CLIWOC 2.1: ship logbook weather observations, 1750-1850

The Climatological Database for the World's Oceans (CLIWOC) was an EU-funded project (final report 2003) that transcribed 287,114 daily entries from 2,005 logbooks (974 distinct ship names). Most come from Dutch, British, Spanish and French ships, plus a few Swedish, American, Danish and Hamburg ships. Each entry has a position, the wind direction and force as the officer wrote it, and often the weather, sea state, clouds and remarks about life on board. The project then turned the wind and position into modern units. This is release 2.1 (25 September 2007), the final one.

This copy keeps every field of the release, gives each one a proper type and adds a few derived columns. The logbook text is kept exactly as it was keyed: Dutch, English, Spanish or French, in upper case, with period spelling.

from datasets import load_dataset
ds = load_dataset("taziksh/cliwoc", "logbooks", split="train")
print(ds[0]["ShipName"], ds[0]["obs_time_utc"], ds[0]["AllWindForces"], ds[0]["wind_speed_ms"])

-- duckdb
SELECT LogbookLanguage, AllWindForces, avg(wind_speed_ms) AS ms, count(*) AS n
FROM 'hf://datasets/taziksh/cliwoc/data/logbooks/*.parquet'
WHERE Duplicate = 0 AND AllWindForces IS NOT NULL
GROUP BY ALL ORDER BY n DESC LIMIT 20;

Configs

config	rows	one row is
logbooks (default)	287,114	one logbook entry: one ship, one (usually noon) observation per day, with time, position, wind, weather and remarks
ship_logbooks	2,007	one logbook (ID), with its ship and whether it duplicates another logbook
wind_force_terms	3,923	one wind-force phrase used in logbooks of one language, with its Beaufort class and m/s
wind_direction_terms	635	one wind-direction phrase used in logbooks of one language, with its direction in degrees
weather_codes	75	one code that stands for a symbol in Dutch extract logbooks (W00 = "Lichte buien", light showers)
magnetic_declinations	222,924	magnetic declination for one 5°x5° cell in one year, 1750-1854

The last five are the lookup tables that ship with the release. The project used them to build the standardised fields.

Source and version

• Data: CLIWOC release 2.1, released 25 Sep 2007 by KNMI. Archived on PANGAEA as CLIWOC21.zip (complete IMMA-format text, 287,114 fixed-width records of 8,569 characters) and CLIWOC21Additional.zip (lookup tables), both at https://doi.org/10.1594/PANGAEA.743343. The dataset series is https://doi.org/10.1594/PANGAEA.611088.
• Retrieved: 2026-10-28. PANGAEA last modified the files on 2010-07-23. The build pins their SHA-256 hashes.
• Original home: projects.knmi.nl/cliwoc (offline now; archived copy).
• Field documentation: Steven Ottens' copy of the KNMI data description, https://stvno.github.io/page/cliwoc/. It is linked from https://www.historicalclimatology.com/cliwoc.html.

About the Ottens conversion. HistoricalClimatology.com links to Steven Ottens' 2018 conversion of the same text file into TSV, ODS and GeoPackage (GitHub stvno/stvno.github.io). His only addition is decimal latitude and longitude columns; he did not clean the data. This dataset is built from the original PANGAEA file because the TSV has conversion artefacts:

• Two records with a stray carriage return in a text field are each split into two broken rows, so the TSV has 287,116 rows.
• A few fields were reformatted as if by a spreadsheet: decimal commas changed, 1.125 became 1125, and leading zeros and quotes were stripped.
• 3 rows have fields shifted by one character.

Every other value matches the TSV. The latitude/longitude columns use Ottens' formula and match his values exactly. The build script and data description were checked against the TSV: once the leading zeros and quoting are normalised, 31 cells across 7 columns differ, and every one is a TSV artefact.

Licence and citation

The CLIWOC 2.1 data is licensed CC BY 3.0 Unported on PANGAEA. PANGAEA asks that its citation be quoted whenever the data is used:

Jones, Philip D; Wheeler, Dennis A; Können, Gunther P; Koek, Frits B; Prieto, Maria del Rosario; García-Herrera, Ricardo (2007): Climatological observations from ship logbooks between 1750 and 1854 (release 2.1) [dataset publication series]. PANGAEA, https://doi.org/10.1594/PANGAEA.611088

The files used here are this member of the series:

García-Herrera, Ricardo (2007): CLIWOC Database 2.1 (final release, original IMMA format) [dataset]. PANGAEA, https://doi.org/10.1594/PANGAEA.743343

The main project paper:

García-Herrera, R., Können, G. P., Wheeler, D. A., Prieto, M. R., Jones, P. D. and Koek, F. B. (2005): CLIWOC: A climatological database for the world's oceans 1750-1854. Climatic Change 73, 1-12. https://doi.org/10.1007/s10584-005-6952-6

Neither the KNMI pages nor Ottens' conversion state a licence. The Ottens GitHub repository has no licence file. This dataset does not use his files. Only his trivial latitude/longitude formula is reproduced, and he is credited here. Changes made here: format conversion, typing and the derived columns listed below.

logbooks columns

There are 188 columns. Source fields keep their original names, including the release's own spelling RefenceCourse. Lowercase names are derived columns added by this build, apart from Duplicate, which is joined in from the ship_logbooks table. Blank fields are null.

Key and derived columns

column	type	meaning
record_number	int32	1-based position of the record in CLIWOC21.txt; a stable row key
latitude, longitude	float64	LAT/100 and LON/100, longitude wrapped to -180..180 (east positive)
Duplicate	int8	from ShipLogbookid21.txt: 0 = original logbook, n = n-th duplicate or extract of a voyage already in the data
obs_date_utc	date	YR-MO-DY (Gregorian, UTC)
obs_time_utc	timestamp (UTC)	the date plus HR. Null when HR is null
wind_speed_ms, slp_hpa, air_temp_c, sst_c	float64	W, SLP, AT, SST divided by 10

IMMA core, standardised by CLIWOC (SI units, Greenwich longitude, UTC)

column	type	meaning
YR, MO, DY	int	UTC date (Gregorian)
HR	int16	UTC hour in hundredths (1300 = 13:00). Always a whole hour
LAT	int16	latitude in hundredths of a degree, north positive
LON	int32	longitude in hundredths of a degree, 0-35999 east of Greenwich
ID	int16	logbook identifier; join to ship_logbooks
C1	str	ship nationality: NL, UK, ES, FR, SE, US, DE (Hamburg), DK
D	int16	true wind direction (degrees, from) 1-360, 361 = calm, 362 = variable
WI	int8	wind speed indicator, always 5 ("Beaufort-like classes")
W	int16	wind speed, tenths of m/s, translated from the wind-force words
SLP	int16	sea-level pressure, tenths of hPa
IT, AT	int	air temperature in tenths °C; IT = original scale (3 Celsius, 7 Fahrenheit, 9 Réaumur)
SI, SST	int	sea surface temperature in tenths °C; SI 10 = implied bucket (only 120 rows)
IM, ATTC, TI, LI, II, ATTI, ATTL, ATTE		IMMA bookkeeping constants
DS, VS, NID, DI, VI, VV, WW, W1, A, PPP, WBTI, WBT, DPTI, DPT, N, NH, CL, HI, H, CM, CH, WD, WP, WH, SD, SP, SH	int16	IMMA fields that CLIWOC marks "Not in use". Always null; kept for completeness

CLIWOC attachment, mostly as written in the logbook

group	columns
archive	InstAbbr, InstName, InstPlace, InstLand, NumberEntry, NameArchiveSet, ArchivePart, Specification, LogbookIdent, LogbookLanguage, ImageNumber, Illustr (bool)
position as logged	dead reckoning drLatDeg/Min/Sec/Hem, drLongDeg/Min/Sec/Hem; observed ("true") LatDeg/Min/Sec/Hem, LongDeg/Min/Sec/Hem; LatInd/LonInd (origin of the final position: 1 dead reckoning, 2 observed, 3 interpolated by hand, 4 interpolated, 5 known place inserted, 6 missing); ZeroMeridian; LongitudeUnits
landmarks	LMname1-3, LMdirection1-3, LMdistance1-3 (float; unit in DistToLandmarkUnits), PosCoastal (bool)
logbook time	Calendar (1 Julian, 2 Gregorian), Year, Month, Day, TimeOB (hour on board), DayOfTheWeek, PartDay, Watch, Glasses, StartDay (ship's day started at 12 noon or 0 midnight)
ship and crew	ShipName, Nationality, ShipType, Company, OtherShipInformation, Name1-3, Rank1-3, DASnumber (Dutch Asiatic Shipping voyage number)
voyage	VoyageFrom, VoyageTo, VoyageIni (voyage start, YYYYMMDD string in the logbook calendar), Anchored (bool), AnchorPlace, CMG (course made good), ShipSpeed, Distance (float; unit in DistTravelledUnits)
encounters	EncName, EncNat, EncRem
free-text remarks	LifeOnBoard, CargoMemo, ShipAndRigMemo, BiologyMemo, WarsAndFightsMemo, OtherRem
instrument readings in original units	TairReading, SSTReading, BaroReading, BarTempReading, PumpWater, HumReading (float), with units in AirThermReadingUnits, SSTReadingUnits, AirPressureReadingUnits, BarTempReadingUnits, HumidityUnits, WaterAtThePumpUnits; BarometerType, BarometerBrand, BarometerIndex, HumidityMethod
compass	RefenceCourse, ReferenceWindDirection, Decl (int16, declination added to the logged wind direction to give D), EstError, ApplError
wind and weather as written	AllWindDirections, AllWindForces, Weather, PrecipitationDescriptor, StateSea, ShapeClouds, DirClouds, Clearness, CloudFrac, CurrentDir, CurrentSpeed
weather flags (bool)	Gusts, Rain, Fog, Snow, Thunder, Hail, SeaIce
bookkeeping	TrivialCorrection (bool), Release (CLIWOC release that last changed the record), DistUnits, UnitsOfMeasurement, WindScale

Raw text and standardised values. Wind is stored twice. AllWindDirections and AllWindForces hold the words from the logbook, for example ONGESTADIGE GEREEFDE MARSZEILSKOELTE, FRESH GALES, FRESCO or BON FRAIS. D (degrees, corrected for magnetic declination) and W / wind_speed_ms hold CLIWOC's translation. The wind_force_terms and wind_direction_terms configs are the per-language dictionaries behind that translation. Each phrase maps to a Beaufort class, a coarse CLIWOC class (1 = Bft 0-1, 2 = 2-4, 3 = 5-7, 4 = >7, 8 = other, 9 = missing) and m/s. For direction, 500 means unknown and 999 means variable. No English translation is provided for Weather, StateSea or the other weather text, and none is added here.

Example rows (abridged):

record	ship	date	language	AllWindForces	W	Weather	StateSea
8956	AMBON	1849-01-07	Dutch	ONGESTADIGE GEREEFDE MARSZEILSKOELTE	93	GOED WEER	HOGE ZEE
17043	ARGO	1766-08-11	English	FRESH GALES	190	CLOUDY	LARGE WESTERN SWELL
51567	DESCONOCIDO-20	1777-07-11	Spanish	FRESCO	123	QUEDAMOS CON HORIZONTES MUI CARGOS ...	GRUESA
131192	LA BELETTE	1784-10-16	French	BON FRAIS	93	COUVERT	FORTE

Quirks and caveats

• Duplicates. Some voyages were keyed from more than one logbook, such as the captain's and an officer's, or an original and an extract. These copies usually complement each other rather than repeat exactly. 9,292 rows come from logbooks with Duplicate > 0. Filter on Duplicate = 0 for one record per voyage, as KNMI recommends.
• Zero meridians. Ships counted longitude from many meridians: Greenwich, London, Tenerife, Paris, Cadiz, Ferro, the Lizard, Lanzarote and more: 719 distinct ZeroMeridian values, including UNKNOWN. CLIWOC converted every position to Greenwich in LAT/LON. The logged LongDeg/LongHem stay relative to the ship's own meridian. Checked here: the offset between the two is constant for each meridian, for example -16.65° for Tenerife, +2.33° for Paris and -6.30° for Cadiz. LongitudeUnits says whether the log counted 0-360° or 180°E/W.
• Calendars and ship's days. Year/Month/Day are the logbook date. YR/MO/DY are the UTC Gregorian date. Julian-calendar logs (Calendar = 1, 4,663 rows) are shifted by +11 days. In Gregorian logs the two dates differ by ±1 day in 357 rows, from ship's days that start at noon and from the time-zone conversion. Nine logbook dates are impossible, such as 31 April or 29 February 1769. They are kept as integers, not parsed.
• Time. Most entries are noon observations (TimeOB = 12 for 94% of rows). HR is that local time converted to UTC using longitude, so HR and obs_time_utc are null exactly when longitude is unknown (26,267 rows).
• Missing positions. 21,891 rows have no latitude and 26,267 have no longitude. 4,584 rows have a latitude but no longitude.
• Coverage beyond 1750-1850. UTC dates run from 1662-10-15 to 1855-07-29. 1,138 rows are before 1750 and 39 are in 1855.
• Wind correction. D was corrected for magnetic declination only for reports with a valid position in 1750-1854. Other directions are uncorrected (KNMI documentation). The wind-force-to-m/s values are mostly Beaufort-class midpoints (WMO-1100), and more specific values were used where available. Swedish ships have no W.
• Pressure. SLP is converted from 20 notations, such as INCHES MERCURY, MILLIMETERS MERCURY, RHINE INCHE MERCURY and Dutch DLS… codes. KNMI called it "still under study", with no temperature correction applied. Only 51,936 rows have pressure, 55,007 have air temperature and 120 have SST.
• Dutch symbol codes. In Dutch extract logbooks, symbols were keyed as codes like W00, V04, Z01 and H00 in Weather, ShapeClouds, StateSea, Clearness and CurrentDir. Decode them with the weather_codes config. The symbol images (CLIWOC21Symbols.zip) are not included.
• Text. Logbook text was keyed in upper case. Decimal commas appear in some free-text numeric fields (CurrentSpeed, LMdirection*, ApplError), which stay strings. Two records contain a stray carriage return inside a text field and are kept verbatim: record_number 4939 LMname1 and 77258 OtherRem. The source is Windows-1252 and was decoded to UTF-8.
• Lookup tables. The lookup tables use mixed case while the logbook text is upper case. 187,889 of the 258,988 non-null AllWindForces values match a dictionary phrase exactly, ignoring case. The rest combine several phrases or vary them.
• Not included. Geodata21.txt (place-name gazetteer) is left out because its column widths are undocumented. The MS Access versions are also left out.

Updates

This is a static snapshot. CLIWOC 2.1 is the final release, and the PANGAEA files have not changed since 2010. build.py rebuilds everything from the two PANGAEA zips (uv run build.py). It checks their SHA-256 hashes and the official record count, and casts every typed column strictly.