BASINGRID

A 40-year (1985–2024) dataset of aligned meteorological and landscape rasters for 9,067 U.S. basins.

Paper Code

Garden City, Kansas, Landsat 5 TM false color, August 15, 2011. NASA Earth Observatory.

BASINGRID is a distributed, machine-learning-ready hydrology dataset, published in Scientific Data. For each of 9,067 U.S. basins it provides 40 years (1985–2024) of meteorological forcings and landscape attributes as rasters on a common grid, aligned with streamflow records.

Why distributed

Most large-sample hydrology datasets provide only basin-averaged values. BASINGRID keeps the spatial detail, so models such as HydroTransformer and Basin2Vec can learn where in a basin rain falls and how the landscape responds.

Figure 1. Forty years of landscape change seen by Landsat, the kind of long record BASINGRID is built from: Garden City, Kansas, in 1972 (Landsat 1, left) and 2011 (Landsat 5, right). Red is healthy vegetation; circles are center-pivot irrigation. NASA Earth Observatory.

Publications

Sci. Data Journal

BASINGRID: A distributed 40-year spatiotemporal hydrology dataset for U.S. basins

Muhammad Talha, A. Pouyan Nejadhashemi, Xiaofeng Liu, and Yuan Yang

Scientific Data · 2026

DOI Code

@article{talha2026basingrid,
  title = {{BASINGRID}: A distributed 40-year spatiotemporal hydrology dataset for {U.S.} basins},
  author = {Talha, Muhammad and Nejadhashemi, A. Pouyan and Liu, Xiaofeng and Yang, Yuan},
  journal = {Scientific Data},
  year = {2026},
  publisher = {Nature Publishing Group},
  doi = {10.1038/s41597-026-08030-4}
}