# Free-source register

No paid account, API subscription, purchase, or negotiated license was used.

| Source | Use in experiment | Reuse and caveats |
| --- | --- | --- |
| USA-NPN Nature's Notebook | Ground observations, 2022–2025; normalized cell-day stages | CC BY 4.0; retain USA-NPN and contributor attribution, plus NEON attribution where relevant. Data terms: https://www.usanpn.org/about/terms . Not permission to reuse unrelated site photographs. |
| Daymet V4 R1, ORNL DAAC | Daily temperature/rainfall and elevation at reporting-cell centers | Free historical scientific data; cite Thornton et al. (2022), DOI https://doi.org/10.3334/ORNLDAAC/2129 . NASA open-data guidance: https://www.earthdata.nasa.gov/engage/open-data-services-software/data-use-policy . Exact service citation retained in every archived CSV. Annual retrospective product, not a live operational feed. |
| USA-NPN MODIS land-surface phenology baseline | Existing median greendown feature | Historical vegetation transition, not observed peak leaf color. Existing dataset and attribution preserved. https://www.usanpn.org/data/maps/land_surface_phenology |
| Existing forest/land-cover masks | Current production context only | The experimental point-coordinate metrics do not reproduce land-mask and bilinear-grid rendering. Do not claim validation of every map pixel. |

## Investigated, not incorporated as forecast truth

- NOAA National Blend of Models: free operational weather guidance; https://vlab.noaa.gov/web/mdl/nbm . Needed for a later real-time model, but future-issue data must never enter historical replay. No newly enabled service or alerts.
- PhenoCam V3: independent canopy color/greenness time series and camera imagery; https://daac.ornl.gov/VEGETATION/guides/Phenocam_Images_V3.html . Could strengthen validation, but greenness transition and red-channel peaks do not automatically mean the app's peak leaf-color event. Published V3 reference series ends in 2023; data access and per-product use terms must be established before incorporation. Not counted as a source used by this experiment.

## Calendar and location details

Daymet includes February 29 but drops December 31 in leap years, so ordinal days must map to the actual Gregorian day within that year. Verified against https://daac.ornl.gov/DAYMET/guides/Daymet_Daily_V4.html . Missing/sentinel or physically invalid weather stays unknown. Feature windows include 30 completed days strictly before the simulated issue date.

NPN inputs are quarter-degree aggregated cells, not exact individual-tree coordinates. A 1-km weather product sampled at their center does not confer 1-km validation accuracy. Original submission timestamps are unavailable in these normalized archives; all historical comparisons are retrospective simulations, not proof of publication-time availability.


## Annual coverage experiment additions

- **NOAA PSL NCEP/NCAR 2m air-temperature normals, 1991–2020:** downloaded, hash verified and converted from Kelvin to Celsius by `decode-climate.py`. Source and hash in `climate/manifest.json`. [Dataset catalog](https://www.psl.noaa.gov/thredds/catalog/Datasets/ncep.reanalysis/Monthlies/surface_gauss/catalog.html). [Public-domain use and attribution](https://psl.noaa.gov/data/obs/download/Citation.html). This coarse historical normal is neither live weather nor a foliage observation. Attribution: Data provided by the NOAA Physical Sciences Laboratory, Boulder, Colorado, USA, from https://psl.noaa.gov/.
- **NALCMS / LANDFIRE Hawaii:** existing downloaded vegetation reused with full class codes for annual research grid. Historical land cover remains distinct from seasonal inference; unresolved cells explicitly retained.
- **USA-NPN full-year leaves observations:** 47,132 cell-days collected for 2023–2025 separately from autumn archives; CC BY4 attribution retained. Measure is site-balanced fraction of observed deciduous broadleaf individuals with leaves present, not canopy fullness or scenic color.
- **NOAA STAR blended vegetation health smoothed NDVI:** 56 historical files downloaded, validated and incorporated into the seasonal model. [Access documentation](https://www.star.nesdis.noaa.gov/smcd/emb/vci/VH/vh_ftp.php). STAR labels its display/download products experimental; operational 1km product is distributed by NOAA CLASS. Coverage and data latency must be checked before use.


## Completed satellite archive

NOAA STAR blended VIIRS smoothed NDVI, NetCDF historical samples 2021–2024. Per-file metadata validates scale0.001, fill-999, valid[-0.1,1.0], and product year/week. [NCEI datasetC00939 metadata and CC0 use terms](https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc%3AC00939). [NOAA CLASS latency documentation](https://www.aev.class.noaa.gov/saa/products/search?datatype_family=JPSS_NGRN) documents an eight-week delay. Future observations enter the smoothing window: these files may be historical training/validation targets, never live as-of predictors at their target week. Coverage stops75.024N; northern Canada needs a separate explicitly modeled fallback. NDVI is greenness, not visible fall color.

## IMS seasonal snow experiment (2026-09-26)
U.S. National Ice Center (2008), IMS Daily Northern Hemisphere Snow and Ice Analysis, G02156 Version 1, NSIDC. https://doi.org/10.7265/N52R3PMC . Official archive: https://noaadata.apps.nsidc.org/NOAA/G02156/netcdf/4km/ . Embedded license: “These data may be redistributed and used without restriction.” Citation retained. 168 daily 4 km files (2021–2024; 14 seasonal anchors, ±3 days), hashes and valid times in snow/manifest.json. Train 2021–2023; test 2024. Archive is historical, not an operational real-time feed. Unknown/water pixels remain unknown. Three-year sample frequencies are not calibrated forecast probabilities.

Map colors are illustrative historical seasonal estimates. Seasonal greenness/snow reference: 2021–2023. Autumn training: 2022–2024. Not a live weather forecast.