Machine Learning • Hydrology • LSTM

Machine Learning for River Forecasting

Leveraging LSTM neural networks to predict river stage levels at Avon and Ballantyne stream gages for up to 48 hours ahead, enhancing flood risk assessment.

Project Overview

This project leverages Machine Learning models to predict river stage levels for up to 48 hours ahead. The goal is to enhance flood risk assessment using real-time data and AI techniques.

Duration: 2022 - 2023 | Location: Genesee River Basin

Project Goals

Forecast river stages in the Genesee River Basin
Improve peak flow prediction for flood management
Enable real-time forecasting with dynamic inputs
Develop scalable models for other river systems

Data Sources

Flow & stage time series from HEC-DSS files
USGS streamgage data (Avon & Ballantyne)
Precipitation forecasts from NOAA
Historical flood event data

USGS Ballantyne USGS Avon HEC-DSSVue

Machine Learning Model

LSTM Primary Model

48h Forecast Horizon

2+ Gauge Stations

Long Short-Term Memory (LSTM) neural networks
Random Forest comparison approach
Feature engineering with lagged variables
Validation using historical flood events

Technologies Used

Programming & Libraries

Python – Core ML development
TensorFlow & Keras – Deep learning frameworks
NumPy & Pandas – Data preprocessing
Scikit-learn – Model evaluation

Visualization

Matplotlib & Seaborn – Data visualization
Plotly – Interactive dashboards
HEC-DSSVue – Time-series analysis

Project Resources

Model Configuration Presentation Watershed Map Streamgages Data Flood Stages