Data and model code used to evaluate a process-guided deep learning approach for in-stream dissolved oxygen prediction

This model archive contains data and code used to assess the use of process-informed multi-task deep learning models for predicting in-stream dissolved oxygen concentrations. Three holdout experiments were run to assess model performance, including a temporal holdout experiment, a spatial holdout experiment with similar sites held out, and a spatial holdout experiment with dissimilar sites held out. This model archive includes data from 10 sites in the lower Delaware River Basin that were used in the model experiments. Model training target data include dissolved oxygen concentrations downloaded from the National Water Information System (NWIS) (U.S. Geological Survey 2023). Model input data include daily meteorological driver variables derived from gridded surface data (gridMET; Abatzoglou 2013); river and catchment characteristics (Wieczorek et al. 2018); and estimates of daily stream metabolism rates (Appling et al. 2018). The contents of this model archive are organized into files or file directories that have been aggregated into zip files.
 

site_month_metrics.csv: Dataset that summarizes model accuracy results for different sites in the study area by month and for different variables, replicates, model versions, partitions, and holdouts.
site_metrics.csv: Dataset that summarizes model accuracy results for different sites in the study area and for different variables, replicates, model versions, partitions, and holdouts.
drb-do-ml.zip: The source code for running the experiments, compiling the results, and making the figures. Additional metadata information is included in a README file as well as file dictionaries within this zip file.
model_inputs_targets.csv: Model inputs and training target data.

Although these data have been processed successfully on a computer system at the U.S. Geological Survey (USGS), no warranty expressed or implied is made regarding the display or utility of the data for other purposes, nor on all computer systems, nor shall the act of distribution constitute any such warranty. The USGS or the U.S. Government shall not be held liable for improper or incorrect use of the data described and/or contained herein.