FAIRDNN
Flood stAge predIction thRough Deep Neural Networks

Overview

This is the official code repository for the paper An End-to-End Flood Detection System Using Deep Neural Networks (2023) by Windheuser, Karanjit, Pally, Samadi, and Hubig.

It consists out of the following two systems:

1. Estimate the depth of an river based on solely a picture with the code in the directory ./estimate_depth/. The CNN and the U-Net is implemented in ./estimate_depth/models/.
2. The code for the predictions of the gauge height based on historical data is in the directory ./predict_flooding/. The different models, which were compared in the paper, are in ./predict_flooding/models/.

Datasets

We used the data provided by the United States Geological Survey (USGS).

For the water depth estimation, we scraped the live webcams of multiple rivers and stored the current image every few minutes. Our scraped images are available here and they are free to use for everyone.

For the flood predictions, the USGS provides various climate data as a csv file for different rivers, which were used as a time-series dataset.

Authors

• Leon Windheuser
• Rishav Karanjit
• Rakshit Pally
• Dr. Vidya Samadi
• Dr. Nina Hubig

Cite Our Work

If you find our code or paper, published at the AGU Earth & Space Science, useful we encourage you to cite our paper.

@article{windheuserend,
  title={An End-to-End Flood Stage Prediction System Using Deep Neural Networks},
  author={Windheuser, L and Karanjit, R and Pally, R and Samadi, S and Hubig, NC},
  journal={Earth and Space Science},
  pages={e2022EA002385},
  publisher={Wiley Online Library}
}