feat(geothermal): normalize OGC view temperatures to Celsius

[jirhiker]

Problem

The per-well geothermal views created in d1e2f3a4b5c6 publish legacy temperatures unconverted and label them with max("TempUnit"). Two consequences:

1. max() picks a unit lexically ('F' > 'C'), so a well holding both C and F readings is labelled 'F' while the values stay mixed.
2. min("Temp") / max("Temp") aggregate across those mixed units — 100 F sorts above 40 C, so the published extremes are meaningless.

Nothing downstream compensates: pygeoapi serves the view columns as-is.

Change

New migration f3a1c2b4d5e6 (revises 2d3c3a268652, single head) adds two helpers and rebuilds the two temperature views.

Helpers:

• public.nmw_temp_unit_code(text) -> text — canonicalizes a legacy unit to 'C' / 'F' / 'K', or NULL when unrecognized. NMW_GtTempDepths."TempUnit" is String(1) and NMW_GtBhtData."TempUnit" is String(5), so both single-letter and spelled-out forms are accepted.
• public.nmw_temp_to_c(double precision, text) -> double precision — converts using that code. Unrecognized unit returns NULL rather than assuming a default, so bad rows are visible instead of silently wrong.

ogc_geothermal_wells_bht and ogc_geothermal_wells_temperature_profile gain:

• min_bht_c / max_bht_c, min_temp_c / max_temp_c — aggregated over normalized values. These are the columns to chart.
• temp_unit — now the constant 'C', describing the *_c columns.
• temp_unit_source — the distinct source units actually present for the well (C, F, C,F, UNKNOWN, ...).
• temp_unit_mixed — flags wells mixing units.
• unconvertible_count — readings whose unit was not recognized.
• profile series objects gain temp_c and temp_unit_source.

Raw columns (min_bht/max_bht, min_temp/max_temp, the series temp key) are unchanged for compatibility. They remain mixed-unit.

Verification

Ran upgrade + downgrade against a local dev database inside a rolled-back transaction — both clean, nothing persisted. Spot checks: 212 F -> 100, 40 C -> 40, 273.15 K -> 0, bogus unit -> NULL, NULL unit -> NULL. Sample wells (all F in dev data) convert correctly: max_bht 138.9 -> 59.39.

One bug surfaced during that run and is fixed here: materialized view population runs with a restricted search_path, so inlining nmw_temp_to_c failed on its unqualified nested call (ERROR: function nmw_temp_unit_code(text) does not exist). All function references are schema-qualified. No SET search_path on the functions, which would disable inlining across the ~370k-row source.

Reviewer notes

• Breaking for anything parsing temp_unit — it now returns 'C' instead of the raw legacy code. Raw codes moved to temp_unit_source.
• Applying this drops and recreates the materialized view, repopulating WITH DATA; expect the usual matview build cost against NMW_GtTempDepths.
• Heat-flow units (HtFlowUnit, GradUnit, TCondUnit, Q_unit, Kpr_unit, Ka_unit) and depth units are not normalized — the summary and interval heat-flow views are untouched. Worth a follow-up.
• SPEC.md lists these views but gains no invariant for unit normalization here.

🤖 Generated with Claude Code