BC wild salmon health · resilient → collapsing

Basin shapes are real BC Freshwater Atlas watershed groups, dissolved into flagship systems. Two are dropped: Nanaimo (no matching group at this scale) and Squamish (its Freshwater Atlas group is a coarse coastal grouping that swallows Metro Vancouver and the Sunshine Coast — every polygon-derived metric for it was contaminated, so honesty means dropping it rather than hand-editing a boundary). That leaves 12 clean basins. The health index is now fully time-aware: for every year 1980–2024 it measures each watershed's spawner level (3-year smoothed, from DFO NuSEDS, per species) against three reference baselines — the early record, the best historical 4-year window, and the long-run 75th percentile. Drag the timeline to move through years; switch the baseline to change the yardstick. The point of three is honesty about uncertainty: a basin shaded red under all three is in robust collapse; a basin the baselines disagree on gets a purple contested ring — the divergence is itself the information. Every stressor correlation below recomputes against the health you've selected, so you can watch a relationship form or dissolve across forty years. Mines and dams are point-in-polygon joins (BC permitted major mines — Mount Polley in Quesnel, Red Chris in the Stikine; and 2,490 registered dams). The new census layers: population and Indigenous share come from StatCan 2021 census subdivisions, assigned by centroid (not areal apportionment, which leaked city populations into wilderness basins), so each is approximate but leak-proof; agricultural footprint is the exact share of each basin inside BC's Agricultural Land Reserve. Indigenous share is context, not a stressor — the high-share basins are the remote, least-developed ones, and the correlation is near zero by design. The new climate and fire layers: summer temperature and its warming trend come from ERA5 reanalysis (Open-Meteo) sampled at each basin centroid over 1965–2024 — real, but a coarse single-point proxy, so read the warming rate as indicative not surveyed (Harrison-Lillooet's high value is centroid noise). Wildfire is the share of each basin burned by fires over 200 ha since 1985, from BC's historical fire perimeters, unioned to handle reburns — the Chilcotin at 26% and Thompson at 19% are the 2017–2021 megafires. Logging is the cumulative cutblock footprint as a share of basin area, summed from BC's 593,000 consolidated cutblocks by filtering each basin server-side in its native projection (no rasters touched) — the interior plateau (Nechako, Thompson, Adams) sits above 25%. Salmon farms are marine — they sit in ocean channels, not freshwater basins — so they're counted by migration-corridor proximity: DFO's 62 currently-licensed salmon net-pens, counted within 30 km of each system's documented smolt route to sea. The six interior-Fraser basins share one route (down the Fraser, north through the Discovery Islands and Johnstone Strait) and so share one farm count; the north-coast Skeena and Nass run farm-free. Still staged: estuary development and water quality (sparse, bespoke data).