A source record under pressure
Entered as Day Sixty-Two
Lake Suwa's published late freeze-date trend is 4.6 days per decade, but the 73 winters with dates alone move at 0.89. Move the missing winters, the censoring day, and the source series yourself, then inspect the record's refusals. The instrument cannot decide which disagreeing historical record is right or explain the weather behind it.
A shrine kept a ridge and a ritual. A climate paper kept a number. Put your hands on the gap between them.
In the 1923 to 2014 window, dated winters move later by 0.77 to 0.95 days per decade across four NSIDC series, while the published ARAI1 Tobit slope is 4.61 because 19 winters are entered as day 62.
The headline is the default 1923 to 2014 comparison. The extended control below tests a different window and is not included in that bound.
The operable base
Watch the rings carry the slope
The selected compilation supplies the plotted rows. A dated freeze below the line is a cyan dot; a source N row or a dated freeze at or beyond the line is an amber ring when entered winters are included.
The timeline is waiting for the local record.
Loading the local record.
The check
The six baseline checks below are recomputed from the frozen NSIDC extract and compare the result with the expected record. Each row can visibly fail if the extract or the engine stops closing.
The baseline check is waiting for the local files.
Free choices named here are the source series, whether censored winters stay in the fit, the entered day, the analysis window, the calendar hypothesis, the season label, how the cited star and DOY are read, and whether the eight-year crossing note is treated as a run. The page refuses values outside the record's support.
I / reproduce before interpreting
First, reproduce the quoted number
Sharma et al.'s published Lake Suwa anchor gives 0.19 days per decade for the early analysis and 4.6 days per decade for the late analysis. Their reason for using a Tobit model is precise: a winter with no full freeze is not an ordinary missing value.
The current ARAI1 extract and the paper's printed no-freeze count are being compared.
The frozen ARAI1 rows are the current NSIDC distribution of that source series, not a copy of the later Ochanomizu table. The engine reads each season's froze field, keeps the source's converted freeze date, and enters N at the declared upper limit.
Engine against publication
Anchor source: Sharma et al., Direct observations of ice seasonality, Results and Methods. The NSIDC distribution and its update history are documented in the G01377 user guide.
II / the choice inside the number
Which winters are the rings?
The late slope is not only a model choice. It also depends on which observer series supplies a winter when several are present. NSIDC keeps the ARAI1 composite beside the shrine, station and Arakawa series. The source guide records the substitution rule, while this page lets the four underlying compilations speak separately.
The four source series are waiting for the local extract.
The table's dated-only column is this page's ordinary least-squares comparison, not a source-published figure. It is shown because the dismissal that dropping rings is always wrong is too simple, while the opposite dismissal that rings are always trustworthy is also too simple.
Counts have a clock too
Sharma et al. count winters by the season-start label. This is not a cosmetic choice when a claim names a span. Switch the label and watch the count, without changing the source rows.
III / disagreement is data
The ledger refuses a verdict
A composite can conceal a disagreement by choosing one date. This ledger returns to the four NSIDC rows and prints their own entries in their own categories. Y means the source says it froze, N means it says it did not, U is uncertain, and a blank date remains a blank date. The page does not decide which observer is right.
Loading the rows.
The page can reproduce the NSIDC series because those rows are the operative record. It does not insert a per-winter Ochanomizu row here: the later file is open access at its repository, but no reuse licence is stated.
IV / the ridge is not the lake
御神渡り is not 全面結氷
The shrine's categories keep two events apart. 全面結氷 is the full-freeze event in the later table's gloss; 御神渡り is the ridge associated with the god's crossing. 明けの海 is the record's no-crossing term. A freeze can be recorded with no crossing, so choosing no crossing as no freeze is a deliberate conflation, not a translation.
Cited freeze column
loadingloading
loading
Cited crossing as no freeze
loadingloading
The choice is shown to expose the error, not to recommend it.
What the later file adds
The cited aggregate is waiting for the local summary.
These are aggregate values derived from the Hasegawa, Mikami and Hirano repository record, version 20260107ver, created 2026-01-07 and available 2026-04-13. The later CSV and workbook are not shipped, and no individual status or date from them is reproduced here.
The old lump note and the new run
The later compilation marks the crossing block for the years 1507 to 1514 with a star. Mikami's account says the source wrote one statement about eight years, not eight yearly observations. Let the note stay unknown, or operate the stronger reading and see exactly what it changes.
Loading the run comparison.
V / the record can say no
Refusal is part of the instrument
A good record does not answer every question. Ask the selected NSIDC compilation for a season, or try to fit the interval the paper itself calls unreliable. The refusal stays visible instead of becoming a guessed value.
The inspector has not been asked a question.
The middle-window refusal has not been tested.
The calendar question
The NSIDC guide documents a correction before its modern release. It does not state whether dates before 1582 should be treated as Julian or proleptic Gregorian for the step test below. Apply the hypothesis as a sensitivity, then compare it with the record's own means.
The calendar sensitivity is waiting for the local files.
VI / what survives
The narrower truth is still a warming record
Under every NSIDC source, no-freeze winters become more common in the late record. The page is not saying the lake is not warming, and it is not saying that the published Tobit model is illegitimate. It is showing what that model's headline contains: dated freezes, plus a judgement about winters that never supplied a date.
What survives the choices is modest and strong at once. In the published 1923 to 2014 window, the dated winters move later by less than one day per decade in each live NSIDC compilation. The extended control is a separate window and can exceed that bound. The frequency of no-freeze and no-crossing winters rises under every source. The size of the quoted late trend depends on which rings are present and where the line is entered.
We searched the Artificial Wasteland corpus, Sharma et al. 2016, the NSIDC Global Lake and River Ice Phenology Database and its user guide, Arakawa 1954, Ishiguro 2001, the 2024 Hasegawa, Hirano and Mikami abstract, the JCDP and JIVX pages, Japanese press coverage of 明けの海, and English and Japanese web searches on 2026-09-14. JIVX already lists nine winter disagreements between two digitizations and keeps ice and crossing as separate columns. We did not find a published interactive that combines the four-series source swap, the entered-day and calendar sensitivities, and the aggregate-only treatment of the later 2026 compilation.
Sources and terms
- NSIDC G01377, Version 1. The data condition requires citation and description of the subset. This page ships the four Suwa series as a frozen extract.
- Sharma, Magnuson, Batt, Winslow, Korhonen and Aono. The published Tobit anchor, the reliability mask, the source rules and the no-freeze counts.
- NSIDC user guide. Freeze means first complete cover in the database, and its Suwa notes document calendar correction and source substitution.
- Hasegawa, Mikami and Hirano. A later repository record used only for aggregate cited comparisons. Its page states open access but no reuse licence.
- Arakawa 1954 and Ishiguro 2001. The older tables and the study of their changing definitions keep the record's categories visible.
- Hasegawa, Hirano and Mikami 2024. The abstract explains why a source-tagged database matters when the entries differ by period.
- Mikami's Omiwatari account, JIVX Atlas and Weathernews. Prior accounts separate the ridge from full freeze and report the modern run; they do not supply this page's fit.
- Nagano Prefecture, page 50. The shrine-table row source was checked but is not transcribed here because no reuse licence was found.