Interpretive deep learning for identifying flooding mechanisms

• Overview
• Quick Start

Overview

The repository contains codes that demonstrate the use of interpretation techniques to gain insights into flooding mechanisms from LSTM-based hydrological models, as described in the papers

Jiang, S., Zheng, Y., Wang, C., & Babovic, V. (2022a). Uncovering flooding mechanisms across the contiguous United States through interpretive deep learning on representative catchments. Water Resources Research, 57, e2021WR030185. https://doi.org/10.1029/2021WR030185

and

Jiang, S., Bevacqua, E., & Zscheischler, J. (2022b). River flooding mechanisms and their changes in Europe revealed by explainable machine learning, Hydrology and Earth System Sciences, 26, 6339–6359, https://doi.org/10.5194/hess-26-6339-2022

Please refer to the file LICENSE for the license governing this code.

Kindly contact us with any questions or ideas you have concerning the code, or if you discover a bug. You may raise an issue here or contact Shijie Jiang through email at shijie.jiang(at)hotmail.com

---

Quick Start

The code was tested with Python 3.7. To use the code, please do:

1. Clone the repository:

   git clone https://github.com/oreopie/hydro-interpretive-dl.git
   cd hydro-interpretive-dl

The study (Jiang et al., 2022a) was implemented based on MOPEX (Model Parameter Estimation Experiment) dataset by NOAA National Weather Service. One can download the data from the official website or HydroShare into mopex.

The study (Jiang et al., 2022b) was implemented based on the following datasets:

• GRDC dataset (https://www.bafg.de/GRDC)
• E-OBS gridded precipitation and temperature dataset (https://www.ecad.eu/download/ensembles/download.php)
• Catchment attributes and boundaries obtained from the Global Streamflow Indices and Metadata Archive (GSIM) (https://doi.pangaea.de/10.1594/PANGAEA.887477) and GRDC (https://www.bafg.de/GRDC/EN/02_srvcs/22_gslrs/222_WSB/watershedBoundaries.html)

We provide a dataset for a sample catchment that contains daily precipitation, temperature, and discharge in data.

2. Install dependencies (conda is recommended to manage packages):

    conda create -n hydrodeepx python=3.6.6
    conda activate hydrodeepx
    pip install innvestigate  tensorflow==1.13.1 h5py==2.10 numpy keras shap  matplotlib jupyter tqdm

   Note for this implementation, tensorflow v1.x is recommended, though tensorflow v2.x may also work.

3. Start Jupyter Notebook and run the Jupyter Notebooks in the repository locally.