Coastline & land cover — extracted from the satellite capture itself: the mesh shoreline, forest/muskeg classification, troll lane and point positions all follow the real coast. (Natural Earth's generalized polygon, used in earlier builds, turned out to be up to ~1 km off on the east and south shores — the imagery corrected it.)
Bathymetry & Graham Island relief — NASA SRTM30+ / Visible Earth global topography (~1-arc-minute), re-registered locally against the coastline by cross-correlation. Dixon Entrance really does drop past 200 m just offshore.
Forest cover — ESA GlobCover satellite classification. The island reads almost wall-to-wall evergreen forest, which matches the ground truth: old-growth Sitka spruce, hemlock and cedar.
Langara's interior heights — real: NRCan MRDEM-30 (the national 30 m elevation model, CDEM's successor), sampled across the island. The 30 m grid reads the summit at ~139 m where charts say 159 m (523 ft) — smoothing, not a different hill — and the surveyed ~42 m lighthouse bluff is kept as a floor where the DEM smears the shoreline. Hill-by-hill shape is now survey-derived.
Langara Point Lighthouse — true position (54.2554°N, 133.0594°W) and true light characteristic: one white flash every 5 seconds, 49 m focal height, est. 1913. Watch it after dark.
Trees — … instanced conifers scattered by the forest classification. Stylized, not a census.
Ground imagery — a real satellite capture of the island (Google Earth, supplied by the owner of this file for personal use), registered to the scene by anchoring the Langara Point lighthouse at its exact coordinates and scaling to the island's documented 33.0 km² area. Outside the capture's coverage, a procedurally synthesized ortho fills in.
Tide — official Canadian Hydrographic Service predictions for Langara Point (station 09964), Sat 13 Jun 2026: low 0.36 m at 06:43, high 3.76 m at 13:14, low 1.97 m at 18:21, building to a 4.9 m spring high overnight. Water level, currents and the scrub-bar curve all run off these real numbers (heights vs chart datum).
Salmon water — the marked troll lanes and points (Cohoe, Andrews, Langara Pt, McPherson, Boulder Bay, Henslung) are the island's documented fishing waters; the bite meter is a guide-lore model (moving water beats slack, west side on the flood, east side on the ebb, dawn and dusk over midday) — not a fish finder. Fish marks are symbolic: colored and sized by species (blue-silver Chinook largest, green-silver Coho, rosy Pink smallest — 2026 is an even-year pink run), with each zone's species mix taken from the same guide lore as the playbooks, not sonar.
Bite model — guide lore: zone weights and flood/ebb bias are hand-set from Langara fishing lore, not data. Log catches and run scripts/fit_model.py to replace these with weights fitted to your own logbook; when fitted data is present the Fish Now header reads "fitted (N=…)" instead of "lore".
Species mix — guide lore, not a DFO feed. We checked: DFO's machine-readable recreational catch data (open.canada.ca "North & Central Coast Recreational Catch") only resolves to Statistical Area 1 — the whole north/west Haida Gwaii region, not these individual points — and the series ends in 2009; the current in-season DFO feed is commercial-only. There is no public per-zone, current species breakdown to drive these percentages, so the chinook/coho/pink split per point stays labeled lore rather than dressed up as data.
No waterfalls. The real Langara has none — its streams are small and its hills top out at 159 m. Any render showing cascades is showing you somewhere else.