Three layers over the mosaic. tiger_states.parquet is the six-state
footprint for clipping the mosaic, which is not masked in the store.
nhd_water_bodies.parquet is the inland and near-shore water, for masking the CTrees
fold. hf435_wildlands.parquet is the one the viewers draw.
The 2022 Wildlands of New England: conserved land that by design lets natural processes run, with no active management. Drawn over the mosaic as boundaries.
Source: Wildlands of New England GIS Data 1900-2022 (HF435), Harvard Forest Data Archive. Foster, D., Johnson, E., Hall, B. (2023). License CC0. Dataset page: https://harvardforest1.fas.harvard.edu/exist/apps/datasets/showData.html?id=HF435
Credit the original when reusing this file: Foster D, Johnson E, Hall B. 2023. Wildlands of New England GIS Data 1900-2022. Harvard Forest Data Archive: HF435.
This is the strict end of the conservation funnel. The project screened over 600 candidate sites against three criteria (wildland intent, management for natural process, permanent protection) and kept 426. That compiles to 1,321,878 acres, 3.3% of New England. Median property is 302 acres, and the list is Baxter State Park, the White Mountain and Green Mountain wildernesses, Acadia, and the TNC reserves rather than town commons and ballfields.
By state: ME 85 properties / 722,838 ac, NH 72 / 233,166, VT 157 / 221,307, MA 57 / 115,933, CT 54 / 28,614, RI 1 / 21.
What changed from the source (see src/landsat_mosaic/wildlands.py):
426 polygons (231 MultiPolygon, 195 Polygon), 1.55 MB. Columns: PropID, PropName, FeeOwner, OwnerType1, OwnerType2, OwnerType3, YearOrig (-999 when unknown, 43 rows), W_Deed, W_Federal, W_State, W_MgmtPlan, W_Other, ProtDocs, ThrdPartYN, ThrdPartID, MgmtPlanTy, State, AcresGIS, geometry.
The snapshot is 2022. Land made Wildland since then is not included.
The six New England states from Census TIGER/Line 2024, for clipping the mosaic to New England in notebooks and viewers. The store itself is not masked: tiles are whole 4096 px squares and keep NY, Quebec, New Brunswick and the ocean. Clip with this layer where the six-state footprint is wanted.
Source: TIGER/Line Shapefiles 2024, State, U.S. Census Bureau. Public domain (U.S. government work). https://www2.census.gov/geo/tiger/TIGER2024/STATE/tl_2024_us_state.zip
Why raw TIGER and not the cartographic boundary file or Overture (compared 2026-09-07): raw TIGER runs to the 3 nautical mile limit, so bays, harbors and the near-shore band are inside and only open ocean is cut. The 500k cartographic file is shoreline-clipped but generalized by about 100 m, several pixels. Overture division areas for the states are triangulated at the coast and miss about 1,300 km2 of land; their land plus maritime union is the same 3 nmi envelope as TIGER but with a jagged internal seam.
What changed from the source (see src/landsat_mosaic/state_boundary.py):
7 rows, 1.1 MB. Columns: kind ("state" or "new_england"), geoid, stusps, name, aland, awater, geometry.
Not tracked in this repo: 18.7 MB. Build it once with
uv run python src/landsat_mosaic/water.py; the viewer says so in its
status line when the file is missing and runs without the mask until then.
Lakes, ponds, reservoirs, estuaries, bays, the near-shore ocean and the
wide-river polygons of the six states, from the USGS National Hydrography
Dataset High Resolution, for masking water out of the CTrees hexagon fold
in pyramid-ctrees-pair.py.
Why this file: CTrees carries a biomass value over open water (zero, or
near it, with an uncertainty), not a nodata flag. Without a mask the fold
draws hexagons over lakes, and a hexagon on the shore averages lake zeros
in with its forest and reads as a low-biomass halo round the water. With
it, water leaves the fold at the pixel level: no hexagon over open water,
and a shore hexagon is the mean of its land pixels. The viewer's WATER MASK
toggle (m) is this layer. It is never drawn.
Source: NHD High Resolution, per-HU4 GeoPackages, U.S. Geological Survey, staged at https://prd-tnm.s3.amazonaws.com/StagedProducts/Hydrography/NHD/HU4/GPKG/ (vintage December 2023). Public domain (U.S. government work). Twelve subregions: 0101 to 0110 (all of hydrologic region 01, New England), 0430 (Lake Champlain, the western half of Vermont) and 0202 (the Hudson, for the Batten Kill, Hoosic and Walloomsac corners of VT and MA).
What changed from the source (see src/landsat_mosaic/water.py):
area_km2 is the source's own figure.permanent_identifier across subregion overlaps.19,379 polygons (3 MultiPolygon), 46,604 km2, 18.7 MB. Of that, 17,080 LakePond (7,120 km2), 255 Reservoir (36 km2), 89 Estuary (3,926 km2, Long Island Sound the largest), 16 SeaOcean (29,931 km2, the band inside the TIGER 3 nmi limit), 4 BayInlet (4,327 km2, the Bay of Fundy) and 1,935 StreamRiver (1,265 km2). Columns: nhd_id, name (GNIS, null for most ponds), ftype, fcode, area_km2, layer ("waterbody" or "area"), huc4, geometry.
The 1 GB of source zips are cached in data/supplemental/nhd/ (gitignored) so a rebuild skips the download.