_embeddings_conus_9067_basins_1985_2024_64d_v1.0.0.npz

The archive contains:

• 64-dimensional Basin2Vec embeddings
• 9,067 CONUS basins
• Basin-year representations covering 1985–2024
• Basin identifiers and corresponding years
• Basin-averaged embeddings computed across available years
• Additional encoder outputs and diagnostic metadata produced during embedding generation

This release contains the embeddings from the complete Basin2Vec model. Embeddings from the ablation variants are not included.

Download

wget https://github.com/Muhammad-Talha-MT/basin2vec/releases/download/v1.0.0/basin2vec_embeddings_conus_9067_basins_1985_2024_64d_v1.0.0.npz

Verify the download

Download the accompanying SHA-256 checksum file and run:

sha256sum -c basin2vec_embeddings_conus_9067_basins_1985_2024_64d_v1.0.0.npz.sha256

Basic loading example

import numpy as np

archive = np.load(
    "basin2vec_embeddings_conus_9067_basins_1985_2024_64d_v1.0.0.npz",
    allow_pickle=True,
)

print(archive.files)

embeddings = archive["embeddings"]
basin_ids = archive["labels"]
years = archive["years"]

print("Embeddings:", embeddings.shape)
print("Basin IDs:", basin_ids.shape)
print("Years:", years.shape)

The archive also includes basin-level averaged representations:

average_embeddings = archive["avg_embeddings"]
average_basin_ids = archive["avg_labels"]

Intended use

These embeddings can be used as static or year-specific basin-conditioning vectors for:

• Streamflow prediction
• Prediction in ungauged basins
• Basin similarity and retrieval
• Hydrologic regionalization
• Basin clustering and visualization
• Transfer learning for basin-level prediction tasks

Citation

A formal citation will be added after publication of the Basin2Vec paper. Until then, please cite this GitHub repository and the corresponding release version.

License

Use of these embeddings is subject to the license provided in the Basin2Vec repository.