Estimating Daily Suspended Sediment Flux from Multiple Data Sources using Deep Learning: Accompanying Data

This repository contains both final trained model weights (model.eqx), and the training, testing, and inference data. Refer to the github repository for information on loading the serialized binary model weights and creating a model for predictions (https://github.com/Hydro-Umass/tss-ml). The data outputs of the training and testing periods, including the observational data used for loss and error calculations, are in 'train_data.parquet' and 'test_data.parquet'.

We also used our final model to predict suspended sediment concentration, suspended sediment flux, and discharge over all large rivers in the contiguous United States. These data are saved as 'all_data.parquet' and identifed by their NHD Plus V2 identifiers. Long-term average data are also saved as a shapefile (average_data.shp) for easier viewing and mapping.