Skip to content

Commit

Permalink
Merge pull request #495 from Anu-Ra-g/fast_grib_doc
Browse files Browse the repository at this point in the history
Added a new documentation page for faster GRIB aggregations
  • Loading branch information
martindurant authored Sep 9, 2024
2 parents 70c5e78 + 73c040b commit af4c5dd
Show file tree
Hide file tree
Showing 2 changed files with 225 additions and 0 deletions.
1 change: 1 addition & 0 deletions docs/source/index.rst
Original file line number Diff line number Diff line change
Expand Up @@ -68,6 +68,7 @@ so that blocks from one or more files can be arranged into aggregate datasets ac
beyond
nonzarr
reference
reference_aggregation
contributing
advanced

Expand Down
224 changes: 224 additions & 0 deletions docs/source/reference_aggregation.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,224 @@
Aggregation special cases
=============================

As we have already seen in this `page <https://fsspec.github.io/kerchunk/test_example.html#multi-file-jsons>`_,
that the main purpose of ``kerchunk`` it to generate references, to view whole archive
of files like GRIB2, NetCDF etc, allowing us for direct access to the data. In
this part of the documentation, we will see some other efficient ways of
combining references.

GRIB Aggregations
-----------------

This reference aggregation method of GRIB files, developed by `Camus Energy <https://www.camus.energy>`_,
and it functions if accompanying ``.idx`` files are present. It involves creating a reference index
for every GRIB message across the files that we want to aggregate.

**But this procedure has some certain restrictions:**

- GRIB files must paired with their ``.idx`` files
- The ``.idx`` file must be of *text* type.
- Only specialised for time-series data, where GRIB files
have *identical* structure.
- Each horizon(forecast time) must be indexed separately.


Utilizing this method can significantly reduce the time required to combine
references, cutting it down to a fraction of the previous duration. The original
idea was showcased in this `talk <https://discourse.pangeo.io/t/pangeo-showcase-optimizations-for-kerchunk-aggregation-and-zarr-i-o-at-scale-for-machine-learning/4074>`_.
It follows a three step approach.

**Three step approach:**

1. Extract and persist metadata directly from a few arbitrary grib
files for a given product such as HRRR SUBH, GEFS, GFS etc.
2. Use the metadata mapping to build an index table of every grib
message from the ``.idx`` files
3. Combine the index data with the metadata to build any FMRC
slice (Horizon, RunTime, ValidTime, BestAvailable)


*How is it faster*

The ``.idx`` file otherwise known as an *index* file contains the key
metadata of the messages in the GRIB files. These metadata include `index`, `offset`, `datetime`,
`variable` and `forecast time` for their respective messages. This metadata
will be used to index every GRIB message. By following this approach, we only have to
``scan_grib`` a single GRIB file, not the whole archive.

Building the index of a time horizon, first requires a single one-to-one mapping of GRIB/Zarr
metadata to the attributes in the idx file. Only constraint is the mapping needs to be
made from a single GRIB file, belonging to the *same time horizon*. The indexing process
primarily involves the `pandas <https://pandas.pydata.org/>`_ library. To confirm this,
see this `notebook <https://gist.github.com/Anu-Ra-g/efa01ad1c274c1bd1c14ee01666caa77>`_.
After indexing a single time horizon, you can combine this index with indexes of
other time horizon and store it.

.. note::
The index in ``.idx`` file indexes the GRIB messages where as the ``k_index``
(kerchunk index), index the variables
in those messages.

The table mentioned below is a *k_index* made from a single GRIB file.

.. list-table:: k_index for a single GRIB file
:header-rows: 1
:widths: 5 10 15 10 20 15 10 20 20 30 10 10 10

* -
- varname
- typeOfLevel
- stepType
- name
- step
- level
- time
- valid_time
- uri
- offset
- length
- inline_value
* - 0
- gh
- isobaricInhPa
- instant
- Geopotential height
- 0 days 06:00:00
- 0.0
- 2017-01-01 06:00:00
- 2017-01-01 12:00:00
- s3://noaa-gefs-pds/gefs.20170101/06/gec00.t06z...
- 0
- 47493
- None
* - 1
- t
- isobaricInhPa
- instant
- Temperature
- 0 days 06:00:00
- 0.0
- 2017-01-01 06:00:00
- 2017-01-01 12:00:00
- s3://noaa-gefs-pds/gefs.20170101/06/gec00.t06z...
- 47493
- 19438
- None
* - 2
- r
- isobaricInhPa
- instant
- Relative humidity
- 0 days 06:00:00
- 0.0
- 2017-01-01 06:00:00
- 2017-01-01 12:00:00
- s3://noaa-gefs-pds/gefs.20170101/06/gec00.t06z...
- 66931
- 10835
- None
* - 3
- u
- isobaricInhPa
- instant
- U component of wind
- 0 days 06:00:00
- 0.0
- 2017-01-01 06:00:00
- 2017-01-01 12:00:00
- s3://noaa-gefs-pds/gefs.20170101/06/gec00.t06z...
- 77766
- 22625
- None
* - 4
- v
- isobaricInhPa
- instant
- V component of wind
- 0 days 06:00:00
- 0.0
- 2017-01-01 06:00:00
- 2017-01-01 12:00:00
- s3://noaa-gefs-pds/gefs.20170101/06/gec00.t06z...
- 100391
- 20488
- None


*What now*

After creating the k_index as per the desired duration, we will use the ``DataTree`` model
from the `xarray-datatree <https://xarray-datatree.readthedocs.io/en/latest/>`_ to view a
part(desired variables) or the whole of the aggregation, using the k_index. Below is a
tree model made from an aggregation of GRIB files produced from **GEFS** model hosted
in AWS S3 bucket.

.. code-block:: bash
DataTree('None', parent=None)
├── DataTree('prmsl')
│ │ Dimensions: ()
│ │ Data variables:
│ │ *empty*
│ │ Attributes:
│ │ name: Pressure reduced to MSL
│ └── DataTree('instant')
│ │ Dimensions: ()
│ │ Data variables:
│ │ *empty*
│ │ Attributes:
│ │ stepType: instant
│ └── DataTree('meanSea')
│ Dimensions: (latitude: 181, longitude: 360, time: 1, step: 1,
│ model_horizons: 1, valid_times: 237)
│ Coordinates:
* latitude (latitude) float64 1kB 90.0 89.0 88.0 87.0 ... -88.0 -89.0 -90.0
* longitude (longitude) float64 3kB 0.0 1.0 2.0 3.0 ... 357.0 358.0 359.0
│ meanSea float64 8B ...
│ number (time, step) int64 8B ...
│ step (model_horizons, valid_times) timedelta64[ns] 2kB ...
time (model_horizons, valid_times) datetime64[ns] 2kB ...
│ valid_time (model_horizons, valid_times) datetime64[ns] 2kB ...
│ Dimensions without coordinates: model_horizons, valid_times
│ Data variables:
│ prmsl (model_horizons, valid_times, latitude, longitude) float64 124MB ...
│ Attributes:
│ typeOfLevel: meanSea
└── DataTree('ulwrf')
Dimensions: ()
│ Data variables:
*empty*
│ Attributes:
│ name: Upward long-wave radiation flux
└── DataTree('avg')
Dimensions: ()
│ Data variables:
*empty*
│ Attributes:
│ stepType: avg
└── DataTree('nominalTop')
Dimensions: (latitude: 181, longitude: 360, time: 1, step: 1,
model_horizons: 1, valid_times: 237)
Coordinates:
* latitude (latitude) float64 1kB 90.0 89.0 88.0 87.0 ... -88.0 -89.0 -90.0
* longitude (longitude) float64 3kB 0.0 1.0 2.0 3.0 ... 357.0 358.0 359.0
nominalTop float64 8B ...
number (time, step) int64 8B ...
step (model_horizons, valid_times) timedelta64[ns] 2kB ...
time (model_horizons, valid_times) datetime64[ns] 2kB ...
valid_time (model_horizons, valid_times) datetime64[ns] 2kB ...
Dimensions without coordinates: model_horizons, valid_times
Data variables:
ulwrf (model_horizons, valid_times, latitude, longitude) float64 124MB ...
Attributes:
typeOfLevel: nominalTop
.. tip::
For a full tutorial on this workflow, refer this `kerchunk cookbook <https://projectpythia.org/kerchunk-cookbook/README.html>`_
in `Project Pythia <https://projectpythia.org/>`_.

.. raw:: html

<script data-goatcounter="https://kerchunk.goatcounter.com/count"
async src="//gc.zgo.at/count.js"></script>

0 comments on commit af4c5dd

Please sign in to comment.