PEP2026: Water Segmentation Project for CO Eagle River near Minturn

This resource is the end product of my project work at the Pixels to Enviro Patterns 2026 workshop, hosted at the University of Nebraska - Lincoln.

For this project, I extracted 3 images per day from 09-05-2025 to 02-20-2026 from GRIME-AI at the USGS site CO Eagle River near Minturn. Using the CVAT annotation tool, I then annotated 15 images spread out at even intervals across that time period for the presence of water, earth, sky, snow, and human infrastructure. I attempted to train GRIME-AI to detect the presence of water in the river by using this annotated set of data to train a water detection model. Finally, I applied the model to the entire initial set of images I extracted and used GRIME-AI's image segmentation tool to produce masks identifying where water was present in each image.

The resource contains all fifteen images I used to train the model with, the file containing my annotations, and a selection of masks demonstrating the output of the model.

This material is based in part upon work supported by the United States Geological Survey.