Skip to content

v0.48.0
Compare
Choose a tag to compare
@zebengberg zebengberg released this 24 Oct 11:55
· 1041 commits to main since this release
v0.48.0
facb65a

This release includes a number of breaking changes and new features. If upgrading from a previous version of pycontrails, please read the changes below carefully. Open an issue if you experience problems.

Breaking changes

  • When running Cocip and other pycontrails models, the met and rad parameter must now contain predefined metadata attributes provider, dataset, and product describing the met source. An error will now be raised in Cocip if these attributes are not present.
  • Deprecate passing arbitrary kwargs into the MetDataArray constructor.
  • No longer convert accumulated radiation data to average instantaneous data in ERA5 and HRES interfaces. This logic is now handled downstream by the model (e.g., Cocip). This change allows for more flexibility in the rad data passed into the model and avoids unnecessary computation in the MetDataSource interfaces.
  • Add new MetDataSource.set_met_source_metadata abstract method. This should be called within the implementing class open_metdataset method.
  • No longer take a finite difference in the time dimension for HRES radiation data. This is now also handled natively in Cocip.
  • No longer convert relative humidity from a percentage to a fraction in ERA5 and HRES interfaces.
  • Require the HRES stream parameter to be one of ["oper", "enfo"]. Require the field_type parameter to be one of ["fc", "pf", "cf", "an"].
  • Remove the steps and step_offset properties in the GFSForecast interface. Now the timesteps attribute is the only source of truth for determining AWS S3 keys. Change the filename method to take in a datatime timestep instead of an int step. No longer assign the first step radiation data to the zeroth step.
  • Change the return type of ISSR.eval, SAC.eval, and PCR.eval from MetDataArray to MetDataset. This is more consistent with the return type of other pycontrails models and more closely mirrors the behavior of vector models. Set output attrs metadata on the global MetDataset instead of the individual MetDataArray in each case.

Features

  • Rewrite parts of the pycontrails.core.datalib module for higher performance and readability.
  • Add optional attrs and attrs_kwargs parameters to MetDataset constructor. This allows the user to customize the attributes on the underlying xarray.Dataset object. This update makes MetDataset more consistent with VectorDataset.
  • Add three new properties provider_attr, dataset_attr, and product_attr to MetDataset. These properties give metadata describing the underlying meterological data source.
  • Add new Model.transfer_met_source_attrs method for more consistent handling of met source metadata on the source parameter passed into Model.eval.
  • No longer require geopotential data when computing tau_cirrus. If neither geopotential nor geopotential_height are available, geopotential is approximated from the geometric height. No longer require geopotential on the met parameter in Cocip or CocipGrid.
  • Remove the Cocip shift_radiation_time parameter. This is now inferred directly from the rad metadata. An error is raised if the necessary metadata is not present.
  • Allow Cocip to run with both instantaneous (W m-2) and accumulated (J m-2) radiation data.
  • Allow Cocip to run with accumulated ECMWF HRES radiation data.

Fixes

  • Correct radiation unit in the ACCF wrapper model [#64]. Both instantaneous (W m-2) and accumulated (J m-2) radiation data are now supported, and the ACCF wrapper will handle each appropriately.
  • Avoid unnecessary writing and reading of temporary files in ERA5.cache_dataset and HRES.cache_dataset.
  • Fix timestep resolution bug in GFSForecast. When the grid parameter is 0.5 or 1.0, forecasts are only available every 3 hours. Previously, the timesteps property would define an hourly timestep.

Internals

  • Include name parameter in MetDataArray constructor.
  • Make the coordinates.slice_domain function slightly more performant by explicitly dropping nan values from the request parameter.
  • Round unwieldy floating point numbers in GeoVectorDataset._display_attrs.
  • Remove the ecmwflibs package from the ecmwf optional dependencies.
  • Add NPY to ruff rules.
  • Add convenience MetDataset.standardize_variables method.
  • Remove the p_settings attribute on the ACCF interface. This is now constructed internally within ACCF.eval. Replace the ACCF._update_accf_config method with a _get_accf_config function.
Assets 2
Loading