RRFS Forecast Ensemble Quick Start Guide

Clone the app’s rrfs_ensemble branch:
```
git clone https://github.com/NOAA-GSL/mpas_app.git -b rrfs_ensemble --recursive
cd mpas_app
```
This branch is kept in sync with the mpas_app main branch, but points to the RRFS development branch in https://github.com/RRFSx/MPAS-Model.
Build the model and components:
```
./build.sh -p <platform>
```

Load the mpas_app conda environment:

source load_wflow_modules.sh <platform>

Create your user YAML in the ush directory. The file can be structured to overwrite any settings in the app, but can be as simple as:
```
user:
  experiment_dir: /path/to/exp/dir
  platform: ursa
platform:
  account: wrfruc
```

Generate the experiment:

cd ush
./experiment_gen.py workflows/ensemble_wflow.yaml <your_user_yaml.yaml>

This generates an experiment directory at the experiment_dir path specified in your user YAML that contains a Rocoto XML file, which is ready to use.

Configuring the RRFS Ensemble

The workflows/ensemble_wflow.yaml defaults to a 2-member, RRFS-DA-ensemble-initialized ensemble with GFS lateral boundary conditions. The forecast configuration otherwise uses the base namelist used by RRFS.

All modifications meant for scientific experimentation can be made in the user’s YAML. Changes meant for contribution to the code base may be made by adding new ush/workflows YAML files, or by modifying existing ones.

YAML configs passed to the ush/experiment_gen.py increase in precedence from left to right, where higher precedence settings will override those set in the default_config.yaml or those provided earlier in the list. To override a setting, simply create the desired key/value pair mirroring that in the original configuration file. Defaults loaded in the ush/experiment_gen.py include the ush/default_config.yaml, the parm/machines/<platform>.yaml, and the relevant sets of parm/wflow YAMLs.

Changing Ensemble Size and Forecast Length

The following is an example of settings could be used in the user YAML in addition to the required settings to change the basic ensemble configuration:

user:
  ensemble_size: 4
forecast:
  mpas:
    length: 48

Any other existing settings can be similarly changed.

Stochastic Perturbations

Stochastic perturbation support is being added in MPAS PR #239. Assuming only namelist changes from that PR are needed in the final implementation, the following section could be added to modify the namelist once, but generate a different seed for each member at run time. This solution leverages the fact that the MPAS App ensemble configuration has the MEM environment variable available to each member, and it’s an integer.

forecast:
  mpas:
    namelist:
      update_values:
        stoch_physics:
          do_sppt: true
          do_skeb: false
          config_spptint: 0
          config_sppt_1: 0.8
          config_sppt_2: 0.0
          config_sppt_3: 0.0
          config_sppt_tau_1: 21600
          config_sppt_tau_2: 86400
          config_sppt_tau_3: 21600
          config_sppt_lscale_1: 450000
          config_sppt_lscale_2: 1000000
          config_sppt_lscale_3: 2000000
          config_sppt_logit: true
          config_sppt_sfclimit: true
          config_iseed_sppt1: '{{ cycle.strftime("%Y%m%d%H") + "MEM" | env }}'
          config_iseed_sppt2: '0'
          config_iseed_sppt3: '0'
          config_sppt_hgt_top1: 15000
          config_sppt_hgt_top2: 27000
          config_stochini: false