부산대학교 도서관

Extreme rainfall events coupled with compromised stormwater infrastructure can result in flooding and erosion. The ability to accurately predict watershed flow behavior during rainfall events can aid stormwater utility professionals in efficiently operating stormwater systems. However, watershed flow behavior can be challenging to predict during rainfall events due to the stochastic behavior of rainfall for a specific location and the nonlinear process characteristics of streams and their respective watersheds. In this work, we propose a machine learning-based approach to model and predict watershed flow behavior during rainfall events to support data driven decision making. In comparison with conventional mathematical model frameworks, machine learning-based models typically provide improved performance for processing large datasets with “noisy” or even missing data points. Several machine learning methods are considered, including long short-term memory (LSTM), which shows promise as a viable method in this application based on the preliminary results with the proposed machine learning framework (e.g., R2 > 0.8).