Data Model

Background: The IAMC timeseries scenario data format

IAMC website

Over the past decade, the Integrated Assessment Modeling Consortium (IAMC) developed a standardised tabular timeseries format to exchange scenario data related to energy systems modelling, land-use change, demand sectors, and economic indicators in the context of climate change and the the Sustainable Development Goals. Previous high-level use cases include reports by the Intergovernmental Panel on Climate Change (IPCC) and model comparison exercises within the Energy Modeling Forum (EMF) hosted by Stanford University.

The table below shows a typical example of integrated-assessment scenario data following the IAMC format from the Horizon 2020 CD-LINKS project. The pyam package was initially developed for analysis and visualization of any scenario data provided in this format.

Illustrative example of IAMC-format timeseries data
via the IAMC 1.5°C Scenario Explorer

Refer to https://data.ece.iiasa.ac.at/ for a selected list of previous projects using the IAMC data format.

The pyam data model

A pyam.IamDataFrame instance is a container for two types of data:

• Scenario timeseries data (‘data’)

• Quantitative and qualitative indicators of scenarios (‘meta’)

Scenario timeseries data (‘data’)

This attribute holds the timeseries data related to an ensemble of scenarios. The data follows the format established by the IAMC shown above. It has the following standard columns (i.e., index dimensions, coordinates):

• model

• scenario

• region

• variable (more info on variables)

• unit

• year/time (more info on the temporal domain)

When initializing an IamDataFrame, the timeseries data can be provided in long format, where each datapoint is one row in a column named ‘value’, or it can follow the standard IAMC-style wide format (see the example above), where each timeseries is one row and the timesteps are represented as columns. Refer to the the data-table tutorial for more information.

The attribute data returns the timeseries data in long format as a pandas.DataFrame, while the method timeseries() returns an indexed pandas.DataFrame in wide format. Exporting an IamDataFrame to file using to_excel() or to_csv() also writes the data table in wide format.

Note

If there are numpy.nan in a pandas.DataFrame or empty cells in xlsx/csv files when initializing an IamDataFrame, it will behave as follows:

column	behaviour
‘value’	ignore/drop ‘nan’
‘unit’	replace ‘nan’ by an empty string ('')
all other	raise an error

Hence, if you are adding variable/region-specific meta information to ‘data’, you need to make sure that you add a value to every single row.

The reason for that implementation is that pandas does not work as expected with ‘nan’ in some situations (see here and here). Therefore, enforcing that there are no ‘nan’s in an IamDataFrame ensures that pyam has a clean dataset on which to operate.

The ‘variable’ column

The ‘variable’ column (or coordinate) describes the type of information represented in the specific timeseries. The variable implements a “semi-hierarchical” structure using the | character (pipe, not l or i) to indicate the depth. Variable names (should) follow a structure like Category|Subcategory|Specification.

Semi-hierarchical means that a hierarchy can be imposed, e.g., one can enforce that the sum of Emissions|CO2|Energy and Emissions|CO2|Other must be equal to Emissions|CO2 (if there are no other Emissions|CO2|… variables).

However, this is not mandatory, e.g., the sum of Primary Energy|Coal, Primary Energy|Gas and Primary Energy|Fossil should not be equal to Primary Energy because this would double-count fossil fuels.

Refer to the variable list in the documentation pages of the IAMC 1.5°C Scenario Explorer to see the full list of variables used in the recent IPCC Special Report on Global Warming of 1.5 ºC (SR15).

The temporal domain

In its original design, the IAMC data format (see above) assumed that the temporal dimension of any scenario data was restricted to full years represented as integer values. In this case, the time index is named ‘year’.

Two additional use cases are supported by pyam:

• continuous-time timeseries data by using an index named ‘time’
  populated by datetime.datetime instances (replacing the ‘year’ index)

• representative sub-annual timeslices via the extra-columns feature

Please reach out to the developers to get more information on this ongoing work.

Custom extra-columns of the ‘data’ table

If an IamDataFrame is initialised with columns that are not in the list above nor interpreted as belonging to the time dimension (in wide format), these columns are included in the ‘data’ table as custom extra-columns. This feature can be used, for example, to distinguish between multiple climate models providing different values for the variable Temperature|Global Mean. It can also be used to describe “representative timeslices”, e.g., “summer-day”, “winter-night”, … by adding an extra-column “subannual”.

Warning

Not all pyam functions currently support the continuous-time format or custom columns in a ‘data’ table. Please reach out via the mailing list or GitHub issues if you are not sure whether your use case is supported.

Quantitative and qualitative indicators of scenarios (‘meta’)

This attribute is intended for categorisation and quantitative indicators at the model-scenario level. Examples in the SR15 context are the warming category (‘Below 1.5°C’, ‘1.5°C with low overshoot’, etc.) and the cumulative CO2 emissions until the end of the century.

When performing operations such as rename() or append(), pyam attempts to keep the information in ‘meta’ consistent with the ‘data’ dataframe.

Note

The ‘meta’ table is not intended for annotations of individual data points. If you want to add meta information at this level (e.g., which stylized climate model provided the variable Temperature|Global Mean, or whether a data point is from the original data source or the result of an operation), this should operate on the ‘data’ table of the IamDataFrame using the custom-columns feature (see extra-columns above).

Filtering

The pyam package provides two methods for filtering scenario data:

An existing IamDataFrame can be filtered using filter(col=...), where col can be any column of the ‘data’ table (i.e., ‘model’, ‘scenario’, ‘region’, ‘unit’, ‘year’/’time’ or any extra-columns), or a column of the ‘meta’ table. The returned object is a new IamDataFrame instance.

A pandas.DataFrame (‘data’) with columns or index [‘model’, ‘scenario’] can be filtered by any ‘meta’ columns from an IamDataFrame (df) using pyam.filter_by_meta(data, df, col=..., join_meta=False). The returned object is a pandas.DataFrame down-selected to those models-and-scenarios where the ‘meta’ column satisfies the criteria given by col=... . Optionally, the ‘meta’ columns are joined to the returned dataframe.