arkenjan / fcn-water-flow

Official PyTorch implementation for the SAIS 2023 paper "Fully Convolutional Networks for Dense Water Flow Intensity Prediction in Swedish Catchment Areas"

Fully Convolutional Networks for Dense Water Flow Intensity Prediction in Swedish Catchment Areas

Official PyTorch implementation of the SAIS 2023 paper Fully Convolutional Networks for Dense Water Flow Intensity Prediction in Swedish Catchment Areas by Aleksis Pirinen, Olof Mogren and Mårten Västerdal.

SAIS 2023 paper | arXiv | Video

Dataset

The dataset used in the paper can be downloaded from this link. Once downloaded, simply unzip the folder in the main folder, which should result in a folder data-extended-smhi.

Code structure overview

Model training is done using training.py. Results are sent to a log folder (see the variable BASE_PATH_LOG), and result plots can then be generated using the file plot_result.py.

Training

Prior to this, ensure you have the dataset and that the variable BASE_PATH_DATA points to this dataset folder.

Model training (and validation on validation data) is performed using training.py. See the file plot_results.py if you are interested in tracking the progress and results throughout training. Models are saved during and upon completion of training, and are sent to the log folder.

Citation

If you find this implementation and/or our paper interesting or helpful, please consider citing:

@article{pirinen2023fully,
  title={Fully Convolutional Networks for Dense Water Flow Intensity Prediction in Swedish Catchment Areas},
  author={Pirinen, Aleksis and Mogren, Olof and V{\"a}sterdal, M{\aa}rten},
  journal={arXiv preprint arXiv:2304.01658},
  year={2023}
}