Models and predictions for "How to deal w___ missing input data"

How to deal w___ missing input data

This repository contains the models, configs, and results files for the paper Gauch et al., "How to deal w___ missing input data".

The corresponding analysis code is available on GitHub: https://github.com/gauchm/missing-inputs.

Contents of this repository

missing-inputs.ipynb -- Jupyter notebook to reproduce figures from the paper.

results/ -- Folder with model weights, configs, and predictions used in missing-inputs.ipynb.

patches/ -- Contains patches for local modifications to reproduce experiments from the paper.

Required setup

Clone neuralhydrology: git clone https://github.com/neuralhydrology/neuralhydrology.git.

Install an editable version of neuralhydrology: cd neuralhydrology && pip install -e ..

Download the following data:

the CAMELS US dataset (CAMELS time series meteorology, observed flow, meta data, version 1.2) from NCAR into some data directory (has to match data_dir in the config files).

the extended Maurer and NLDAS forcings set available on HydroShare: Maurer, NLDAS.

the models, results, and config files from this paper avaliable on this Zenodo repository.

Note that to reproduce the experiments, local modifications to NeuralHydrology are necessary. To do so, apply the patches in the patches/ directory: git apply patches/experiment-N.patch.