What’s New in Iris#

v3.16 (13 Aug 2026)#

This document explains the changes made to Iris for this release (View all changes.)

✨ Features#

#6799: @stephenworsley introduced the function guess_2D_bounds() for guessing the bounds of pairs of 2D coordinates. ( #6097)
#7098: @pp-mo, @trexfeathers, @ukmo-ccbunney and @pt331 improved NetCDF character-array handling so Iris now supports string data in cube data arrays (as well as label coordinates), honours _Encoding for decode/encode, and saves string arrays back to char variables. Read the detail here: Character and String datatypes. ( #6309)
#7113: @trexfeathers and @stephenworsley added support for loading and saving of Zarr files formatted in Zarr Storage Spec Version 2 via the NetCDF NCZarr implementation. ( #6967)
#7146: @bjlittle extended the cell() and cells() methods to allow users to specify that they want datetime compatible objects returned within each generated Cell from a temporal coordinate. This is achieved by passing the keyword argument pydate=True. ( #7112)
#7185: @pt331 and @trexfeathers and @pp-mo allowed Iris’ Mesh Support, including saving, to handle coordinates without standard_name. ( #4863)
#7185: @pt331 and @trexfeathers and @pp-mo improved iris.mesh.components.MeshXY.coords() and coord() to better detect the AXES of constituent coordinates. It previously depended on guess_coord_axis().
#7210: @trexfeathers, @stephenworsley and @pt331 implemented an initial draft of the UGRID Location Index Set: iris.mesh.components._MeshIndexSet. This remains a work in progress; the implementation will evolve as we keep learning, and notable current absences include: I/O support, tracking of the original MeshXY content, and detailed documentation. But we wanted to give users the earliest possible chance to try this valuable addition to the Iris mesh/UGRID data model. You can track the ongoing progress of this project at: #6120.
#7210: @trexfeathers, @stephenworsley added the iris.experimental.mesh_coord_indexing module, allowing the use of the new iris.mesh.components._MeshIndexSet class to index meshes and cubes-with-meshes. The new indexed mesh can be converted to a separate MeshXY instance - trusted reliable behaviour - OR as a _MeshIndexSet instance - for the latest experimental functionality. Please read more and give it a try! ( #6120)
#7237: @HGWright updated the CF standard names table to version 94.

🐛 Bugs Fixed#

#7119: @gaoflow fixed iris.analysis.cartography.wrap_lons() so that it preserves the floating-point dtype of its input (e.g. float32) instead of always promoting the result to float64. Integer inputs are still returned as float64. ( #4119)
#7120: @gaoflow fixed iris.coords.Coord.cell() so that a masked bound is preserved as masked, instead of revealing the value stored underneath the mask. This also corrects the cube/coordinate printout for such bounds. ( #5158)
#7121: @gaoflow made iris.Constraint raise a TypeError when used in a boolean context (e.g. with the and/or/not keywords) instead of silently discarding one of the constraints. Use the & operator to combine constraints. ( #4337)
#7122: @gaoflow fixed coordinate summaries so that length-0 string and date-like coordinates print as empty arrays instead of raising ValueError. ( #6531)
#7123: @gaoflow fixed iris.cube.Cube.rolling_window() so that it no longer raises a TypeError when the cube has a lazy auxiliary coordinate along the windowed dimension. ( #6480)
#7124: @gaoflow fixed regridding so that horizontal coordinates of the result cube are copies of, rather than references to, the target grid’s coordinates. Previously, modifying a result coordinate also modified the target cube’s coordinate, and vice versa. ( #6844)
#7125: @gaoflow fixed iris.cube.CubeList.concatenate() so that cubes with string coordinates of differing width, such as dtypes <U1 and <U5, can be concatenated. The result coordinate is promoted to the wider dtype. ( #6676)
#7126: @gaoflow fixed an error when computing (e.g. saving) a scalar lazy cube whose units had been converted with convert_units(). The unit conversion could yield a plain Python scalar for the 0-dimensional block, which Dask was then unable to store. ( #6965)
#7197: @trexfeathers and @kwilliams-mo fixed the saving of PP fields to now always write lbft in hours in the no-bounds forecast period case, including when forecast_period uses non-hour units such as seconds.

💣 Incompatible Changes#

#7113: @trexfeathers altered iris.io.decode_uri() to separate out URI fragments (the section following the # character) - decode_uri() now returns a 3-part tuple: scheme, path, and fragment.

🚀 Performance Enhancements#

#7089: @trexfeathers improved the speed of field iteration when reading PP files. Up to 3x speed up has been seen, dependending on the circumstances.

🔥 Deprecations#

#7102: @bjlittle deprecated the iris.analysis.calculus module containing the following public functions:
- cube_delta()
- curl()
- differentiate()
- spatial_vectors_with_phenom_name()
Native Cube calculus functionality will not be replaced and is scheduled for removal in Iris 4.0.0. ( #6262)

🔗 Dependencies#

#7053: @trexfeathers and @tkknight removed the maximum pin for the PyData Sphinx Theme (used in the docs). ( #6885)
#7060: @tkknight added a minimum pin for the PyData Sphinx Theme as we use the collapse sidebar feature introduced in 0.17.0.
#7084: @tkknight updated a dependency in the Read The Docs configuration file to use the latest python.
#7100: @tkknight added a dependency named sphinx-sitemap to generate sitemap.xml for the documentation.
#7105: @tkknight added a dependency named sphinx-llm to generate summaries that LLMs can understand, llms.txt and llms-full.txt.
#7223: @trexfeathers updated Iris to be compatible with Matplotlib version 3.11.

📚 Documentation#

#7053: @trexfeathers and @tkknight made the docs compatible with the latest versions of PyData Sphinx Theme (>=0.16). ( #6885)
#7060: @tkknight enabled the theme option to collapse the sidebar. Note, it only appears once you click on a link away from the landing page. Also moved the search box to the top navigation bar.
#7072: @trexfeathers switched to using the official URL of the cf-checker, after our previous URL of choice was taken down.
#7079: @tkknight updated the voted table that uses datatables to not highlight the sorted column or row as is uses the incorrect theme color (light). Also updated the datatables version from 2.3.2 to 2.3.8.
#7096: @trexfeathers simplified the pull request checklist and moved it directly into the GitHub pull request template (read more here: Pull Request Checklist). Also updated several associated pages of more detailed guidance.
#7100: @tkknight updated the documentation to generate a sitemap.xml files for the stable version.
#7105: @tkknight updated the documentation to generate summaries that LLMs can understand, llms.txt and llms-full.txt.
#7118: @trexfeathers and @hdyson clarified that iris.analysis.cartography.area_weights() requires 1-dimensional lat and lon coordinates on the input Cube.
#7131: @bjlittle Added an explanation section for Temporal Coordinates.
#7150, #7157: @bjlittle and @trexfeathers Added the custom sphinx readingtime directive to automatically estimate the audiance reading time of a page and render a branded banner in-situ.
#7159: @bjlittle Updated the Temporal Coordinates section.
#7166: @bjlittle migrated to towncrier for changelog management.
#7175: @tkknight Added AGENTS.md for AI-assisted development guidelines across the project root, docs/, changelog/ and tests/ directories.
#7202: @bouweandela Fixed the type hints for the iris.cube.Cube.__init__().

💼 Internal#

#6976: @ESadek-MO and @pp-mo removed unit test reliance on all optional dependencies except for mo_pack. ( #6832)
#7036: @trexfeathers altered the messaging for ‘stale’ issues and pull requests, to reduce the negative connotations. We now use needs-checkin for the initial prompt, and not-resourced if the issue/PR ends up closed. ( #6993)
#7046: @trexfeathers fixed the benchmarking asv_delegated.py to work with Nox release 2026.04.10 (which adds more files to the environment parent directory, breaking previous assumptions).
#7087: @SgtVarmint migrated codebase from os.path to pathlib.Path where possible. ( #4523)
#7098: @pp-mo, @trexfeathers, and @ukmo-ccbunney refactored NetCDF character bytecoding into dedicated dataset handlers, added optional load-time string decoding, and enabled saving from either string or character cube content (with explicit string dimensions for char output). ( #6309)
#7113: @trexfeathers and @stephenworsley added functionality to the NetCDF loader and saver to handle Zarr file I/O using the NetCDF NCZarr implementation. ( #6967)
#7134: @HGWright changed the default of the private switch _LAZY_DERIVED_LOADING (formerly .CONCRETE_DERIVED_LOADING) for controlling laziness of coordinates from pp loading, now the switch must be set to True for lazy loading to be enabled. Note: this object is temporary and is likely to be replaced by a permanent solution or else be renamed. ( #7094)
#7138: @bjlittle and @trexfeathers added the Zizmor pre-commit hook to monitor for security vulnerabilities in Iris’ GitHub Actions workflows, and then actioned Zizmor’s recommendations to harden the workflows.
#7148, #7165: @trexfeathers set the link checking workflow to accept redirect HTTP codes, as the reports were getting too noisy.
#7163: @rcomer fixed the capitalisation of a pytest marker.
#7185: @pt331 and @trexfeathers and @pp-mo improved type hints Type Hinting in iris.common.metadata and iris.mesh.components
#7187: @2319bli has removed assignments to class level variables used in test_cell.py. ( #6625)
#7188: @GitRylee moved duplicated code in iris.tests.unit.analysis.trajectory.test__nearest_neighbour_indices_ndcoords to a setup fixture. ( #6625)
#7189: @king56180468-droid removed redundant commented codes from test_CFReader. ( #6625)
#7190: @2319bli has partially improved the names of various test classes. ( #6625)
#7203: @bouweandela enabled Ruff type checking (TC) rules.
#7206: @tkknight added CLAUDE.md that for now just includes AGENTS.md.