Add abandoned-mine features from USGS USMIN

[scottwmccoy]

Puts historic mining on the same map as the rain. Post-fire debris flows run out
of steep burned catchments, and in the Great Basin those catchments are full of
it — mine dumps and tailings are loose, often contaminated material sitting on
the channel network, exactly what a convective cell mobilises.

Why USMIN and not an AML hazard database

There is no public point-level abandoned-mine hazard database, by policy.
USGS FS 2025-3003 says the national
abandoned-mine-feature database being assembled under USMIN "will not publish
specific location information of any abandoned mine workings, and the detailed
national abandoned mine feature database will not be publicly available" —
the locations could be used to enter hazardous workings or vandalise historic
structures — and only aggregated derivatives are planned.

Nevada matches. The Division of Minerals' operational AML layers are real and
findable (NVPoints, NVSites, InternFieldDataCaptureLayers, AMLCracForms),
but every one answers 499 Token Required anonymously.

What is public is USMIN's compilation of mine symbols digitised from published
topo sheets, which reveals nothing the printed maps did not. National, uniform,
live; all 11 western states covered. Read it as a historical map compilation,
not a hazard inventory — topo_date runs 1950–1994, no hazard ranking, no
securing status, no confirmation a feature still exists.

Sources are a registry, not a hard-coded URL: NDOM is already entered with
its service URLs and a $STORMSCAPE_NDOM_TOKEN hook, so credentialed access is
--source ndom --token … rather than a code change.

What's here

• stormscape/mines.py — six feature groups over USMIN's 55-type vocabulary
  (exact map + prefix families, so a later Tailings - X still lands in waste);
  the kind filter pushed into the service query where expressible and re-applied
  locally regardless; equal-area density as both a vector grid and an
  mrms-style raster, cross-checked against each other in tests.
• CLI mines writes the layer, a per-group tally, a per-km² raster and a map.
  --mines overlays it on map, run, nexrad, zoom, burn, export.
• 75 offline tests (480 total, no network).

Three findings worth reviewing

Default geometry is both, not points. USMIN records mine waste almost
entirely as polygons — 14,815 against 413 points nationally, 778 against 4 in
Nevada — so a points-only waste query returns nearly nothing and looks exactly
like an AOI with no mining.

A latent paging bug in refdata. ArcGIS reports truncation at the top level
on a MapServer but only under properties on a hosted FeatureServer, so the
old check returned exactly 2,000 of 3,134 USMIN features with no error. The
existing layers are all MapServers and were unaffected — verified by paging
26,037 NHD flowlines correctly — so nothing already published changes. Separately,
an ArcGIS error arrives as HTTP 200 with an {"error": ...} body, which read as
an empty AOI; it now says so on stderr.

drape_i15 accepts i15=None (terrain + overlays, no colour bar). The first
mines figures draped the density raster and drew the features — blocky per-km²
cells under graduated symbols of the same counts, one quantity encoded twice. The
figure now carries the features alone; --density-map drapes the raster on request.

One bug got past the unit tests and was caught only by rendering: _cmd_mines
read args.vmax, which only burn's parser defines, so the command fetched,
wrote three files and then died at the last step. Fixed, plus a test that
AST-walks the command for args.<name> against the parsed namespace.

Note on the second commit

This branch also carries d7d24fd ("Warn that severity_colors bands raw dNBR"),
which was committed to burn-severity after PR #1 was cut and so never reached
main. It is unrelated to this work but would otherwise be stranded on a merged
branch — flagging it rather than dropping it silently.

🤖 Generated with Claude Code