Skip to content

Coverage & Confidence

Garbage in, garbage out. If we don't have coverage, we don't have a high-confidence model.

The Problem

IONIS learns from real observations. But the ionosphere doesn't care about our data collection — paths exist whether anyone is listening or not. The model can only be as good as the data that feeds it.

32,400 possible grids. The Maidenhead grid system has 18 longitude fields (A-R) × 18 latitude fields × 10 longitude squares × 10 latitude squares = 32,400 unique 4-character grid locators.

Not all grids are equal:

  • Dense coverage: FN31 (Northeast US), JN48 (Central Europe) — thousands of active stations, high observation counts
  • Sparse coverage: Remote land areas with few stations — limited observations
  • Permanent gaps: Mid-ocean grids, Antarctica, uninhabited regions — no stations possible (but propagation still happens)

Interpolation vs Observation

When IONIS has direct observations for a path (e.g., FN31→JN48 on 20m at 14 UTC), confidence is high. The model learned from actual measurements.

When IONIS has no observations for a path, it must interpolate from nearby grids, similar hours, adjacent bands. The further the interpolation, the lower the confidence should be.

Analogy: Weather forecasts are more accurate where there are weather stations. Predictions for the middle of the ocean rely on satellite data and models, not ground truth. Same principle here.

Coverage as a Confidence Metric

A coverage analysis would provide:

  1. Observation density map: For each of the 32,400 grids, count observations (as TX, as RX, or both). Visualize as a heatmap.

  2. Gap classification:

  3. Permanent gaps: Ocean, uninhabited — mark as "interpolation only"
  4. Sparse coverage: < N observations — mark as "low confidence"

  5. Dense coverage: > M observations — mark as "high confidence"

  6. Interpolation distance: For any query path, calculate the "distance" to the nearest observed signatures. Longer distance = lower confidence.

  7. Per-band breakdown: 10m coverage differs from 40m coverage. Report confidence per band.

Why This Matters

When IONIS predicts "20m open from FN31 to JN48 at 14 UTC with SNR -8 dB", a coverage metric could add:

  • Confidence: HIGH — 50,000 observations for this path/band/hour combination
  • Confidence: MEDIUM — 500 observations, reliable pattern
  • Confidence: LOW — 5 observations, use with caution
  • Confidence: INTERPOLATED — No direct observations, prediction based on nearby grids

This helps users understand when to trust the prediction and when to be skeptical.

Implementation Notes

The coverage analysis is a roadmap candidate, not yet implemented. When built:

  • Query wspr.signatures_v2_terrestrial for grid observation counts
  • Cross-reference with RBN and contest data for additional coverage
  • Generate static heatmaps for documentation
  • Consider adding confidence scores to IONIS inference output

Last updated: 2026-02-08