Model Output Standards
Importance of Standardizing Model Output
Agreed-upon model standards and conventions, such as variable names and data structures, are essential for achieving model interoperability and developing shared, reusable tools. Establishing these conventions enhance data consistency, facilitate dissemination, and support efficient data analysis.
By leveraging existing community standards and conventions within the broader modeling community, researchers can enhance engagement and make their products more valuable to a wider audience. We recommend adopting the model output standard described by (Dietze et al. 2023) for all ecological forecasts. As a community-driven convention, contributions to further develop and improve this standard are highly encouraged.
Model Output Standards
The model output standards outlined by (Dietze et al. 2023) build upon the Climate and Forecast (CF) conventions (Eaton et al. 2003) and Cooperative Ocean/Atmosphere Research Data Service (COARDS) conventions.
For all model output, the recommended order of dimensions should be T, Z, Y, X, U, where:
- T: Time
- Z: Vertical dimension
- Y: Latitude
- X: Longitude
- U: Uncertainty (e.g., ensemble member or summary statistic)
The following dimension names should be used as listed in order to describe T, Z, Y, X, U:
| Dimension | Description |
|---|---|
| reference_datetime | REQUIRED. ISO 8601 (ISO, 2019) datetime the forecast starts from (aka issue time). Datetimes are allowed to be earlier than the reference_datetime if a reanalysis/reforecast is run before the start of the forecast period. |
| datetime | REQUIRED. ISO 8601 (ISO, 2019) datetime being predicted and follows CF convention. |
| duration | REQUIRED. Specifies the time step of the prediction. Use P1D for a daily prediction, P1W for a weekly prediction, and PT30M for a 30-minute prediction. The format should adhere to ISO 8601 duration. |
| depth or height | REQUIRED IF STORING MULTIPLE Z DIMENSIONS. No single standard name for the Z dimension. Where possible, use CF conventions for vertical dimension names and attributes. |
| lat or Y | REQUIRED IF PREDICTING ON GRID. Latitude (units = degrees_north). |
| lon or X | REQUIRED IF PREDICTING ON GRID. Longitude (units = degrees_east) is the default spatial coordinate. The alternative use of Y, X for spatial coordinates should conform to the CF convention and requires additional metadata about grids and projections. |
| site_id | REQUIRED IF NOT PREDICTING ON A GRID. For predictions that are not on a spatial grid, use a site dimension that maps to a more detailed geometry (points, polygons, etc.). site_id should be used when making predictions that map to a stream network geofabric, for example. |
| family | REQUIRED. Describes the distribution of the prediction (e.g., normal for a normal distribution). For ensemble predictions, use ensemble. If the predictions yield a single realization (i.e., there is no associated uncertainty), it is preferred to specify ensemble with the ensemble size set to 1. If this dimension remains constant, it is acceptable to define family as a variable attribute, provided that the file format supports this (e.g., netCDF). Refer to appendix S1 of (Dietze et al. 2023) for supported family and parameter names. |
| parameter | REQUIRED. For ensemble predictions, specify integers from 1 to Ne (where Ne represents the total size of the ensemble). For named distributions, indicate the parameter or statistic being specified (e.g., a normal distribution would have mean and sigma as the values in the parameter column). |
| variable | REQUIRED. Standardized variable name of what is being predicted. We recommend using CSDMS Standard Names. For example, maximum stream temperature would be named channel_water_surface_water__max_of_temperature. However, existing widely-used variable names (e.g., max_temp_c) are also acceptable. |
| prediction | REQUIRED. Predicted value for the parameter in the parameter column. For an ensemble forecast, the value is the prediction for the ensemble member. For a distribution forecast, the value corresponds to the parameter associated with the distribution in the parameter column. |