-
Notifications
You must be signed in to change notification settings - Fork 35
Home
Welcome to the summaWorkflow_public wiki! Here we outline some common questions and how to deal with them. For questions specific to the computational infrastructure available at the University of Saskatchewan, you can also look here: FAQ specific to University of Saskatchewan.
Content:
- Re-use data for different domain setup
- Change SUMMA outputs
- SUMMA initial conditions
- Calibrate SUMMA
- Parallelizing SUMMA
- Summarize multiple SUMMA output logs
- Merge parallelized SUMMA outputs
- Parallelizing mizuRoute
You can specify the path to an existing data folder in the control file for your new domain and selectively running scripts from the workflow. I.e. do not run the download script for the existing data but start at the relevant pre-processing step.
Determine which simulaiton variables SUMMA outputs by changing the outputControl.txt file in the SUMMA settings folder. A full overview of SUMMA variables can be found in the SUMMA source code, in file SUMMA/build/source/dshare/var_lookup.f90. See here for more details: https://summa.readthedocs.io/en/latest/input_output/SUMMA_input/#output-control-file
Run SUMMA with the -r e argument for as long a period as you want to use to determine initial states. This will give a (r)estart file at the end of your simulation. Other restart options are year, month, day or never, giving you restart files at the start of each new year, month or day. Optionally use the script to be added in summaWorkflow_public/0_tools to:
- Ensure canopy is empty
- Remove any existing snow layers
to be added
SUMMA uses a hierarchical approach to the reading of parameter files (https://summa.readthedocs.io/en/latest/input_output/SUMMA_input/#attribute-and-parameter-files). In essence this means it reads the parameter files in sequence. First, it reads the GRU and HRU parameters from basinParamInfo.txt and localParamInfo.txt respectively. Next, certain HRU parameter values are overwritten with values read from the lookup tables for soil and vegetation parameters. Finally, SUMMA checks the trial parameters .nc file and overwrites any existing GRU or HRU-level parameters with the values found in this value. If the trial parameter file is empty, no existing values are changed.
Connect SUMMA to a calibration algorithm by (1) determining which parameters you want to calibrate and (2) letting the calibration algorithm write the values it wants to test to the trialParams.nc file. A full overview of SUMMA parameters can be found in the SUMMA source code, in file SUMMA/build/source/dshare/var_lookup.f90. The existing code to create a trial parameter file in summaWorkflow_public/4_model_input/SUMMA/3e_trial_parameters can form the basis for step 2.
Run the SUMMA executable with the optional argument -g. -g takes two inputs: start_gru and num_gru which are respectively the index of the GRU to start this specific run at (counted as indices in the attributes .nc file, so this variable has a minimum value of 1 and a maximum value of the nubmer of GRUs in your domain), and the number of GRUs to run from that starting index. Some examples:
summa.exe -g 1 1 -m path/to/filemanager.txt
Starts the SUMMA run at GRU 1 and runs only that GRU.
summa.exe -g 1 50 -m path/to/filemanager.txt
Starts the SUMMA run at GRU 1 and runs 50 GRUs (including the first), ending with GRU 50.
summa.exe -g 50 1 -m path/to/filemanager.txt
Starts the SUMMA run at GRU 50 and runs only that GRU.
summa.exe -g 50 50 -m path/to/filemanager.txt
Starts the SUMMA run at GRU 50 and runs 50 GRUs, ending with GRU 100.
Parallelizing such runs on a High Performance Computing cluster can look as follows (SLURM example). Submit a job with the --array option (technically these arrays are 0-indexed but starting them from 1 may be easier to conceptualize):
sbatch --array=1-10 run_summa.sh
Inside the run_summa.sh script, use the $SLURM_ARRAY_TASK_ID variable to set which GRUs to run:
#!/bin/bash
#SBATCH --ntasks=1
# ... more SBATCH stuff if needed
# Define the GRU settings
gruMax=198 # Total number of GRUs in the domain
gruCount=20 # Number of GRUs per chunk
# Get the array ID for further use
offset=$SLURM_ARRAY_TASK_ID
# Start at 1 for array ID 1, 21 for array ID 2, etc
gruStart=$(( 1 + gruCount*(offset-1) ))
# Check that we don't specify too many basins
if [ $(( gruStart+gruCount )) -gt $gruMax ]; then
gruCount=$(( gruMax-gruStart+1 ))
fi
# Run SUMMA
summa.exe -g $gruStart $gruCount -m /path/to/filemanager.txt
SUMMA produces log files for each set of GRUs when the GRUs are run in parallel with option -g. These files can be summarized into an easy-to-read overview by using the script to be added in the summaWorkflow_public/0_tools folder.
SUMMA runs split into multiple chunks with the -g argument results in multiple output files that each contain the simulations for a single chunk of GRUs. These output files can be merged into a single file by using the script to be added in the summaWorkflow_public/0_tools folder.
to be added