Skip to content

README.md

Latest commit

 

History

History
48 lines (26 loc) · 2.04 KB

README.md

File metadata and controls

48 lines (26 loc) · 2.04 KB
Raw

CAMELS Graphwavenet

Code repo for graphwavenet on CAMELS

These are most of the files used in my CAMELS WRR work (The CAMELS data loaders were too large to be uploaded).

The original WRR paper is

Sun, Alexander Y., Peishi Jiang, Maruti K. Mudunuru, and Xingyuan Chen. "Explore Spatio‐Temporal Learning of Large Sample Hydrology Using Graph Neural Networks." Water Resources Research 57, no. 12 (2021): e2021WR030394.

The original GraphWaveNet paper is

Wu, Z., Pan, S., Long, G., Jiang, J., & Zhang, C. (2019). Graph wavenet for deep spatial-temporal graph modeling. arXiv preprint arXiv:1906.00121.

To use for your own work, just implement the data loaders as requested.

For lookback = 30 days

  • readcamels.py, this is the main data preparation file for using CAMELS data

  • trainwavenetwu.py, this is the main driver of GraphWaveNet

  • gwnetmodel.py, the original GraphWaveNet model by Wu et al.

  • util_gtnet.py, contains utility functions

  • utils_wnet.py, contains utility functions

  • trainwavenetwuimpute.py, main driver for gage imputation (aka prediction at ungaged basins)

  • gwnetmodel_impute.py, modified GraphWaveNet model for gage imputation experiments

For lookback = 365 days [note: the datasets for 365 days are significantly larger. I had to switch to HDF format for data loading]

  • readcamels365.py, this is the main data preparation file for using CAMELS data
  • trainwavenetwu365.py, this is the main driver of GraphWaveNet

To run experiments on your own system, you need to download the CAMELS datasets and Freddie Kratzert's NLDAS from HydroSHARE and put them under proper folers for the data generator to find.

To reproduce my paper experimental results for lookback=30 days,

  • run batch_datagen to generate datasets for lookback=30 days
  • run runsample to train all 10 ensembles [this will take a long time]

To reproduce my paper experimental results for lookback=365 days,

  • run batchgen_365 to generate datasets for lookback=365 days
  • run runsample365 to train all 10 ensembles [this will take a long long time]