NetCDF and NcML Digital Twin Architecture

Scientific datasets in earth observation, weather forecasting, and oceanography are commonly delivered as binary NetCDF (Classic, 64-bit offset, NetCDF-4/HDF5) files containing gigabytes of dense numerical arrays.

The Binary Problem

Manipulating binary files directly in an engineering modeling environment poses several challenges:

• Binary files cannot be natively queried or cross-referenced with EMF (Eclipse Modeling Framework) tools.
• Modifying or adding metadata (e.g. documentation, authoring units, calibration factors) typically requires rewriting the entire multi-gigabyte binary file.
• Version control and diffing binary files in Git is impractical.

NcML: The Digital Metadata Twin

NcML (NetCDF Markup Language) is an XML dialect designed by UCAR Unidata that provides an XML representation of NetCDF dataset metadata without duplicating the heavy payload data.

Capella Datascience uses an EMF-based NcML 2.2 meta-model as a first-class digital twin:

<?xml version="1.0" encoding="UTF-8"?>
<netcdf xmlns="http://www.unidata.ucar.edu/namespaces/netcdf/ncml-2.2"
        location="/path/to/dataset.nc">
    <dimension name="time" length="24" isUnlimited="true"/>
    <dimension name="lat" length="180"/>
    <dimension name="lon" length="360"/>
    <variable name="temperature" shape="time lat lon" type="float">
        <attribute name="units" value="degC"/>
        <attribute name="long_name" value="Surface Air Temperature"/>
        <attribute name="missing_value" type="float" value="-999.0"/>
    </variable>
</netcdf>

Key Benefits of the Architecture

• Zero Data Duplication: The NcML model references the binary dataset via its location attribute (local file path, HTTP OpenDAP URL, or S3 object URI).
• Lightweight Storage: NcML files are only a few kilobytes and can be committed to source control and diffed effortlessly.
• Non-Destructive Overrides: Users can add annotations, rename variables, reshape dimensions, or adjust scale/offset equations without touching the original binary data.
• Full EMF Ecosystem Integration: The dataset structure participates in EMF Search, Parsley properties, OCL validation, and Capella traceability links.