Machine learning-driven river stage predictions: Using game camera imagery
Abstract
The recent floodings in Vermont have highlighted the need for increased monitoring of river levels to provide timely information to riverside communities and support hydraulic modeling tasks that aim to improve water resource management. River stage and discharge are crucial to understand river flow dynamics, flood prediction, and produce accurate flood inundation mapping. Machine learning (ML) has become a popular and effective method to model environmental parameters with minimal input data. A convolutional neural network (CNN) for regression was created to accurately estimate river discharge from imagery using machine learning techniques. Three months of hourly trail camera imagery was gathered from the Shacksboro site; imagery from July 15th, 2023 to September 5th, 2023 was used to train and validate the model. A basic supervised convolutional neural network was created, trained on the trail camera imagery and stage values.The model scored well in multiple evaluation methods, when tested on a new dataset, it had a R2 score of 0.9914 and mean absolute error of .0806. CNNs provide a skillful means of estimating discharge, relying on affordable game cameras for imagery of river stages.
Primary Faculty Mentor Name
Kristen Underwood
Status
Undergraduate
Student College
College of Engineering and Mathematical Sciences
Program/Major
Computer Science
Primary Research Category
Life Sciences
Machine learning-driven river stage predictions: Using game camera imagery
The recent floodings in Vermont have highlighted the need for increased monitoring of river levels to provide timely information to riverside communities and support hydraulic modeling tasks that aim to improve water resource management. River stage and discharge are crucial to understand river flow dynamics, flood prediction, and produce accurate flood inundation mapping. Machine learning (ML) has become a popular and effective method to model environmental parameters with minimal input data. A convolutional neural network (CNN) for regression was created to accurately estimate river discharge from imagery using machine learning techniques. Three months of hourly trail camera imagery was gathered from the Shacksboro site; imagery from July 15th, 2023 to September 5th, 2023 was used to train and validate the model. A basic supervised convolutional neural network was created, trained on the trail camera imagery and stage values.The model scored well in multiple evaluation methods, when tested on a new dataset, it had a R2 score of 0.9914 and mean absolute error of .0806. CNNs provide a skillful means of estimating discharge, relying on affordable game cameras for imagery of river stages.
