Machine Learning Methods for Downscaling Stream Temperature Predictions, Tested in Eight Watersheds of the Conterminous United States

This model archive provides all data, code, and model outputs used in the corresponding manuscript (Barclay and others, 2026) to test machine learning (ML) methods for downscaling and multi-scale modeling of stream temperature to combine an ML model and/or input data at coarse spatial resolution with an ML model and/or input data at fine spatial resolution to predict stream temperatures at fine spatial resolution in a watershed. The models were tested in eight watersheds across the conterminous United States for the period of 1979 through 2021.
The data are organized into these components:

1. Model Inputs - Meteorological data, river network matrices, and stream temperature observations
2. Model Code - Python files and README for reproducing model training and evaluation
3. Coarse Model - Trained coarse stream temperature model to be downscaled
4. Model Outputs - Model simulation outputs and evaluation metrics

 
The publication associated with this model archive is: Barclay, J.R., Koenig, L.E., Fan, Y., Jia, X., Appling, A.P., (in review). Big data for small streams: Leveraging a coarse national model to enhance fine-resolution stream temperature predictions, http://doi.org/TBD/TBD.
This data compilation was supported by the U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research, Environmental System Science Data Management Program, as part of the ExaSheds project, under Award Number 89243021SSC000068. Work was also supported by the U.S. Geological Survey, Water Availability and Use Science Program.