Basin2Vec

Basin-centric embeddings that let hydrologic models transfer to basins they have never seen.

Code

Chesapeake Bay watershed, Landsat 8 mosaic, 2015. NASA Earth Observatory.

Basin2Vec learns a compact embedding for each river basin from its distributed meteorological and landscape data. Basins with similar hydrologic behavior end up close together in the embedding space, so a model trained on gauged basins can be transferred to basins it has never seen.

Accepted at the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2027).

Publications

WACV Conference

Basin2Vec: Learning Basin-Centric Embeddings for Scalable Hydrological Modeling

Muhammad Talha, Xiaofeng Liu, Yuan Yang, Pang-Ning Tan, and A. Pouyan Nejadhashemi

Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) · 2027

Code

@inproceedings{talha2026basin2vec,
  title = {Basin2Vec: Learning Basin-Centric Embeddings for Scalable Hydrological Modeling},
  author = {Talha, Muhammad and Liu, Xiaofeng and Yang, Yuan and Tan, Pang-Ning and Nejadhashemi, A. Pouyan},
  booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year = {2027},
  publisher = {IEEE}
}