New Machine Learning Method Hopes To Enhance Flood Prediction in Vermont

Big DataBig Data

Posted: February 29, 2024

New Machine Learning Method Hopes To Enhance Flood Prediction in Vermont

"The application of machine learning in this study offers an alternative approach to studying floods. By identifying different floodplain functional classes, it becomes possible to predict flood behavior with more precision, aiding in better flood management planning."

Researchers in Vermont are exploring a machine learning approach to improve the understanding and prediction of flood wave behavior. This method, focusing on channel-floodplain interactions, contributes to advancements in flood management and environmental science.

The approach involves using the Ward hierarchical clustering method to classify hydraulically relevant topographic features into distinct 'floodplain functional classes.' These classes represent different interactions between floodplains and flood waves. This classification simplifies the dynamics of flood wave routing, a complex process in river networks.

Flood waves, or the surge of water in rivers during floods, pose challenges in prediction. Traditional 2D hydrodynamic models, while detailed, require extensive computational resources and are often limited to small-scale studies. Simpler models like Muskingum-Cunge, used for larger areas, sometimes do not predict flood behaviors accurately.

In this study, researchers used 1-meter digital elevation models to create geomorphon maps. These maps detail the land's surface features, such as hills, valleys, and the connections between different areas. These features are key to understanding how flood waves move through floodplains.

The application of machine learning in this study offers an alternative approach to studying floods. By identifying different floodplain functional classes, it becomes possible to predict flood behavior with more precision, aiding in better flood management planning.

This research was conducted across over 3,000 river areas in Vermont. It demonstrates the usefulness of combining hydrological studies with machine learning to provide new insights into flood management. This approach could serve as a model for similar studies in other regions, contributing to the broader field of environmental management and disaster response.


Functional Floodplain Classes Emerge from Regional Dataset of Hydraulically-Relevant Topographic Features” was presented as part of “Advances in Applied Watershed Science: Integration of Watershed Processes, Ecosystem Functions, and Societal Systems I” on December 15, 2023 during the American Geophysical Union Annual Fall Meeting


  • Scott Lawson, University of Vermont
  • Rebecca Manners Diehl, University of Vermont
  • Kristen Underwood, University of Vermont
  • Julianne Scamardo, Colorado State University
  • Beverley Wemple, University of Vermont