FIDLAR

Code for the paper "FIDLAR: Forecast-Informed Deep Learning Approaches for Flood Mitigation" accepted by AAAI'25.

Repository description

• data folder includes data sets used
• baseline folder includes baseline models used
• model folder includes our proposed models
• loss folder includes loss functions used
• preprocess folder includes data pre-processing
• postprocess folder includes the programs for experiment results, visualization, and ablation study
• training_WaLeF_models folder includes training programs for Flood Evaluator with all models
• training_optimization_models folder includes training programs for Flood Manager with frozen Flood Evaluator

Requirements

conda create -n env_name python=3.8
conda activate env_name
pip3 install -r requirements.txt

Running

• Download the entire repository and install the required packages (see requirements above).
• For training,
  • Flood Evaluator, go to the training_WaLeF_models folder and run cells in the ipynb files
  • Flood Manager, go to the training_optimization_models folder and run cells in the ipynb files
• For testing and experiment analysis, go to the postprocess folder and run cells in the ipynb files.

Main Results

Flood Prediction

Flood Mitigation

Citation

If you find this work interesting and useful, please cite our paper:

@inproceedings{shi2025fidlar,
  title={FIDLAR: Forecast-Informed Deep Learning Architecture for Flood Mitigation},
  author={Shi, Jimeng and Yin, Zeda and Leon, Arturo and Obeysekera, Jayantha and Narasimhan, Giri},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={27},
  pages={28377--28385},
  year={2025}
}