Hands-on LSTM and Transformer for Operational Streamflow Prediction
Day 2 Session 1 (10:45AM)
Presenters:
Jiangtao Liu (Pennsylvania State University)
This hands-on workshop presents a complete operational workflow for training, evaluating, and comparing deep-learning models for streamflow prediction. Participants will implement a standardized modeling pipeline using CAMELS data, train an LSTM baseline model, and perform a controlled comparison with a lightweight Transformer architecture.
The session emphasizes practical considerations for operational forecasting, including model stability and performance optimization. Participants will apply techniques such as learning-rate scheduling, regularization, and gradient clipping, and will work with reusable templates designed for rapid experimentation and deployment in real-world hydrologic forecasting systems.
Learning Outcomes:
- Implement end-to-end workflows for operational streamflow prediction
- Train and evaluate LSTM and Transformer models using standardized datasets
- Apply optimization and stability techniques to improve deep-learning model performance
- Adapt provided modeling templates to operational hydrologic forecasting tasks