Michael Rosati (2024)

Project Title: Enhancing Water Quality Monitoring Through Regional Machine Learning Models

Fellow: Michael Rosati

Adviser(s): Yeo Howe Lim

Project summary: This project tested sequence-based neural networks (LSTMs) for predicting hydrology and water quality in data-limited watersheds. Using climate, hydrologic, and nutrient data, both site-specific and regional models were built across four basins.

Results showed strong performance, with regional models matching or slightly outperforming site-specific ones, demonstrating they can share information across watersheds while maintaining accuracy.

The work also advanced interpretable neural network design, offering insights into model behavior and supporting ongoing research on hydrologic processes.