Metrics

Metrics are computed by dctools during evaluation and aggregated in the result JSON files.

Core metric

The main score family is RMSD (Root Mean Square Deviation):

$$ \mathrm{RMSD} = \sqrt{\frac{1}{N}\sum_{i=1}^{N} (\hat{x}_i - x_i)^2} $$

where $\hat{x}_i$ is the model value interpolated at observation location/time and $x_i$ is the reference value.

Metric set in current DC1 configs

In dc1/config/dc1_wasabi.yaml and dc1/config/dc1_edito.yaml, dataset-level metric lists typically include:

  • rmsd

  • mae (for several observation datasets)

The pipeline also supports leaderboard display labels for additional metrics via dc1/config/leaderboard_config.yaml.

Surface-only evaluation

DC1 metrics are computed on surface data only. If depth is present in inputs, the evaluation extracts the surface level before interpolation/scoring.

Spatial per-bin outputs

When enabled, the pipeline exports per-bin metric summaries using per_bins_resolution (configured in YAML). This is used for map-based diagnostics.

Practical interpretation

  • Lower values indicate better agreement with references.

  • Compare scores by variable and lead time rather than only a single global value.

  • Validate stability across references (altimetry, Argo, gridded reanalysis).