A clear line between
history and a forecast.
FirstSnowOdds tracks one event: the season’s first measurable snowfall. FirstSnowOdds pairs observed historical timing with a separate near-term NWS snow signal.
Who runs FirstSnowOdds?
FirstSnowOdds is maintained by Joe Ruch. It is an independent project, with no NOAA or National Weather Service endorsement.
For a correction, include the city, date and source: joeruch23@gmail.com. Each city’s Sources link names the historical reference, coverage, forecast source and calculations. Find your city’s sources →
Data / media kit · original ten Gold cities
First-snow dates you can trace
Ten city references, not a national census of first snow. Primary averages, normals, project medians, sample windows and official extremes keep separate definitions. Status comes from the same saved-evidence model as the tracker and city pages.
Version 1.0.0 · reviewed 2026-10-17. Saved-evidence snapshot: .
UTF-8 CSV. Unknown values are the literal null; arrays use JSON. The download reads current saved records when requested and can be newer than this HTML snapshot. Issue, ingestion, check and verification times are separate.
| City | Typical timing / reference | Current saved status | Full data record |
|---|---|---|---|
| Colorado Springs, CO | October 27Published NWS historical meanColorado Springs city / airport record | Snow Watch: watchingPublished result: none-published | Fields for Colorado Springs
|
| Denver, CO | October 19Published NWS city averageDenver official city record | Snow Watch: watchingPublished result: none-published | Fields for Denver
|
| Chicago, IL | November 18Published NWS normal first dateChicago O’Hare | Snow Watch: quietPublished result: none-published | Fields for Chicago
|
| Boston, MA | December 2Project historical medianBoston Logan Airport | Snow Watch: quietPublished result: none-published | Fields for Boston
|
| Minneapolis, MN | November 8Project historical medianMinneapolis–St. Paul International Airport | Snow Watch: quietPublished result: none-published | Fields for Minneapolis
|
| Buffalo, NY | November 9Project historical medianBuffalo Niagara International Airport | Snow Watch: watchingPublished result: none-published | Fields for Buffalo
|
| New York City, NY | December 14Project historical medianCentral Park | Snow Watch: quietPublished result: none-published | Fields for New York City
|
| Cleveland, OH | November 11Project historical medianCleveland Hopkins International Airport | Snow Watch: quietPublished result: none-published | Fields for Cleveland
|
| Philadelphia, PA | December 16Project historical medianPhiladelphia International Airport | Snow Watch: quietPublished result: none-published | Fields for Philadelphia
|
| Salt Lake City, UT | November 5Project historical medianSalt Lake City International Airport | Snow Watch: watchingPublished result: none-published | Fields for Salt Lake City
|
Data dictionary · 64 fields
A season is named by its July starting year. Historical month/day labels are not timestamps. YYYY-MM-DD values are observation/reporting dates, without a timezone shift. ISO timestamps retain their source offsets. Snowfall amounts are inches unless a field explicitly says otherwise.
- city
- City label; the observation describes the named reference, not every neighborhood.
- state
- Two-letter U.S. state abbreviation.
- location_key
- Existing curated FirstSnowOdds identity. Original ten only; inclusion is not indexing approval.
- city_url
- Canonical city experience URL.
- reference_name
- Reference for the PRIMARY typical-date statistic.
- reference_scope
- Geographic and record scope of the primary statistic.
- station_id
- Primary statistic station ID, or null for a composite/changing-site city statistic.
- statistic_type
- published_nws_city_average, published_nws_historical_mean, published_nws_normal, or project_historical_median.
- statistic_label
- Exact consumer method label used by the city and national comparison.
- baseline_start
- First year in the primary source period label, or null when not stated.
- baseline_end
- Last year in the primary source period label. Not a promise that all seasons are complete.
- included_seasons
- JSON array of included starting years for a project primary median. Null if the NWS source does not supply membership.
- excluded_seasons
- JSON array of excluded starting years for a project primary median; null when not established by the primary source.
- typical_date
- Historical month/day label, never this year’s prediction.
- window_start
- Month/day at the project sample’s lower quartile; null without a supported distribution window.
- window_end
- Month/day at the project sample’s upper quartile. Snowy-season window; not a forecast interval.
- window_reference
- Named distribution reference; distinct from the primary average/normal where applicable.
- window_period
- Distribution period label, distinct from an official primary-statistic period.
- distribution_kind
- complete-station or official-recent; separate from the primary statistic type.
- distribution_reference
- Reference for the Winter Line sample.
- distribution_station_id
- Station ID only for a station distribution; null for Denver’s separately supplied recent city dates.
- distribution_period
- July–June sample period label. A 1995–2025 station baseline considers starting seasons 1995 through 2024.
- distribution_included_seasons
- JSON array of all included sample starting years. Complete snow-free seasons remain included.
- distribution_excluded_seasons
- JSON array of completeness-excluded station seasons; null for an official recent series with no exclusion registry.
- earliest_sample_date
- Earliest seasonal calendar position in the displayed sample, as its observed YYYY-MM-DD date.
- latest_sample_date
- Latest seasonal calendar position in the displayed sample. January belongs to the preceding starting season.
- extreme_scope
- Scope of the two sample-date fields: sample. Never an all-time city record claim.
- published_extreme_early
- Separate official-table earliest date, when used by the page; otherwise null.
- published_extreme_late
- Separate official-table latest date, when used by the page; otherwise null.
- published_extreme_scope
- Exact official-table record/update scope. Cleveland retains its August 2020 vintage.
- published_extreme_source
- Source for the separate official extremes, not the sample extremes.
- previous_season
- Previous July–June season’s starting year, even if its observation is unavailable.
- previous_arrival_date
- Reviewed previous-season reporting date, or null. Older observations are not relabelled as last season.
- previous_amount
- New snowfall in inches on that reporting date, or null.
- previous_reference
- Reference for the previous-season observation.
- previous_station_id
- Known previous observation station; null when the source is an official city table without a fixed station ID.
- previous_source
- Authoritative previous-season observation source.
- source_urls
- JSON array of historical timing, sample, story and discrepancy source URLs.
- primary_source_sha256
- Retained source hash for the primary statistic; not a current forecast hash.
- distribution_source_version
- Sample method/source version used by the historical interaction.
- distribution_snow_rows_sha256
- Retained station SNOW-row hash, or null for an official recent city series.
- method_version
- Dataset, national presentation and Gold context method versions.
- dataset_version
- Historical dataset/schema release version.
- dataset_date
- Date this dataset/schema release was reviewed; not a weather issue time.
- as_of
- When this response read its saved evidence; not the NWS issue time.
- snow_watch_state
- Shared presentation state: quiet, watching (sub-threshold hint), signal, stale, incomplete, unavailable or retrieval-failure.
- snow_watch_method_version
- Version of the saved Snow Watch method, or null without a saved check.
- forecast_source_url
- Named NWS grid source for the saved forecast, distinct from historical and observation sources.
- candidate_start
- Unmodified forecast interval start timestamp, only with a usable active hint/signal.
- candidate_end
- Exclusive forecast interval end timestamp; not an exact snowfall onset.
- issue_time
- NWS source issue timestamp, including an old issue when known.
- retrieval_time
- Stored first ingestion time of this guidance; null if unknown. Not replaced with check time.
- last_check_time
- When the saved location check was made, distinct from issue and ingestion.
- freshness_state
- fresh-usable, fresh-usable-partial, stale, incomplete, unavailable or retrieval-failure.
- current_season
- Current local snow season’s July starting year. January 2027 is season 2026.
- season_label
- Human-readable July–June season.
- result_status
- none-published, confirmed, corrected-date, corrected-amount, corrected, retracted or unavailable.
- result_date
- Verified reporting YYYY-MM-DD date; null for none, retracted or unavailable.
- result_amount
- Published new snowfall in inches; null without a verified amount.
- result_reference
- Named observation reference in the published result ledger, independent of the historical statistic.
- result_station_id
- Published observation station ID; never silently reused as the historical composite reference.
- verification_time
- Owner review/publication timestamp, including a retraction revision when present.
- result_source
- Authoritative published observation source; no unreviewed candidates.
- result_source_stage
- Official preliminary/final observation status. Revisions remain possible.
Citation, methods and version policy
Joe Ruch. FirstSnowOdds original Gold city first-snow data, version 1.0.0 (2026-10-17). Saved-evidence snapshot 2026-11-21T12:11:20.408Z. https://firstsnowodds.com/methodology#gold-data. Cite each row’s original sources for scientific use.
Historical/schema changes require a new dataset version. Saved forecasts and reviewed results can change within a dataset version; retain the downloaded file and its as_of value when citing a status snapshot. The primary period and station do not silently become the explorer’s period/station. Null means unknown or inapplicable, not zero or no snowfall.
Snow Watch describes NWS guidance, not an official NWS watch or a first-event probability. Forecast evidence never confirms arrival. Published records can be preliminary, corrected or retracted. Export inclusion does not grant indexing approval.
For corrections, include the city, field, dataset version and source: joeruch23@gmail.com. Each city’s Sources page preserves the reference, sample and original source links.
What counts as first snow?
At least 0.1 inch of new snowfall at the named reference in a daily station record. Reported snowfall can include sleet or ice pellets under NWS measurement conventions. A trace of flakes does not meet that definition. Neither does leftover snow on the ground. The NWS uses separate first dates for trace, measurable snowfall and one inch.
Our season runs July 1 through June 30. This is a tracking convention. A single “winter arrival” date can be less meaningful at locations with summer snow. Station reporting dates are not always the exact calendar day or hour when flakes began.
What does the historical window mean?
The window spans the 25th to 75th percentiles of first-snow dates in complete snowy seasons. The historical percentage is the number of complete seasons with a first snow by that calendar date, divided by all complete seasons. Fully observed seasons without snow stay in the denominator.
We consider 30 seasons, July 1995 through June 2025. Every daily snowfall observation must be present and pass the parser’s quality checks. At least 20 complete seasons are required to display a distribution. Excluded years are listed on each page. Strict completeness can select an unrepresentative subset, and older weather patterns do not guarantee today’s climate.
These are unconditional historical frequencies, not this year’s forecast odds. They are not automatically updated merely because another day passes without snow. The middle-50% window is not a promised arrival window.
Source: NOAA NCEI GHCN Daily. Its data format defines snowfall in integer millimeters. For US observations, 0.1 inch is stored as approximately 3 mm. We exclude missing, quality-flagged, and presumed-zero data; trace values do not qualify.
Why does Denver show an average?
Denver’s raw airport dataset had fewer than 20 complete seasons under our current check. Its distribution is withheld. Instead, the page shows the NWS Boulder published October 19 average and recent official dates. That source does not identify the averaging period beside the average, so we do not label it as our 1995–2025 baseline or infer probabilities from it.
One observed discrepancy also needs review: the GHCN daily record dates Denver’s 2022 first qualifying accumulation November 3, while the NWS city table lists November 4. Denver’s displayed recent dates follow the cited NWS city table. We do not silently merge those records.
What does “Snow by when?” show?
The Winter Line plots first-snow date horizontally and July–June season vertically, newest at the top. Each dot is one included season at the named historical reference. A separate None rail keeps complete snow-free seasons visible without assigning a date. Filled dots had snow by your call; open dots reached it later. Your call is the vertical date line; any Snow Watch band is current guidance, not a confirmed event or a climate trend. The individual-season table provides every exact date. The date control counts how many reviewed seasons had their first measurable snow on or before your selected calendar date. It runs locally and makes no forecast request when you move it. Denver’s explorer is separately labeled as ten recent NWS-reported seasons, 2016–2025. It is a short descriptive sample, not the withheld long-term station distribution. Its percentages do not come from the October 19 average.
A personal call is your own date choice. “Early,” “middle,” and “late” describe its position in the displayed sample: fewer than a quarter of seasons recorded snow before the date, between a quarter and three quarters, or more than three quarters. Snow-free complete seasons remain in that denominator. These labels do not judge your call’s chances. Forecast candidate comparisons use local calendar dates and an exclusive interval end.
What is a snow signal?
We request the NWS point and raw grid forecast for a place’s coordinates. A future forecast interval entirely inside the next 72 hours with at least 2.54 mm of forecast snowfall triggers a measurable-snow signal. Forecast intervals are shown as periods, not exact onset times.
Numeric zero-snow guidance requires valid, non-overlapping, non-null snowfall intervals across the full 72 hours. Separately, complete structured NWS weather guidance can establish that no snow is indicated even when numeric snowfall coverage is shorter; that does not create a numeric total. A null weather value is missing, while a decoded no-significant-weather condition is valid qualitative guidance. Weather indicating snow without a usable amount remains a qualitative possibility. Gaps, outdated guidance and failed requests stay unresolved. Positive amounts below 0.1 inch in individual intervals are described separately; we do not infer a daily accumulation from incomplete intervals.
Snow Watch is a FirstSnowOdds summary, not an official NWS watch product. Snow Watch has three atmospheric states: quiet (complete zero-snow guidance), watching (a positive amount below 0.1 inch in a future interval), and signal (an interval reaching 0.1 inch). These are descriptions of one NWS forecast, not probabilities or verified first-snow predictions. Persistent evidence means the same state and overlapping period appeared in two source updates at least three hours apart; it does not establish independent confidence.
Current guidance follows NWS HTTP cache expiry and successful retrieval or conditional revalidation. An unchanged issue time can remain current after a valid NWS check. Validation never extends forecast intervals or fills missing hours; the 72-hour scope is not a guarantee of forecast accuracy. Forecast grid snow is not evidence that a station has already measured its first snow.
A failed live refresh is shown as unavailable. We never substitute design fixtures or bundled source captures. We reuse forecasts only within NWS-provided HTTP freshness, use conditional revalidation where available, and recheck point mappings at least daily. Failures honor Retry-After with a minimum one-minute cooldown. Visiting a location checks its guidance. Open views schedule polling separately, at intervals between one and 30 minutes; an expired response is not labeled current while waiting. NWS checked names the upstream validation time; Last app check names the application evaluation time. Neither changes the original NWS issue time. A private collection job can check curated locations in four-place batches on six-hour collection slots. Automatic scheduling is not enabled yet.
The source update time and the time we retrieved it are separate. “Latest guidance” shows the source issue time. Stale data expires visibly even on a page left open. Original NWS accumulation intervals are preserved: we do not add or prorate compressed intervals, infer snowfall from precipitation probability, or turn forecast text into snow odds.
Why aren’t live first-snow odds shown?
Combining historical frequency with ensemble guidance needs calibration and backtesting. We have not validated that model. Publishing a precise probability or a most-likely future date would imply skill we have not established.
A future model will need independent verification by lead time and region, a trustworthy observation registry, and tests of probability reliability. Raw ensemble agreement and generic precipitation probability are not interchangeable with the probability of first measurable snow.
National map geography
Map boundaries use Natural Earth public-domain data. Alaska and Hawaii appear as insets and are shown at different scales. The map covers curated locations. Its Snow to watch filter uses 72 hours of guidance, while city outlooks can show up to seven days.
How local is the answer?
Search uses a refreshable GeoNames US place index: 168,541 populated places. The maintained GeoNames postal catalog contains 41,706 codes, including separately published U.S. territory files. A dated USPS source reconciliation adds 637 recognized identifiers, for 42,343 in the combined inventory. All 41,810 identifiers in the reviewed October 2026 USPS source union have an explicit destination or postal disposition. Recognition can mean a closed, uncertain or nongeographic mailing identity, not a physical forecast or universal active-USPS certification. An exact place, state and county crosswalk reuses existing identities where unambiguous; otherwise the named postal locality keeps its own identity and representative point. GeoNames is not a certified current USPS ZIP inventory. Overseas military codes identify mail routing, not a verified physical forecast location. A ZIP may resolve to a named place rather than an exact postal polygon or your address. We display the resolved name and coordinates. We never silently assign a suburb a nearby city’s climate history.
Selected destinations have named reference stations. Distance, elevation, coastal exposure and lake-effect bands can all matter. Locations without approved local timing can explore a separately named nearby station record where one qualifies. Nationwide location lookup does not imply nationwide validated local climatology.
How can additional local references qualify?
New local relationships use the same historical quality classifier with additional evidence: at least 20 complete seasons, at least 80% snowy seasons, a complete 2024–25 baseline season, consistent NOAA/HOMR/ACIS identity, a current station period and no documented or undated relocation since 1995. The consumer place and station must share state, county and NWS forecast-zone context, be within 15 km, and pass actual USGS place/station/midpoint elevation checks with a maximum 50 m corridor range. Existing mountain, lake and coastal review screens remain, with additional Appalachian, Black Hills, northern-Indiana and northeastern-Ohio lake-effect safeguards. Conflicting qualified timing candidates stay under review. These point checks do not prove identical snowfall at every address or rule out undocumented station changes.
Approved references attach to the consumer identity; they do not change its name, route or local forecast coordinates. Newly saved calls can bind the named reference and relationship version. Old unbound calls remain unscored. A confirmed station observation is always attributed to that station.
What if only nearby history is available?
A nearby station can provide clearly labeled regional context without establishing your town’s usual first-snow date. The reference name, distance, usable seasons and limitations stay visible. Your local call remains a hunch unless a separate reviewed observation reference was bound when you saved it. A nearby station’s confirmation is not silently assigned to your community.
Regional context may use either complete July–June seasons or a separately labeled first-event cohort. In the latter, every daily SNOW value from July 1 through the first qualifying event must be valid. A missing day before that event excludes the season; later missing days cannot change its first date. A snow-free season still requires the entire season. These descriptive station samples do not replace the complete-season baselines or their historical frequencies. Missing-prefix exclusions can bias the retained sample.
Regional selection is computed offline using source identity, recency, sample size, distance, elevation and geographic screens. Closer, comparatively suitable references rank first. Other validated station histories can be explored within 250 km, a product boundary rather than a claim of climate equivalence. Beyond the closer 40 km sensitive-area or 80 km general screen, or with known elevation/region mismatches, the primary answer prominently labels regional reference history. Mountain terrain, lake effects and coastal exposure can differ within these broad screens; no exploratory match grants local scoring approval. Coarse GeoNames DEM heights from SRTM/GTOPO30 are screening evidence, not surveyed municipal heights or proof of common terrain. They cannot grant local representative approval. When reviewed history is still unavailable, the locality retains its current Snow Watch, an unscored hunch where appropriate, and nearby exploration. Missing guidance never means no snow.
Has first snow already happened?
Check the location page for this season’s confirmed reference-station result. Until official observations are reviewed, completion remains unverified. A later forecast snow signal may not be the first of the season. Forecast guidance never confirms completion; reviewed observations, reporting dates and corrections determine the season ledger.
How well does the record match my place?
Gold cities retain their reviewed city sources. Standard references need at least 20 complete seasons, at least 25 candidate seasons, at least 15 snowy seasons, accepted source hashes, and two-thirds completeness. For other places, we check geography before station distance. A lowland representative match must be in the same state, within 15 km when place elevation is unknown, or within 25 km with a known elevation difference no greater than 50 m. Those distances are conservative product limits, not validated measures of climate equivalence.
Approved representative relationships are generated offline, then looked up by consumer place identity. Public pages do not search the station inventory. Retained USGS 3DEP point elevations support a point-height comparison; they do not establish terrain or neighborhood equivalence. Coverage does not promote a place to Gold or make it indexable.
Unreviewed mountain, Great Lakes snowbelt, coastal and terrain-sensitive matches keep historical timing withheld. A nearby station does not automatically establish a match. Representative windows and Winter Line marks describe the named station; they do not become invented town records. We disclose unknown elevation, distance, coverage and other limitations on Sources. Rare-snow places have a specialized explanation rather than an annual prediction. Unsupported history is different from a known rare-snow climate.
How does a season end?
Only official observation evidence can establish first measurable snow. A protected owner workflow collects a NCEI candidate or records a reviewed NWS CLI/CF6 observation, then reviews station identity, snowfall type, reporting date, observing window and first-of-season coverage. GHCN candidates use at least 3 whole millimeters; trace, missing, presumed-zero and failed-quality observations never confirm a result. CF6 snowfall can include hail, so a positive column alone is insufficient. Partial-day reports and omitted snowfall remain unresolved.
Confirmations, corrections and retractions are stored separately from forecasts with their source identifiers, retrieval and issue times, review method and revision history. Official preliminary observations can be corrected later. A result confirms the named reference station, not every neighborhood. An accepted representative place can use that same station result, with the reference disclosed. Only calls made before the observed reporting date and tied to the same place, season and station receive an absolute date-error score. Retraction removes the score; a corrected date recalculates it. Basic calls still need no account.
Observation reporting dates are not exact snowfall onset times. NWS climate days usually follow local standard time; cooperative reports can describe the preceding 24 hours. The owner reviews and retains that window. See the NWS climate-product instruction and the GHCN format and flags.
Your calls, saved places & privacy
Your calls stay on this device and are never forecast inputs or sent to our server. We retain up to 16 location-and-season calls, with up to three earlier selections for each. Each records your chosen date, selection time, historical count and source version, and the Snow Watch state and candidate period when selected. New calls also record their reviewed station identifier and quality-method version where available. Calls older than 730 days are ignored and are pruned on the next save; clearing browser data removes them immediately. Basic calls need no account, while private review access is controlled separately.
Saved places and previous checks stay in this browser, with no account requirement for that functionality. We keep up to eight saved places and twelve location visit records. Visit records older than 180 days are ignored. Clearing browser data removes them.
Our server stores weather snapshots and resolved locations without visitor identifiers. We use Cloudflare Web Analytics to measure visits and page performance. This is separate from the optional engagement measurement described on our Privacy page. Consumer pages are ad-free. Hosting and upstream services may retain their standard operational logs. Queries are searched in our server’s GeoNames index; precise device geolocation is not requested.
Private pool keys also stay on this device. Clearing them can remove your ability to edit or organize a pool. Save your private organizer link, or delete the server pool in its organizer controls first.
Owner-only Snow Pools preview
Public pool creation and joining are not currently available.
A pool stores its name, named observation reference, cutoff, optional off-platform prize description, display names, dates, optional first-day amount and submission times. No email, account profile, entry fee or payment is collected. Names and calls are visible to anyone with the private join link and site access. The primary pool surface remains owner-only; a join link does not override its access boundary.
Join, organizer and edit keys are random capabilities. Only their hashes are stored on our server. Links carry keys after # so they are not sent in a page URL; the browser moves them into local storage when possible. Keep organizer links private. Device storage holds at most 16 pool keys for up to 730 days. Clearing browser data does not delete a server board; organizers can delete their board and calls. Boards remain until organizer deletion.
Closest absolute date error wins; an optional first-day amount breaks date ties, then remaining ties share the result. Calls made on or after the observed event day cannot win. Rules and observation reference are fixed at creation. The organizer attests an exact date using a dated NOAA/NWS source. This is not automatic or independent observation confirmation and does not change Snow Watch. No-event pool closure is deferred; do not invent an event date for a snow-free season.
Snow Stories & the experimental arrival engine
Stories are computed from reviewed station cohorts or explicitly reviewed official city records. Sample extremes are not all-time city records. First-day amounts are daily observations; multi-day storm totals retain their interval. Weather whiplash compares quality-controlled daily highs at one station across the previous three days and the first-snow day. It never establishes an hourly temperature drop or record warmth. Missing temperature observations suppress that story.
Snow DNA uses the middle-half date span and arrival counts in its displayed cohort. Its variability labels are editorial: at most 14 days, 15–35 days, or more than 35 days. They are not forecast confidence. Anniversaries use local civil dates. Every historical story exposes its source, record scope and method.
The private arrival research engine stores empirical daily arrival mass, cumulative frequency, hazard and complete snow-free-season mass with source hashes. It remains unconditional until observations verify this season’s eligibility. A historical leave-one-season-out diagnostic is not current-year operational validation. NBM, PPP and GEFS ingestion/calibration remain R&D, and proprietary current-year odds are withheld.
The national GeoNames index is used under CC BY 4.0 with attribution. Open-Meteo helped resolve the initial curated city records during research; live search does not depend on that service.