ML Integration:
- Load trained model at app startup, cache compiled predict fn in persistent_term
- Grid worker uses batched ML prediction (10K chunks) when model loaded,
falls back to algorithm scorer when not
- ML score replaces composite, algorithm factor scores preserved for detail view
- Fix process explosion: single EXLA call per chunk instead of per-grid-point
QSO Features:
- Callsign search (ILIKE on station1/station2) with trigram indexes
- Reciprocal QSO grouping (same pair, same band, same hour)
- Wider layout (max-w-7xl) for data table pages
- QSO Training Data link on map page
Infrastructure:
- Re-enable hourly propagation grid worker in dev
- Track ML model weights in git for Docker builds
- Add btree indexes on qsos (timestamp, band, distance_km)
- Remove nav icons from layout header
Add SFI, Kp max (solar), K-index, lifted index (sounding stability),
and ducting_detected (HRRR) as model features. Training now joins to
solar_indices and nearest sounding (within 6 hours) for both phases.
Model can learn solar/geomagnetic effects if they exist in the data.
- 15 features: add surface_refractivity and latitude
- Bigger network: 128→64→32 (3 hidden layers)
- Phase 1: pretrain on 500K stratified algorithm scores (all seasons/locations)
- Phase 2: fine-tune on 57K real QSO-HRRR matched data (percentile target)
- Lower LR (0.0003) for fine-tuning to preserve pretrained knowledge
- Model.train accepts :initial_state option for transfer learning
- 3 hidden layers instead of 2 for better feature interaction learning
- Target is within-band distance percentile (0-1) instead of raw
normalized distance — reduces noise from operator/equipment variation
Raw features had vastly different scales (pressure ~1013, sin/cos ~[-1,1])
causing gradient explosion. Normalize all atmospheric features to ~[0,1]
using known physical bounds. Add Polaris dep for optimizer.
Solar time (longitude/15) replaces fixed CDT/CST offset for time-of-day
scoring. Correlation analysis shows dramatic improvement at higher
frequencies: 24 GHz rho jumps from 0.056 (UTC) to 0.188 (solar), and
75 GHz corrects from spurious -0.39 to physically correct +0.24.
Score time-of-day per grid point using longitude/15 solar offset instead of
hardcoded CST/CDT. Add PWAT as 10th scoring factor. Refine pressure thresholds.
Update ML model and training pipeline to use local solar time.
- point_detail response includes forecast array (score per valid_time)
- SVG sparkline shows score trend across all forecast hours
- Trend indicator: Improving/Declining/Steady based on first vs last score
- Add covering index (band_mhz, lat, lon, valid_time) INCLUDE (score) for
point_forecast queries
- HrrrClient.hrrr_url accepts forecast_hour param (wrfsfcfHH.grib2)
- PropagationGridWorker fetches all 19 forecast hours per run
- Propagation.scores_at/3 queries scores at specific valid_time
- Propagation.available_valid_times/1 returns all forecast times for timeline
- Pruning keeps scores with valid_time >= now - 2h (forecast-aware)
- MapLive: select_time event, timeline data pushed to JS
- JS: forecast timeline bar at bottom of map with clickable hour buttons
- PubSub broadcast sends list of valid_times instead of single time
Previous thresholds (-500 to -60) were calibrated for radiosonde data.
HRRR profiles have coarser vertical resolution, with gradients clustering
between -40 and -130 N/km (median -70). Nearly all grid points were
falling through to the default score of 42, wasting the refractivity
factor. New thresholds (-200 to -40) spread across HRRR percentiles.
- Replace piecewise knife-edge loss with P.526-16 Eq. 31 single formula
- Fix diffraction parameter ν to use standard formula instead of ad-hoc approximation
- Implement Deygout 3-edge method for multiple obstacle diffraction
- Add dynamic k-factor from HRRR refractivity gradient (falls back to 4/3)
- Terrain worker now looks up nearest HRRR profile for atmospheric correction
- Update algo.md with P.526-16 methods and k-factor table
- Fix pre-existing map_live_test antenna height default (33 ft, not 8)
Deletes grid-aligned profiles (0.125 degree) older than 48 hours
while preserving QSO-linked profiles at arbitrary positions.
Called after each PropagationGridWorker run.
458K-record upsert held a connection for the entire transaction,
exceeding the 15s Postgrex timeout on prod. Set transaction timeout
to infinity and increase prod pool from 10 to 20.
Req default 15s receive_timeout too short for GRIB2 byte-range
downloads on prod. Also reduce HRRR queue concurrency from 20 to 5
to avoid NOAA rate limiting (was causing burst-then-stall pattern).
- Viewshed uses propagation score to set max range (ducting/NLOS distance)
- Terrain checked within radio horizon; clear paths get full atmospheric range
- Detail panel moves from map popup to sidebar for better visibility
- Panel shows instantly on click with loading state
- Dark opaque background for readability over map
- Factor analysis section explains each score with contextual descriptions
Replace generic range circles with actual LOS coverage polygon computed
from SRTM elevation data. Casts 180 rays (every 2 degrees) from the
clicked point, runs Fresnel/diffraction analysis on each, and renders
the reachable area as a Leaflet polygon.
- Viewshed module with haversine forward, terrain sweep, async compute
- Antenna height control (default 8 ft) in map panel
- LiveView start_async/handle_async for non-blocking computation
- Remove signal icon from band selector
- Add enqueue_for_qso/1 to directly enqueue weather/HRRR/terrain/IEMRE
jobs for a single user-submitted QSO (no cron, no bulk processing)
- Submit flow calls enqueue_for_qso instead of generic enqueue worker
- Add enrichment queues to prod config for on-demand processing
- Guard against HRRR fill values in store_hrrr_profiles (fixes badarith)
- Filter QSOs without pos2 in build_terrain_jobs
Replaces the one-shot startup Task with a GenServer that checks every
5 minutes whether propagation scores are older than 2 hours. If stale,
enqueues a PropagationGridWorker job (with dedup check to avoid double
queuing). Disabled in test env to avoid SQL Sandbox conflicts.
On app start, check if latest scores are older than 2 hours. If so,
immediately enqueue a PropagationGridWorker job. Covers app restarts,
deploys, and missed cron ticks.
Skip grid points where surface_temp_c or surface_dewpoint_c are
physically impossible (< -80°C or > 60°C). HRRR returns -273.15
(absolute zero) for ocean/missing points which caused division by
zero in absolute_humidity calculation.
- Wrap upsert_scores in Repo.transaction for all-or-nothing visibility
- Prune scores older than the 2 most recent valid_times after each upsert
- Add band-specific latest_valid_time/1 to eliminate N+1 query
- Add require Logger to Propagation module
- Scores now include factors and valid_time in the viewport query,
eliminating the server round-trip for popups
- Click shows two range circles: solid inner (typical range) and
dashed outer (max estimated range), colored by score tier
- Circles disappear when popup closes
- Band info pushed to client on band switch for accurate range estimates
Click anywhere on the propagation map to see a detailed popup with:
- Overall score and tier label with color
- Estimated range for the selected band (CW mode)
- All 9 scoring factors with visual bar charts, individual scores,
and weight percentages
- Grid point coordinates and data timestamp
Factors are displayed in weight order so users can immediately see
which atmospheric conditions are driving the prediction.
New AsosAdjustmentWorker runs every 10 minutes:
- Fetches latest ASOS observations from all ~2900 US stations via
IEM bulk currents API (parallel fetch across 51 state networks)
- For each grid point within 75km of a reporting station, re-scores
using fresh ASOS data (temp, dewpoint, wind, sky, pressure, precip)
with HRRR refractivity gradient from the last hourly computation
- Pushes updated scores to the map via PubSub
Also stores HRRR profiles in the database during grid computation
so the data persists for reference and ASOS blending.
The JS hook sends map bounds on load and on pan/zoom. The server
queries only scores within those bounds, dramatically reducing the
payload for band switches and map updates. At zoom 7 (DFW area)
this sends ~2k scores instead of ~95k.
Add a separate wgrib2-builder stage that compiles wgrib2 from source.
Docker layer caching means this only builds once — subsequent deploys
reuse the cached layer. The binary is copied into the final runtime
image. Also fixed wgrib2 path lookup to be runtime instead of
compile-time so it works in the Docker build pipeline.
Replace circle markers with a canvas tile layer that renders smooth,
flowing colored regions using bilinear interpolation between grid
points. Colors interpolate between tiers for gradients. ~95k grid
points at 0.125 degree resolution with wgrib2 extraction.
Add Wgrib2 module that shells out to wgrib2 binary for fast GRIB2
grid extraction using -lola (nearest-neighbor to regular lat-lon grid).
Falls back to pure-Elixir decoder if wgrib2 is not installed.
Also: parallel GRIB2 range downloads, merge adjacent byte ranges,
skip corrupt messages instead of failing, pressure fetch is optional.
Both extract_points and extract_grid now skip messages with missing
data sections instead of halting. Pressure product fetch is now
optional for both grid and single-point modes — if pressure data
has corrupt messages, surface-only profiles are still stored. Fixes
historical HRRR backfill failing on old data with incomplete files.
95k points at 0.125 degree resolution caused the GRIB2 extraction to
take too long. 0.5 degree (~55 km) resolution gives 6k points which
completes in under a minute. Can increase resolution later once the
extraction is optimized.
The hrrr/weather/iemre/terrain backfill queues (20+ concurrent jobs)
were competing for bandwidth with the propagation grid HRRR download.
Now pauses those queues before the grid fetch and resumes them after,
ensuring the hourly propagation update completes reliably.
QSO enrichment now groups all path points by HRRR hour and creates
one batch job per hour instead of one job per point. The batch job
downloads the GRIB2 data once and extracts all needed points from
the same binary. Legacy single-point jobs are still supported for
backward compatibility.
