We follow here the guidance presented by `intake-esm` on its [repository](https://intake-esm.readthedocs.io/en/latest/user-guide/cmip6-tutorial.html).
## Motivation of intake-esm
> Simulations of the Earth’s climate and weather generate huge amounts of data. These data are often persisted on different storages in a variety of formats (netCDF, zarr, etc...). Finding, investigating, loading these data assets into compute-ready data containers costs time and effort. The data user needs to know what data sets are available, the attributes describing each data set, before loading a specific data set and analyzing it.
> `Intake-esm` addresses these issues by providing necessary functionality for **searching, discovering, data access and data loading**.
%% Cell type:markdown id: tags:
For intake users, many data preparation tasks **are no longer necessary**. They do not need to know:
- 🌍 where data is saved
- 🪧 how data is saved
- 📤 how data should be loaded
but still can search, discover, access and load data of a project.
%% Cell type:markdown id: tags:
<aclass="anchor"id="features"></a>
## Features of intake and intake-esm
Intake is a generic **cataloging system** for listing data sources. As a plugin, `intake-esm` is built on top of `intake`, `pandas`, and `xarray` and configures `intake` such that it is able to also **load and process** ESM data.
- display catalogs as clearly structured tables 📄 inside jupyter notebooks for easy investigation
- browse 🔍 through the catalog and select your data without
- being next to the data (e.g. logged in on dkrz's luv)
- knowing the project's data reference syntax i.e. the storage tree hierarchy and path and file name templates
- open climate data in an analysis ready dictionary of `xarray` datasets 🎁
%% Cell type:markdown id: tags:
All required information for searching, accessing and loading the catalog's data is configured within the catalogs:
- 🌍 where data is saved
* users can browse data without knowing the data storage platform including e.g. the root path of the project and the directory syntax
* data of different platforms (cloud or disk) can be combined in one catalog
* on mid term, intake catalogs can be **a single point of access**
- 🪧 how data is saved
* users can work with a *xarray* dataset representation of the data no matter whether it is saved in **grb, netcdf or zarr** format.
* catalogs can contain more information an therefore more search facets than obvious from names and pathes of the data.
- 📤 how data should be loaded
* users work with an **aggregated***xarray* dataset representation which merges files/assets perfectly fitted to the project's data model design.
* with *xarray* and the underlying *dask* library, data which are **larger than the RAM** can be loaded
%% Cell type:markdown id: tags:
In this tutorial, we load a CMIP6 catalog which contains all data from the pool on DKRZ's mistral disk storage.
CMIP6 is the 6th phase of the Coupled Model Intercomparison Project and builds the data base used in the IPCC AR6.
The CMIP6 catalog contains all data that is published or replicated at the ESGF node at DKRZ.
%% Cell type:markdown id: tags:
<aclass="anchor"id="terminology"></a>
## Terminology: **Catalog**, **Catalog file** and **Collection**
We align our wording with `intake`'s [*glossary*](https://intake.readthedocs.io/en/latest/glossary.html) which is still evolving. The names overlap with other definitions, making it difficult to keep track. Here we try to give an overview of the hierarchy of catalog terms:
- a **top level catalog file** 📋 is the **main** catalog of an institution which will be opened first. It contains other project [*catalogs*](#catalog) 📖 📖 📖. Such catalogs can be assigned an [*intake driver*](#intakedriver) which is used to open and load the catalog within the top level catalog file. Technically, a catalog file 📋 is <aclass="anchor"id="catalogfile"></a>
- is a `.yaml` file
- can be opened with `open_catalog`, e.g.:
```python
intake.open_catalog(["https://dkrz.de/s/intake"])
```
-**intake driver**s also named **plugin**s are specified for [*catalogs*](#catalog) becaues they load specific data sets. There are [many driver](https://intake.readthedocs.io/en/latest/plugin-directory.html) libraries for intake. <aclass="anchor"id="intakedriver"></a>.
%% Cell type:markdown id: tags:
- a **catalog** 📖 (or collection) is defined by two parts: <aclass="anchor"id="catalog"></a>
- a **description** of a group of data sets. It describes how to *load***assets** of the data set(s) with the specified [driver](#intakedriver). This group forms an entity. E.g., all CMIP6 data sets can be collected in a catalog. <aclass="anchor"id="description"></a>
- an **asset** is most often a file. <aclass="anchor"id="asset"></a>
- a **collection** of all [assets](#asset) of the data set(s). <aclass="anchor"id="collection"></a>
- the collection can be included in the catalog or separately saved in a **data base** 🗂. In the latter case, the catalog references the data base, e.g.:
The term *collection* is often used synonymically for [catalog](#catalog).
```
%% Cell type:markdown id: tags:
- a *intake-esm***catalog** 📖 consists of a `.json` file (the **description**) and the underlying data base. The data base is either provided within the `.json` file or as a `.csv.gz` formatted list.
The intake-esm catalog can be opened with intake-esm's function `intake.open_esm_datastore()` where the `.json` part is the argument, e.g:
#note that intake_esm is imported with `import intake` as a plugin
importintake
#to find out the version of intake-esm, you can do:
importintake_esm
intake_esm.__version__
```
%% Cell type:markdown id: tags:
<aclass="anchor"id="browse"></a>
## Open and browse through catalogs
intake (not intake-esm) **opens** catalog-files in `yaml` format. These contain information about additonal sources: other catalogs/collections which will be loaded with specific *plugins*/*drivers*. The command is `open_catalog`.
<mark> You only need to remember one URL as the *single point of access* for DKRZ's intake catalogs: The DKRZ top level catalog can be accessed via dkrz.de/s/intake . Intake will only follow this *redirect* if a specific parser is activated. This can be done by providing the url in a list.</mark>
Right now, two versions of the top level catalog file exist: One for accessing the catalog via [cloud](https://gitlab.dkrz.de/data-infrastructure-services/intake-esm/-/raw/master/esm-collections/cloud_access/dkrz_catalog.yaml), one for via [disk](https://gitlab.dkrz.de/data-infrastructure-services/intake-esm/-/raw/master/esm-collections/disk_access/dkrz_catalog.yaml). They however contain **the same content**.
```
%% Cell type:markdown id: tags:
We can look into the catalog with `print` and `list`
%% Cell type:markdown id: tags:
Over the time, many collections have been created. `dkrz_catalog` is a **main** catalog prepared to keep an overview of all other collections. `list` shows all sub **project catalogs** which are available at DKRZ.
%% Cell type:code id: tags:
``` python
list(dkrz_catalog)
```
%% Cell type:markdown id: tags:
All these catalogs are **intake-esm** catalogs. You can find this information via the `_entries` attribute. The line `plugin: ['esm_datastore']
` refers to **intake-esm**'s function `open_esm_datastore()`.
%% Cell type:code id: tags:
``` python
print(dkrz_catalog._entries)
```
%% Cell type:markdown id: tags:
The DKRZ ESM-Collections follow a name template:
`dkrz_${project}_${store}[_${auxiliary_catalog}]`
where
-**project** can be one of the *model intercomparison project*, e.g. `cmip6`, `cmip5`, `cordex`, `era5` or `mpi-ge`.
-**store** is the data store and can be one of:
-`disk`: DKRZ holds a lot of data on a consortial disk space on the file system of the High Performance Computer (HPC) where it is accessible for every HPC user. Working next to the data on the file system will be the fastest way possible.
-`cloud`: A small subset is transferred into DKRZ's cloud in order to test the performance. swift is DKRZ's cloud storage.
-`archive`: A lot of data exists in the band archive of DKRZ. Before it can be accessed, it has to be retrieved. Therefore, catalogs for `hsm` are limited in functionality but still convenient for data browsing.
-**auxiliary_catalog** can be *grid*
%% Cell type:markdown id: tags:
**Why that convention?**:
-**dkrz**: Assume you work with internation collections. Than it may become important that you know from where the data comes, e.g. if only pathes on a local file system are given as the locations of the data.
-**project**: Project's data standards differ from each other so that different catalog attributes are required to identify a single asset in a project data base.
-**store**: Intake-esm cannot load data from all stores. Before data from the archive can be accessed, it has to be retrieved. Therefore, the opening function is not working for catalog merged for all stores.
%% Cell type:markdown id: tags:
**Best practice for naming catalogs**:
- Use small letters for all values
- Do **NOT** use `_` as a separator in values
- Do not repeat values of other attributes ("dkrz_dkrz-dyamond")
%% Cell type:markdown id: tags:
We could directly start to work with **two intake catalog** at the same time.
Let's have a look into a master catalog of [Pangeo](https://pangeo.io/):
While DKRZ's master catalog has one sublevel, Pangeo's is a nested one. We can access another `yaml` catalog which is also a **parent** catalog by simply:
%% Cell type:code id: tags:
``` python
pangeo.climate
```
%% Cell type:markdown id: tags:
Pangeo's ESM collections are one level deeper in the catalog tree:
%% Cell type:code id: tags:
``` python
list(pangeo.climate)
```
%% Cell type:markdown id: tags:
### The `intake-esm` catalogs
We now look into a catalog which is opened by the plugin `intake-esm`.
> An ESM (Earth System Model) collection file is a `JSON` file that conforms to the ESM Collection Specification. When provided a link/path to an esm collection file, intake-esm establishes a link to a database (`CSV` file) that contains data assets locations and associated metadata (i.e., which experiment, model, the come from).
Since the data base of the CMIP6 ESM Collection is about 100MB in compressed format, it takes up to a minute to load the catalog.
%% Cell type:markdown id: tags:
```{note}
The project catalogs contain only valid and current project data. They are constantly updated.
If your work is based on a catalog and a subset of the data from it, be sure to save that subset so you can later compare your database to the most current catalog.
```
%% Cell type:code id: tags:
``` python
esm_col=dkrz_catalog.dkrz_cmip6_disk
print(esm_col)
```
%% Cell type:markdown id: tags:
`intake-esm` gives us an overview over the content of the ESM collection. The ESM collection is a data base described by specific attributes which are technically columns. Each project data standard is the basis for the columns and used to parse information given by the path and file names.
The pure display of `esm_col` shows us the number of unique values in each column. Since each `uri` refers to one file, we can conclude that the DKRZ-CMIP6 ESM Collection contains **6.1 Mio Files** in 2022.
%% Cell type:markdown id: tags:
The data base is loaded into an underlying `panda`s dataframe which we can access with `esm_col.df`. `esm_col.df.head()` displays the first rows of the table:
%% Cell type:code id: tags:
``` python
esm_col.df.head()
```
%% Cell type:markdown id: tags:
We can find out details about `esm_col` with the object's attributes. `esm_col.esmcol_data` contains all information given in the `JSON` file. We can also focus on some specific attributes.
%% Cell type:code id: tags:
``` python
dict(esm_col.esmcat)["description"]
```
%% Cell type:code id: tags:
``` python
print("What is this catalog about? \n"+dict(esm_col.esmcat)["description"])
#
print("The link to the data base: "+dict(esm_col.esmcat)["catalog_file"])
```
%% Cell type:markdown id: tags:
Advanced: To find out how many datasets are available, we can use pandas functions (drop columns that are irrelevant for a dataset, drop the duplicates, keep one):
You will browse the collection technically by setting values the **column names** of the underlying table. Per default, the catalog was loaded with all cmip6 attributes/columns that define the CMIP6 data standard:
%% Cell type:code id: tags:
``` python
esm_col.df.columns
```
%% Cell type:markdown id: tags:
These are configured in the top level catalog so you <mark> do not need to open the catalog to see the columns </mark>
Most of the time, we want to set more than one attribute for a search. Therefore, we define a query `dict`ionary and use the `search` function of the `esm_col` object. In the following case, we look for temperature at surface in monthly resolution for 3 different experiments:
Note that only the combination of a `variable_id` and a `table_id` is unique in CMIP6. If you search for `tas` in all tables, you will find many entries more:
Be careful when you search for specific time slices. Each frequency is connected with a individual name template for the filename. If the data is yearly, you have YYYY-YYYY whereas you have YYYYMM-YYYYMM for monthly data.
%% Cell type:markdown id: tags:
### How to load more columns
Intake allows to load only a subset of the columns that is inside the **intake-esm** catalog. Since the memory usage of **intake-esm** is high, the default columns are only a subset from all possible columns. Sometimes, other columns are of interest:
If you work remotely away from the data, you can use the **opendap_url**'s to access the subset of interest for all files published at DKRZ. The *opendap_url* is an *additional* column that can also be loaded.
We can define 3 different column name types for the usage of intake catalogs:
1.**Default** attributes which are loaded from the main catalog and which can be seen via `_entries[CATNAME]._open_args`.
2.**Overall** attributes or **template** attributes which should be defined for **ALL** catalogs at DKRZ (exceptions excluded). At DKRZ, we use the newly defined **Cataloonie** scheme template which can be found via `dkrz_catalog.metadata["parameters"]["cataloonie_columns"]`. With these template attributes, there may be redundancy in the columns. They exist to simplify merging catalogs across projects.
3.**Additional** attributes which are not necessary to identify a single asset but helpful for users. You can find these via
You may find *variable_id*s in the catalog which are not obvious or abbrevations for a clear variable name. In that cases you would need additional information like a *long_name* of the variable. For CMIP6, we provided the catalog with this `long_name` so you could add it as a column.
```
%% Cell type:markdown id: tags:
So, this is the instruction how to open the catalog with additional columns:
1. create a combination of all your required columns:
- ⭐ The customization of catalog columns allows highest flexibility for intake users.
- ⭐ In theory, we could add many more columns with additional information because ot all have to be loaded from the data base.
%% Cell type:markdown id: tags:
```{warning}
The number of columns determines the required memory.
```
%% Cell type:markdown id: tags:
```{tip}
If you work from remote and also want to access the data remotely, load the *opendap_url* column.
```
%% Cell type:markdown id: tags:
<aclass="anchor"id="dataaccess"></a>
## Access and load data of the ESM collection
With the power of `xarray`, `intake` can load your subset into a `dict`ionary of datasets. We therefore focus on the data of `MPI-ESM1-2-LR`:
%% Cell type:code id: tags:
``` python
#case insensitive?
query=dict(
variable_id="tas",
table_id="Amon",
source_id="MPI-ESM1-2-LR",
experiment_id="historical")
cat=esm_col.search(**query)
cat
```
%% Cell type:markdown id: tags:
You can find out which column intake uses to access the data via the following keyword:
%% Cell type:code id: tags:
``` python
cat.esmcat.assets.column_name
```
%% Cell type:markdown id: tags:
As we are working with the *_disk* catalog, **uri** contains *pathes* to the files on filesystem. If you are working from remote, you would have
- to change the catalog's attribute `path_column_name` to *opendap_url*.
- to reassign the `format` column from *netcdf* to *opendap*
as follows:
%% Cell type:code id: tags:
``` python
#cat.esmcat.assets.column_name="opendap_url"
#newdf=cat.df.copy()
#newdf.loc[:,"format"]="opendap"
#cat.df=newdf
```
%% Cell type:markdown id: tags:
**Intake-ESM** natively supports the following data formats or access formats (since opendap is not really a file format):
- netcdf
- opendap
- zarr
You can also open **grb** data but right now only by specifying xarray's attribute *engine* in the *open* function which is defined in the following. I.e., it does not make a difference if you specify **grb** as format.
You can find an example in the *era5* notebook.
%% Cell type:markdown id: tags:
The function to open data is `to_dataset_dict`.
We recommend to set a keyword argument `xarray_open_kwargs` for the chunk size of the variable's data array. Otherwise, `xarray` may choose too large chunks. Most often, your data contains a time dimension so that you could set `xarray_open_kwargs={"chunks":{"time":1}}`.
If your collection contains **zarr** formatted data, you need to add another keyword argument `zarr_kwargs`.
**Unfortunately, this has changed in new versions**:
- DEPRICATED The trick is: You can just specify both. Intake knows from the `format` column which *kwargs* should be taken.
%% Cell type:code id: tags:
``` python
help(cat.to_dataset_dict)
```
%% Cell type:code id: tags:
``` python
xr_dict=cat.to_dataset_dict(
#xarray_open_kwargs=
#dict(
# chunks=dict(
# time=60
# ),
#consolidated=True
#decode_times=True,
#use_cftime=True
#)
)
xr_dict
```
%% Cell type:markdown id: tags:
`Intake` was able to aggregate many files into only one dataset:
- The `time_range` column was used to **concat** data along the `time` dimension
- The `member_id` column was used to generate a new dimension
The underlying `dask` package will only load the data into memory if needed. Note that attributes which disagree from file to file, e.g. *tracking_id*, are excluded from the dataset.
%% Cell type:markdown id: tags:
How **intake-esm** should open and aggregate the assets is configured in the *aggregation_control* part of the description:
%% Cell type:code id: tags:
``` python
cat.esmcat.aggregation_control.aggregations
```
%% Cell type:markdown id: tags:
Columns can be defined for appending or creating new dimensions. The *options* are keyword arguments for xarray.
They **keys** of the dictionary are made with column values defined in the *aggregation_control* of the **intake-esm** catalog. These will determine the **key_template**. The corresponding commands are:
You can work with these keys **directly** on the **intake-esm** catalog which will give you an overview over all columns (too long for the web page):
%% Cell type:code id: tags:
``` python
#cat["CMIP.MPI-ESM1-2-HR.historical.Amon.gn"]
```
%% Cell type:markdown id: tags:
If we are only interested in the **first** dataset of the dictionary, we can *pop it out*:
%% Cell type:code id: tags:
``` python
xr_dset=xr_dict.popitem()[1]
xr_dset
```
%% Cell type:markdown id: tags:
### Troubleshooting
The variables are collected in **one** dataset. This requires that **the dimensions and coordinates must be the same over all files**. Otherwise, xarray cannot merge these together.
For CMIP6, most of the variables collected in one **table_id** should be on the same dimensions and coordinates. Unfortunately, there are exceptions.:
- a few variables are requested for *time slices* only.
- sometimes models use different dimension names from file to file
Using the [preprocessing](https://tutorials.dkrz.de/tutorial_intake-4-preprocessing-derived-vars.html#use-preprocessing-when-opening-assets-and-creating-datasets) keyword argument can help to rename dimensions before merging.
For Intake providers: the more information on the dimensions and coordinates provided already in the catalog, the better the aggregation control.
%% Cell type:markdown id: tags:
### Pangeo's data store
Let's have a look into Pangeo's ESM Collection as well. This is accessible via cloud from everywhere - you only need internet to load data. We use the same `query` as in the example before.
There are differences between the collections because
- Pangeo provides files in *consolidated*, `zarr` formatted datasets which correspond to `zstore` entries in the catalog instead of `path`s or `opendap_url`s.
- The `zarr` datasets are already aggregated over time so there is no need for a `time_range` column
If we now open the data with `intake`, we have to specify keyword arguments as follows:
`dset_dict` and `xr_dict` are the same. You succesfully did the intake tutorial!
%% Cell type:markdown id: tags:
### Making a quick plot
The following line exemplifies the ease of the intake's data processing library chain. On the web page, the interactivity will not work as all plots would have to be loaded which is not feasible.
For more examples, check out the **use cases** on that web page.
* [Part 4: Use preprocessing and create derived variables](https://data-infrastructure-services.gitlab-pages.dkrz.de/tutorials-and-use-cases/tutorial_intake-4-preprocessing-derived-variables.html)
* [Part 5: How to create an intake catalog](https://data-infrastructure-services.gitlab-pages.dkrz.de/tutorials-and-use-cases/tutorial_intake-5-create-esm-collection.html)
- You can also do another [CMIP6 tutorial](https://intake-esm.readthedocs.io/en/latest/user-guide/cmip6-tutorial.html) from the official intake page.
