Camera-Based Streamflow Monitoring – Blacksmith Fork Site (Example Dataset)

Overview

This HydroShare resource provides a complete example of camera-based streamflow monitoring data collection and automated segmentation processing for the Blacksmith Fork site. The dataset demonstrates one full day of real image data and outputs generated through a containerized, cloud-based inference pipeline.

The workflow leverages the Segment Anything deep learning model, deployed in a serverless environment using AWS Lambda and S3. Each image is automatically segmented to identify water regions, compute region-of-interest (ROI) pixel statistics, and generate stage estimates. Outputs are validated with ground truth comparisons using Intersection over Union (IoU) scores and written to a PostgreSQL database for hydrologic analysis.

This resource supports reproducibility of the modeling approaches described in manuscripts submitted to Environmental Modelling & Software and provides transparency into the complete data pipeline—from raw camera imagery to processed hydrologic outputs.

Folder and File Structure

├── Images/ # Original camera images (one day of data)
├── Segmented_Images/ # Segmented overlays generated by SAM
├── Videos/ # Video clips corresponding to image captures
├── Blacksmithfork2024-12-12.csv # Segmentation results and stage estimates
├── FileStatus.db # SQLite database used for file tracking

File Descriptions

• Images/
  Contains raw input images captured at the Blacksmith Fork site for one day of monitoring.

• Segmented_Images/
  Contains model outputs showing water segmentation overlays and ROI boundaries.

• Videos/
  Video files captured alongside images, supporting visual validation and velocity analysis.

• Blacksmithfork2024-12-12.csv
  A CSV file containing:

• Timestamps of captured images
• ROI pixel counts
• Estimated stream stage values
• Segmentation verification flags
• IoU scores against ground truth

• Links to both original and segmented images

• FileStatus.db
  A lightweight SQLite database used on the field computer for file tracking and data integrity verification.
  It records metadata such as file name, type, capture time, upload status, cloud storage destination, and integrity checks.

Applications

• Serves as a reference implementation of end-to-end cloud-integrated camera-based monitoring.
• Demonstrates the role of serverless AI pipelines for operational hydrologic monitoring.
• Provides example datasets for reproducibility of methods and validation of segmentation-based water level estimation.
• Supports research on scalability and automation in environmental monitoring networks.