OpenQuake Engine 3.25.1

[Michele Simionato (@micheles)]

• Disabled prefiltering across the International Date Line
• Listed h3 as a dependency in setup.py
• Making sure save_rates uses the processpool distribution
• Listed AvgSA as primary IMT

OpenQuake Engine 3.25.0

[Michele Simionato (@micheles)]

• Reduced the memory consumption with multifault sources by splitting
  in more tiles

[Christopher Brooks (@CB-quakemodel)]

• Make source/path adjustment coeff files used in modifications of CY14
  as described in Boore et al. (2022) more flexible to avoid need for
  dummy values for IMTs with unknown coeffs for source scaling but known
  for path scaling or vice versa.

[Michele Simionato (@micheles)]

• Fixed avg_losses_by in case of a single realization
• Optimized the saving of the rates in large classical calculations
• Fixed the rup_id <-> event_id associations in openquake.engine.global_ses

[Yen-Shin Chen (@vup1120)]

• FDHA: added primary surface displacement package (base, Youngs et al. 2003
  AD/MD, tests)
• FDHA: added primary surface rupture package (base, Youngs et al. 2003
  ExC/GB/nBR, tests)

[Michele Simionato (@micheles)]

• Changed the task distribution in classical calculations and reduced the
  slow tasks, resulting in huge speedups in same cases (like 2.5x for USA)
• Bug fix: using sites=lon1 lat1, ... with a parent calculation
  did not result in the specified sites to be used

[Yen-Shin Chen (@vup1120)]

• Added get_x_l_ratio() method to BaseSurface for computing normalized
  along-strike position (x/L) for PFDHA applications
• Vectorized get_x_l_ratio() implementation using NumPy broadcasting
  for improved performance with large site meshes
• Added _get_tor() helper method to extract the top rupture
  trace with NaN handling for KiteSurface compatibility

[Christopher Brooks (@CB-quakemodel)]

• Added the EMME24 site model to the existing OQ implementation of the
  EMME backbone model for ASCR (files describing the site model were
  provided by Abdullah Sandıkkaya).

[Michele Simionato (@micheles)]

• Changed large classical calculations to store the rates in ds_calc_dir;
  removed the need to specify custom_tmp for the SLURM distribution
• Internal: instantiating a GSIM automagically generates a ._toml
  attribute from the GSIM name and arguments, associated to the GSIM hash

[Christopher Brooks (@CB-quakemodel)]

• Fix small bug in Hashash et al. (2020) site term implementation within the
  NGA East models and regenerate the test tables (very small differences are
  observed and only for SA(0.4)).

[Michele Simionato (@micheles)]

• Removed the long time deprecated method get_mean_and_stddevs from the
  GSIMs; you can use the helper function contexts.mean_stds instead
• Internal: changed cmaker.from_srcs to return a context array and not a
  list with a single context array
• Fixed AssertionError on len(mags) for GenericGmpeAvgSA with table-based
  GMPEs
• complex_fault_mesh_spacing is now mandatory in the job.ini file if
  the model contains complex fault sources

[Nicholas Clemett (@nc-hsu)]

• Refactored the BaseAvgSACorrelationModel to make avg_periods an optional
  argument
• Added EmpiricalAvgSACorrelationModel
• Added classes for the Clemett correlation models for active shallow crust,
  subduction interface, subduction slab and vrancea regions in europe based
  on the ESHM20 GMMs.

[Michele Simionato (@micheles)]

• Internal: introduced subtasks in classical calculations
• Added endpoint validate_ini to validate local files
• Optimized preclassical by gridding the point sources in parallel

[Christopher Brooks (@CB-quakemodel)]

• Added support for GMPE tables to be specifiable as the underlying
  GSIMs within conditional GMPEs.

[Paolo Tormene (@ptormene)]

• Refactored the WebUI, separating its logic into multiple files
• Added Playwright tests to test the WebUI frontend in different
  application modes
• Added an extractor exposure_by_location

[Michele Simionato (@micheles)]

• Turned tiling into an explicit parameter (i.e. you must pass it to
  enable the full tiling algorithm, before it was automagic)
• Optimized the management of the RateMap in the classical calculator
• Optimized the reading of the rates by vectorizing on the sites
• Bug fix: use_rates was not overridden by the child calculation
• Extended install.py devel to infer the version from the branch name

[Christopher Brooks (@CB-quakemodel)]

• Added the option of using a modified form of the Campbell and Bozorgnia
  (2014) GMM's sigma model within the Kuehn et al. (2020) GMMs as required
  for the 2023 USGS Alaska model.
• Refactored the indirect AvgSA GMPEs capabilities to permit the
  calculation of non-AvgSA IMTs using the underlying GMPE (added
  new unit test of classical/case_91 for checking this works within
  GmpeIndirectAvgSA and updated classical/case34 for checking this
  works within GenericGmpeAvgSA).

[Michele Simionato (@micheles)]

• Internal: removed configuration parameter pmap_max_gb
• Implemented the "ilabel" feature i.e. region-dependent gsim logic trees
• Reduced the number of chunks generated in classical calculations, which
  has a huge effect on performance if you have 192 cores or more

[Christopher Brooks (@CB-quakemodel)]

• Fixing class inheritances in MorikawaFujiwara2013 GMMs.

[Amirhossein Mohammadi (@amirxdbx)]

• Added Mohammadi2023Turkiye ML-based GSIM for Turkey.

[Christopher Brooks (@CB-quakemodel)]

• Added the Alaska bias adjustments for the NGASUB GMMs
  as required for the USGS 2023 Alaska model.

[Antonio Scala (@antonio-scala)]

• Added Scala et al (2025) GMM

[Michele Simionato (@micheles)]

• Show the GSIM class warnings only once
• Extended oq engine --lhc and oq engine --lrc to show the
  calculation_mode

[Christopher Brooks (@CB-quakemodel)]

• Refactored conditional GMPEs to be specifiable within ModifiableGMPE
  on a per-IMT basis (safer handling, more flexibility).
• Added documentation for the implementation of conditional GMPEs within
  the OpenQuake Engine.

[Michele Simionato (@micheles)]

• Added a consistency check on the effective time when starting from SES
• Added a new exporter hmaps-stats producing one file per IMT,
  associated to the display name "Hazard Maps Statistics"

[Maoxin Wang (@MaoxinWang)]

• Added a new gsim file wang_2025_turkey.py to implement the Turkey-specific
  ground motion models for Arias Intensity (IA), Cumulative Absolute Velocity
  (CAV), 5-75% Significant Duration (RSD575), and 5-95% Significant Duration
  (RSD595) developed by Wang et al. (2025)

[Michele Simionato (@micheles)]

• For scenario_risk calculations expose the aggrisk-stats output even if
  there is a single gsim; the stats are associated to the simulations
• Use PYTHONUTF8 mode, i.e. assume the default encoding to be
  UTF8 also on Windows
• Internal: extended oq purge
• Included .onnx and .onnx.gz files in MANIFEST.in
• Added command oq db keep <workflow_id>
• Added command oq checkout workflow.toml
• Extended oq check_input to accept workflow files
• Added an output "Average Losses By Taxonomy"
• Extended oq run to accept .toml files and to use the job cache
• Replaced dbserver.cache parameter in openquake.cfg with a cache
  parameter in job.ini
• If the site parameters are more distant than ASSOC_DIST raise an error
  and not a warning
• Added parameter siteid in job.ini to specify the custom_site_id
• Internal: extended baselib.sap to manage command classes
• Saved memory in ebrisk, now most calculations can be run with 2 GB
  of RAM per core

[Paolo Tormene (@ptormene)]

• Fixed the management and layout/styling of navigation buttons in the
  WebUI, with and without authentication
• Improved the default configuration of the TOOLS_ONLY application mode

[Michele Simionato (@micheles)]

• Supported site-specific gsim logic trees via site_labels
• Added source_info to the global SES file
• Optimized _save_events for large calculations over 100,000 events
• Parallelized the rupture prefiltering
• Fixed count_ruptures producing more than 32768 tasks
• Removed the distance filtering when sampling the ruptures; removed also the
  mosaic model filtering

[Paul Henshaw (@pslh), Paolo Tormene (@ptormene)]

• Force use of utf8 encoding when writing non-binary files.
• Update AreaSource demo to use UTF8 chars that are obviously not ASCII.
• Use utf-8-sig when reading so as to handle BOM

[Anirudh Rao (@raoanirudh)]

• Add missing axis argument in recent bugfix in classical_risk
  with nontrivial weights in the taxonomy mapping

[Michele Simionato (@micheles)]

• Extended the global_ses script to accept job.ini files
• Implemented oq engine --run job1.ini .. jobN.ini --hc parent.hdf5
• Extended --hc to accept paths to datastore files and removed the
  experimental feature rupture_model_file = rups.hdf5
• Using the filtered site collection, thus changing numbers in OQImpact
• Implemented rupture prefiltering for scenario and event based calculators

[Paolo Tormene (@ptormene)]

• Added the possibility to assign tags to jobs and to set a job as the
  "preferred" one for a tag

[Michele Simionato (@micheles)]

• Internal: extended oq reduce_sm to accept a source model directory
• Implemented event_based_risk from ses.hdf5
• Raised an early error if the vulnerability functions are not
  strictly increasing
• Fixed a bug in classical_risk with nontrivial weights in the
  taxonomy mapping
• Removed the ebrisk calculation mode, since it can be replaced
  with event_based_risk with ground_motion_fields = false
• Turned LSD into a primary IMT, so that the Youd and ZhangZhao GMPE
  can be used

[Lana Todorovic (@LanaTodorovic93)]

• Added Abrahamson and Bhasin (2020) GMPE
• Fixed the Macedo2019 GMPE with underlying GMPEs requiring extra parameters
  (like the KuehnEtAl2020SInter requiring ztor)

[Michele Simionato (@micheles)]

• Filtering GMFs associated to low secondary intensities to make
  secondary peril calculations ...

OpenQuake Engine 3.23.4

[Paul Henshaw (@pslh), Paolo Tormene (@ptormene)]

• Force use of utf8 encoding when writing non-binary files.
• Update AreaSource demo to use UTF8 chars that are obviously not ASCII.
• Use utf-8-sig when reading so as to handle BOM

[Michele Simionato (@micheles)]

• Backported the fix for small negative damages (#10772)
• Fixed the documentation about the installation directories

OpenQuake Engine 3.24.1

[Paul Henshaw (@pslh), Paolo Tormene (@ptormene)]

• Force use of utf8 encoding when writing non-binary files.
• Update AreaSource demo to use UTF8 chars that are obviously not ASCII.
• Use utf-8-sig when reading so as to handle BOM

[Michele Simionato (@micheles)]

• Removed the BASE33489 hack and made the first branch of the source
  model logic tree regular, with a single letter abbreviation

[Paolo Tormene (@ptormene)]

• When running a job via the WebUI it is possible to select multiple input
  files or a single .zip file
• Added a whitelist of resources that can be extracted from the datastore
  via the '/v1/calc/<calc_id>/extract/' and the
  '/v1/calc/<calc_id>/extract_html_table/' API endpoints. All
  resources can still be extracted if the WebUI authentication is disabled
  or if the current user has a high profile level (>1) or has the permission
  'auth.can_view_'

[Michele Simionato (@micheles)]

• Reduced the number of tiles
• Optimized the management of cluster of sources (the array loop)
• Changed the ps_grid_spacing approximation to use average occurrence rates

[Claudia Mascandola (@mascandola)]

• Added class "LanzanoEtAl2019_RJB_OMO_NESS2" to lanzano_2019 gsim

[Michele Simionato (@micheles)]

• Reduced the memory consumption of multifault sources for the USA model
• Improved the parallelization of disagg_source in AELO calculations
• Fixed the OverflowError: Python integer out of bounds for uint32
  happening for extra-large calculations when saving source_info

[Guillermo Aldama-Bustos (@aldabus)]

• Fixed the unit of measure in the Douglas et al (2024) GMM and updated
  coefficients for the 3- and 5-branch models to fix an error

[Michele Simionato (@micheles), Kendra Johnson (@kejohnso)]

• Internal: parallelized the AELO mosaic tests and made it possible to pass
  the asce_version and vs30 parameters

[Michele Simionato (@micheles)]

• Extended the source_id-filtering feature to use .startswith
• Added a flag "aggregate_exposure" (false by default) to aggregate
  the assets with the same taxonomy on the same hazard site
• Optimized "building cmakers" and the RateMap in the case of clusters
  of sources (260x for the latest USA model)

[Christopher Brooks (@CB-quakemodel)]

• Added BSSA14 site term capability to Modifiable GMPE + unit tests

[Richard Styron (@cossatot)]

• Added method from_moment to the TaperedGRMFD class
• Fixed KiteSurface.get_middle_point to manage NaNs

[Michele Simionato (@micheles)]

• Doing the check on the export_dir only if --exports is passed
• Fixed a NameError: Missing liquefaction:LiqProb in gmf_data
  when using the TodorovicSilva2022NonParametric model
• gzipping the avg_losses reducing the disk space by half or so
• Added an extractor losses_by_location
• Filtering the assets around the rupture according to the maximum distance
  in scenario calculations

[Claudia Mascandola (@mascandola)]

• Added a Ramadan et al (2023) gsim

[Michele Simionato (@micheles)]

• Disabled the threads in PicklableInferenceSession, making it possible
  to use the TodorovicSilva2022NonParametric model in high concurrency
  situations
• Made it possible to merge pre_job.ini and job.ini in ShakeMap calculations
• Internal: changed Exposure.read_around to work in terms of h3 hexagons;
  changed the exposure.hdf5 accordingly
• Added a private endpoint /v1/calc/run_ini to run remote calculations
• Added a private endpoint /v1/calc/jobs_from_inis

[Myles Turner (@mylesturner99), Guillermo Aldama-Bustos (@aldabus)]

• Added GMM of Douglas et al. (2024)

[Michele Simionato (@micheles)]

• Added a parameter config.dbserver.cache to avoid repeating calculations
• Removed the --reuse-input flag, since it was not used (use --hc instead)
• Changed install.py to exit if any of the tools is not installed
• Fixed small negative damages due to numeric errors
• Fixed the MIMETYPE to application/x-zip when needed in the WebUI
• Removed the parameter site_effects since it is useless
• Changed num_cores to be equal to the pool size when explicit
• On Windows use as default num_cores
  num_threads/2 if num_threads > 8 else num_threads
• Extended the aggregation by site_id to multiple tags
• Fixed an extract bug in disagg-traditional with realizations
• Fixed a bug with the quantiles in event_based_damage
• Improved loggin the progress in event based risk
• Reduced slow tasks in the USA model
• Added output mce_default in AELO calculations
• Extended oq export_mean_disagg_by_src to multiple sites
• Extended oq export_mag_dst_eps_sig to multiple sites
• Extended aggexp_tags and aggrisk_tags to multi-tags

[Paolo Tormene (@ptormene)]

• The OQ-Impact workflow option 'Use finite rupture from the USGS' was
  replaced by two options 'Use ShakeMap fault rupture from the USGS'
  and 'Use finite-fault model from the USGS', making it possible to convert
  a wider set of ruptures from the USGS data
• A sanity check was added to make sure that the hypocenter is inside the
  surface of the rupture. Otherwise, it is relocated to the middle point of
  the surface and a warning indicates the original and translated
  coordinates.

[Christopher Brooks (@CB-quakemodel)]

• Fixed bug in ParkerEtAl2020 epistemic uncertainty scaling factors.

[Michele Simionato (@micheles)]

• Fixed indentation errors in nearly all secondary peril classes
• Reduced the data transfer in avg_losses (20x for Philippines),
  thus solving an out-of-memory error; also improved the performance
  (25% for Philippines) and reduced the memory occupation on the workers
• Print a warning if there are vs30 values distant more than 10% from
  DEFINED_FOR_REFERENCE_VELOCITY

[Anirudh Rao (@raoanirudh)]

• Fixed a bug when generating the aggregate curves for affectedpop and
  injured loss types, due to missing units

[Michele Simionato (@micheles)]

• Extended override_vs30 to multiple values
• Forbidden GMF correlation with more then max_sites_correl=1000
• Made minimum_intensity and minimum_magnitude mandatory with event based
  calculations with many sites

[Paolo Tormene (@ptormene)]

• When a job is deleted, also the corresponding calc_CALCID_tmp.hdf5 file is
  removed, if present

[Astha Poudel (@asthapoudel), Catarina Costa (@catarinaquintela)]

• Modified infrastructure risk to avoid calculating efficiency loss for
  large networks

[Paolo Tormene (@ptormene)]

• AELO: the management of warnings was made more robust

[Michele Simionato (@micheles)]

• OQImpact: reading the sites around the rupture and not around the hypocenter
• Raised an error if the ShakeMap does not contain SA(0.6) when required
• Fixed the multi-peril implementation (imt_by_lt)
• Raised an error for risk IMTs not covered by the secondary IMTs
• Optimized build_global_exposure

[Michele Simionato (@micheles), Paolo Tormene (@ptormene)]

• OQ-Impact: fixed aggregation by ID_2, which was using indices instead
  of strings

[Marco Pagani (@mmpagani)]

• Storing also tau in the mean spectrum calculator

[Christopher Brooks (@CB-quakemodel)]

• Added ability to specify z1pt0 and z2pt5 as -999 within site model
  to inform (basin term considering) GMMs to use their own vs30 to
  z1pt0 or z2pt5 relationship to estimate these site parameters, in
  effect preventing additional adjustment to GMM basin terms based
  on unknown site information. QA tests (classical case_89) + unit
  tests (utils_infer_basin_params_test.py) added
• Added validation checks for core site params (lon, lat, vs30, z1pt0
  and z2pt5) when building the site collection within a calculation

[Paolo Tormene (@ptormene)]

• In the API endpoint listing the outputs for a job, the first available
  export_type is specified in each output url

[Michele Simionato (@micheles)]

• Added a new uncertainty abMaxMagAbsolute to the logic tree
• Internal: refactored lt.Branch, lt.BranchSet and lt.CompositeLogicTree
• Avoided out of memory in calculations starting from the global SES

[Lana Todorovic (@LanaTodorovic93)]

• Updated units of slope (in m/m) in site model.

[Michele Simionato (@micheles)]

• Improved the validation of area source polygons
• Internal: added method CompositeLogicTree.apply_all
• Fixed downloading of uncertainty.xml in OQImpact calculations

[Michele Simionato (@micheles), Paolo Tormene (@ptormene)]

• Added a route /v1/calc/XXX/job_zip and a button to download the archive
  from the WebUI, visible only in OQImpact for level 2 users
• Fixed an error in oq plot examples

[Baran Güryuva (@BaranGuryuva), Christopher Brooks (@CB-quakemodel)]

• Implemented the EMME24 backbone GMC for crustal
  earthquakes (emme24.py)

[Marco Pagani (@mmpagani), Michele Simionato (@micheles)]

• Added uncertainties for area geometry and seismogenic depths

[Michele Simionato (@micheles)]

• Added a check on duplicate branches in the source model logic tree
• Internal: added methods CompositeLogicTree.to_node and
  CompositeLogicTree.to_nrml

[Paolo Tormene (@ptormene), Michele Simionato (@micheles)]

• Added extractor for 'damages-stats'

[Kendra Johnson (@kejohnso)]

• Updated the mosaic geometries for AELO year 4

[Matteo Nastasi (@nastasi-oq)]

• Allow free mapping between standalone apps and their url path prefix
  (regardless of their name)

[Claudia Mascandola (@mascandola)]

• Added GMPEs of ambraseys_1996 and sabetta_pugliese_1996

[Graeme Weatherill (@g-weatherill), Michele Simionato (@micheles)]

• Implemented the MacedoEtAl2019SInter and MacedoEtAl2019SSlab conditional
  GMMs

[Anirudh Rao (@raoanirudh)]

• In the stations parser, make sure to skip all "vertical" ('z' or 'Z') or
  "up" ('u' or 'U') components when extracting the max horizontal component

_[Manuela ...

OpenQuake Engine 3.23.3

[Michele Simionato (@micheles)]

• Disabled the low level threads in TodorovicSilva2022NonParametric model

[Christopher Brooks (@CB-quakemodel), Michele Simionato (@micheles)]

• Improved error management for missing tools in install.py
• Backported fix to Parker (2020)

OpenQuake Engine 3.23.2

[Michele Simionato (@micheles)]

• Backported fix to taxonomy mapping with consequences
• Backported fix to 64 bit poes critical for multifault sources
• Backported fix to workerpool critical for zmq clusters

OpenQuake Engine 3.23.1

[Marco Pagani (@mmpagani)]

• Backported fix to Thingbaijam et al. (2017) scaling relation

[Michele Simionato (@micheles)]

• Backported fix to logs.dbcmd: querying the calculations caused the DbServer
  to break in server installations
• Backported fix to oq plot hcurves
• Backported fix to Arteta (2021) missing the backarc parameter

OpenQuake Engine 3.23.0

[Michele Simionato (@micheles)]

• Fixed contexts.py: we were incorrectly discarding ruptures with magnitude equal
  to the minimum magnitude
• Hiding spurious "divide by zero" warnings generated by shapely in
  get_joyner_boore_distance

[Fatemeh Alishahiha (@FatemehAlsh)]

• Added GMM Campbell-Bozorgnia (2003)

[Christopher Brooks (@CB-quakemodel)]

• Another small bug fix for NZ 2022 inslab variants of the Kuehn et
  al. (2020) GMM - the sigma_mu_model and sigma_mu_epsilon params
  were sometimes missing from GSIM object + added a sensitivity test

[Marco Pagani (@mmpagani), Michele Simionato (@micheles)]

• Added an occurrence_rate column in median_spectrum_disagg and fixed
  the exporter to order the ruptures by ID

[Michele Simionato (@micheles)]

• Fixed a bug in classical_damage with sampling reported by Evi Riga
• Added utils/fix_consequences to fix consequence.csv files
  containing risk_ids rather than taxonomies, now invalid
• Implemented quantiles in scenario_risk, visible in aggrisk_tags
• Added checks on the consequence functions across taxonomies and perils
• Fixed the taxonomy field in consequences.csv to be the exposure
  taxonomy and not a fragility function ID
• Internal: renamed aristotle -> impact
• Changed the sourcewriter to save the rupture_idxs as compressed datasets,
  thus reducing the size of some models a lot (i.e. from 400M to 17M)

[Christopher Brooks (@CB-quakemodel)]

• Small bug fix for NZ 2022 inslab variants of the Kuehn et al. (2020)
  and Parker et al. (2020) GMMs - backarc param was not always in the
  req. site params but is needed given included in NZ 2022 site model

[Savvinos Aristeidou (@Savvinos-Aristeidou)]

• Added new ANN-based GMM in the gsim library, named aristeidou_2024
• Added new IMs, that is the FIV3, Sa_avg2, and Sa_avg3

[Michele Simionato (@micheles)]

• Extended rupture_dict to the msr and aspect_ratio parameters
• Fixed the instantiation of file-dependent GMPEs from the datastore
• Optimized slow tasks both in the preclassical and the classical phase
• Reading the site model in calculations with ruptures.hdf5 and making
  minimum_intensity mandatory
• Added parameter minimum_engine_version

[Paolo Tormene (@ptormene), Michele Simionato (@micheles)]

• Storing the reduced asset collection and fixing a bug when exporting
  the average losses

[Richard Styron (@cossatot), Michele Simionato (@micheles)]

• Extended the preclassical calculator to work in absence of a site collection

[Michele Simionato (@micheles)]

• Added a new output "Aggregated Exposure Values" (aggexp_tags) and a new extractor
  aggrisk_tags;

OpenQuake Engine 3.22.1

[Michele Simionato (@micheles)]

• Fixed HDF5 bug when storing the avg_losses, visible on MacOS with a large
  number of assets (say over a million). Fixed the same bug for damages too.

[Christopher Brooks (@CB-quakemodel)]

• Added ability to specify period-dependent bias adjustment for USGS
  classes of NGAEast GMMs (US 2023 NSHMP). Unit tests also added.
• Added ability to specify Chapman and Guo (2021) Coastal Plains site
  amplification adjustment for USGS classes of NGAEast GMMs (US 2023
  NSHMP). Unit tests also added.
• Added sediment depth site param (z_sed) to sites module as required
  for Chapman and Guo (2021) Coastal Plains site amp. adjustment. QA
  is provided through unit tests and qa_tests_data/classical/case_87.

[Michele Simionato (@micheles)]

• Removed the deprecated function openquake.commonlib.datastore.hdf5new
  and changed build_dstore_log into create_job_dstore;
• Extended the ShakeMap parser to read the intensities associated to SA(0.6)
• Fixed a bug while exporting realizations.csv for scenario calculations
• Added webapi.calc_timeout configuration parameter
• Reduced conditioned_gmfs_gb to 8 GB by default
• Parallelized get_mean_covs when conditioning the GMFs

[Christopher Brooks (@CB-quakemodel)]

• Added Seattle Basin classes of Kuehn et al. (2020) NGASUB GMM (uses
  Cascadia coeffs except for basin term).
• Added USGS basin scaling and CyberShake adjustments to the NGAWest2
  GMMs + ability to pass base gsim arguments within the NSHMP2014 gsim class.

[Michele Simionato (@micheles)]

• Saving the rupture geometries as 32 bit floats to save memory
• Internal: added a command oq filter_around lon lat csvfile
• Added a command oq plot ebruptures?min_mag=XXX
• Added a script utils/build_global_ses
• Removed the ability to read GMFs generated with engine <= 3.11
• Fixed the filtering of the site collection in event_based with --hc
• Internal: writeability check for datadir and scratch_dir (if any)

[Christopher Brooks (@CB-quakemodel)]

• Added instantiation-level arguments for contextually applying
  the M9 basin term, the CB14 basin term and the USGS basin scaling
  model to the AbrahamsonGulerce2020, KuehnEtAl2020, ParkerEtAl2020,
  ZhaoEtAl2006 and AtkinsonMacias2009 GMMs as required for the 2023
  US NSHM model's subduction interface GMC. Unit tests are also
  provided for these GMM adjustments

[Michele Simionato (@micheles)]

• Internal: forcing the signature (C, ctx, region, ...) for _get_basin_term
  in hazardlib
• Extended gmfs_file to multiple .hdf5 files, assuming they correspond to
  disjoint sites and events

[Paolo Tormene (@ptormene)]

• Added a duplicates_strategy argument to read_df and changed the
  approach used while reading station data, to calculate the average values
  of stations having the same coordinates instead of raising an error

[Christopher Brooks (@CB-quakemodel)]

• Added epistemic uncertainty scaling capabilities to
  the Parker et al. (2020) NGASUB GMM.
• Added M9 basin term and associated tests to ModifiableGMPE
• Added CB14 basin term and associated tests to ModifiableGMPE

[Michele Simionato (@micheles)]

• Added parameter with_betw_ratio
• Added command oq info peril
• Extended consequences to perils
• Replaced taxonomy mapping by loss type with taxonomy mapping by peril
• Internal: changed the ordering in the composite risk model from
  (loss_type, riskid) -> (riskid, loss_type)
• Added an exporter for trt_gsim
• Internal: added utility function readinput.read_source_models

[Manuela Villani (@ManuelaVillani)]

• Improved the "Governing MCE" plot

[Michele Simionato (@micheles)]

• Raised an error in case of non-invertible hazard curves, affecting
  disaggregation and site-specific hazard spectrum calculations
• Changed the ruptures.csv exporter to also export the source IDs
• Added support for consequence=losses for liquefaction and landslides
• Added a check for missing secondary perils
• Added loss types liquefaction and landslide
• Removed support for XML consequences, after 3 years of deprecation
• Fixed the stored calculation_mode in oq sensitivity_analysis
• Added command oq info loss_types

[Paolo Tormene (@ptormene)]

• Added the possibility to export the asset collection via command line,
  keeping it private from the webui

[Michele Simionato (@micheles)]

• Fixed median_spectrum with multiple sites
• Extended Aristotle calculation to manage multiple countries at once
• Extended taxonomy_mapping.csv with a loss_type field
• Modernized the regionalization of Chiou Youngs (2014)
• Modernized the regionalization of Campbell Bozorgnia (2014)
• Internal: made it possible to override CoeffsTable
• Added a memory check in disaggregation calculations
• Made scientific_format resilient against encoding errors

[Ilaria Oliveti (@IlariaOliveti)]

• Fixed the GMPE Tusa-Langer-Azzaro (2019) table of coefficients, the
  IMTs were incorrectly using Hz instead of seconds

[Michele Simionato (@micheles)]

• Fixed the assert(losses) error in the view delta_loss in the case of few events

OpenQuake Engine 3.21.0

[Paolo Tormene (@ptormene)]

• Added commands oq plot "rupture?" and oq plot "rupture_3d?"

[Michele Simionato (@micheles)]

• Extended oq reset to also remove the custom_tmp directory, if any
• Added classes CampbellBozorgnia2019, CampbellBozorgnia2019HighQ, CampbellBozorgnia2019LowQ
• Reduced the size of the large CSV files in hazardlib and added a check
  to forbid files larger than 600k
• Added a check on the signatures of get_std_dev_mag and get_median_area
  and fixed the signature of StrasserIntraslab.get_std_dev_mag

[Enrico Abcede (@emabcede30), Francis Bernales (@ftbernales)]

• Implemented Campbell and Bozorgnia (2019) IA and CAV to campbell_bozorgnia_2014

[Michele Simionato (@micheles)]

• Fixed scenario_risk from GMFs when the intensity measure types were
  incorrectly identified

[Marco Pagani (@mmpagani), Michele Simionato (@micheles)]

• Added a median spectrum post-processor

[Michele Simionato (@micheles)]

• Added a parameter mea_tau_phi in the job.ini to save mean, tau and phi
  for each rupture, site, gsim and imt in a GMF calculation
• Internal: changed oq run to automatically generate the db when possible
• Added a check on missing risk files
• Reduced the space used by the CollapsedPointSources (2.7x)
• Fixed the site model association procedure to work in conditioned
  scenario calculations
• Used oq engine --run to submit asynchronous jobs to SLURM and
  oq run to submit interactive jobs

[Christopher Brooks (@CB-quakemodel)]

• Added the Japan region versions of the NGAWest2 relations used to obtain
  z1pt0 and z2pt5 from vs30 in prepare sites command

[Michele Simionato (@micheles)]

• Fixed oq reduce_sm for calculations with nonparametric sources
• Fixed logging the classical time with --sample-sources
• Improved the progress log in classical calculations

[Kendra Johnson (@kejohnso)]

• Fixed an error invalid literal for int() with base 10 affecting
  event based calculations using mutex sources and a nontrivial logic tree

[Michele Simionato (@micheles)]

• Optimized the --sample-sources feature
• Honored the custom_tmp in classical calculations and saved data transfer
  by using TileGetters, then parallelized the saving of the rates

[Lana Todorovic (@LanaTodorovic93)]

• Implemented Nowicki Jessee et al. (2018) landslide geospatial model that
  computes the areal coverage by landslide occurrence.

[Paolo Tormene (@ptormene)]

• Updated extractor for gmf_data for a single event id (used by the IRMT
  QGIS plugin), including data for secondary perils

[Michele Simionato (@micheles)]

• Refined the tiling calculator (partial tiling, task weighting, saving memory)
• Optimized "computing pnes" in classical calculations (3x in the common case)
  and fully switched to 32 bit rates, thus saving memory and improving performance

[Nicolas Schmid (@schmidni)]

• Fixed a regression in the ShakeMap to_gmfs code, a forgotten sigma^2/2 term

[Marco Pagani (@mmpagani)]

• Fixed aliases for Kuehn2020 GMPEs (signalled by Eric Thompson)

[Michele Simionato (@micheles)]

• Internal: splitting in groups with homogeneous temporal occurrence model
• Optimized the calculation of mean and stdevs in event based calculations, with
  a speedup of 13x for the EUR model

[Chris di Caprio (@chrisdicaprio)]

• Allowed extrapolation in the Kuehn (2020) GMPEs to solve numeric issues

[Michele Simionato (@micheles)]

• Fixed an encoding error when exporting the realizations
• Fixed a bug with multifault sources raising a cryptic error
  "object has no attribute msparams"
• Fixed a critical memory bug causing over 80 GB per core to be needed for
  event based calculations with ~6 million sites
• Made minimum_intensity mandatory in event based calculations
• Reduced the memory consumption in event_based calculations: now even calculations
  with 5 million sites can be run with ~2 GB per core
• Reduced the memory occupation in gen_poes
• Using half the memory in postclassical by using 32 bit arrays
• Using half the memory on Windows by using half the threads by default
• Fixed a bug in conditional spectrum calculations with a non-contributing TRT

[Savvinos Aristeidou (@Savvinos-Aristeidou)]

• Added new GMM in the gsim library, named aristeidou_2023
• Added new IM, that is the inelastic spectral displacement, SDi,
  which takes as input one additional parameter (i.e., strength ratio, R)

[Fatemeh Alishahiha (@FatemehAlsh)]

• Added a new gsim module named zafarani_2018.
• Added a new gsim module named ambraseys_2005.

[Kyle Smith (@kslytherin)]

• Added a new gsim file sandikkaya_akkar_2017.py to implement the
  Arias Intensity and Cumulative Absolute Velocity ground motion
  models from Sandikkaya and Akkar (2017).

[Michele Simionato (@micheles)]

• Making sure that the tiles contains at least 100 sites in classical tiling
  calculations
• Rewritten the SLURM algorithm to work around the submission limit
• Added uniqueness check for the station coordinates
• Internal: removed parameter [dbserver]listen in openquake.cfg
• Added command oq submit n job_ini for usage in supercomputers
• Reimplemented the sensitivity analysis in a supercomputer-friendly way
• Fixed export realizations which was truncating the branch_path
• Extended logic tree filtering via smlt_branch
• Worked on disaggregation by multifault source

[Michele Simionato (@micheles), Christopher Brooks (@CB-quakemodel)]

• Internal: added a function for filtering sites around a rupture

[Michele Simionato (@micheles)]

• Internal: added method GsimLogicTree.to_node
• Internal: added commands oq compare oqparam and oq plot_sites csvfiles
• Fixed exporting the ruptures when imported from a CSV
• Fixed the issue of small negative eigenvalues in conditioned GMFs
  calculations by adding a correlation_cutoff
• Internal: extended Starmap.apply to split DataFrames
• Fixed the avg_gmf exporter in presence of a filtered site collection

[Lana Todorovic (@LanaTodorovic93)]

• Improved secondary perils documentation

[Michele Simionato (@micheles)]

• Fixed the ordering of the realizations to depend on the source model
  branch path also in the case of full enumeration: this is crucial to
  get consistent results with the two approaches
• Saved data transfer in the context makers in classical calculations
• Internal: added utility oq info geohash:<lon>,<lat>
• Changed the scenario calculators to discard the sites far away from
  the rupture, with the effect of generating different GMFs
• Cached get_realizations and optimized the memory occupation; now
  302_990_625 realizations (i.e. EUR) require only 90 GB

[Anna Rood (@AnnaRood)]

• Updates to the AELO workflow: added additional spectral periods ranging
  from 0.02 to 10 s and added option to use ASCE7-22. NB: minor changes in
  the results of year 1 are expected

[Michele Simionato (@micheles)]

• Changed the GMF-based calculators to always set a custom_site_id
• Internal: added variable OQ_SITE_DEBUG