5. Nearest single radar for max resolution; ranked list for failover; no blending

Status

Accepted

Context

The tool must work for any CONUS location and show the highest-resolution data for it. Level 2 super-resolution (0.5° × 250 m) is the resolution ceiling, and a radar's data quality at a point degrades with distance: earth curvature puts the beam progressively higher above ground farther out, so a closer radar sees nearer the surface and resolves low-level detail better.

The question raised: should we combine multiple nearby radars — e.g. a proximity-weighted blend of every station that covers the location?

Decision

• Derive the active radar from a configured lat/lon against a bundled static NEXRAD site table (~160 sites, ICAO + lat/lon — fixed public data, no API).
• Use the single nearest covering radar for rendering. That gives the best data for the location.
• Compute and keep the full ranked list by distance, but use it only for failover — if the nearest site has no recent data (maintenance/outage), drop to the next.
• Do not blend multiple radars into a single rendered frame.
• If wide-area context is ever wanted, use MRMS (the national mosaic) at low zoom rather than rolling our own multi-radar composite. (Tracked in ROADMAP "Later".)

Consequences

• Highest available resolution at any location, with no compositing artifacts.
• Resilience to single-radar outages via the ranked list, cheaply.
• Works anywhere in CONUS from config alone.
• Wide-area views (when we add them) come from a purpose-built, already-reconciled product instead of a homegrown blend.

Alternatives considered

• Proximity-weighted multi-radar blend. Rejected: would most likely reduce quality at a point — beam-height mismatches, time skew between non-synchronized scans, and seams — and it destroys the crisp native polar geometry we render from. Correct multi-radar reconciliation is exactly the hard problem NOAA already solved and distributes for free as MRMS; re-implementing it badly is strictly worse than using it.