1. NOAA historical freeze data
Our station records come from NOAA’s 1991–2020 annual/seasonal US Climate Normals archive. We retain the agricultural freeze variables, original station IDs, coordinates, source quality flags, reference years and source-file hashes. There are 7,285 agricultural station references in this snapshot; precipitation-only records are not treated as freeze records.
The threshold shown is 32°F (0°C) station air temperature. Visible frost on a leaf can occur under different conditions. Source station pages link directly to the underlying record.
NOAA product overview · Live variable catalogue · Calculation methodology, page 26
2. Date probabilities are conditional
This distinction matters. NOAA’s freeze-date probabilities are conditional on a cold season containing a freeze at the chosen threshold. A 10% date does not mean “10 out of every 100 years” at a mild station where freezing is uncommon.
- Last spring freeze: among freezing cold seasons, the probability of the last occurrence on that date or later.
- First fall freeze: among freezing cold seasons, the probability of the first occurrence on that date or earlier.
- Annual freeze occurrence: a separate model value indicating how often a cold season contains a freeze. We display it separately from the conditional dates.
The spring and fall directions are intentionally different. NOAA’s older variable PDF contains inconsistent wording for last-spring direction; we follow the live catalogue and the ordered source values. Earlier/later calendar-year offsets are preserved, including mild-climate dates that wrap into a neighboring year.
3. Quality gates and unusual climates
Ordinary seasonal reference calculations require suitable ordered source dates and standard (S) or representative (R) quality with at least 10 years in the source fields. There are 6,448 references meeting the current seasonal-model gate and 6,833 source-rich references meeting our data-page gate; those are different tests.
Infrequent freezing, no modeled freeze occurrence, year-round risk and limited-quality records keep their actual status. We do not substitute a colder station or manufacture dates. Reliable conditional date facts may remain visible even where routine yearly calendar calculations are disabled.
4. Census places and nearby stations
The location lookup covers 32,363 places in the 2026 Census Gazetteer: the 50 states, the District of Columbia and Puerto Rico. These are incorporated places and Census-designated places, not ZIP codes, addresses or neighborhood boundaries.
We calculate straight-line distance from the published Census internal point to agricultural NOAA stations, except for explicitly labeled, reviewed city reference overrides such as San Francisco’s mainland point. Those place pages retain the original Census coordinates and cite the replacement; distances, comparisons and forecast links use the same effective city point. Neither point represents an individual garden. The nearest agricultural station remains the nearest even if its climate or source quality does not support a calendar. Nearby alternatives are shown for comparison, not silently substituted.
Place reference points have no verified elevation in this input. We cannot validate the elevation difference to the station. Proximity labels are editorial descriptions, not scientific claims that two sites share a climate. Land and water areas describe Census geography, not garden suitability.
Original Census Gazetteer download. NOAA station coverage extends to additional US territories that are not part of this Census place file; those areas are shown as station-only coverage.
5. Regional crop evidence stays regional
The 48 crop guides bring together reviewed facts from Colorado State University Extension, NC State Extension, University of Florida IFAS, University of Minnesota Extension and University of Maryland Extension. A source can describe a particular region, crop grouping or establishment method. Those qualifiers stay attached to the facts.
Colorado frost-relative references, Central North Carolina calendar bands and North/Central/South Florida month windows are separate tools. Selecting Florida does not automatically choose its north, central or south region. Central North Carolina guidance is not applied to the whole state. Ambiguous source cells are withheld rather than guessed.
Maturity measured from direct sowing cannot be swapped for maturity measured from transplanting. Nursery lead time may be additional. A minimum germination temperature is not an optimum growing temperature or a safe transplant threshold.
6. The fall count-back calculator
For 18 reviewed crop scenarios, the tool subtracts the selected variety’s days to first harvest plus a 14-day slower-fall-growth allowance from a harvest date you choose. The allowance is a Maryland Extension rule of thumb, not a universal law.
The calculator does not automatically use the first 32°F freeze as a harvest deadline. Some crops tolerate freezing; other crops need warm growing conditions well before it. The arithmetic does not establish present planting viability, local suitability, germination success, heat accumulation or future weather.
7. Reuse, provenance and maintenance
NOAA states that nearly all its data and information are public domain, subject to identified exceptions and no implied endorsement. Federal Census data are used with their source identity and vintage retained. University extension prose and photographs are not assumed to have an open reuse license: our explanations are original, selected factual values are attributed, and the original sources are linked.
Raw inputs remain separate from transformed records. Repeatable download, parsing, cleaning, validation and refresh scripts preserve source hashes and detect changed or removed inputs. Checks cover duplicate identities, missing or sentinel values, conditional probability semantics, year wrap, quality flags, nearest-station joins and geographic coverage.
Planned maintenance is a monthly source-change check and a review when NOAA publishes corrections or a new normal period; horticultural references should be reviewed at least annually. Automatic refresh scheduling is not enabled. Source updates are reviewed before publication.
8. Pages, previews and corrections
Development previews remain noindex. The production sitemap separates source-rich canonical content from geographic reference lookups and unsupported combinations; appearing in a lookup does not mean a page should be indexed. We do not create a city–crop page just because two records can be combined.
There are no affiliate links or ad tags on this site. We use Google Analytics to understand site usage; see our privacy policy. Email hello@seasonplot.comwith corrections. See what to include.