Lenient Cube Maths

This section provides an overview of lenient cube maths. In particular, it explains what lenient maths involves, clarifies how it differs from normal or strict cube maths, and demonstrates how you can exercise fine control over whether your cube maths operations are lenient or strict.

Note that, lenient cube maths is the default behaviour of Iris from version 3.0.0.

Introduction

Lenient maths stands somewhat on the shoulders of giants. If you’ve not already done so, you may want to recap the material discussed in the following sections,

In addition to this, cube maths leans heavily on the resolve module, which provides the necessary infrastructure required by Iris to analyse and combine each Cube operand involved in a maths operation into the resultant Cube. It may be worth while investing some time to understand how the Resolve class underpins cube maths, and consider how it may be used in general to combine or resolve cubes together.

Given these prerequisites, recall that lenient behaviour introduced and discussed the concept of lenient metadata; a more pragmatic and forgiving approach to comparing, combining and understanding the differences between your metadata (Table 1). The lenient metadata philosophy introduced there is extended to cube maths, with the view to also preserving as much common coordinate (Table 2) information, as well as common metadata, between the participating Cube operands as possible.

Let’s consolidate our understanding of lenient and strict cube maths through a practical worked example, which we’ll explore together next.

Lenient Example

Consider the following Cube of air_potential_temperature, which has an atmosphere hybrid height parametric vertical coordinate, and represents the output of an low-resolution global atmospheric experiment,

>>> print(experiment)
air_potential_temperature / (K)             (model_level_number: 15; grid_latitude: 100; grid_longitude: 100)
    Dimension coordinates:
        model_level_number                                     x                  -                    -
        grid_latitude                                          -                  x                    -
        grid_longitude                                         -                  -                    x
    Auxiliary coordinates:
        atmosphere_hybrid_height_coordinate                    x                  -                    -
        sigma                                                  x                  -                    -
        surface_altitude                                       -                  x                    x
    Derived coordinates:
        altitude                                               x                  x                    x
    Scalar coordinates:
        forecast_period                     0.0 hours
        forecast_reference_time             2009-09-09 17:10:00
        time                                2009-09-09 17:10:00
    Attributes:
        Conventions                         CF-1.5
        STASH                               m01s00i004
        experiment-id                       RT3 50
        source                              Data from Met Office Unified Model 7.04

Consider also the following Cube, which has the same global spatial extent, and acts as a control,

>>> print(control)
air_potential_temperature / (K)     (grid_latitude: 100; grid_longitude: 100)
    Dimension coordinates:
        grid_latitude                             x                    -
        grid_longitude                            -                    x
    Scalar coordinates:
        model_level_number          1
        time                        2009-09-09 17:10:00
    Attributes:
        Conventions                 CF-1.7
        STASH                       m01s00i004
        source                      Data from Met Office Unified Model 7.04

Now let’s subtract these cubes in order to calculate a simple difference,

>>> difference = experiment - control
>>> print(difference)
unknown / (K)                               (model_level_number: 15; grid_latitude: 100; grid_longitude: 100)
    Dimension coordinates:
        model_level_number                                     x                  -                    -
        grid_latitude                                          -                  x                    -
        grid_longitude                                         -                  -                    x
    Auxiliary coordinates:
        atmosphere_hybrid_height_coordinate                    x                  -                    -
        sigma                                                  x                  -                    -
        surface_altitude                                       -                  x                    x
    Derived coordinates:
        altitude                                               x                  x                    x
    Scalar coordinates:
        forecast_period                     0.0 hours
        forecast_reference_time             2009-09-09 17:10:00
        time                                2009-09-09 17:10:00
    Attributes:
        experiment-id                       RT3 50
        source                              Data from Met Office Unified Model 7.04

Note that, cube maths automatically takes care of broadcasting the dimensionality of the control up to that of the experiment, in order to calculate the difference. This is performed only after ensuring that both the dimension coordinates grid_latitude and grid_longitude are first leniently equivalent.

As expected, the resultant difference contains the HybridHeightFactory and all it’s associated auxiliary coordinates. However, the scalar coordinates have been leniently combined to preserve as much coordinate information as possible, and the attributes dictionaries have also been leniently combined. In addition, see what further rationalisation is always performed by cube maths on the resultant metadata and coordinates.

Also, note that the model_level_number scalar coordinate from the control has be superseded by the similarly named dimension coordinate from the experiment in the resultant difference.

Now let’s compare and contrast this lenient result with the strict alternative. But before we do so, let’s first clarify how to control the behaviour of cube maths.

Control the Behaviour

As stated earlier, lenient cube maths is the default behaviour from Iris 3.0.0. However, this behaviour may be controlled via the thread-safe LENIENT["maths"] runtime option,

>>> from iris.common import LENIENT
>>> print(LENIENT)
Lenient(maths=True)

Which may be set and applied globally thereafter for Iris within the current thread of execution,

>>> LENIENT["maths"] = False  
>>> print(LENIENT)  
Lenient(maths=False)

Or alternatively, temporarily alter the behaviour of cube maths only within the scope of the LENIENT context manager,

>>> print(LENIENT)
Lenient(maths=True)
>>> with LENIENT.context(maths=False):
...     print(LENIENT)
...
Lenient(maths=False)
>>> print(LENIENT)
Lenient(maths=True)

Strict Example

Now that we know how to control the underlying behaviour of cube maths, let’s return to our lenient example, but this time perform strict cube maths instead,

>>> with LENIENT.context(maths=False):
...     difference = experiment - control
...
>>> print(difference)
unknown / (K)                               (model_level_number: 15; grid_latitude: 100; grid_longitude: 100)
    Dimension coordinates:
        model_level_number                                     x                  -                    -
        grid_latitude                                          -                  x                    -
        grid_longitude                                         -                  -                    x
    Auxiliary coordinates:
        atmosphere_hybrid_height_coordinate                    x                  -                    -
        sigma                                                  x                  -                    -
        surface_altitude                                       -                  x                    x
    Derived coordinates:
        altitude                                               x                  x                    x
    Scalar coordinates:
        time                                2009-09-09 17:10:00
    Attributes:
        source                              Data from Met Office Unified Model 7.04

Although the numerical result of this strict cube maths operation is identical, it is not as rich in metadata as the lenient alternative. In particular, it does not contain the forecast_period and forecast_reference_time scalar coordinates, or the experiment-id in the attributes dictionary.

This is because strict cube maths, in general, will only return common metadata and common coordinates that are strictly equivalent.

Finer Detail

In general, if you want to preserve as much metadata and coordinate information as possible during cube maths, then opt to use the default lenient behaviour. Otherwise, favour the strict alternative if you require to enforce precise metadata and coordinate commonality.

The following information may also help you decide whether lenient cube maths best suits your use case,

  • lenient behaviour uses lenient equality to match the metadata of coordinates, which is more tolerant to certain metadata differences,

  • lenient behaviour uses lenient combination to create the metadata of coordinates on the resultant Cube,

  • lenient behaviour will attempt to cover each dimension with a DimCoord in the resultant Cube, even though only one Cube operand may describe that dimension,

  • lenient behaviour will attempt to include auxiliary coordinates in the resultant Cube that exist on only one Cube operand,

  • lenient behaviour will attempt to include scalar coordinates in the resultant Cube that exist on only one Cube operand,

  • lenient behaviour will add a coordinate to the resultant Cube with bounds, even if only one of the associated matching coordinates from the Cube operands has bounds,

  • strict and lenient behaviour both require that the points and bounds of matching coordinates from Cube operands must be strictly equivalent. However, mismatching bounds of scalar coordinates are ignored i.e., a scalar coordinate that is common to both Cube operands, with equivalent points but different bounds, will be added to the resultant Cube with but with no bounds

Additionally, cube maths will always perform the following rationalisation of the resultant Cube,

previous

Lenient Metadata

next

Developers Guide