The observation is the starting point.

One source, one station, one local day.

The sole weather source is the NOAA Aviation Weather Center data service, using METAR observations. This pilot never fills gaps with forecasts, modeled weather or another station. NOAA does not endorse YesterSky, and these recent observations are not certified records.

Each selected date starts at midnight in the selected place’s named time zone (older station-only links retain station-local days) and ends at the following midnight. The first instant is included; the ending instant belongs to the next date. Daylight-saving changes can produce 23- or 25-hour days. Phoenix uses America/Phoenix and remains on 24-hour local days.

The date range is yesterday through 30 local days ago. A date remains selectable even if it is partial or unavailable. The oldest dates can have incomplete coverage when the service first begins retaining records.

Samples tell you what was reported.

Temperature lows and highs are the lowest and highest retained samples, not official daily minimum and maximum temperatures. Wind speed and decoded gusts are reported fields. “Highest decoded gust” excludes peak-wind remarks in raw reports, which can describe stronger winds. Those remarks remain visible without an unproved claim about their occurrence time or local day.

The timeline uses actual observation times. Coverage shows occupied elapsed-hour intervals and the largest time gap, including the day’s boundaries. An occupied interval does not prove continuous observation or that every field was present. Missing temperature, wind or precipitation values remain missing.

Source temperatures are in degrees Celsius, wind speeds in knots and decoded precipitation amounts in inches. The display converts these to the selected units without changing the source record.

Rain evidence is not a daily rain total.

Hourly, three-hour, six-hour and 24-hour precipitation fields remain separate. A missing hourly field does not mean all other periods are missing. Trace, unavailable sensors and missing amounts have different meanings; none is silently converted to zero.

Reporting periods may overlap or cross the selected day’s boundaries. YesterSky does not add them together, invent interval boundaries or turn a reported 24-hour amount into a local midnight-to-midnight total. No local-day total is reconstructed.

Recognized present-weather precipitation codes are summarized by the searched place’s local overnight (midnight–6 a.m.), morning (6 a.m.–noon), afternoon (noon–6 p.m.) and evening (6 p.m.–midnight). Light, moderate and heavy describe reported intensity, never a sum of amounts or a count of reports. Freezing, mixed precipitation and showers keep their distinctions. Nearby or recent weather and accumulation-only evidence do not establish precipitation at the observation time. That is evidence of a reported condition at an instant, not continuous rain between reports or a measured amount at your property. See the AWC METAR decoding guide. These reports cannot establish soil moisture or watering needs.

Duplicates, corrections and retained records.

Reports stay tied to their station and observation timestamp. Exact duplicates collapse. For competing reports at the same station and time, a corrected report takes precedence; later receipt orders otherwise equally ranked revisions. An unresolved equal-rank conflict is excluded and disclosed. Different observation timestamps are never merged.

Weather retrieval starts only when someone requests a location and date. A first uncached lookup can request one local day through the bounded executor. A requested cached result is eligible for a background recheck after one hour for the two most recent completed local days, or after 24 hours for older days. Unused locations and dates cause no observation collection. Shared saved answers remain available when new retrieval is busy or the observation source is unavailable, provided hosting and storage reads remain healthy. New work is bounded by source pacing, safe concurrency and actual hosting capacity. A shortened upstream response does not erase fuller retained coverage, while corrections are still reconciled.

The displayed retrieval time is the actual snapshot time, not the time you opened the page. A retained record may contain reports preserved from an earlier retrieval. Provisional status remains visible because upstream reports can change. If a refresh fails, the service may show an older retained snapshot with its true age; if that snapshot cannot be retrieved, it shows an unavailable response.

Distance and responsible use.

New lookup distances use verified AWC station coordinates and GeoNames place/ZIP/full-postal reference points. Old station-only links retain their original Census reference distances. They are straight-line reference distances, not distance from your device or address. Candidates are ranked by great-circle distance and station ID from the versioned U.S. and Canadian METAR catalogs, within each supported country. Up to three candidates within 50 km are tried; the first with usable observations is retained. Empty results can advance to the next candidate, but source failures stop retrieval. Cached suitable observations can be served without new source work. We label the station nearby, not unconditionally nearest. Mountains, coasts and local rain can make conditions at your property different.

Use YesterSky to understand the available recent observations. It does not replace certified records, safety alerts, a forecast or a site-specific measurement. To flag a source or calculation concern, send the result link to engineering@sparkandspoolco.com.

Current-location matching.

When you choose Use current location, YesterSky matches your temporary device position against offline official boundaries. U.S. matching uses Census Bureau TIGER/Line 2026 place and county-subdivision geography, with official water polygons excluded. Canada uses municipal census subdivisions, with coast and water guards. A boundary match must have a trustworthy canonical place relationship. Otherwise a nearby reference is eligible only inside the same subdivision, within 5 km and clearly separated from other candidates. Ambiguous boundaries, water points and unsafe matches ask for manual entry. This selects a nearby weather reference; it does not establish weather at your exact location.

Adapted from Statistics Canada, Census Subdivision Boundary File, January 1, 2025, and the 2021 cartographic province, territory and Ontario census-subdivision land boundaries, under the Statistics Canada Open Licence. This does not constitute an endorsement by Statistics Canada of this product. Canadian water guards use Natural Resources Canada CanVec 50K hydrography (2019). Contains information licensed under the Open Government Licence – Canada. U.S. geometry is derived from the U.S. Census Bureau TIGER/Line 2026 release. No official endorsement is implied.

Location references and local time.

Location data is from GeoNames, downloaded September 16, 2026 for the U.S. and September 23, 2026 for Canada, under Creative Commons Attribution 4.0. YesterSky normalizes aliases, builds indexes, joins postal localities to place time zones and applies documented quality guards. The postal readme states CC BY 4.0 while its older linked label refers to BY 3.0; both source notices are retained with the project evidence.

The station catalog is from NOAA AWC. US DOT/BTS standard-zone boundaries cross-check U.S. place zones. Canadian places use verified Canadian IANA zone identifiers; current IANA rules handle daylight saving, half-hour offsets and permanent standard/daylight time. The selected place always determines the day. Uncertain points and generalized boundary cases may require a named locality or show an availability explanation. ZIP+4 uses the base five-digit code. Postal coordinates do not locate a property; military, unlocated and conflicting codes may not resolve. U.S. territories remain outside the 50-state/DC lookup boundary. Canadian city references include all 13 provinces and territories. Supported full Canadian postal codes retain their actual reference coordinates; they require consistent source assignments, address/shape-centroid accuracy and a uniquely matched nearby locality time zone. Codes failing those checks ask for city/province instead; no six-character code is silently reduced to an FSA. GeoNames reference accuracy does not establish property-level precision.

A station across a time-zone boundary still uses the searched place’s day. An explicit result link preserves the place reference, station, date and UTC interval. The public picker covers yesterday through 30 local days ago. Stored snapshots expire at local midnight 33 days after the selected date, so a bookmark does not promise permanent availability. Expired dates are never replaced with a different day.

Optional nearby water observations

Nearby water uses NOAA CO-OPS and NOAA NDBC, separately from weather. A qualifying summary appears immediately after the temperature graph. The named station measures one site; it need not represent the water surface, a beach or an entire water body.

The value is the arithmetic mean of accepted observations from one station and sensor during the same selected place-local day. Providers and stations are never combined. It is neither the midpoint of low/high readings nor an official certified daily average. Celsius values retain full precision until conversion; only display is rounded. Equal timestamps count once; conflicting readings suppress the day. Gaps are never filled with zeros or estimates.

CO-OPS uses six-minute observations with all three temperature flags zero. NDBC uses the standard meteorological WTMP stream in Celsius and UTC; missing MM values, nonnumeric values and implausible sentinels are excluded. NDBC applies automated source checks, but these files have no per-reading quality flags and are not final historical verification. We infer a stable 10, 20, 30 or 60 minute reporting cadence from all timestamps across at least a day of the rolling file, including missing water readings.

At least 90% of expected slots and elapsed-hour bins must be represented, with observations within 30 minutes of both boundaries and no retained gap over an hour. Day duration follows the selected place’s current IANA rules. A change over 3°C within 12 minutes suppresses the summary. These conservative display guards can omit otherwise useful records; they do not certify accuracy.

Selection ranks eligible stations within 40 kilometers by distance, then provider and station ID. At most two candidates are considered; insufficient data can advance to the next, while source failures stop new retrieval. CO-OPS retains compatible civil-day screening; NDBC UTC timestamps are restricted directly to the selected place’s bounds. Water-body labels require documented official site evidence. The CO-OPS registry was reviewed September 22 and NDBC September 23, 2026; both expire after 90 days without a reviewed update. Coverage includes proved Lake Michigan observations near Chicago, but is not nationwide or lake-wide. Canadian weather does not promise Canadian water coverage.

Results are shared by provider, station/sensor and exact interval, rechecked only on demand after six hours for the two most recent days or 24 hours for older dates, and expire 33 days after the selected date. NDBC files usually cover 45 days, but some partners retain less. Source trouble leaves weather intact; a sufficient saved water record keeps its true last-checked time.

NOAA and station operators supply provisional data without warranty or endorsement of this derived summary. Partner attribution remains visible. Reviewed IISG/Purdue Lake Michigan data permit free reuse; NOAA hosting alone does not establish every partner’s rights. See the CO-OPS disclaimer and NWS data disclaimer. This is historical station context, not swimming safety or water quality. Read the Nearby water guide.

When no qualifying nearby reading is available, a verified U.S. coastal location may show Regional coastal water from one compatible open-coast station within 120 km (about 75 miles). Offline NOAA shoreline and named-water geometry, together with Census locality and areal-water boundaries, establish coastal eligibility and reject protected-water or intervening-land matches. The reference point must be within 10 km of the verified coast; a containing Census locality touching that coast can establish frontage, otherwise the land connection must avoid other water. Postal points do not establish ZIP boundaries. Ambiguous or unverified geography is excluded. The station name, distance, selected local date and one arithmetic mean remain visible; this is a regional station observation, not an exact beach measurement. Regional rules do not extend Canadian or Great Lakes coverage.

Coastal sensor review: October 12, 2026. Additional NOAA-distributed thermometers are admitted only with verified station identity, compatible water geography and actual WTMP observations. Missing temperature evidence is not proof of a missing sensor. Data courtesy of CDIP, Scripps Institution of Oceanography, with station operators identified in each reading. Ocean Observatories Initiative observations are supported by the U.S. National Science Foundation; OOI data-use guidance. COMPS observations are supplied by the University of South Florida. Displayed averages are YesterSky calculations from unchanged source readings.

Coastal reference attribution: NOAA Office of Coast Survey, ENC Direct to GIS, Coastline and Sea Area / Named Water Area layers, retrieved October 11, 2026; U.S. Census Bureau TIGER/Line 2026. These derived geographic associations are for historical-data context, not navigation. NOAA data may be downloaded and used under its ENC User Agreement. Coverage is incomplete and selected-day observations are checked on demand.

Use the record for a practical question.

The recent weather guide explains the main limits in plain language. For a focused walkthrough, see how to check yesterday’s weather by ZIP code or city, review whether rain was reported yesterday, check sampled cold from last night, or interpret yesterday’s wind observations.

The tools share public observation and correction rules: one nearby station per result, a selected calendar day or completed overnight window, visible gaps and no invented values.

Overnight cold assessment.

The morning date ends a product-defined interval: 18:00 on the preceding evening to 06:00 on that morning, in the trusted local time zone of the selected place, with the end excluded. This is not a meteorological definition of nighttime or a sunset/sunrise calculation. Independent civil-time boundaries handle 11-, 12- and 13-hour daylight-saving intervals. Before 06:00 locally, the default is the previous completed night; at or after 06:00 it is the night ending this morning. Incomplete and future nights are rejected.

Accepted instantaneous temperature observations come from one suitable station throughout the interval, using the same normalization, correction precedence, duplicate handling and ambiguous-correction exclusion as full weather. Celsius values are classified before display rounding. Any accepted value at or below 0°C establishes an at-or-below-freezing reading. A negative assessment needs at least 80% of elapsed hourly bins to contain a temperature sample and no boundary or inter-sample gap greater than 90 minutes. Other report fields cannot fill a missing-temperature bin. Sparse samples above zero remain visible with partial coverage; insufficient coverage cannot establish a negative assessment.

This is the lowest sampled station air temperature in the window, never an official daily minimum, frost-free declaration, freeze duration or assurance about plants, pipes, ice or a property. Compatible retained observations are reused; otherwise only the validated bounded interval is requested from NOAA AWC on demand. No overnight collection schedule is added. The rule is checked against zero-degree thresholds, sparse readings, exactly 90-minute gaps, daylight-saving changes and corrected reports.