This data resource contains photographs of stream stage used in the manuscript titled "Leveraging Artificial Intelligence to Process Citizen Science Photos of Water Levels" by Abhinna Manandhar and Christopher S. Lowry.

The files are organized into five folders and were collected in the summer of 2025 on Tonawanda Creek, Western New York.  Descriptions of files in each folder are given below.

File: step 1 - original images: This file contains the raw photos of a staff gauge with variable water levels.  This file contains 141 photographs and an associated spreadsheet listing the image number and the manually measured water level on the gauge.   

File: step 2 - detect gauge using Yolov8: This file represents a subset of the original data set consisting of 41 images used to train a computer vision model (Yolov8) to detect the location of the staff gauge within the photo.  

File: step 2 - detect stationLabels using Yolov8: This file represents a subset of the original data set consisting of 41 images used to train a computer vision model (Yolov8) to detect the location of the station label at the time of the staff gauge within the photo.  

File: step 3 - preprocessing to remove noise:  These files show the results of the preprocessing of the 41 test images to reduce the noise in the photos.  

File: step 4 - preprocessing for waterline detection: These files present the results of preprocessing the 41 test images to convert them to black and white in preparation for identifying the water level on the gauge.  

File: step 5 - enhanced image with detected waterline to send to Gemini Model: These files present the results of preprocessing the 41 test images, where the water level is marked with a red horizontal line.  The base image is converted back into a color image.  These images are then provided to the Gemini Model to determine the water-level value. 

The source code of the web application, along with the model and the LLM prompts, is freely available on GitHub (https://github.com/cslowry/CrowdHydrology.