Skip to content

Commit

Permalink
Shapefile masking (#5470)
Browse files Browse the repository at this point in the history
* Working draft of shapefile masking

* Version of shapefile masking with tests and ready for preliminary review

* Updated tests with proper paths and skip_tests decorator

* fixed some paths and removed broken code

* Added more tests and split into integration and unit tests. Testing warnings/errors aswell.

* responces to comments on utils.py for shapefile masking

* tests actually pass now

* Moved tests to correct locations and strted changes on _shapefiles.py

* some changes to _shapefiles to match review

* added setUp cases to tests

* moved test names to lower_case and added acknoledgment

* removed seperate guess_bounds function

* updated structure to properly call coord names/coords when optimal

* sphnix improvements to docstring

Co-authored-by: Martin Yeo <[email protected]>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* commited dask map_blocks approach and some test improvements

* replaced bounds rebasing via modulus with vectorized version

* Dask chunk control and some docstrings

* reverted behaviour of modulus function to ASCEND and switcher argument order in utils

* edied tests to work with flipped argument order

* Improved optimisation by reading shapely docs properly and just using a dask.delayed call

* Docstring updates and a 4d integration test

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Update lib/iris/_shapefiles.py

Co-authored-by: Martin Yeo <[email protected]>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* improving readability from martin

* removed dask.delayed call

* Update lib/iris/_shapefiles.py

Co-authored-by: Martin Yeo <[email protected]>

* Update lib/iris/_shapefiles.py

Co-authored-by: Martin Yeo <[email protected]>

* Update lib/iris/util.py

Co-authored-by: Martin Yeo <[email protected]>

* Added warning for possible mismatch of mask/cube coords

* test for new warning

* added test

* Update lib/iris/_shapefiles.py

Co-authored-by: Martin Yeo <[email protected]>

* Added licenses

* fixed doctest failures in example

* Improved test coverage

* fixed doctest

* doctest again

* Docstring tidy up.

* fixed prime meridian bug

* Update lib/iris/_shapefiles.py

Co-authored-by: Martin Yeo <[email protected]>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Added first draft of user guide page

* Add What's New entry.

---------

Co-authored-by: Martin Yeo <[email protected]>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Martin Yeo <[email protected]>
  • Loading branch information
4 people authored Feb 13, 2024
1 parent 117dd76 commit 1b11d74
Show file tree
Hide file tree
Showing 7 changed files with 605 additions and 2 deletions.
24 changes: 24 additions & 0 deletions docs/src/userguide/plotting_examples/masking_brazil_plot.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,24 @@
"""Global cube masked to Brazil and plotted with quickplot."""
import cartopy.io.shapereader as shpreader
import matplotlib.pyplot as plt

import iris
import iris.quickplot as qplt
from iris.util import mask_cube_from_shapefile

country_shp_reader = shpreader.Reader(
shpreader.natural_earth(
resolution="110m", category="cultural", name="admin_0_countries"
)
)
brazil_shp = [
country.geometry
for country in country_shp_reader.records()
if "Brazil" in country.attributes["NAME_LONG"]
][0]

cube = iris.load_cube(iris.sample_data_path("air_temp.pp"))
brazil_cube = mask_cube_from_shapefile(cube, brazil_shp)

qplt.pcolormesh(brazil_cube)
plt.show()
37 changes: 36 additions & 1 deletion docs/src/userguide/subsetting_a_cube.rst
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,8 @@ Subsetting a Cube
=================

The :doc:`loading_iris_cubes` section of the user guide showed how to load data into multidimensional Iris cubes.
However it is often necessary to reduce the dimensionality of a cube down to something more appropriate and/or manageable.
However it is often necessary to reduce the dimensionality of a cube down to something more appropriate and/or manageable,
or only examine and analyse a subset of data in a dimension.

Iris provides several ways of reducing both the amount of data and/or the number of dimensions in your cube depending on the circumstance.
In all cases **the subset of a valid cube is itself a valid cube**.
Expand Down Expand Up @@ -329,6 +330,36 @@ on bounds can be done in the following way::
The above example constrains to cells where either the upper or lower bound occur
after 1st January 2008.

Cube Masking
--------------

.. _masking-from-shapefile:

Masking from a shapefile
^^^^^^^^^^^^^^^^^^^^^^^^

Often we want to perform so kind of analysis over a complex geographical feature - only over land points or sea points:
or over a continent, a country, a river watershed or administrative region. These geographical features can often be described by shapefiles.
Shapefiles are a file format first developed for GIS software in the 1990s, and now `Natural Earth`_ maintain a large freely usable database of shapefiles of many geographical and poltical divisions,
accessible via cartopy. Users may also provide their own custom shapefiles.

These shapefiles can be used to mask an iris cube, so that any data outside the bounds of the shapefile is hidden from further analysis or plotting.

First, we load the correct shapefile from NaturalEarth via the `Cartopy`_ instructions. Here we get one for Brazil.
The `.geometry` attribute of the records in the reader contain the shapely polygon we're interested in - once we have those we just need to provide them to
the :class:`iris.util.mask_cube_from_shapefile` function. Once plotted, we can see that only our area of interest remains in the data.


.. plot:: userguide/plotting_examples/masking_brazil_plot.py
:include-source:

We can see that the dimensions of the cube haven't changed - the plot is still global. But only the data over Brazil is plotted - the rest is masked.

.. note::
While Iris will try to dynamically adjust the shapefile to mask cubes of different projections, it can struggle with rotated pole projections and cubes with Meridians not at 0°
Converting your Cube's coordinate system may help if you get a fully masked cube from this function.


Cube Iteration
--------------
It is not possible to directly iterate over an Iris cube. That is, you cannot use code such as
Expand Down Expand Up @@ -440,3 +471,7 @@ Similarly, Iris cubes have indexing capability::
# Get the second element of the first dimension and all of the second dimension
# in reverse, by steps of two.
print(cube[1, ::-2])


.. _Cartopy: https://scitools.org.uk/cartopy/docs/latest/tutorials/using_the_shapereader.html#id1
.. _Natural Earth: https://www.naturalearthdata.com/
12 changes: 11 additions & 1 deletion docs/src/whatsnew/latest.rst
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,12 @@ This document explains the changes made to Iris for this release
NetCDF chunking with the use of the :data:`iris.fileformats.netcdf.loader.CHUNK_CONTROL`
context manager. (:pull:`5588`)

#. `@acchamber`_ and `@trexfeathers`_ (reviewer) added
:func:`iris.util.mask_cube_from_shapefile`. This builds on the original work
of `@ckmo`_, `@david-bentley`_, `@jmendesmetoffice`_, `@evyve`_ and
`@pelson`_ for the UK Met Office **ASCEND** library. See
:ref:`masking-from-shapefile` for documentation. (:pull:`5470`)


🐛 Bugs Fixed
=============
Expand Down Expand Up @@ -110,7 +116,7 @@ This document explains the changes made to Iris for this release
#. `@bouweandela`_ changed :func:`iris.coords.Coord.cell` so it does not realize
all coordinate data and only loads a single cell instead. (:pull:`5693`)

#. `@rcomer`_ and `@trexfeathers`_ (reviewer) modified
#. `@rcomer`_ and `@trexfeathers`_ (reviewer) modified
:func:`~iris.analysis.stats.pearsonr` so it preserves lazy data in all cases
and also runs a little faster. (:pull:`5638`)

Expand Down Expand Up @@ -242,6 +248,10 @@ This document explains the changes made to Iris for this release
.. _@scottrobinson02: https://github.com/scottrobinson02
.. _@acchamber: https://github.com/acchamber
.. _@fazledyn-or: https://github.com/fazledyn-or
.. _@ckmo: https://github.com/ckmo
.. _@david-bentley: https://github.com/david-bentley
.. _@jmendesmetoffice: https://github.com/jmendesmetoffice
.. _@evyve: https://github.com/evyve


.. comment
Expand Down
243 changes: 243 additions & 0 deletions lib/iris/_shapefiles.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,243 @@
# Copyright Iris contributors
#
# This file is part of Iris and is released under the BSD license.
# See LICENSE in the root of the repository for full licensing details.

# Much of this code is originally based off the ASCEND library, developed in
# the Met Office by Chris Kent, Emilie Vanvyve, David Bentley, Joana Mendes
# many thanks to them. Converted to iris by Alex Chamberlain-Clay


from itertools import product
import warnings

import numpy as np
import shapely
import shapely.errors
import shapely.geometry as sgeom
import shapely.ops

from iris.exceptions import IrisDefaultingWarning, IrisUserWarning


def create_shapefile_mask(
geometry,
cube,
minimum_weight=0.0,
):
"""Make a mask for a cube from a shape.
Get the mask of the intersection between the
given shapely geometry and cube with x/y DimCoords.
Can take a minimum weight and evaluate area overlaps instead
Parameters
----------
geometry : :class:`shapely.Geometry`
cube : :class:`iris.cube.Cube`
A :class:`~iris.cube.Cube` which has 1d x and y coordinates
minimum_weight : float, default 0.0
A float between 0 and 1 determining what % of a cell
a shape must cover for the cell to remain unmasked.
eg: 0.1 means that at least 10% of the shape overlaps the cell
to be unmasked.
Requires geometry to be a Polygon or MultiPolygon
Defaults to 0.0 (eg only test intersection)
Returns
-------
:class:`np.array`
An array of the shape of the x & y coordinates of the cube, with points
to mask equal to True
"""
from iris.cube import Cube, CubeList

try:
msg = "Geometry is not a valid Shapely object"
if not shapely.is_valid(geometry):
raise TypeError(msg)
except Exception:
raise TypeError(msg)
if not isinstance(cube, Cube):
if isinstance(cube, CubeList):
msg = "Received CubeList object rather than Cube - \
to mask a CubeList iterate over each Cube"
raise TypeError(msg)
else:
msg = "Received non-Cube object where a Cube is expected"
raise TypeError(msg)
if minimum_weight > 0.0 and isinstance(
geometry,
(
sgeom.Point,
sgeom.LineString,
sgeom.LinearRing,
sgeom.MultiPoint,
sgeom.MultiLineString,
),
):
minimum_weight = 0.0
warnings.warn(
"""Shape is of invalid type for minimum weight masking,
must use a Polygon rather than Line shape.\n
Masking based off intersection instead. """,
category=IrisDefaultingWarning,
)

# prepare 2D cube
y_name, x_name = _cube_primary_xy_coord_names(cube)
cube_2d = cube.slices([y_name, x_name]).next()
for coord in cube_2d.dim_coords:
if not coord.has_bounds():
coord.guess_bounds()
trans_geo = _transform_coord_system(geometry, cube_2d)

y_coord, x_coord = [cube_2d.coord(n) for n in (y_name, x_name)]
x_bounds = _get_mod_rebased_coord_bounds(x_coord)
y_bounds = _get_mod_rebased_coord_bounds(y_coord)
# prepare array for dark
box_template = [
sgeom.box(x[0], y[0], x[1], y[1]) for x, y in product(x_bounds, y_bounds)
]
# shapely can do lazy evaluation of intersections if it's given a list of grid box shapes
# delayed lets us do it in parallel
intersect_template = shapely.intersects(trans_geo, box_template)
# we want areas not under shapefile to be True (to mask)
intersect_template = np.invert(intersect_template)
# now calc area overlaps if doing weights and adjust mask
if minimum_weight > 0.0:
intersections = np.array(box_template)[~intersect_template]
intersect_template[~intersect_template] = [
trans_geo.intersection(box).area / box.area <= minimum_weight
for box in intersections
]
mask_template = np.reshape(intersect_template, cube_2d.shape[::-1]).T
return mask_template


def _transform_coord_system(geometry, cube, geometry_system=None):
"""Project the shape onto another coordinate system.
Parameters
----------
geometry: :class:`shapely.Geometry`
cube: :class:`iris.cube.Cube`
:class:`~iris.cube.Cube` with the coord_system to be projected to and
a x coordinate
geometry_system: :class:`iris.coord_systems`, optional
A :class:`~iris.coord_systems` object describing
the coord_system of the shapefile. Defaults to None,
which is treated as GeogCS
Returns
-------
:class:`shapely.Geometry`
A transformed copy of the provided :class:`shapely.Geometry`
"""
y_name, x_name = _cube_primary_xy_coord_names(cube)
import iris.analysis.cartography

DEFAULT_CS = iris.coord_systems.GeogCS(
iris.analysis.cartography.DEFAULT_SPHERICAL_EARTH_RADIUS
)
target_system = cube.coord_system()
if not target_system:
warnings.warn(
"Cube has no coord_system; using default GeogCS lat/lon",
category=IrisDefaultingWarning,
)
target_system = DEFAULT_CS
if geometry_system is None:
geometry_system = DEFAULT_CS
target_proj = target_system.as_cartopy_projection()
source_proj = geometry_system.as_cartopy_projection()

trans_geometry = target_proj.project_geometry(geometry, source_proj)
# A GeogCS in iris can be either -180 to 180 or 0 to 360. If cube is 0-360, shift geom to match
if (
isinstance(target_system, iris.coord_systems.GeogCS)
and cube.coord(x_name).points[-1] > 180
):
# chop geom at 0 degree line very finely then transform
prime_meridian_line = shapely.LineString([(0, 90), (0, -90)])
trans_geometry = trans_geometry.difference(prime_meridian_line.buffer(0.00001))
trans_geometry = shapely.transform(trans_geometry, _trans_func)

if (not isinstance(target_system, iris.coord_systems.GeogCS)) and cube.coord(
x_name
).points[-1] > 180:
# this may lead to incorrect masking or not depending on projection type so warn user
warnings.warn(
"""Cube has x-coordinates over 180E and a non-standard projection type.\n
This may lead to incorrect masking. \n
If the result is not as expected, you might want to transform the x coordinate points of your cube to -180-180 """,
category=IrisUserWarning,
)
return trans_geometry


def _trans_func(geometry):
"""Pocket function for transforming the x coord of a geometry from -180 to 180 to 0-360."""
for point in geometry:
if point[0] < 0:
point[0] = 360 - np.abs(point[0])
return geometry


def _cube_primary_xy_coord_names(cube):
"""Return the primary latitude and longitude coordinate names, or long names, from a cube.
Parameters
----------
cube : :class:`iris.cube.Cube`
Returns
-------
tuple of str
The names of the primary latitude and longitude coordinates
"""
latc = (
cube.coords(axis="y", dim_coords=True)[0]
if cube.coords(axis="y", dim_coords=True)
else -1
)
lonc = (
cube.coords(axis="x", dim_coords=True)[0]
if cube.coords(axis="x", dim_coords=True)
else -1
)

if -1 in (latc, lonc):
msg = "Error retrieving 1d xy coordinates in cube: {!r}"
raise ValueError(msg.format(cube))

latitude = latc.name()
longitude = lonc.name()
return latitude, longitude


def _get_mod_rebased_coord_bounds(coord):
"""Take in a coord and returns a array of the bounds of that coord rebased to the modulus.
Parameters
----------
coord : :class:`iris.coords.Coord`
An Iris coordinate with a modulus
Returns
-------
:class:`np.array`
A 1d Numpy array of [start,end] pairs for bounds of the coord
"""
modulus = coord.units.modulus
# Force realisation (rather than core_bounds) - more efficient for the
# repeated indexing happening downstream.
result = np.array(coord.bounds)
if modulus:
result[result < 0.0] = (np.abs(result[result < 0.0]) % modulus) * -1
result[np.isclose(result, modulus, 1e-10)] = 0.0
return result
Loading

0 comments on commit 1b11d74

Please sign in to comment.