How to index and load a time-stacked NetCDF file with Open Data Cube?

Question

Our use case is to make a time-stacked NetCDF file accessible via Open Data Cube.

Each NetCDF file (CMEMS data) has a time coordinate variable, with 8 time points evenly distributed over a range of one day. When trying to datacube.Datacube().load() such a NetCDF file, the following error arises:

Error opening source dataset: NETCDF:"C:\CMEMS\mfwamglocep_2021013000_R20210122_mini.nc":VHM0
Traceback (most recent call last):
...
DeprecationWarning: Stacked netcdf without explicit time index is not supported anymore

How to achieve such an "explicit time index"?

Regarding indexing, it seems that such a time-stacked NetCDF file has to be covered by 8 time-specific dataset definition files (according to an issue discussion), but how to provide the specific time index?

---

Used resources:

• Open Data Cube version:

(cubeenv) C:\> datacube --version
Open Data Cube core, version 1.8.3

• Product definition:

name: mfwamglocep_mini
description: GLOBAL_ANALYSIS_FORECAST_WAV_001_027
metadata_type: eo

metadata:
  format:
    name: NETCDF
  instrument:
    name: NA
  platform:
    code: NA
  product_type: mfwamglocep_mini

storage:
  crs: EPSG:4326
  resolution:
          longitude: 0.08332825
          latitude: -0.08333588

measurements:
- name: VHM0
  dtype: float32
  units: m
  nodata: -32767
  add_offset: 0.0
  scale_factor: 0.01
- name: VMDR
  dtype: float32
  units: degree
  nodata: -32767
  add_offset: 180.0
  scale_factor: 0.01

• Dataset definition:

creation_dt: '2021-01-22T08:19:00Z'
extent:
  center_dt: '2021-01-30T13:30:00Z'
  coord:
    ll: {lon: -180, lat: -80}
    lr: {lon: 179.9167, lat: -80}
    ul: {lon: -180, lat: 90}
    ur: {lon: 179.9167, lat: 90}
  from_dt: '2021-01-30T03:00:00Z'
  to_dt: '2021-01-31T00:00:00Z'
format: {name: NETCDF}
grid_spatial:
  projection:
    geo_ref_points:
      ll: {x: -180, y: -80}
      lr: {x: 179.9167, y: -80}
      ul: {x: -180, y: 90}
      ur: {x: 179.9167, y: 90}
    spatial_reference: EPSG:4326
id: 272302c9-1449-4a33-8166-4b6083a8a711
image:
  bands:
    VHM0: {path: mfwamglocep_2021013000_R20210122_mini.nc, layer: VHM0}
    VMDR: {path: mfwamglocep_2021013000_R20210122_mini.nc, layer: VMDR}
instrument: {name: NA}
lineage:
  source_datasets: {}
product_type: mfwamglocep_mini
platform: {code: NA}

• Original NetCDF file, reduced to a publicly downloadable small subset (86 MB):

import xarray
ds = xarray.open_dataset("mfwamglocep_2021013000_R20210122.nc")
to_drop = []
for name in ds.data_vars:
  if name!="VHM0" and name!="VMDR": to_drop.append(name)
mini = ds.drop_vars(to_drop)
mini.to_netcdf("mfwamglocep_2021013000_R20210122_mini.nc")

Answer 1

An explicit time index can be achieved by appending a fragment (#part=<index>) to the uri in the path attribute of the metadata document. It starts from 0. The uri also needs to start with file://. In your case

image:
  bands:
    VHM0: {path: file://mfwamglocep_2021013000_R20210122_mini.nc#part=0, layer: VHM0}
    VMDR: {path: file://mfwamglocep_2021013000_R20210122_mini.nc#part=0, layer: VMDR}

should work for the first time slice. As far as I know you need to create a separate dataset document for each time slice (updating also the time stamps in the extent attribute).

The approach was extracted from here https://github.com/opendatacube/datacube-core/commit/5723c040fc5aeacc638756ea21e395d491202bf8.