Deploying the Power of Machine Learning and Near Real Time Modeling for Stage Forecasting in Streams

Forecasting stream flows and flood depths is an essential task in flood management and water resource planning.  Hydrologic/hydraulic simulation models or data-driven approaches can be used to generate these real-time predictions.  Modern advances in artificial intelligence and state-of-the-art machine learning (ML) techniques have provided new opportunities for quicker streamflow prediction processes with more efficient and accurate forecasts.

In this presentation, we will dig into the use of ML to develop flow depth prediction models. Prediction models using different ML algorithms and various prediction settings are trained and validated based on the historic precipitation data and stream flow depth observations.  These models apply different ML techniques, including multiple linear regression, support vector regression, random forest regression, artificial neural network multi-layer perceptron, long short-term memory and ensemble learning.

Case examples will illustrate how radar and forecasted rainfall datasets can be applied to traditional H&H models and ML models to provide lead times of several hours or days in advance of a flood event.  These forecasts can be critical for issuing flood warnings and the real-time operation of flood control systems.