From e490f65359dbcaf634142ba877cda574981bd63a Mon Sep 17 00:00:00 2001 From: Graham McIntire Date: Sat, 1 Aug 2026 14:39:10 -0500 Subject: [PATCH] docs(about): update for HRDPS, Rust pipeline, perf work, remove ML model refs MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - Add Canadian HRDPS model coverage (to 60°N) - Document Rust pipeline (prop-grid-rs) and its performance characteristics - Add .sgrid/.pgrid binary format details - Note ~6s chain step and 23-band fused pass - Remove Nx/Axon/ExLA ML model references (scaffolding was removed) - Update roadmap: drop model training, add PSKR calibration, propagation alerts - Fix factor count (was 9, is 10) in roadmap item --- lib/microwaveprop_web/live/about_live.ex | 59 ++++++++++++++---------- 1 file changed, 35 insertions(+), 24 deletions(-) diff --git a/lib/microwaveprop_web/live/about_live.ex b/lib/microwaveprop_web/live/about_live.ex index a07fb07c..6118df44 100644 --- a/lib/microwaveprop_web/live/about_live.ex +++ b/lib/microwaveprop_web/live/about_live.ex @@ -56,10 +56,11 @@ defmodule MicrowavepropWeb.AboutLive do @@ -72,13 +73,16 @@ defmodule MicrowavepropWeb.AboutLive do
  1. The weather. - Hourly 3 km HRRR forecasts give surface fields and pressure-level - profiles. We derive refractivity profiles, minimum refractivity - gradient, ducting detection, boundary-layer depth, and precipitable - water for every grid cell. ASOS surface obs, 12-hourly upper-air - soundings, and gridded IEMRE reanalysis fill in the gaps. A - 10-factor composite score is written out for every 0.125° cell - across CONUS, for each hour of a 48-hour forecast. + Hourly 3 km NOAA HRRR forecasts give surface fields and pressure-level + profiles across CONUS, while the Canadian HRDPS model (0.125° + resolution, 4× daily) extends coverage north to 60°N. We derive + refractivity profiles, minimum refractivity gradient, ducting + detection, boundary-layer depth, and precipitable water for every + grid cell. ASOS surface obs, 12-hourly upper-air soundings, and + gridded IEMRE reanalysis fill in the gaps. A 10-factor composite + score is written for every 0.125° cell (~95k across CONUS) across + 23 microwave bands (902 MHz–241 GHz), for each hour of a 48-hour + forecast.
  2. The contacts. @@ -154,6 +158,16 @@ defmodule MicrowavepropWeb.AboutLive do (HRRR pulls, terrain, ASOS, soundings, IEMRE, solar indices, commercial links).
    3. +
  3. + A Rust pipeline (prop-grid-rs) handles the heavy lifting: + fetching GRIB2 data from NOAA S3, decoding via wgrib2, deriving + weather scalars, and scoring all 23 bands in a single fused pass + over ~95k cells. Writes dense binary artifacts (.pgrid profiles, + .sgrid weather scalars, and per-band .prop score files) to NFS + for sub-millisecond single-cell reads from Elixir. A full + forecast-hour chain step runs in ~6 seconds — ~3× faster than + the previous Elixir-only pipeline. +
  4. Leaflet + Canvas tile layers for the propagation heatmap — we render 0.125° cells at interactive frame rates. Tailwind v4 + @@ -167,16 +181,12 @@ defmodule MicrowavepropWeb.AboutLive do
  5. Data sources: NOAA HRRR (AWS S3, hourly analysis + 18 h + forecasts), Canadian HRDPS (MSC Datamart, 4× daily 48 h forecasts), Iowa Environmental Mesonet (ASOS and upper-air soundings), IEMRE gridded reanalysis, SRTM 90 m terrain, SNMP polling on seven commercial microwave links near DFW at 5-minute intervals, and NCEP NARR for pre-2014 contact enrichment.
    6. -
  6. - There's Nx/Axon/EXLA scaffolding for a small feed-forward model - (13 features → 64 → 32 → 1 sigmoid). Not trained yet — waiting - on enough calibrated data to not overfit. -
    7. @@ -191,20 +201,21 @@ defmodule MicrowavepropWeb.AboutLive do
  7. Weight calibration. - Fit the 9 factor weights against recorded distances / QSO counts - per band so the score reflects real propagation, not my - intuition. -
    8. -
  8. - Training the model. - Once there are enough clean (contact, conditions) pairs, train - the Axon model to augment or replace the hand-tuned scoring. + Re-fit the 10 factor weights against recorded distances / QSO + counts per band so the score reflects real propagation, not hand + tuning. The PSKR spot ingestion pipeline (sampling every 5 min) + is building the calibration dataset automatically.
  9. Better inputs. MRMS for precipitation at 24+ GHz where rain attenuation dominates, RTMA/URMA surface analysis blending, GOES total precipitable water.
    10. +
  10. + Expanded coverage. The Canadian HRDPS model gives + us southern Canada (to 60°N). Alaska, Hawaii, Mexico, and the + Caribbean need their own data sources and scoring pipelines. +