Sediment Modeling with LSTM

LSTM-based deep learning model for suspended sediment prediction at river gauging stations

This project applies a Long Short-Term Memory (LSTM) neural network to predict daily suspended sediment concentrations, using climate forcing and watershed attributes. Built on top of the hydroDL framework.

Repository Structure

├── Sediment.py                      # Main model training & evaluation
├── LR.ipynb                         # Linear regression baseline
├── demo-LSTM-Tutorial.ipynb         # Tutorial notebook
├── test_hydroDL.ipynb               # hydroDL integration tests
├── compare_CAMEL_DAYMET_HYDRODATA.py  # Dataset comparison utility
├── hydroDL/                         # hydroDL submodule
├── SedDemo/                         # Demo data
└── output/                          # Model outputs

Branches

Branch	Description
main	Whole-CONUS model
Local	Local (per-basin) model

Input Data Format

File type	Description
Forcing data	Daily climate variables (precipitation, temperature, etc.)
Attributes data	Static watershed characteristics
DNP	Date and period configuration

Usage

pip install torch numpy pandas matplotlib
python Sediment.py

Context

Developed at Penn State Water Resources Engineering. Companion to the Erosion_rate_estimator project.