6. Running the Model

The preparation manifest and step inputs are now complete. Continue with the PROJECT_DIR variable set in the preprocessing chapter. The remaining command executes the open loop and ensemble, performs the configured updates and validates the complete output set.

Review runtime configuration

The setup YAML selects the 100 m openAMUNDSEN domain and its model grid format. The project YAML selects the ensemble size, event sequence and compact result variables. The tutorial reference uses ensemble_size: 30 and eight events. Selected scientific fields are shown below; the shipped YAML remains the complete configuration.

data_assimilation:
  prior_forcing:
    ensemble_size: 30
    random_seed: 1415935400
    sigma_t: 1.0
    mu_p: -0.055
    sigma_p: 0.6
    sigma_rh: 1.2
    sigma_sw: 0.15
  resampling:
    ess_threshold_ratio: 0.7
    seed: 1415935400
  rejuvenation:
    sigma_t: 1.0
    mu_p: -0.055
    sigma_p: 0.6
    sigma_rh: 1.2
    sigma_sw: 0.15
    seed: 1415935400
  likelihood:
    wet_snow_line:
      obs_sigma: 200.0
      min_model_finite_fraction: 0.95
  uncertainty:
    scf:
      enabled: true
      assimilation:
        sigma_mode: uncertainty_layer
        aggregate_metric: unc_mean
  output:
    retention: full
    grids:
      format: netcdf
      variables:
        - var: snowdepth_daily
          metrics: [open_loop, ens_mean, ens_std, ens_min, ens_max, increment, analysis_mean, analysis_increment]
        - var: swe_daily
          metrics: [open_loop, ens_mean, ens_std, ens_min, ens_max, increment]

The fixed seeds and keyed random-number scheme make the scientific ensemble reproducible in the pinned release image. Worker count changes scheduling and runtime but not the configured random sequence. The precipitation parameters describe a lognormal multiplicative factor; mu_p=-0.055 and sigma_p=0.6 give an arithmetic mean factor of about 1.133.

Run the project

Purpose: execute the prepared single-domain project, benchmark it, render every configured plot/map/report and finalize cleanup only after validation succeeds.

The positional argument is the project directory. Replace <MAX_WORKERS> with a number no larger than the <CPU_COUNT> exposed to the container in Chapter 2; do not type the angle brackets. For this ES30 project, at most 31 propagations can run at once. On a 24-core machine, use 24; on a machine exposing 48 CPUs, use at most 31. Use a lower value when memory is limited. The numerical-library thread variables remain set to 1.

--json emits the typed run result as one JSON envelope. There is intentionally no run-level --overwrite: completed projects are immutable and interrupted projects resume only from hash-identical inputs.

🟢 Run command:

openamundsen-da run "$PROJECT_DIR" --max-workers <MAX_WORKERS>

The runtime can take from tens of minutes to several hours depending on host, Docker allocation and storage. Keep the container open. Performance monitoring is automatic; do not start a separate monitor.

Completion contract

The command returns success only when all of these are valid:

all open-loop and member propagations;
all eight assimilation updates and their diagnostics;
results/grids/da_output_grids.nc;
benchmark tables and score plots;
configured plots and maps;
results/reports/project_report.pdf; and
automatic removal of package-owned restart state.

The atomic results/run_manifest.json always records the openAMUNDSEN-DA software version. It also records image and image_digest when the runtime supplies OPENAMUNDSEN_DA_IMAGE and OPENAMUNDSEN_DA_IMAGE_DIGEST; otherwise those optional fields are empty. Completed/reused members, the output inventory and cleanup record are included independently of this optional image metadata. Interrupted or failed projects keep restart state.

Before interpretation, confirm that the manifest reports status: success and that the performance plot, result overview, compact NetCDF and PDF report exist. The next chapter explains these outputs without repeating the run command.

Continue with 7. Results and Diagnostics.