Methodology & sources

How planting windows are computed

Every planting window comes from one well-established, citeable method: count forward or back from the average frost date — the scheduling approach taught by U.S. Cooperative Extension services. The anchor dates are NOAA's median (p50) last spring frost (LSF) and first fall frost (FFF) at the 32°F threshold; the per-crop offsets are a documented synthesis of extension planting guides.

LSF, FFF        = NOAA 1991-2020 normals, 32 °F, p50 (median)
start indoors   = [ LSF − weeks_max×7 ,  LSF − weeks_min×7 ]
plant out       = [ LSF + weeks_lo×7  ,  LSF + weeks_hi×7 ]   (negative weeks = before LSF)
first harvest   = plant-out start + days_to_maturity range
fall planting   = count back from FFF by (days_to_maturity + frost-tolerance buffer)

The NOAA data and its probability grammar

Frost, freeze, growing-season, and growing-degree-day figures are read directly from NOAA's 1991–2020 Annual/Seasonal Climate Normals for the weather station nearest each location that carries a full 30-year record. Nothing is modeled or smoothed: the numbers on a location page are the station's own normals.

Freeze dates are published at three probability levels. For the last spring freeze, p10 is the roughly 90%-safe “plant after” date (the freeze has passed in about 9 years out of 10), p50 is the average/median, and p90 is the optimistic early date; the first fall freeze is the mirror image. Each location page shows the median plus the 90%-safe date so you can choose your risk tolerance.

Missing data is never invented: NOAA's -9999 sentinel and blank cells become a genuine “no value”, and locations that essentially never freeze are flagged frost-free rather than assigned a fabricated date.

Hardiness zones

Hardiness zones come from the USDA Plant Hardiness Zone Map 2023, matched to each location's ZIP. The map is distributed as static, USDA-derived per-ZIP JSON (via phzmapi.org; Last-Modified November 2023, the 2023 edition), so each location's zone is the USDA value for its ZIP — not an estimate. A handful of ZIPs with no PHZM entry fall back to the nearest ZIP that does; none are fabricated.

Because the zone is matched per ZIP rather than per metro area, a dense-urban page and its surrounding suburbs can legitimately read different half-zones — for example a low-desert urban core reading one half-zone warmer than its metro reference. That granularity is the point.

Frost-free locations

Some locations are effectively frost-free — for example Miami never reaches 32°F in the 1991–2020 record (so its freeze dates are null and its season is the full 365 days), and Phoenix has only a vanishing early-January freeze risk. For these, the engine does not fabricate frost-relative planting dates. It flags the location frost-free and notes that the practical limit is summer heat, not cold — warm-season crops can be grown across much of the year, and cool-season crops do best in the mild winter (roughly October–February).

One page per distinct climate (canonicalization)

The site publishes one page per unique (nearest-station, hardiness-zone) cluster, not one page per ZIP. Thousands of ZIPs share a nearest weather station and therefore share identical frost/GDD/season data; those collapse to a single canonical page, and every other ZIP in the cluster resolves to it. A station's territory is split by hardiness half-zone where a valley and a hillside genuinely differ, and a further safety-net merge collapses two adjacent stations whose normals are effectively identical. The result is that every indexed page carries genuinely unique data — the guardrail against thin, duplicated content.

Limits and accuracy

These are planning estimates, not guarantees. Frost dates are 30-year averages; any single year can freeze weeks earlier or later than the median. Planting windows use species-average offsets and days-to-maturity that vary by cultivar, soil, and weather. Microclimates matter — a low spot, a south wall, or an urban core can shift your real frost dates by a week or more from the nearest official station (in dense cities the nearest full-record station is often a warm downtown or airport site). Treat the dates as a starting calendar and adjust to what you observe in your own garden.

Update policy

NOAA's 1991–2020 normals are a static release (the next, 2001–2030, is expected around 2031), so there is no scheduled data refetch — only an annual reachability and checksum check. The planting-calendar dates are year-agnostic (MM/DD), so the yearly rollover is a re-render, not a re-fetch. The USDA PHZM revises on a multi-year cadence and is refreshed on announcement. Corrections reported by readers are checked against the primary source and applied promptly.

PRISM (not used here)

A future scale-up step could fill grid cells far from any weather station using PRISM 800m normals. PRISM is not public domain — Oregon State University asserts copyright and requires prominent attribution (name, URL, and access date) wherever it is used. The sample does not use PRISM (every location maps to a real station), so no attribution is due yet; if grid-cell fill is ever adopted, the required PRISM attribution ships with it on this page and in the footer.