glenyuyuyu / sr-lstm Goto Github PK

This example of SR-LSTM only works on the following modules:

1. NeuralHydrology Python library in version 1.3.0,
2. Raven Hydrologic Modelling Framework in version 3.7,
3. RDRS forcing data in version 2.1 downloaded from CaSPAr,
4. Streamflow/discharge observations that in same format as the CSV files downloaded from WSC

• basinmaker -- Folder that contains the DEM and HRU data used for BasinMaker routing delineation
• data
  • discharge_obs/ -- Folder that contains the streamflow/discharge observation CSV files
  • USGS_discharge_data.ipynb -- Jupyter notebook to convert the format of USGS data to WSC format
  • forcing_csvs/ -- Folder that will contain the lumped forcings for each subbasins, as well as the lumped forcings for the basin
  • gridded/ -- Folder that contains the gridded source dataset for calculating the static attributes
    • dem/ -- DEM data (HydroSHEDS DEM 3s, merged via rio merge *.bil na_ca_dem_3s.tif)
    • landcover/ -- Landcover data (NALCMS)
    • soil/ -- Soil data (GSDE netCDF files (BD, CLAY, GRAV, OC, SAND, SILT))
  • rdrs_downloads/ -- Folder that contains the gridded RDRS forcing data
  • routing_networks -- Folder that will contain the BasinMaker-created routing network files
• model
  • time_series/ -- Folder that will contain the netCDF files with forcings and discharge
  • attributes/ -- Folder that will contain the CSV files with static attributes
  • basins/ -- Folder that will contain basin-IDs txtfile train_basins.txt
  • trained_model -- Folder that contains the trained LSTM lumped model files
• raven -- Folder that contains the Raven files
• results -- Folder that will contain the simulation results
• scripts -- Folder that contains the SR-LSTM codes
  • derive_grid_weights.py -- Script of Grid-Weights-Generator
  • ensemble2netcdf.py -- Script that takes the pickled ensemble reults file and creates one netCDF submission file with predictions

1. Create a virtual environment in Anaconda3 using the provided srlstm_environment.yml
2. Download the RDRS v2.1 forcings from CaSPAr and place them in data/rdrs_downloads/
3. Download the gridded static attribute data and place them in corresponding folder under data/gridded/ and basinmaker/
4. Place streamflow observation CSVs in data/discharge_obs/, if routing network is provided, rename the folder as *gaugeID_routing and place it in data/routing_networks
5. Place validated NeuralHydrology LSTM model files in model/trained_model/
6. Create a txtfile named astrain_basins.txt which contains the IDs of basins used to train the LSTM model, and place it in model/basins/
7. Edit the input parameters in scripts/run.py, and run the script to start the simulation

• --watershed -- the gauged basin ID
• --experiment -- define the delineation scheme
  • 'default' -- default mode, the threshold for minimum drainage area of subbasins will be 10% of the total basin area, lakes smaller than 5km^2 will be removed
  • 'lumped' -- lumped mode, there will be no discretization
  • 'allsublake' -- all-in mode, all subbasins and lakes in the routing product will be preserved
  • 'MDAx_LAy' -- change x and y to define the threshold for subbasin area and lake area. e.g., MDA200_LA10 uses 200km^2 as the threshold to merge subbasins and 10km^2 to remove lakes
• gauge_lat -- latitude of the basin gauge under the WGS 84 geospatial reference system. Note that Google Maps uses this geospatial reference system.
• gauge_lon -- longitude of the basin gauge under the WGS 84 geospatial reference system.
• start_date end_date -- the time range for simulation, use the format yyyy-mm-dd
• save_forcing -- optional, used if you want to save the generated subbasin-level forcings

Example: os.system(f"python ./scripts/main.py --watershed 02KB001
--experiment MDA200_LA5
--gauge_lat 45.886111
--gauge_lon -77.315278
--start_date 1980-01-01
--end_date 1981-12-31
--save_forcing")