Skyfield: HomeTable of ContentsChangelogAPI Reference

Installing Skyfield

Skyfield has only a single binary dependency, the NumPy vector library, and is designed to install cleanly with a single invocation of the standard Python package tool:

pip install skyfield

This should install Skyfield, NumPy, and the small collection of pure-Python astronomy libraries that Skyfield depends on. If you lack pip and need to install each dependency by hand, consult Skyfield’s file for the full list.

If trying to install Skyfield gives you errors about NumPy, there are several other ways to get NumPy installed:

Read the Changelog below to learn about recent fixes, changes, and improvements to Skyfield. You can protect your project from any abrupt API changes by pinning a specific version of Skyfield in your requirements.txt or or install instructions:


By preventing Skyfield from getting accidentally upgraded until you are ready to advance the version number yourself, you can avoid even slight changes in behavior and output coordinates that might result from an upgrade. If you find any problems or would like to suggest an improvement, simply create an issue on the project’s GitHub page:

Good luck!

Checking your Skyfield version

The Skyfield package offers a VERSION tuple that your code can use to test which version of Skyfield is running. For example, this code checks whether Skyfield is at least at version 1.24:

import skyfield

if skyfield.VERSION < (1, 24):
    print('Too old')

A new featured added in Skyfield 1.36 (released January 2021) is that you can invoke the skyfield module from the command line to display its version:

$ python -m skyfield

Skyfield version: 1.37
jplephem version: 2.14
sgp4 version: 2.17
Built-in leap seconds table ends with leap second at: 2016-12-31 23:59:60 UTC
Built-in ∆T table from finals2000A.all covers: 1973-01-01 to 2022-01-29

Also displayed are the versions of the libraries Skyfield depends on, and the start and end dates of its built-in timescale tables.

Citing Skyfield

If Skyfield has proven useful in your research, the project welcomes your citation to bring it to the attention of other academics who might benefit from it.


Released versions

v1.48 — 2024 February 7

  • Times now support the < operator, so Python can sort them.
  • For convenience, geoids like wgs84 have a new attribute polar_radius.
  • You can no longer subtract two positions unless they have the same .center. Otherwise, a ValueError is raised. This check has always been performed when you subtract vector functions, but it was missing from the position subtraction routine.
  • On days that the Sun fails to rise and set in the Arctic and Antarctic, the new rising and setting routines now correctly set the value False not only for sunrise but also for sunset.

v1.47 — 2024 January 13

v1.46 — 2023 April 9

  • The oppositions_conjunctions() routine now measures ecliptic longitude using the ecliptic of each specific date, rather than always using the J2000 ecliptic, which should improve its accuracy by several seconds.
  • Skyfield’s internal table for the ∆T Earth orientation parameter has been updated, so that its predictions now extend to 2024-04-13.
  • Bugfix: Skyfield was giving values several kilometers off when computing the elevation above ground level of a target that was positioned directly above the Earth’s north or south pole.
  • Bugfix: the is_behind_earth() method was incorrectly returning True if the Earth was on the line that joins the two satellites, but over on the far side of the other satellite where it wasn’t really in the way.
  • Internals: the altaz() method now lives on the main position class instead of in two specific subclasses. If the user mistakenly tries to call .altaz() on an instance of the Astrometric position subclass — which previously lacked the method — then a friendly exception is raised explaining their error.

v1.45 — 2022 September 15

  • Bugfix: minor planets and comets in Skyfield 1.44 would raise an exception if asked for a position in the half of their orbit where they are inbound towards their perihelion.

v1.44 — 2022 September 12

  • Skyfield’s internal table for the ∆T Earth orientation parameter has been updated, so that instead of including measurements only through December 2021 it now knows Earth orientation through September 2022.
  • Distance and velocity objects can now be created by calling their unit names as constructors, like d = and v = Velocity.km_per_s(0.343).
  • Updated the URL from which the Hipparcos database hip_main.dat is downloaded, following a change in the domain for the University of Strasbourg from to

v1.43.1 — 2022 July 6

  • An attempt at overly clever scripting resulted in a Skyfield 1.43 release without a in its .tar.gz; within an hour, a Python 2.7 user had reported that Skyfield could no longer install. This release is identical to 1.43 but (hopefully) installs correctly for everyone!

v1.43 — 2022 July 6

  • Fixed planetary_magnitude() so it works for Saturn even when the time is an array rather than a single time; also, improved its calculation slightly with respect to Uranus. #739
  • Improved load_comets_dataframe() so that parsing CometEls.txt with the most recent version of Pandas doesn’t stumble over the commas in the final field of (for example) Halley’s Comet and give the error ParserError: Error tokenizing data. C error: Expected 12 fields…saw 13. #707

v1.42 — 2022 February 6

  • Added two new position methods phase_angle() and fraction_illuminated() that, given an illuminator (usually the Sun) as their argument, compute whether the observer is looking at the bright side or the dark side of the target body. They replace a pair of old functions in the almanac module.
  • The almanac routine moon_nodes() would sometimes skip nodes that were closer together than 14.0 days. It has been tightened down and should now detect all lunar nodes. #662
  • Time objects now feature a to_astropy() method.
  • The position method to_skycoord() now sets the frame attribute of the sky coordinate it returns, and for now only supports barycentric and geocentric positions. #577

v1.41 — 2021 December 16

  • Times now support arithmetic: you can add or subtract from a time either a number representing days of Terrestrial Time (TT) or a Python timedelta which Skyfield interprets as TT days and seconds. #568
  • Fixed the .itrs_xyz vector of the geographic position returned by the subpoint_of() method. #673
  • Skyfield now uses HTTPS instead of FTP to download JPL ephemeris files like de421.bsp. This does risk raising an error for users whose machines have out-of-date root certificates. But it protects the connection from outside tampering, and will keep working if the FTP service is ever shut down — as happened earlier this year to FTP on NASA’s server. #666

v1.40 — 2021 November 14

v1.39 — 2021 April 14

v1.38 — 2021 April 3

  • Replaced the old historic ∆T table from the United States Naval Observatory with up-to-date splines from the 2020 release of the extensive research by Morrison, Stephenson, Hohenkerk, and Zawilski and also adjusted the slope of Skyfield’s near-future ∆T estimates to make the slope of ∆T much less abrupt over the coming century.
  • Added a full reference frame object for the TEME reference frame used by SGP4 Earth satellite elements.

v1.37 — 2021 February 15

  • Added a frame_latlon_and_rates() method that can compute the rates at which angles like altitude and azimuth, or right ascension and declination, are changing.
  • Accepted a contributor’s helpful fix for a rounding error that had slightly shifted a few constellation boundaries. #548
  • The Time tuple utc and method utc_strftime() are now backed by the same math, so they always advance to the next calendar day at the same moment. This makes it safe to mix values returned by one of them with values returned by the other. #542
  • Vector subtraction now returns the position subclass specific to the resulting vector’s center. #549

v1.36 — 2021 January 26

  • Tweaked several lines of code that build NumPy arrays to avoid a new deprecation warning Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. NumPy no longer wants to accept a simple constant like 0.0 where the resulting array needs a whole row of zeros. #536
  • Added an hadec() position method that returns hour angle and declination. #510
  • The default str() and repr() strings for geographic positions have been streamlined, and no longer raise ValueError when elevation is an array. They now show simple decimals instead of splitting degrees of longitude and latitude into minutes and seconds; always show elevation, even if zero; properly format NumPy arrays; and abbreviate long arrays. #524
  • Fixed Angle.dstr() and Angle.hstr() to return an array of strings when the angle itself is an array. #527

v1.35 — 2020 December 31

  • Deprecated the old Topos class, which not only featured a clunky interface but hid from users the fact that Skyfield was generating IERS2010 positions from latitude and longitude when in fact nearly all users want WGS84 positions. Users are now encouraged to supply latitude and longitude to the latlon() method of either the wgs84 object or the iers2010 object. Related discussion: #372
  • The two new geoid objects wgs84 and iers2010 have also provided a happy new home for the subpoint() method — which was previously stranded over on the Geocentric class, where it couldn’t be used with positions of other classes that might be centered at the geocenter. (The old method will remain in place to support legacy code, but is discouraged in new applications.)
  • The effects of Precession and Nutation — if configured — are now included both when computing the position in space of an Earth latitude and longitude, and when determining the latitude and longitude beneath a celestial position.
  • Added load_constellation_names().
  • The utc_jpl() method now correctly designates its return value as UTC instead of the ambiguious UT. #515

v1.34 — 2020 December 10

  • The position classes have gained methods frame_xyz(), frame_xyz_and_velocity(), frame_latlon(), and from_time_and_frame_vectors() that work with a new library skyfield.framelib to offer a number of familiar reference frames. These replace the existing ad-hoc position methods for ecliptic and galactic coordinates, which are now deprecated (but will continue to be supported). See Coordinates in other reference frames. #476
  • Added an official itrs reference frame.
  • Added support for IERS Precession and Nutation 𝑥 and 𝑦.
  • Added a method lst_hours_at() that computes Local Sidereal Time.
  • A new almanac routine moon_phase() returns the Moon phase as an angle where 0° is New Moon, 90° is First Quarter, 180° is Full, and 270° is Last Quarter. #282
  • Almanac search routines that previously returned a Boolean true/false array now return an integer 0/1 array instead, to work around a new deprecation warning in NumPy which, for example, would have outlawed using the Boolean array from moon_nodes() to index into the MOON_NODES list that provides a name for each node. #486
  • The undocumented columns magnitude_H and magnitude_G in the Minor Planet Center comets dataframe have been renamed magnitude_g and magnitude_k following further research on the file format (which does not itself document which magnitude model is intended). #416

v1.33 — 2020 November 18

  • Fix: running load.timescale(builtin=False) was raising an exception FileNotFoundError if the finals2000A.all file was not already on disk, instead of downloading the file automatically. #477

v1.32 — 2020 November 16

  • A new lunar_eclipses() routine finds lunar eclipses and determines their degree of totality. #445
  • The almanac module’s new meridian_transits() routine can find the moments at which a body transits the meridian and antimeridian. #460
  • Fix: the find_minima() function was ignoring its epsilon and num arguments and always using the default values instead. #475
  • Fix: the .epoch attribute of Earth satellite objects that were built using from_satrec() was, alas, a half-day off. #466
  • Fix: the Topos constructor arguments x and y, which never worked properly anyway, have been deprecated and are now ignored.

1.31 — 2020 October 24

  • Skyfield now uses the International Earth Rotation Service (IERS) file finals2000A.all for updated ∆T and leap seconds. The USNO is no longer updating the files and deltat.preds that previous versions of Skyfield used, and the server from which they were fetched will discontinue anonymous FTP on 2020 October 31. See UT1 and downloading IERS data. #452 #464
  • The comets dataframe built from the MPC file CometEls.txt now includes the reference column, so users can tell which orbit is most recent if there are several orbits for a single comet. (For example, the file currently lists two C/2020 F3 (NEOWISE) orbits.) The comet examples in the documentation now build a dataframe that only includes the most recent orbit for each comet. #463
  • Two new methods days_old() and download() make it simple to download a fresh copy of a file if the copy on disk is older than you would like.

1.30 — 2020 October 11

  • The various strftime() Skyfield methods now support the %j day-of-year format code.
  • Fix: the new Julian calendar support broke support for out-of-range month numbers, wrapping them into the current year instead of letting them overflow into subsequent years. #461
  • Fix: a stray debugging print() statement was stranded in t.dut1. #455
  • The Time object, if manually instantiated without a Julian date fraction, now provides a fraction array with dimensions that match the Julian date argument. #458

1.29 — 2020 September 25

  • Fix: the new Julian calendar feature was raising an exception in the calendar methods like tt_calendar() if the time object was in fact an array of times. #450
  • Fix: trying to iterate over a time object would raise an exception if the time was created through ut1().

1.28 — 2020 September 24

  • Broken URL: Because the VizieR archive apparently decided to uncompress their copy of the hip_main.dat.gz Hipparcos catalog file, the old URL now returns a 404 error. As an emergency fix, this version of Skyfield switches to their uncompressed hip_main.dat. Hopefully they don’t compress it again and break the new URL! A more permanent solution is discussed at: #454
  • To unblock this release, removed a few deprecated pre-1.0 experiments from April 2015 in skyfield.hipparcos and skyfield.named_stars that broke because the Hipparcos catalog is no longer compressed; hopefully no one was using them.
  • In a sweeping internal change, the Timescale() and Time() objects now offer support for the Julian calendar that’s used by historians for dates preceding the adoption of the Gregorian calendar in 1582. See Ancient and modern dates if you want to turn on Julian dates in your application. #450

1.27 — 2020 September 15

  • The printed appearance of both vectors and of vector functions like Earth locations and Earth satellites have been rewritten to be more informative and consistent.
  • Added compute_calendar_date() which lets the caller choose the Julian calendar for ancient dates instead of always using the proleptic Gregorian calendar. This should be particularly useful for historians.
  • Added J() that builds a time array from an array of floating point years. #436
  • Added four new strftime methods for the non-UTC timescales (#443). All four of them support %f for microseconds, and provide a reasonable default format string for callers who don’t wish to concoct their own:
  • Thanks to several fixes, comets and asteroids with parabolic and hyperbolic orbits should now raise fewer errors.
  • The prototype planetary_magnitude() can now return magnitudes for Uranus without raising an exception. The routine does not yet take into account whether the observer is facing the equator or poles of Uranus, so the magnitude predicted for the planet will only be accurate to within about 0.1 magnitudes.

1.26 — 2020 August 1

  • The official ∆T files on NASA’s FTP server have stopped receiving updates — they have no new data beyond February, the start of the global pandemic. Unless they are updated by next February, older versions of Skyfield will unfortunately download the files all over again every time timescale() is called (unless the builtin=True parameter is provided). To make Skyfield less fragile going forward:
    1. The loader’s timescale() method now defaults to builtin=True, telling it to use the ∆T and leap second files that ship with Skyfield internally. To download new ∆T files from NASA and the leap second file from the International Earth Rotation Service, specify builtin=False.
    2. The concept of an “expired” file has been removed from load(). Skyfield is now much simpler: if a file with the correct name exists, Skyfield uses it. See UT1 and downloading IERS data if you still want your application to check the age of your timescale files and automatically download new ones.
  • The ICRF.separation_from() method now officially supports the combination of an array of positions with a single reference position! Its previous support for that combination was, alas, accidental, and was broken with the 1.23 release. #414 #424
  • A prototype planetary_magnitude() routine has been added with support for several planets. #210
  • The utc timezone that Skyfield returns in Python datetimes is now either the Python Standard Library’s own UTC object, if it supplies one, or else is defined by Skyfield itself. Skyfield no longer silently tries importing the whole pytz package merely to use its UTC object — which also means that the timezone returned by Skyfield longer offers the non-standard localize() method. #413

1.25 — 2020 July 24

  • Added parse_constellations() and parse_star_names() to load Stellarium star names and constellation lines. Constellation lines are featured in a new example script Drawing a finder chart for comet NEOWISE that produces a finder chart for comet C/2020 F3 NEOWISE.
  • The Hipparcos star catalog should now load faster, having switched behind the scenes to a higher performance Pandas import routine.
  • Fixed the ability of utc() to accept a Python object as its argument. #409
  • Slightly lowered the precision of two tests when they detect that Python is compiled for a 32-bit processor, so the test suite can succeed when contributors package Skyfield for 32-bit Linux. #411

1.24 — 2020 July 20

  • Added methods from_datetime() and from_datetimes() to the Timescale class, to better advertise the ability to build a Skyfield time from a Python datetime — an ability that was previously overloaded into the year parameter of the utc() method (where it is still supported for backwards compatibility, but no longer documented).
  • Fix: improved the accuracy with which velocity is converted between the Earth-fixed ITRF frame that rotates with the Earth and the inertial GCRS frame that does not. In particular, this should make Earth satellite velocities more accurate.

1.23 — 2020 July 9

  • Added Kepler Orbits support for generating the positions of comets and asteroids from Minor Planet Center data files.
  • Added is_behind_earth() to determine whether a celestial object is blocked from an Earth satellite’s view by the Earth itself.
  • Replaced the awkward and hard-to-explain rough_period search parameter with the conceptually simpler step_days parameter, and updated the instructions in Searching for the dates of astronomical events to match.
  • Made the tle_file() import method less strict about Earth satellite names: any text on the line before two lines of TLE data is now saved as the satellite name. A parameter skip_names=True turns this off if, for particular TLE files, this leads to unwanted text being saved.

1.22 — 2020 Jun 8

  • Skyfield’s improved time precision (stored internally as two floats) is now used in computing ephemeris positions, Earth orientation, and light-travel time, producing position angles which change much more smoothly over time on a sub-milliarcsecond scale.
  • Searching for the dates of astronomical events is now documented for custom events that users define themselves, instead of only being documented for the official pre-written Almanac Computation functions. Not only discrete events but also maxima and minima are now officially supported and documented, thanks to a rewrite of the underlying code.
  • Time objects no longer cache the nutation and precession matrices, since they are never used again after being multiplied together to create the equinox-of-date rotation matrix. This should save 144 bytes for each time in a Time array.
  • It is now possible to Build a satellite with a specific gravity model thanks to a new Earth satellite constructor method. #384
  • Added build_url() that returns the URL from which Skyfield will download a file. #382
  • Added close() to support applications that need to do fine-grained resource management or whose testing framework check for dangling open files. #374
  • Skyfield’s dependency list now asks for “jplephem” version 2.13 or later. Skyfield 1.21, alas, could incur a Module not found error when importing jplephem.exceptions if a user had an old “jplephem” version already installed. #386

1.21 — 2020 May 29

  • Added is_sunlit() to determine whether Earth satellites in orbit are in Earth’s shadow or not, thanks to a pull request from Jesse Coffey.
  • Added position_of_radec() to replace the poorly designed position_from_radec().
  • Skyfield Time objects now have microsecond internal accuracy, so round trips to and from Python datetimes should now preserve all the microsecond digits.
  • The utc_strftime() method now rounds to the nearest minute or second if it sees that either minutes or seconds are the smallest unit of time in the format string.
  • The 6 numbers in the sequence t.utc can now be accessed by the attribute names year, month, day, hour, minute, and second.
  • Nutation routines should now be faster and have a smaller memory footprint, thanks to a rewrite that uses more optimized NumPy calls. #373
  • Thanks to Jérôme Deuchnord, the exception raised when asking for a position out-of-range of a JPL ephemeris now shows the calendar dates for which the ephemeris is valid and carries several useful attributes. #356

1.20 — 2020 April 24

  • Erik Tollerud contributed a fix for a deprecation warning about SSL from the most recent versions of Python (“cafile, cpath and cadefault are deprecated, use a custom context instead”). The file download routine now auto-detects which mechanism your Python supports. #363
  • Added an elevation_m argument to build_latlon_degrees().

1.19 — 2020 April 23

  • To hopefully fix the SSL: CERTIFICATE_VERIFY_FAILED errors that some users encounter when downloading timescale files, Skyfield has taken the risk of switching away from your system’s SSL certificates to the certificate bundle from the certifi package. #317
  • Added a new almanac routine for finding Lunar Nodes. #361
  • Gave geographic location objects a new itrf_xyz() method that returns their raw ITRF coordinates. #354
  • Fixed the sign of the velocity vector when two vectors are directly geometrically subtracted. #355

1.18 — 2020 March 26

  • Deprecated the old hybrid-key satellite dictionary returned by load.tle() in favor of a simple list returned by the new tle_file() routine. #345
  • The almanac find_discrete() routine no longer returns extraneous values in its second return value if no changes of state were found. #339 #351
  • Added documentation and support for computing lunar libration. #80

1.17 — 2020 February 2

  • Upgraded to a new version of the sgp4 Python library that, when possible, uses the fast official C++ implementation of SGP4.
  • Added a find_events() Earth satellite method that finds the times at which a satellite rises, culminates, and sets.
  • Improved the logic behind the Almanac Computation routines to avoid rare situations in which a cluster of nearly identical times would be produced for what should really be considered a single event. #333
  • Fixed the utc_strftime() method so it does not report that every day in all of recorded history is a Monday. #335

1.16 — 2019 December 20

1.15 — 2019 November 20

  • Changed the URL for the Hipparcos catalog, because the VizieR archives FTP server is no longer responding. #301
  • Added a dark_twilight_day() function that not only handles sunrise and sunset but also all three kinds of twilight. #225

1.14 — 2019 November 1

  • Changed the URL from which leap second files are downloaded; the server that previously provided them is no longer responding. Thanks to Richard Shaw for the pull request. #296 #297
  • Added a risings_and_settings() function for computing rising and setting times. #271

1.13 — 2019 October 10

  • Provided a constellation lookup routine through load_constellation_map().
  • Added a position_from_radec() function.
  • Fixed the apparent() method in the case where a single observer position is observing an entire vector of target positions. #229

1.12 — 2019 September 2

  • Fix: an exception was being thrown when creating a Loader pointed at a Windows directory for which Python’s os.makedirs() function returned a spurious error. #283
  • The internal reverse_terra() routine can now be given an iterations=0 argument if the caller wants geocentric latitude and longitude.

1.11 — 2019 July 22

  • You can now call load.timescale(builtin=True) to use time scale files that Skyfield carries internally, instead of downloading them. Note that the time scale files distributed with any given version of Skyfield will gradually fall out of date.
  • Fix: indexing a position now returns a position with an actual velocity. #241
  • Fix: the Star method from_dataframe() now correctly pulls stellar parallax data from the dataframe if available. #266
  • Fix: find_discrete() was generating empty arrays of search dates, upsetting the astronomy code, if the start and end dates were very close together. #240

1.10 — 2019 February 2

  • Fix: teach Skyfield the new format of the Naval Observatory ∆T data file deltat.preds, whose change in format caused Skyfield to start throwing an exception for new users. #236

1.9 — 2018 September 23

  • Added seasons() to the Almanac Computation module that can be used to predict solstices and equinoxes.
  • Fix: the ecliptic coordinate routines no longer raise ValueError: too many values to unpack if they are passed a time array. #207 #208

1.8 — 2018 September 12

  • There is now an Almanac Computation module that can compute the times of sunrise, sunset, and the phases of the moon, based on the search algorithms announced at my recent PyBay talk “An Import Loop and a Fiery Reentry.”
  • Two new methods cirs_xyz() and cirs_radec() have been contributed which provide support for rotating a position into the Celestial Intermediate Reference System (CIRS). #192

1.7 — 2018 September 3

  • Skyfield now supports loading the Hipparcos star catalog as a Pandas dataframe, providing the user with convenient mechanisms for looking up a single star by HIP number or filtering the entire catalog by magnitude. See Stars and Distant Objects for details.
  • Ecliptic coordinates can now be produced for epochs other than J2000 thanks to a new optional parameter specifying the desired epoch for the ecliptic_latlon() method.
  • A position that gives a position, velocity, and time can now be converted into full osculating orbital elements through the routine osculating_elements_of().
  • A couple of bugs in the load() routine have been fixed. #193 #194

1.6 — 2018 July 25

  • Both of the loader methods open() and tle() now accept not just URLs but also plain local file paths; they correctly re-download a remote file if “reload=True” is specified; and they allow specifying a different local “filename=” than the one at the end of the URL.
  • Earth satellite objects no longer try to instantiate a timescale object of their own, which often kicked off an unexpected download of the three files needed to build a timescale.
  • Satellite names are now correctly loaded from Space-Track TLE files.
  • The ability to create times using Julian Dates is now better advertised, thanks to dedicated timescale methods whose names end in …_jd().

1.5 — 2018 July 4

  • The subpoint() method now normalizes the longitude values it returns into the range −180° to 180° #182 and returns an actual elevation instead of zero. #185
  • Earth satellites now return a real velocity vector instead of zero. #187
  • Earth satellites now offer an ITRF_position_velocity_error() method that returns raw ITRF coordinates for users interested in them. #85

1.4 — 2018 May 20

  • You can now specify the distance to an object when generating a position from altitude and azimuth coordinates. #158
  • The dictionary of satellites returned when you read a TLE file now supports lookup by integer satellite ID, not just by name, and now knows how to parse TLE files from Space-Track. #163 #167
  • Star coordinates can now be offered for any epoch, not just J2000. #166
  • You can now create a time object given the UT1 date. #91
  • Fractional Julian years are now available on Time objects as .J.
  • The parameter DUT1 is now available on Time objects as .dut1. #176

1.3 — 2018 April 15

  • Geocentric coordinates now have a subpoint() method that computes the latitude and longitude of the point beneath that body.
  • All of the Timescale time constructor methods now accept arrays.
  • Emergency fix to stop Skyfield from endlessly downloading new copies of deltat.preds, since the file has gone out of date at the USNO site.
  • Fixed ability of a Star to be initialized with a tuple that breaks units into minutes and seconds (broke in version 1.2).
  • Issues fixed: #170 #172

1.2 — 2018 March 29

  • The documentation now describes how to create an excerpt of a large JPL ephemeris without downloading the entire file. Several Skyfield tests now run much faster because they use an ephemeris excerpt instead of waiting for a download.
  • For load_file() a leading ~ now means “your home directory”.
  • You can now initialize a velocity from kilometers per second with Velocity(km_per_s=...).
  • Empty time and angle objects no longer raise an exception when printed. (Thanks, JoshPaterson!)
  • Issues fixed: #160 #161 #162

1.1 — 2018 January 14

  • Positions can now be converted to AstroPy with to_skycoord().
  • You can now provide a timescale of your own to an EarthSatellite() instead of having it trying to load one itself.
  • Downloaded files are no longer marked as executable on Windows.
  • A friendly error message, rather than an obscure traceback, is now returned if you try converting a position to alt/az coordinates but the position was not measured from a position on the Earth’s surface.

1.0 — 2017 March 15

  • Brought the core API to maturity: replaced the narrow concept of building a “body” from several ephemeris segments with the general concept of a vector function that is the sum of several simpler vector functions.
  • Added support for adding and subtracting vector functions.
  • Deprecated the Earth topos() method in favor of vector addition.
  • Deprecated the Earth satellite() method in favor of vector addition.
  • Deprecated the body geometry_of() method in favor of vector subtraction.
  • Celestrak satellite files can now be opened with load.tle(url_or_filename).

0.9.1 — 2016 December 10

  • Attempted to speed up Earth satellite calculations by caching a single time scale object instead of creating a new one each time.
  • Fixed a possible divide-by-zero error when applying deflection to an apparent position.


  • The observe() method of an observer on the Earth’s surface now correctly accounts for the way that the Earth’s gravity will deflect the apparent position of objects that are not exactly overhead, bringing Skyfield’s agreement with the Naval Observatory’s NOVAS library to within half a milliarcsecond.
  • The time method tt_calendar() method no longer raises a TypeError when its value is an array.
  • Running repr() on a Time array now produces a more compact string that only mentions the start and end of the time period.
  • The api.load() call no longer attempts to animate a progress bar if the user is running it under IDLE, which would try to accumulate the updates as a single long line that eventually hangs the window.


  • Added an API Reference document to the project, in reverent imitation of the Pandas API Reference that I keep open in a browser tab every time I am using the Pandas library.
  • New method ICRF.separation_from() computes the angular separation between two positions.
  • Fixed == between Time objects and other unrelated objects so that it no longer raises an exception.


  • Introduced the Timescale object with methods utc(), tai(), tt(), and tdb() for building time objects, along with a load.timescale() method for building a new Timescale. The load method downloads ∆T and leap second data from official data sources and makes sure the files are kept up to date. This replaces all former techniques for building and specifying dates and times.

  • Renamed JulianDate to Time and switched from jd to t as the typical variable used for time in the documentation.

  • Deprecated timescale keyword arguments like utc=(…) for both the Time constructor and also for all methods that take time as an argument, including and

  • Users who want to specify a target directory when downloading a file will now create their own loader object, instead of having to specify a special keyword argument for every download:

    load = api.Loader('~/ephemeris-files')


  • Users can now supply a target directory when downloading a file:

    load('de421.bsp', directory='~/ephemerides')
  • Fix: removed inadvertent dependency on the Pandas library.

  • Fix: load() was raising a PermissionError on Windows after a successful download when it tried to rename the new file.


  • Skyfield now generates its own estimate for delta_t if the user does not supply their own delta_t= keyword when specifying a date. This should make altitude and azimuth angles much more precise.
  • The leap-second table has been updated to include 2015 July 1.
  • Both ecliptic and galactic coordinates are now supported.


  • Skyfield has dropped the 16-megabyte JPL ephemeris DE421 as an install dependency, since users might choose another ephemeris, or might not need one at all. You now ask for a SPICE ephemeris to be downloaded at runtime with a call like planets = load('de421.bsp').

  • Planets are no longer offered as magic attributes, but are looked up through the square bracket operator. So instead of typing planets.mars you should now type planets['mars']. You can run print(planets) to learn which bodies an ephemeris supports.

  • Ask for planet positions with instead of body(t).
  • Per IAU 2012 Resolution B2, Skyfield now uses lowercase au for the astronomical unit, and defines it as exactly 149 597 870 700 meters. While this API change is awkward for existing users, I wanted to make the change while Skyfield is still pre-1.0. If this breaks a program that you already have running, please remember that a quick pip install skyfield==0.4 will get you up and running again until you have time to edit your code and turn AU into au.



  • The floating-point values of an angle a.radians, a.degrees, and a.hours are now attributes instead of method calls.