Paper

Reading Shallow Flood Depth from Real USGS Gauge-Camera Footage with a Learned Staff-Plate Reader. Kalp Shah and Udit Bhatia, IIT Gandhinagar (2026). Four pages, NeurIPS-style preprint format.

Download PDF LaTeX source

Every number, table and figure in the paper is regenerated from results/ by scripts/make_paper_assets.py, and the numbers on this website come from the same generated file.

Your browser cannot show the PDF inline. Open it directly.

Cite

@misc{shah2026floodcam,
  author       = {Shah, Kalp and Bhatia, Udit},
  title        = {Reading Shallow Flood Depth from Real {USGS} Gauge-Camera Footage
                  with a Learned Staff-Plate Reader},
  year         = {2026},
  note         = {IIT Gandhinagar},
  howpublished = {\url{https://github.com/kalpshah18/FloodCam}}
}

References used in the paper

  • R. Vandaele et al. River-level monitoring from camera images. HESS 25, 2021.
  • Y. Liang et al. V-FloodNet: urban flood video segmentation. Environmental Modelling and Software, 2023.
  • USGS NIMS images and NWIS gage height; GitHub projects waterline and stagesight.
  • D. Lowe. Distinctive image features from scale-invariant keypoints (SIFT). IJCV 60(2), 2004.
  • F. Pedregosa et al. Scikit-learn: machine learning in Python. JMLR 12, 2011.