Regridding an ancillary field to a target grid#

Producing an ancillary usually means taking source data on one grid and regridding it onto a target model grid. ANTS provides regridding capability that goes beyond what Iris offers directly, in particular schemes that are useful when moving between very different grid types (for example a lat-lon source and a rotated pole or cubesphere target). See ants.regrid for the full picture; Iris itself is the right place to look for capability that ANTS does not add to (see iris.analysis).

Horizontal and vertical regridding#

ANTS provides three groups of regridding scheme:

  • ants.regrid.rectilinear - horizontal regridding/interpolation schemes for rectilinear grids, for example Linear, AreaWeighted and a TwoStage scheme that performs an intermediate linear regrid before an area weighted regrid, which is useful when the source and target coordinate reference systems differ substantially.

  • ants.regrid.interpolation - vertical, points-based interpolation schemes.

  • ants.regrid.esmf - regridding schemes that use the ESMF framework, useful for more complex source/target grid combinations (for example unstructured or cubesphere grids).

Rather than choosing a horizontal or vertical scheme directly in your application code, use ants.regrid.GeneralRegridScheme. This lets the regridding method be swapped later via configuration, without changing your application:

import ants

scheme = ants.regrid.GeneralRegridScheme(horizontal_scheme="TwoStage")

Worked example#

The following mirrors the approach taken by the ancil_general_regrid application (see Worked examples: the ANTS command line tools). We build a small source cube and a target grid, both using ants.tests.stock, and regrid the source onto the target:

import numpy as np
import ants
import ants.tests.stock as stock

# source data on a coarse grid; data must already be shaped to match
# the cube's shape, so reshape the flat array of values first
source = stock.geodetic((4, 4), data=np.arange(16).reshape(4, 4))

# target grid to regrid onto
target = stock.geodetic((8, 8))

scheme = ants.regrid.GeneralRegridScheme(horizontal_scheme="TwoStage")
regridding_result = source.regrid(target, scheme)

regridding_result is a cube on the target grid. In an application dealing with larger, real datasets you would typically pass the regridding operation through ants.decomposition.decompose() rather than calling regrid directly, so that the regrid can be split into pieces that fit into memory - see ANTS Decomposition Framework for details.

Note

For the example above using the TwoStage scheme is redundant as the source and target are on the same grid - it achieves the same result as directly calling the AreaWeighted scheme.

Regridding onto a land sea mask#

Where the target has a land sea mask, you will usually want the regridded result to be consistent with it, so that land points do not end up with sea sourced values and vice versa. Use ants.analysis.make_consistent_with_lsm() after regridding to fill any points that need correcting:

target_lsm = ants.io.load.load_landsea_mask("/path/to/target_lsm.nc")

ants.analysis.make_consistent_with_lsm(
    result, target_lsm, invert_mask=True, method="kdtree"
)

See Merging datasets and filling missing data for more on the fill algorithms make_consistent_with_lsm uses under the hood, including which one is recommended.

Zonal mean behaviour#

If the source and target both have global extent in the x axis, or the source has only a single column in x, the target is treated specially so that the result is a proper zonal mean, regardless of how many longitude points the target actually has.

Key Points#

See Also#

On this page
Show Source