v2026.11.1

Added

• uataq.gsv: gsv.get_obs reads the Google Street View cars' Salt Lake observations (May 2019 to March 2020) as one time-sorted table with positions, for chosen pollutants, cars and a time range. The QC flags are kept, and good_only=True keeps only rows with a good position and empties values whose flag is not 0. It reads the Lin group's merged table on CHPC or the published copy (doi:10.5281/zenodo.7605718). A module of its own like uataq.sodar, not a site of the network. (#195, #196)
• TRX02's temperature, humidity and pressure sensors (2016-02-04 to 2017-06-20) are readable through a new Met instrument. (#160, #193)
• get_obs(group="horel") reads TRX02's LGR NO2 from the Horel group's pilot archive (2016-06-27 to 2017-06-20), where each row is the mean of the minute before its time. The default group is still lin, so existing calls do not change. (#160, #194)
• UATAQ_HOREL_DIR points the package at the Horel group's archive root. It is read each time a path is needed, so setting it after import uataq takes effect, and a value that is not a directory raises FileNotFoundError naming the variable. (#110, #192)

v2026.11.0

Added

• Site.get_obs and MobileSite.get_obs label each row with its instrument in an instrument column after the measurement columns, built the same way for both. The column holds text, so a plain .mean() on the table now needs numeric_only=True (or the measurement columns picked first). Row counts and values are unchanged. (#163)

Changed

• Breaking: Network.sites lists only the sites the network kept, in the order given, not every name that was asked for (a misspelled site or one that does not measure the pollutant is no longer in it). The "none of the sites given" error still names the sites the caller gave. (#165, #167)
• Breaking: a site ID given as a string is one site: Network("WBB", "CO2") and get_network_obs("TRX01", "CO2") are the network of that site, and so is get_availability(lvls="final"). A string used to be iterated as letters and never made a network. "all" is unchanged. (#174)
• Data change: read_data(..., "lgr_ugga", lvl="raw") before 2015-12-29 returns UTC times. Those raw files were logged in Mountain time, so times move later by 6 h in summer and 7 h in winter, as the pipeline's lgr_ugga_init.r does, and the raw rows of 2015-12-29 (89,144 rows at the five sites) are left out. A raw read that starts before 2015-12-30 opens files from two days before its start. (#175, #178)
• The teledyne_t500u qaqc level names its sample temperature Internal_T_C, as its raw level and the T300 and T400 do, where it came out as Samp_T_C. The values are the same. (#185, #188)
• WBB is named the William Browning Building. (#172)

Fixed

• A horel GPS read at final keeps Altitude_msl, equal row for row to the qaqc read. (#176, #177)
• The horel final level drops an ES642 row that has an ambient temperature or pressure but no PM2.5. Every other horel final read is the same. (#186, #189)
• The ES405 values that horel's qaqc and final files name for the ES642 at BUS01 and BUS02 are read. (#171)
• The first row of an air-trend raw file that has no header line is kept. (#187)
• Removal dates are corrected where the last reading shows a later one: BUS03's 2b_405 (#156), TRX01's lgr_ugga (#162) and FRU's lgr_ugga (#170). TRX01's metone_es642 is installed on 2014-12-05, the day of the field team's note. (#169)

v2026.10.2

Added

• LINGROUP_MEASUREMENTS_DIR sets the Lin group's measurements root when it is
  set and not empty. It is read each time a path is needed, so setting it after
  import uataq works, except for the pipeline configuration files, which are
  read at import. groupspaces.lin.MEASUREMENTS_DIR, CONFIG_DIR and DATA_DIR
  keep their names and follow the variable. A value that is not a directory
  raises FileNotFoundError naming the variable.
• MobileSite.get_obs has an instrument column naming the instrument that
  logged each row, as get_network_obs already has. At TRX01 the rows of
  lgr_ugga and lgr_ugga_manual_cal share columns, so this is the only way to
  tell them apart. The column follows the measurement columns and comes before
  the GPS ones, in wide and long format.
• MobileSite.get_obs warns EmptyResultWarning when it read rows but none
  found a GPS fix (TRX01's methane in 2018-09 and 2018-10, its lin ozone in
  2017-08), says how many rows were read and points to include_gps=False. It
  returned an empty table without a word.

Changed

• The default Lin group measurements root is the group's dedicated measurements
  space, not a link to it in another group space. The data read are the same.
  LinGroup.data_dir is now a property.
• uataq declares minimum versions: geopandas 1.1, numpy 1.24, pandas 2.0 and
  xarray 2022.3. geopandas 1.0 and older import shapely.geos, which warns on
  shapely 2.1.
• A bad TimeRange (TimeRange("2024", start="2023-01-01")), a bad data
  level in Instrument.read_data, and the raw-only check in
  LGR_UGGA_File.get_files raise ValueError with the same messages. They were asserts, so under
  python -O the check was skipped and start was dropped without a word.
• Docs: the front page cites Mendoza et al. (2024), the electric bus paper, for
  the bus data, and no longer cites Mitchell et al. (2018, PNAS), the long-term
  CO2 record. From a new student's read: the README's installation link works,
  the pin example names the newest release, wide tables say they are shortened,
  the loop that skips sites with no data shows its log lines, docstrings say
  time_range is UTC and what read_data and Network.sites hold, the SODAR
  example uses a month USDR1 has data for, num_processes is said to speed up
  final-level reads too, and large reads, the Slurm script, the dev setup, the
  horel clock and the raw-only months each have one page.

Removed

• uataq.filesystem.groupspaces.lin.dms2dd. Nothing called it, and
  it was wrong for negative degrees (dms2dd(-111, 53, 24) gave -110.11, not
  -111.89); lair.geo.dms2dd is the one to use.

Fixed

• Rows that share a timestamp keep their file order in every session. A read
  sorts its rows stably, and so does the list of files in one period. The
  default sort could swap them, as it can for TRX01's lgr_ugga_manual_cal
  qaqc files, which repeat about 1,400 timestamps a month.
• A row with no valid GPS time of day gets a NaT with the ns unit. The unitless
  NaT warns on NumPy 2.5 and will become an error in a future NumPy.
• Site.get_obs gives each pollutant only its own columns. "NO" also
  returned NO2_ppb and NOx_ppb, and "PM1" returned PM10_ugm3; in the long
  format their values were rows of the same table. WBB's teom_1400ab
PM2.5_ugm3_30min and PM2.5_ugm3_1hr (2016-09 to 2017-03) are no longer in
  get_obs("PM2.5").
• Raw LGR files with other column layouts are read by their header. Ten files
  (ROO and HPL, 2015-11) have 77 columns and gave no rows: every name landed on
  the wrong field. A layout that is neither 23 nor 24 columns takes the 21
  measurement columns, Fit_Flag by name and the MIU description; one without a
  Fit_Flag raises ParserError.
• A raw read of the horel pilot archive keeps only the rows before the pilot
  phase ends. In 2018-11 TRX01's and TRX02's pilot files run on with a
  once-a-minute copy of the main archive's rows, so they came back twice
  (TRX01 2b_205: 14,685 of 521,690 rows).
• uataq.gps haversine distances are never NaN. Rounding could put the
  haversine term just above 1 for near-antipodal points.

v2026.10.1

Added

• Raw horel reads take the *.h5_prior405upgrade file of a month beside its
  .h5. horel split the raw files of the month of each MetOne ES642-to-ES405
  swap (BUS02 2025-03, BUS03 2024-12, TRX01, TRX02 and TRX03 2025-01): the rows
  before the swap are in the second file, which uataq did not read. The
  ES642's last 3 to 13 days before each swap can now be read (raw only; horel's
  qaqc and final files do not have them), and so can TRX01-03's raw ozone
  (2b_205) and GPS from 1 January 2025 to the swap, which locate those rows.
  The ES405 does not read these files.

Changed

• Docs: rewritten for a new student and organized by task, with no redirects
  from the old addresses. The navbar has four sections, Getting started,
  Guides, Reference and Contributing, each a folder of pages, and the front page
  says what UATAQ is, shows a first read and how to cite the data. Where the
  2026.10.0 pages went: "Quick Start" is "Read data" (guides/read_data.html),
  rewritten; "General" is "Input Parameters"
  (reference/input_parameters.html), rewritten, with format and recent;
  "Installation" is getting_started/installation.html, covers Jupyter on Open
  OnDemand, and its development setup is on Contributing; "Configuration" is
  reference/config.html and no longer prints config.json; "Pollutants" is in
  "Columns" (reference/columns.html); the module pages ("Laboratory",
  "Filesystem", "Sites", "Instruments", "GPS", "SODAR") are in the API
  reference. New pages: "Your first data" (a month of CO2 and CH4 in local time,
  hourly means, a daily-cycle plot, and what to write down), "Find sites and
  data" (which sites measure what, when they have data, the newest data), "Data
  levels and flags" (the levels, ID, what final keeps, the flags, empty
  _cal), "Mobile sites" (positions, the two groups' clocks, TRX01's two LGR
  instrument names, the months only at raw), "When there is nothing to read",
  "Sites and instruments" (tables made from the configuration) and "Columns"
  (the columns of each instrument and level). Each fact is on one page, and the
  others link to it.
• Docs: the API reference is generated, with a page for each module, class,
  function and class member, as in the docs of the other packages (the page
  "Functions and modules", reference/functions_and_modules.html, lists them,
  the public functions first and the internals last).
  gps.html, instruments.html, pollutants.html, sodar.html and
  filesystem.html no longer exist: the module pages are
  api/uataq.gps.html, api/uataq.instruments.html,
  api/uataq.pollutants.html, api/uataq.sodar.html and
  api/uataq.filesystem.html, and each class and function is under api/
  (api/uataq.instruments.GPS.html). What the GPS, Pollutants and File System
  pages said by hand is the docstring of the module, shown at the top of its
  page. The File System page's DataFile, GroupSpace and functions are now
  under uataq.filesystem.core, and the group spaces' pages under
  api/uataq.filesystem.groupspaces.horel.html and .lin.html have one page
  for each class. read_data and get_obs moved from "Input Parameters"
  (input_parameters.html#uataq.read_data and #uataq.get_obs), and
  get_recent_obs, get_network_obs and Network from the closing
  "Reference" section of "What data exists" (what_data_exists.html#reference,
  #uataq.get_recent_obs, #uataq.get_network_obs and #uataq.Network), to
  api/uataq.read_data.html, api/uataq.get_obs.html and so on. TimeRange
  (api/uataq.timerange.html) and the exceptions and warning
  (api/uataq.errors.html), which had no page, are documented.
• Docs: the "Laboratory" and "Sites" pages are generated like the rest of the
  API reference, so laboratory.html and sites.html no longer exist.
  laboratory, get_site and the Laboratory class have pages under api/
  (api/uataq.laboratory.html, api/uataq.get_site.html and
  api/uataq.Laboratory.html; the class is documented as uataq.Laboratory,
  no longer as uataq._laboratory.Laboratory), and Site and MobileSite are
  listed on the module page api/uataq.sites.html, each with its own page
  (api/uataq.sites.Site.html). The anchors laboratory.html#uataq.laboratory,
  #uataq.get_site, #uataq._laboratory.Laboratory,
  sites.html#uataq.sites.Site and #uataq.sites.MobileSite moved to those
  pages. uataq.laboratory hides its value, which would be the whole
  configuration. Site.pollutant_instruments has a docstring.
• Docs: the levels are called "data levels" on every page and docstring.
  "Input Parameters" retitles its "Processing Level" section "Data Level" (the
  anchor input_parameters.html#processing-level is now #data-level).
  "Reading data" says the mobile sites are on TRAX trains and buses, as the
  front page does.
• Instrument.model, SensorMixin.pollutants, Sodar.model and
  Sodar.species_measured have docstrings, and the Attributes of
  LinDatFile and LGR_UGGA_File name the ones their __init__ sets
  (instrument and lvl; meta).
• Configuration: HDP, HEB and SUG were taken down, so they are marked
  is_active: false and their instruments now have removal dates (HDP
  lgr_ugga 2025-02-08, HEB licor_6262 2025-05-30, SUG licor_7000
  2025-08-23). Each is the midnight after the last sample in the archive,
  except HEB's licor_6262, which is the day HEB switched to a licor_7000.
  A read after one now reports the instrument as inactive instead of finding
  no files. HEB's licor_7000, which replaced its licor_6262 from
  2025-05-30 to 2025-09-16, is added.
• Configuration: WBB's magee_ae33 and TRX02's lgr_no2 now have removal
  dates, 2024-06-20 and 2017-06-21, the midnight after their last samples
  (2024-06-19 17:41:53 and 2017-06-20 05:28:52 UTC, at raw, qaqc and final).
  No row is read differently: neither has data after its last sample. A
  get_obs of WBB's BC, or of TRX02's NO2 before its teledyne_t500u was
  installed (2020-12-22), for a range wholly after the date now raises
  ReaderError saying the instrument is inactive, instead of returning an
  empty frame.
• The error for a read that finds no data at a site says what you asked for and
  which files exist. Site.get_obs("CO2", time_range="2021-08") at WBB, which
  has no final file that month, raised No data found for ['lgr_ugga'] at WBB (lgr_ugga in lin groupspace). It now raises No final data found for CO2 at WBB: lgr_ugga in lin groupspace: No files found within the specified time range (files run from 2014-02 to 2026-10, with gaps). The first and last
  period come from the files the read had already listed, so the error costs
  no extra read. read_data names
  its instruments where get_obs names the pollutants, a read with no lvl
  leaves the level out, and the warning for each instrument and Network's
  warning for a site carry the same periods. ReaderError.detail holds what
  follows the colon (#63).
• A read with files for the range but no rows in them now warns. get_obs
  of WBB's CO2 for 2024-03-04, a day with no rows in the March 2024 file,
  returned an empty table and said nothing. It still returns the empty table,
  and now warns No final data found for CO2 at WBB (2024-03-04 00:00:00 to 2024-03-05 00:00:00): the files for lgr_ugga exist but hold no data in that range. get_recent_obs and Network.get_obs (and so get_network_obs)
  warn the same way when their table has no rows. It is a Python warning,
  uataq.errors.EmptyResultWarning (a UserWarning), so it shows without
  turning the log on. A table with rows from any instrument or site does not
  warn, and read_data is unchanged (#63).
• A read of several pollutants now warns when it leaves one out. get_obs(["CO2", "O3"], time_range="2021-08") at WBB, which has no final CO2 file that month,
  returned only the O3 column and said nothing. It still returns that table,
  and now warns
  No final data found for CO2 at WBB (2021-08-01 00:00:00 to 2021-09-01 00:00:00): lgr_ugga gave none, so the table leaves out CO2. The warning names
  every pollutant left out and the instruments that gave none, whether they had
  no files for the range or were not installed then. It is a Python warning,
  uataq.errors.DroppedPollutantWarning (a UserWarning, apart from
  EmptyResultWarning so each can be silenced alone). get_recent_obs and
  MobileSite.get_obs warn the same way. A pollutant that another instrument
  still measured is not left out, get_obs() with no pollutants named does not
  warn, and a read where no pollutant has data still raises the error (#63).
• Raw lgr_ugga files are parsed with pandas' C engine instead of its
  python engine, about three times faster: 0.2 s to 0.07 s for a day's file,
  and 76 to 26 minutes of CPU for the whole archive (19,554 files at six
  sites). Every one of those files gives the same table as before.
• A raw lgr_ugga read (WBB, CSP, FRU, HDP, HPL, ROO) opens only the files
  of the days in its range, plus the day on either side as before. It opened
  every raw file the site has to read its header, before the range was
  applied: a one-day read at WBB, the first on a node, took nearly 5 minutes,
  4 of them opening its 4,472 files; it now takes about 30 s, most of it
  listing the folders. The same files are read and the same rows returned; a
  file outside the range whose header cannot be read is no longer reported.
• A raw lgr_ugga read lists only the day folders that overlap its range (the
  range plus the day on either side), where it walked every day folder the site
  has (4,336 at WBB) and read the date from every file name. The first read of
  a day at WBB in a new process took 2.1 to 3.3 s, 1.1 to 2.2 s of it listing
  the folders (22 to 26 s on a node where the file server had not cached them),
  and takes 0.4 to 0.5 s, 0.1 s of it listing. The same files ar...

v2026.10.0

The first release since 2025.11.0.

Added

• Research groups per instrument. group can map each instrument to a
  group, and with no group each part of a time range is read from the group
  that has it (group_dates in config.json): TRAX ozone comes from lin before
  2017-10-28 and from Horel after.
• uataq.pollutants: helpers for UATAQ's concentration column names
  (CO2d_ppm_cal and the like).
• uataq.gps: great-circle helpers and estimate_speed_course. GPS.read_data
  fills missing Speed_m_s and Course_deg from the positions and marks them
  in Speed_Estimated and Course_Estimated. Recorded values are never
  overwritten.
• Horel raw GPS reads carry GPS_Time_UTC, the receiver's own time. The Horel
  logger's clock runs 1 to 20 s ahead of GPS time. Rows without a fix get NaT.
• Mobile observations carry GPS_Group, the group whose GPS located each row.
• Network.get_availability and Network.plot_availability: data availability
  by site and level.

Changed

• Mobile observations are located only with the GPS of the group they were
  read from (breaking): lin rows on the Pi's clock, Horel rows on the logger's
  clock. Naming another group's GPS for some rows raises ValueError. Horel rows
  used to be joined to lin's GPS, which put moving TRAX rows a median 120 m from
  where they were measured and dropped rows.
• Time ranges are half-open (breaking): a range's stop is excluded, in row
  slicing and in TimeRange membership. A date string still covers its whole
  day.
• Network observations keep _cal and _raw columns and have an
  instrument column (breaking), so two instruments measuring the same species
  no longer produce duplicate columns.
• A time range that only touches an instrument's installation or removal date
  raises InactiveInstrumentError. It used to return empty data.
• uataq no longer turns on pandas' copy-on-write option. Setting it changed
  pandas for the caller's whole session.
• Notices are logged, not printed.
• Off CHPC, import uataq fails after 10 s with a clear message when it cannot
  fetch the pipeline's configuration. It used to hang.
• Requires Python 3.11 or newer (breaking): Python 3.10 reaches its end of
  life in October 2026. Tested on 3.11 through 3.14.
• The version now comes from git tags (setuptools-scm). An install from git
  between releases reports a development version such as
  2025.11.1.dev3+g1a2b3c4 rather than the last release's number.
• The documentation no longer describes class attributes and methods twice: the
  API pages drop a class docstring's Methods section and the Attributes
  entries that the page documents anyway. Entries that described methods that no
  longer exist (Site.read_obs) are gone from the pages.

Fixed

• Horel GPS speed is read as m/s. It was treated as knots, so speeds came out at
  about half their value.
• lin pipeline files: the header line is skipped. It was read as a row, which
  made every column text, and a repeated value in the first row of a headerless
  file was altered.
• Horel finalized rows with no observations are dropped.
• lin MetOne air-trend files are read with their own column layout. They were
  matched to the met sensor's, which lost the PM2.5 column.
• Raw LGR files are dated from the style of their file name, not from a table
  of analyzer software versions, so an unknown version no longer stops a read.
• TimeRange keeps the minutes and seconds of a string and rejects malformed
  strings.
• Site.get_obs matches pollutant names in any case.
• Reads are confined to the time an instrument was installed.
• A read raises ReaderError when none of its files can be parsed, and skips a
  level directory the instrument does not have.
• Raw MetOne pressure columns are named in hPa, as the qaqc and final levels
  name them.
• num_processes="max" sizes its pool by the CPUs the process may use.
• Parsing a file no longer fails on Windows when the file and the archive root
  are on different drives.

uataq 2025.11.0

First tagged release since the initial v2024.08 snapshot, and the first release archived on Zenodo for a DOI. Highlights since v2024.08: installable/editable packaging, per-group filesystem readers (Horel/Lin groupspaces), Network for aggregating a pollutant across sites, the Horel-group SODAR reader (moved from lair.mesowest), and a growing pytest suite with Codecov coverage reporting.

See the full commit history for details: v2024.08...v2025.11.0