Method

What Nephos does, how it decides, and what it refuses to say.

This page is written for collection staff, curators and lecturers deciding whether to link, cite, teach with or correct a Nephos record. It states the question, names the sources, defines every verdict label, lists the kinds of disagreement the records contain, and explains how versions and corrections work.

Who makes this.

Nephos is built and maintained by Vedang Sharma as an independent research project. It is not affiliated with, funded by or endorsed by any holding institution, and no museum has reviewed or approved the analysis on these pages. Museum catalogue statements, historical documents and Nephos inferences are attributed separately wherever they appear.

The project began in August 2026 with one record and has been shaped by replies from art historians, curators, climate historians and digital humanities editors. Their suggestions, including the register of dated studies, the plain-words introduction, the taxonomy of disagreement and the machine-readable records, are the reason those things exist. Questions and corrections go to the enquiry page.

The question.

When a painting, its museum record or the artist’s own letters say what the weather was, how far do independent records of those days support each claim?

The question is deliberately narrow. It is not whether the painting is an accurate picture of the sky, which none of the sources can test, and which several correspondents have rightly called the least interesting thing to ask of a painting. It is whether the date, hour, wind, rain, warmth and view that were written down hold against records the writer never saw, and what the pattern of agreement, near-agreement and silence shows about both the painter and the records.

Twenty-seven studies, from Turner’s Whalley sketches of about 1808 to Homer’s Bahamas of 1899, are read the same way. Seventeen are Constable object records, 1821 to 1833, held by six museums; three retain disputed dates. The others are Turner, Caillebotte, Pissarro, Monet twice, Van Gogh’s letters, Perry, Homer and two further Constable studies. Van Gogh’s letters are artist accounts: their weather phrases are claims to test, not independent confirmation. The register page lists the dated Constable sky studies found so far in public museum records, whether or not Nephos has read them.

One method for every study.

Start from what is written. The words on the canvas, in the museum’s record or in a dated letter, quoted exactly and cited. Place and date come first; without them nothing can be tested.
Gather the records of those days. Registers, station files, printed tables, newspapers, storm tracks and the computed sun, each with its distance from the place and a fingerprint of the file. What was searched and not found is listed too.
Test each claim, one source at a time. A claim is set against each record separately; no record stands in for another, and none is averaged into a single score.
Give a verdict, and say what cannot be known. Agrees, Partially agrees, Contradicts or Unknown, each with its reason, and a plain note of what the records cannot tell.

Every study page opens with a card that lays out these four steps for that study. The close readings of paintings and the reading of Van Gogh’s letters are second looks: they read the picture or the words, give no verdict, and link to the weather study that does.

The sources, and what kind of thing each one is.

A record needs a date of making stated by the museum or in the artist's hand. Calendar leaves, publication dates, exhibition dates and accession dates are not dates of making and do not qualify.

Every record separates three kinds of evidence, because they cannot be compared as if they were the same.

The inscriptionDescribed: Words the artist wrote in a particular moment, as the holding museum transcribes them. Nephos never corrects a museum’s transcription. Where a museum has published none, Nephos may read the inscription from a photograph the museum supplies and labels the reading as its own, as on the Perry record. Each phrase becomes one claim.
Luke Howard's registerMeasured and described: The Climate of London (2nd ed., 1833) prints daily wind, barometer and thermometer readings for the London area with a short note for each day. The station for these years was Tottenham or Stratford, not Hampstead. Every cell Nephos cites is read from the scanned page image, and the page URL is given.
20th Century Reanalysis, version 3Estimated: A weather reconstruction built from historical pressure observations; no painting was used. Its grid square nearest London is about 111 by 68 kilometres. The model is run many times, and Nephos reports the average of those runs (the ensemble mean) together with how far they disagree (the spread). The exact queries are listed in each record’s sources.
Central England TemperatureMeasured composite: The Met Office Hadley Centre daily mean series, a regional composite. It is used only for context around the word "warm", never as a measurement at the painter's position.
The computed sunEstimated: Altitude and azimuth for the inscribed hour at an assumed site, computed with Skyfield and the JPL DE440 ephemeris. Where a sky has no horizon or landmark, the computed sun cannot be tested against the picture, and the record says so.
Artist correspondenceDocumentary account: Selected letters retain short quotations, scholarly-edition citations and a source receipt. A weather phrase in a letter is a claim, tested like an inscription; the letter does not independently corroborate the same artist’s painting. Van Gogh’s letters against the gauges.
Météo-France daily recordsMeasured: Daily rain and lowest and highest temperature from the national archive, under the Licence Ouverte: Paris-Montsouris from 1872, Arles from 1881, and gauges in the Gard and Normandy. Quality codes are kept, a missing value stays missing, and each station’s distance from the painted place is given. Each day’s rain is the total from 06:00 UTC that day to 06:00 UTC the next.
Printed tables of the timeMeasured, transcribed: The Annales of the Bureau central météorologique (the Saint-Rémy rain gauge, Marseille wind six times a day) and the U.S. Weather Bureau daily maps. Values are read by eye from the page images and cited to the page; the tables behind the Van Gogh verdicts were checked row by row a second time.
Newspapers and reportsDescribed: Paris papers of January 1880 on the Seine, read on Gallica, and other reports of their day. They describe what writers saw in their own place; a Paris paper is not a report from Vétheuil.
HURDAT2Estimated reconstruction: NOAA’s Atlantic hurricane database: its “best track”, the official reconstruction of the centre of every known tropical storm and hurricane, every six hours since 1851. For the nineteenth century it is rebuilt from ship and coast reports; a small storm far from shipping can be missing, and ordinary winter gales are not included.
Climate summariesMeasured averages: O. L. Fassig’s 1905 climate of the Bahama Islands, from the Nassau record of 1898 to 1902. Averages and extremes describe a place; they cannot date a day.
The museum captureEstimated, capture-specific: Any measurement of the image is a measurement of one photograph of the object, identified by a capture id and file hash. It describes paint colours in that digitisation, not the atmosphere. A second capture could measure differently.

Four weather verdicts; findings first.

Verdicts are labels, not scores. Comparison findings appear first. Unresolved claims remain in an expandable section, with their exact wording, sources and limits. A record without a resolved comparison is presented as a source dossier. A weather phrase in an artist’s letter is tested like an inscription; the close reading of the letters themselves gives no verdict.

Agrees: An independent source reports the same thing on the same scale. No claim in the current corpus has earned this label, because none of the witnesses stood where the painter stood.
Partially agrees: The independent evidence is compatible with the claim but not identical to it: an adjacent compass sector, a value inside a wide uncertainty band, a strength word with no shared scale. Near-agreement is reported as a finding in its own right, never rounded up to a match.
Contradicts: The calendar, or an independent record of the place and days, rules out the claim as written. There are two cases so far: a date that does not exist (the Ashmolean sheet’s “31 Sep.r”, 31 September) and the rain in Van Gogh’s “days of mistral and rain”, when the Saint-Rémy gauge recorded none. A contradiction of the words is not a judgment about the painting.
Unknown: No located source can test the claim. This includes claims that are consistent with the evidence but untestable, such as a stated hour in a sky with no horizon, and claims where the only agreement is between two notes by the same hand.

Six kinds of disagreement.

The cases where the evidence resists are more useful than the cases where it agrees. Each Nephos verdict is filed under one of these, and each type has at least one worked case in the discrepancy packet.

1 · Source correction: A reading changes when the source itself is inspected. Nephos first read Howard’s wind for the 21st by optical character recognition (OCR) and marked it provisional. The page image confirmed NE. Because NE is not Constable’s East, and Howard kept his register at Tottenham or Stratford rather than Hampstead, the comparison was lowered from agreement to partial agreement; both states are kept. Case
2 · Apparent agreement: Two numbers sit close but their uncertainties or their natures differ. A 58% paint segmentation and a 61% ensemble mean with ±33 points of spread are compatible, not matching. Case
3 · Same-hand consistency: Two notes by one author agree. That is internal consistency, one witness rather than two. The Yale and Courtauld sheets agree on date, view, wind and warmth, and are still both Constable's. Case
4 · Impossible statement: A claim contradicts the calendar or physics. The Ashmolean sheet's "31 September" cannot be literal; the museum's correction to 1 October is a catalogue proposal that the weather record supports without proving, since 30 September was also fine. Case
5 · Underdetermined attribution: The evidence is compatible with a proposed cause but cannot single it out. Tate's "may" about a Turner sunset and the Tambora eruption is kept as a "may". Case
6 · Untestable claim: The source that could test the claim does not exist for this object. A stated hour or a stated bearing in a sky with no horizon is consistent with the computed sun and cannot be checked against the picture. Case

What Nephos will not claim.

That a regional weather cell or a London-area register speaks for a hilltop in Hampstead. That a value inside a wide uncertainty band confirms anything. That consistent numbers confirm a felt quality like "warm". That a painted cloud is a measurement of the atmosphere, or that paint colours in one photograph are the colours of the sky.

That Constable always painted exactly what was in front of him. Every record lists its production limits: where the easel stood, whether every passage was executed outdoors in the stated interval, and what preparatory or later work the sources do not establish. "Unknown" in those rows is not evidence that a stage did not happen.

That the register of dated studies is complete. It contains what public museum records make discoverable, with the search method recorded per holder.

Corrections and versions.

Each record carries a presentation version date in its citation. Verdicts and values change only through a dated entry in that record's revision trail, which keeps the original claim, the conflicting evidence, the conclusion now, and links to the retained earlier machine records. Presentation changes that do not touch the evidence are logged in the same trail and earlier pages are kept on the earlier versions page. Their dates name the saved copies, not the exact times of editing.

Anyone can suggest a correction from the "Suggest a correction" link beside a claim, which prepares an email carrying the claim's own URL. Corrections from holding institutions about their own catalogue statements are applied as soon as they are received and are attributed to the institution.

Citing a record.

Every record has a suggested citation with its version date and a copy button. Cite an individual claim by its link, for example …/record/ycba/b1981-25-116/#claim-wind. When you use a museum's catalogue statement or Howard's register, cite those sources directly; each record's sources list gives the URLs. Image rights are separate from the research text: the Yale capture is CC0, the Courtauld image is CC BY-NC 4.0 with the museum's credit, and the Ashmolean image is not reproduced here.

A record's "Print summary" button produces a two-page version with the claims table and the sources appended.

Machine-readable records.

Each record is published as JSON-LD: the object's identity in Linked Art terms, each claim and its verdict as a W3C Web Annotation, and the derivation of every verdict from its sources in PROV-O, including the correction from the OCR reading to the page-image reading. Where the holding museum publishes a IIIF manifest, an annotation page targets its canvas so the evidence can be attached to the museum's own image, carrying the capture id and hash it was measured on.

JSON-LD context and the record documents under /data/records/
IIIF annotation page for the Yale record
Verified analysis receipt and the pre-verification record it revised
Discrepancy packet · Howard, September 1822, transcribed · Register of dated studies
How the records are structured, with a glossary of Nephos terms

Verdict labels are carried as the same four strings. No document contains a confidence number derived from a label.

Using Nephos in teaching.

The records were built so that the reasoning is visible. Three exercises, tried out with correspondents, are offered here for anyone to use.

Read one inscription against the witnesses. Give students the Yale sheet's five phrases and the four witness columns with the verdict column hidden. Ask them to assign a label to each claim and defend it, then compare with the record. The wind row and the warmth row usually split a room.

Follow a correction. The Yale revision trail links the pre-verification machine record, the verified record and the page image. Ask what changed, what did not, and why the verdict changed when the reading itself was confirmed.

Two notes, one hand. The afternoon comparison shows two of Constable's own notes agreeing. Ask whether that is evidence, and of what. Then ask what a third, independent witness would have to be.

Lecturers who use a record are welcome to say so through the enquiry page; it helps to know which pages are read and where they are unclear.

What has already been done.

Nephos follows earlier readers of these skies. Each entry below was checked against its catalogue or web link on 7 September 2026; the note says only what was actually read. Thornes dated undated studies from weather records in the 1970s, Robbins read Constable and Howard together in 2021, and solar-geometry dating has been applied to Monet and Turner. The list is also published as JSON with per-item reading notes in the repository.

Constable and Howard

1950Badt, Kurt. John Constable's Clouds. Routledge & K. Paul, 1950. Locator
Its accessible contents list sections on Goethe and Luke Howard, Constable's cloud studies, Constable and meteorologists, and relations between science and art, but no chapter text was available. Existence checked at the link; contents not read.
1979Thornes, John E. “Constable's Clouds.” The Burlington Magazine 121 (1979), pp. 697-699, 701-704. JSTOR
Thornes examined Constable's annotations in Thomas Forster's meteorological book against Luke Howard's classification, finding strong evidence of meteorological knowledge while stopping short of claiming that Howard inspired the 1821-22 studies. Read in full.
1984Thornes, John E. “Luke Howard's Influence on Art and Literature in the Early Nineteenth Century.” Weather 39 (1984), pp. 252-255. DOI
The verified publisher record establishes an article on Luke Howard's influence on early nineteenth-century art and literature; no abstract or article text was available at the link. Existence checked at the link; contents not read.
2001Hamblyn, Richard. The Invention of Clouds: How an Amateur Meteorologist Forged the Language of the Skies. Farrar, Straus and Giroux, 2001. Locator
The publisher synopsis presents a history of Luke Howard's cloud nomenclature and describes its influence on contemporaries including Constable, Goethe, and Shelley. Abstract or synopsis read.
2008Broglio, Ron. Technologies of the Picturesque: British Art, Poetry, and Instruments, 1750-1830. Bucknell University Press, 2008. Locator
The publisher synopsis says the book traces how meteorology, cartography, and animal breeding shaped picturesque representation in works by Constable and other British artists and writers. Abstract or synopsis read.
2021Robbins, Nicholas. “John Constable, Luke Howard, and the Aesthetics of Climate.” The Art Bulletin 103 (2021), pp. 50-76. DOI
Robbins read Constable's inscribed studies and paintings alongside Howard's observations, tables, and graphs as parallel attempts to represent climate across immediate sensation and aggregated time rather than as a simple line of influence. Read in full.

Constable's skies

1937Bonacina, L. C. W. “John Constable's Centenary: His Position as a Painter of Weather.” Quarterly Journal of the Royal Meteorological Society 63 (1937), pp. 483-490. DOI
The verified record establishes a journal article assessing Constable specifically as a painter of weather; no abstract or article text was available at the link. Existence checked at the link; contents not read.
1969Hawes, Louis. “Constable's Sky Sketches.” Journal of the Warburg and Courtauld Institutes 32 (1969), pp. 344-365. DOI
The verified JSTOR record establishes an article devoted to Constable's sky sketches; no abstract or article text was available at the link. Existence checked at the link; contents not read.
1978Thornes, John B. The Accurate Dating of Certain of John Constable's Cloud Studies 1821-22 Using Historical Weather Records. Department of Geography, University College London, Occasional Papers 34, 1978. Locator
The national bibliography identifies a 30-page report on dating Constable's 1821-22 cloud studies with historical weather records; no abstract or full text was available. Existence checked at the link; contents not read.
1979Thornes, John E. “The Weather Dating of John Constable's Cloud Studies.” Weather 34 (1979), pp. 308-315. DOI
The publisher metadata identifies this as a weather-dating study of Constable's cloud studies; no abstract or article text was available at the link. Existence checked at the link; contents not read.
1998Boudreau, Beverley A. “John Constable's 1821 and 1822 Cloud Studies: The Artist as a Scientist?” Weather 53 (1998), pp. 357-362. DOI
The verified publisher record establishes an article asking whether Constable's 1821 and 1822 cloud studies can be understood as a scientist's work; no abstract or article text was available at the link. Existence checked at the link; contents not read.
1999Thornes, John E. John Constable's Skies: A Fusion of Art and Science. University of Birmingham Press, 1999. Locator
The catalogue identifies a 288-page book devoted to Constable's skies and explicitly framed as a fusion of art and science; no abstract or book text was available. Existence checked at the link; contents not read.
2000Morris, Edward. Constable's Clouds: Paintings and Cloud Studies by John Constable. Trustees of the National Galleries of Scotland and Trustees of the National Museums and Galleries on Merseyside, 2000. Locator
The museum catalogue record identifies this as the 2000 exhibition catalogue devoted to Constable's paintings and cloud studies; no essay text was available at the link. Existence checked at the link; contents not read.
2007Wood, Gillen D'Arcy. “Constable, Clouds, Climate Change.” The Wordsworth Circle 38 (2007), pp. 25-33. DOI
The institutional record establishes a peer-reviewed article focused on Constable's clouds and climate change; no abstract or article text was available at the link. Existence checked at the link; contents not read.
2020Jankovic, Vladimir. La vérité dans l'atmosphère: lire les ciels de Constable. Revue de l'Art, 2020. Locator
Jankovic argued that Constable approached cloud studies systematically but transformed atmospheric phenomena into idealized visual and emotional forms rather than straightforward documentary records. Abstract or synopsis read.

Comparable methods on other painters

2006Baker, Jacob; Thornes, John E. “Solar Position within Monet's Houses of Parliament.” Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 462 (2006), pp. 3775-3788. DOI
Baker and Thornes measured the painted Sun against architectural reference points and calculated solar tracks to derive dates, times, and Monet's vantage point for nine Houses of Parliament paintings. Abstract or synopsis read.
2010Khan, Soraya; Thornes, John E.; Baker, Jacob; Olson, Donald W.; Doescher, Russell L. “Monet at the Savoy.” Area 42 (2010), pp. 208-216. DOI
The authors used solar geometry to date four Waterloo Bridge paintings and one Charing Cross Bridge painting, then checked the probable dates against Monet's letters and contemporary weather records. Abstract or synopsis read.
2014Mohan, Geoffrey. “Physicist Puts Time on Timeless Monet Painting.” Los Angeles Times, 2014. Locator
Mohan reported Donald Olson's dating of Monet's Impression, Sunrise by combining the reconstructed viewpoint, solar geometry, tide levels, wind direction, and historical weather observations. Read in full.
2022Olson, Donald W. Investigating Art, History, and Literature with Astronomy: Determining Time, Place, and Other Hidden Details Linked to the Stars. Springer Cham, 2022. Locator
Olson set out a practical workflow combining astronomy software with archives, maps, letters, and diaries and presented art case studies involving Vermeer, Monet, and Turner's early oil paintings. Abstract or synopsis read.

Turner, Monet and the atmosphere

2007Zerefos, C. S.; Gerogiannis, V. T.; Balis, D.; Zerefos, S. C.; Kazantzidis, A. “Atmospheric Effects of Volcanic Eruptions as Seen by Famous Artists and Depicted in Their Paintings.” Atmospheric Chemistry and Physics 7 (2007), pp. 4027-4042. DOI
The authors measured red-to-green ratios in dated sunset paintings, including works by Turner, and used radiative-transfer modelling to reconstruct aerosol optical depth after major eruptions from 1500 to 1900. Abstract or synopsis read.
2014Zerefos, C. S.; Tetsis, P.; Kazantzidis, A.; Amiridis, V.; Zerefos, S. C.; Luterbacher, J.; Eleftheratos, K.; Gerasopoulos, E.; Kazadzis, S.; Papayannis, A. “Further Evidence of Important Environmental Information Content in Red-to-Green Ratios as Depicted in Paintings by Great Masters.” Atmospheric Chemistry and Physics 14 (2014), pp. 2987-3015. DOI
The authors tested painting-image colour measurements against precision photography, independent aerosol proxies, and a co-located painting experiment before using red-to-green ratios to estimate historical aerosol optical depth. Abstract or synopsis read.
2023Albright, Anna Lea; Huybers, Peter. “Paintings by Turner and Monet Depict Trends in 19th Century Air Pollution.” Proceedings of the National Academy of Sciences 120 (2023), e2219118120. DOI
Albright and Huybers quantified contrast and intensity in 60 Turner and 38 Monet paintings and modelled their association with historical sulfur dioxide emissions while controlling for time trends and subject matter. Read in full.