GardenMethodology · sources
Data Sources and Licenses
GardenTrack's growing advice comes from 1,023 university extension publications across 230 extension services, used as facts under citation. The frost tool runs on public-domain NOAA, NASA Daymet and USGS terrain data. This page lists every source, its license, what it powers — and the datasets we turned down on license grounds.
Two different bodies of data run this site. The growing advice comes from university extension services — thousands of individual publications, cited field by field. The frost and weather work comes from public climate and terrain datasets. They have different licenses and different rules, so they are listed separately below, along with the ones we decided we were not allowed to use. The method that sits on top of all of it is on the how we research page.
The plant data
Every field on a crop or variety page records the publication it came from — 3,728 sources across 730 organizations. They are not one kind of thing, and treating them as one would be the misleading way to present this:
| Kind | Sources |
|---|---|
| University extension services | 1,023 |
| Seed catalog listings | 2,368 |
| Other horticultural references | 280 |
| Wikipedia and Wikispecies | 21 |
| USDA and government records | 15 |
| Community and grower records | 14 |
| Peer-reviewed research | 6 |
| Patent records | 1 |
The split matters. Extension publications carry the growing advice — when to sow, how deep, how far apart, what a plant should be doing and when. Seed catalogs carry variety-level facts and nothing else, because a catalog is the primary record of what a cultivar is and a marketing document about how to grow it. Keeping those two jobs separate is a rule the data enforces, not a preference.
How we are allowed to use them
Almost all of it is under ordinary copyright:
| License | Sources |
|---|---|
| All rights reserved | 3,680 |
| CC BY-SA 4.0 | 21 |
| Public domain | 19 |
| CC BY 4.0 | 2 |
| Other or non-commercial | 6 |
So the posture is the one that copyright actually allows: facts, not text. A fact — that a tomato germinates in six to twelve days — cannot be copyrighted, and taking it with a citation and a link is legitimate. Copying the paragraph it appeared in is not. 3,718 of 3,728 sources are flagged facts-only in the registry, and the build refuses to publish a value whose source is missing or mislicensed.
It also means we send traffic back. Every cited publication is linked from the field that uses it, which is the least we owe the extension services this whole site is built on.
Climate, weather and terrain data
The frost tool and the app's alerts run on public data. Every one of these was license-checked before a byte was downloaded:
| Dataset | License | What it powers |
|---|---|---|
| NOAA NCEI U.S. Climate Normals 1991–2020 | US public domain | Official freeze probabilities for roughly 7,300 stations. The cross-check beside every frost answer, the fallback where our surface has no data, and the held-out ground truth for our accuracy numbers. |
| Daymet V4 R1 (NASA / ORNL DAAC) | Public domain, CC0-equivalent; citation requested and given | The base climate record behind the frost surface: 26 years of daily minimum temperatures on a 1 km grid. The open archive we bake from ends 30 December 2020, which is why the window stops there. |
| USGS 3DEP elevation | US public domain, no use restrictions | The terrain behind the ~250 m elevation correction, and the cold-air-drainage research. |
| Copernicus DEM GLO-30 | Free to use, attribution required | Fallback elevation at a small number of documented gaps in 3DEP coverage, mostly coastal. Because we derive values from it, the modified-data notice applies and is carried on the frost methodology page. |
| nClimGrid-Daily | US public domain | Cross-checking the frost surface. Its 1951-onward record is the long baseline for climate-shift work. |
| gridMET | CC0 | Cross-checking the frost surface. It also carries reference evapotranspiration, which is the honest engine behind a watering calculator — the reason "one inch a week" is the most repeated wrong number in gardening. |
| US Census Geocoder | US public domain; caching permitted | Turning the address or ZIP you type into coordinates. It is the only external service in the live path of the frost tool, and it never sees anything but the location you asked about. |
| National Weather Service API | US public domain, free, no key | The live forecast side of frost and heat alerts in the app. Climatology says the watch season is open; the forecast fires the actual warning. |
The full method these feed — how the surface is built, and how accurate it measured against 7,006 held-out stations — is on the frost dates methodology page.
What we refused to use
A license policy is only checkable if the refusals are published too. These were all available, useful, and turned down:
- Google Earth Engine. Its terms of service bar unpaid commercial use even over public-domain data. We bake from the original archives instead.
- CGIAR-CSI void-filled SRTM. Non-commercial license. Where SRTM data is needed we use the NASA releases.
- Google Geocoding API. Does not permit the persistent caching a local-first app needs.
- Public Nominatim. Its usage policy bars commercial apps that use it as a primary geocoder.
- PRISM free tiers. Commercial use is unstated rather than granted. Treated as reference-only.
Each of these cost real work to route around — the Earth Engine one in particular, since it is the convenient way to do almost everything we do with climate data. Using it would have been faster and invisible to you.
What we publish back
Taking this much public data and releasing none of it would be poor form. The first release is live: the frost-station normals extract — all 7,305 stations of NOAA's official 1991–2020 freeze probabilities decoded into a single documented CSV, with a plain-text column dictionary covering the traps that make the raw files hard to use, including the inverted spring percentiles and the censoring flags.
NOAA's underlying facts are US public domain. Our compilation is CC BY 4.0 with a copy-paste citation provided, and the download is free.
Next, in order: the per-station validation tables behind our published accuracy numbers, so the math can be checked independently, then a downsampled copy of the frost surface itself. Each gets linked here the day it is actually live — no placeholder links.
FAQ
Can I reuse GardenTrack’s plant data?
The underlying horticultural facts are not ours to license — they belong to the extension services that published them, and we use them as facts under citation. Our compilation work is a different matter, and the first open release is already out: the frost-station extract is free under CC BY 4.0. A wider open release of the plant database is planned.
Why are most sources “all rights reserved”?
Because university extension publications generally are. 3,680 of our 3,728 registered sources sit under all-rights-reserved terms, and we use them the only way that is legitimate: we take the facts, cite the publication, and link to it. We do not reproduce their text, and we do not redistribute their documents.
Do you use seed catalogs as a source?
Yes, for one job only. 2,368 catalog listings carry variety-level facts — days to maturity, color, growth habit — because the catalog is the primary record of what a cultivar actually is. They are not used for growing advice, which comes from extension publications.
What data does GardenTrack publish back?
The frost-station extract is live: 7,305 stations of NOAA freeze normals, decoded into one documented CSV with a plain-text column dictionary covering the traps. NOAA’s facts are public domain and our compilation is CC BY 4.0. The per-station validation tables and a downsampled copy of the frost surface follow, in that order.
Spotted a source used wrongly, or a license we have read incorrectly? Tell us — corrections go on the record, and the page changes.
The method that sits on top of all this data is on the how we research page, including the audit findings on our own work. The plant database is where the data actually lands.