Australian Alpine Weather Hub — source catalogue

Half-hourly temp/wind/precip at the alpine stations — the verification layer the obs threads run on.

P2-1 VERIFIED all five station identities on 2026-07-26 against BoM's own station metadata list IDCJMC0014 (climate/data/lists_by_element/stations.txt), then confirmed the join against each live feed's wmo and name fields. Two identifier namespaces are in play and both are preserved: the seven-character BoM station number in resorts.yaml (071075) and the five-digit WMO/feed id that addresses the JSON product (94915). The crosswalk lives in src/lib/obs/stations.ts with its evidence.

Official 7-day point forecasts and severe-weather/blizzard warnings for the alpine districts.

Anonymous FTP (IDN/IDV product XML) is the sanctioned machine-readable route — no scraping.

Dominant for short-range "what will it look like at the resort" checks across EC/GFS/ICON/AROME layers.

Official embed widget exists — the one third-party viewer we can legitimately embed. Lazy-load it (heavy iframe).

The season-long snow-depth tracking thread built on the Snowy Hydro readings.

Live precipitation echo over the Alps — the only genuinely real-time panel on the site.

Relevant radars — Captains Flat (IDR403), Yarrawonga (IDR493), Melbourne (IDR023). Alpine radar coverage is poor (beam blocking); label that honestly.

Daily depths and overnight new-snow numbers; marketing-inflected but universally checked.

THE verification dataset for Australian snow — weekly snow-course readings back to 1954 plus newer near-daily instrumentation.

Ground truth, and the emotional centre of the season for this community ("Spencers Creek 2026" thread). The historical series makes a current-vs-1954–present percentile chart the single best panel on the site — IF the audit clears reuse. Otherwise: link-out card, no exceptions.

Cloud structure over the SE Australian domain; fills radar's alpine blind spots.

Live visual confirmation of what the models promised.

The overnight-verification ritual — did the forecast snow actually land on the stake.

Pair each stake cam with last night's forecast snowfall for that resort — the highest-value "did it verify?" panel.

Resort forecasts, snow depths and cams — the benchmark the community scores its own daily comparisons against.

Obs, lightning and model summaries; commercial (DTN).

Alternative flow visualiser; good for explaining a cutoff low to non-specialists.

The community's reference model; run-to-run continuity of "EC" is the core currency of the threads.

Two explicitly separate point products are live. The 0.25° feed supplies the pressure-level profile used for snow-level and resort-elevation phase analysis. Native deterministic IFS supplies an approximately 9 km Australian surface precipitation and snowfall-water-equivalent reference. The native surface track is unphased and never substituted for, or silently merged into, the 0.25° profile. Community views IFS mostly through third-party chart sites; we render directly from Open-Meteo point data and ECMWF open GRIB fields.

Second reference model; the over-eager one, used to cross-check EC.

Fully open via NOMADS and AWS Open Data. No auth, no rate paywall.

Official 7-day point forecasts and severe-weather/blizzard warnings for the alpine districts.

Anonymous FTP (IDN/IDV product XML) is the sanctioned machine-readable route — no scraping.

The official synoptic chart the community screenshots inside ~4 days.

BoM material is CC-BY but the site ToS restricts automated retrieval — audit required (P0-6).

Dominant for short-range "what will it look like at the resort" checks across EC/GFS/ICON/AROME layers.

Official embed widget exists — the one third-party viewer we can legitimately embed. Lazy-load it (heavy iframe).

Free multi-model point API (including ECMWF ensembles, GEFS, GEPS, MOGREPS-G and ICON-EPS) — the data backbone for every own-rendered panel.

MANDATORY ATTRIBUTION: "Weather data by Open-Meteo.com", linked, on every panel carrying its data — not footer-only. CONSTRAINT: free tier is CC BY 4.0 non-commercial, ~10k calls/day; project is committed to staying non-commercial. Migration path if that changes is Open-Meteo's paid endpoint (customer-api.open-meteo.com + API key, same API shape) — NOT raw model ingestion. All calls sit behind the WeatherProvider interface (ADR 0001 §7).

Community-built resort snow-level/precip dashboard on Open-Meteo with transparent thickness-based methodology.

PRIOR ART, not a competitor. Our Phase 1 is the same idea done wider. Credit him explicitly on the snow-level panel and in /credits. Do not scrape.

The local model; leaned on hard inside 48–72 h for snow-level and precip detail, largely ignored at range.

Direct gridded ACCESS-G needs NCI registered access. Open-Meteo exposes ACCESS-G point forecasts — use that for Tier A meteograms; skip raw NCI.

Third-opinion cross-check, usually consulted via Windy.

Free ride — already in the Open-Meteo multi-model call.

Snow-level and MSLP+precip loops plus Kosciuszko atmosphere/sounding panels, built on EC/AIFS/GFS open data.

Free. Its snow-level product is a second cross-check for our own panel.

Optional global deterministic comparison from Météo-France, not a primary resort model.

Météo-France produces ARPEGE. This application obtains only point forecasts from Open-Meteo; it does not download, republish, or ingest direct Météo-France files. A live Falls Creek probe on 2026-08-03 returned usable ARPEGE World values for about 109 hours of a requested seven-day window, so valid-through is derived from returned values rather than the requested days.

Fifth-opinion global deterministic centre, carried as a comparison track rather than on the primary panel.

Free ride — added to the existing Open-Meteo multi-model call. Rendered only as a comparison track. The headline licence stays "unverified" until JMA's own GSM terms have been read.

Fourth-opinion global deterministic centre, carried as a comparison track rather than on the primary panel.

Free ride — added to the existing Open-Meteo multi-model call, so it costs no extra request. Rendered only as a comparison track; it is not one of the four models the primary precipitation split is built from. The headline licence stays "unverified" until someone reads the Met Office's own terms for the global deterministic 10 km product.

Australian multi-model aggregator graphing EC/GFS/ACCESS side by side per location; linked from the KB sticky.

Closest existing analogue to our Phase 1 comparison view. Worth a card and worth studying.

The "modez" (ECMWF) Australian panels, sub-seasonal anomaly charts and rapid-update animations.

The KB sticky's canonical snow-level links — EC snowline and GFS 0 °C isotherm height.

DUAL ROLE: also the ground-truth cross-check for our own snow-level panel. QA ticket P1-7 sanity-checks our thickness-derived snow level against this product for the same run.

Resort forecasts, snow depths and cams — the benchmark the community scores its own daily comparisons against.

MetService's commercial arm; supplies several resort forecasts.

Context card only — explains who actually writes some resort forecasts.

Obs, lightning and model summaries; commercial (DTN).

Official gridded AU forecast layer (7-day, includes snow) — the authoritative comparison for our own snow-level panel.

MetEye is a heavy JS app — link-out card. ADFD gridded access is registered-only.

Alternative flow visualiser; good for explaining a cutoff low to non-specialists.

An independent deterministic centre in the model-comparison view, publishing the full nine-level pressure profile at half the panel models' cadence.

15 km grid. Verified live 2026-08-07: usable horizon of ~229 h (~9.5 days) at Perisher coordinates, and all nine requested pressure levels (1000/950/925/900/850/800/700/600/500 hPa) non-null. Storage domain is cmc_gem_gdps_15km, not the more obvious cmc_gem_gdps — that domain returns HTTP 200 but is an abandoned copy, last run initialised 2026-05-26. It is a comparison track only, not one of the panel models: it updates 2×/day against their 4×/day.

Free official EC/AIFS/ENS charts incl. meteograms and anomaly products — the primary-source alternative to third-party chart sites.

Highest-value Tier B source: authoritative, CC-BY, replaces the ToS-murky Pivotal/Meteologix panels for synoptic context. Prefer over all of them.

Raw fields for own-rendered MSLP/precip/thickness maps over an Australian domain.

GRIB decoding in a Worker is impractical — use the standalone scheduled Python/ecCodes producer, emitting a canonical-grid JSON artifact for the static site.

Ensemble-at-a-glance widget; useful benchmark for our own multi-model meteogram.

Cleared for embedding ONLY via meteoblue Plug&Play widgets with a visible METEOBLUE reference close to the data. Scraping/framing the plain page is not covered.

The trans-Tasman flank of the same synoptic systems.

Raw GFS/GEFS GRIB with byte-range subsetting for the Australian domain.

Use the `.idx` byte-range filter — retrieve only the selected full-grid message ranges, then interpolate the Australian domain to the canonical output grid.

Resort-level multi-elevation forecasts; widely used outside the forum.

Cleared for embedding ONLY via their free official widget (snow-forecast.com/pages/configure_feed1) with attribution. Framing the raw resort page is explicitly banned — swap `url` to the widget endpoint before any Phase 5 embed.

Alternative multi-model visualiser with a good snow-depth layer.

Cleared for embedding ONLY via embed.ventusky.com with an active link back to ventusky.com. No API exists.

The community's reference model; run-to-run continuity of "EC" is the core currency of the threads.

Two explicitly separate point products are live. The 0.25° feed supplies the pressure-level profile used for snow-level and resort-elevation phase analysis. Native deterministic IFS supplies an approximately 9 km Australian surface precipitation and snowfall-water-equivalent reference. The native surface track is unphased and never substituted for, or silently merged into, the 0.25° profile. Community views IFS mostly through third-party chart sites; we render directly from Open-Meteo point data and ECMWF open GRIB fields.

Second reference model; the over-eager one, used to cross-check EC.

Fully open via NOMADS and AWS Open Data. No auth, no rate paywall.

Where the analysis this site serves actually happens — the primary audience and the reason the hub exists.

Persistent "discuss this on the forum" link in the header. This is a companion to the forum, not a replacement.

The official synoptic chart the community screenshots inside ~4 days.

BoM material is CC-BY but the site ToS restricts automated retrieval — audit required (P0-6).

Dominant for short-range "what will it look like at the resort" checks across EC/GFS/ICON/AROME layers.

Official embed widget exists — the one third-party viewer we can legitimately embed. Lazy-load it (heavy iframe).

Early-signal tiebreaker at long lead — "the EC robot is also beginning to show some interest".

Available in Open-Meteo as a deterministic model and as a separate 51-member AIFS 0.25° ensemble. Ensemble point values retain native-grid semantics, remain separate from the other ensemble systems, and are not described as snowfall at a resort elevation or at 1500 m.

Free multi-model point API (including ECMWF ensembles, GEFS, GEPS, MOGREPS-G and ICON-EPS) — the data backbone for every own-rendered panel.

MANDATORY ATTRIBUTION: "Weather data by Open-Meteo.com", linked, on every panel carrying its data — not footer-only. CONSTRAINT: free tier is CC BY 4.0 non-commercial, ~10k calls/day; project is committed to staying non-commercial. Migration path if that changes is Open-Meteo's paid endpoint (customer-api.open-meteo.com + API key, same API shape) — NOT raw model ingestion. All calls sit behind the WeatherProvider interface (ADR 0001 §7).

Community-built resort snow-level/precip dashboard on Open-Meteo with transparent thickness-based methodology.

PRIOR ART, not a competitor. Our Phase 1 is the same idea done wider. Credit him explicitly on the snow-level panel and in /credits. Do not scrape.

The local model; leaned on hard inside 48–72 h for snow-level and precip detail, largely ignored at range.

Direct gridded ACCESS-G needs NCI registered access. Open-Meteo exposes ACCESS-G point forecasts — use that for Tier A meteograms; skip raw NCI.

The probabilistic backbone; posters police the norm that beyond ~5 days you read spread, not runs.

Resort-point spread and independent liquid-equivalent precipitation probability are live via Open-Meteo's global ECMWF IFS 0.25° ensemble. Native approximately 9 km ensemble delivery listed for Europe is a geographic forecast domain, not an account or citizenship entitlement, and does not cover Australian resort points. Domain anomaly/map context remains P3-5; UI shows separate spread and probability, never a single member.

Fully open ensemble system — our best shot at own-rendered spread/probability panels.

Open-Meteo exposes GEFS members at points — enough for resort-level probability-of-snow plumes without touching GRIB.

Third-opinion cross-check, usually consulted via Windy.

Free ride — already in the Open-Meteo multi-model call.

Narrative synthesis of the model picture; well received in-thread.

Substack exposes a public RSS feed — headline + date + link only, never full text.

Levi Cowan's free site — source of most GFS/EPS/AIFS Australia-domain panels posted in-thread.

Highest-frequency chart source in the threads (20+ uploads). Ships as a deep-linked Tier C card until the audit says otherwise — never hotlink the .gif files.

Snow-level and MSLP+precip loops plus Kosciuszko atmosphere/sounding panels, built on EC/AIFS/GFS open data.

Free. Its snow-level product is a second cross-check for our own panel.

Optional global deterministic comparison from Météo-France, not a primary resort model.

Météo-France produces ARPEGE. This application obtains only point forecasts from Open-Meteo; it does not download, republish, or ingest direct Météo-France files. A live Falls Creek probe on 2026-08-03 returned usable ARPEGE World values for about 109 hours of a requested seven-day window, so valid-through is derived from returned values rather than the requested days.

Fifth-opinion global deterministic centre, carried as a comparison track rather than on the primary panel.

Free ride — added to the existing Open-Meteo multi-model call. Rendered only as a comparison track. The headline licence stays "unverified" until JMA's own GSM terms have been read.

NOAA's experimental AI global model, retrieved direct from NOMADS as GRIB2 and own-rendered rather than borrowed as someone else's chart.

Reclassified from "not worth own-rendering". That verdict was about the Tropical Tidbits chart route and was correct for it; the direct NOMADS GRIB2 route is a different proposition — public directory, .idx sidecars, byte-range retrieval, our own rendering, no third party's terms involved. Wired into produce_synoptic.py as a third, deterministic system alongside ECMWF ENS and GEFS (P3-5), and included in the twice-daily scheduled refresh's core-anomaly views.

NOAA's hybrid AI-and-physics ensemble, retrieved direct from NOMADS as GRIB2. Carried as the centre's own published ensemble MEAN, which is a real product — the HGEFS equivalent of the GEFS `geavg` this project already prefers over a mean it computes itself.

Publishes `ensstat` only — an `avg` and a `spr`, with no individual members served (verified 2026-08-30). No percentile range can therefore be computed from it, and none is: deriving a P10–P90 from a mean and a standard deviation would assume the members are normal about the mean, which precipitation, bounded at zero and skewed, is not. Carried as one line named as an ensemble mean, with nothing drawn around it. Publishes no snowfall field, and no snow level is derived — a wet-bulb level taken from the ensemble MEAN profile is not any member's snow level.

Fourth-opinion global deterministic centre, carried as a comparison track rather than on the primary panel.

Free ride — added to the existing Open-Meteo multi-model call, so it costs no extra request. Rendered only as a comparison track; it is not one of the four models the primary precipitation split is built from. The headline licence stays "unverified" until someone reads the Met Office's own terms for the global deterministic 10 km product.

Australian multi-model aggregator graphing EC/GFS/ACCESS side by side per location; linked from the KB sticky.

Closest existing analogue to our Phase 1 comparison view. Worth a card and worth studying.

NOAA's experimental AI ensemble, retrieved direct from NOMADS as GRIB2. Every one of the 31 members is fetched and retained, so percentiles are computed from the members themselves rather than inferred from the published mean and spread.

Publishes no snowfall field — the surface stream carries 10 m winds, 2 m temperature, MSLP and total precipitation only — and no member-level pressure profiles are retrieved, so this source can offer no snow depth and no elevation-specific phase probability. Precipitation distribution only, stated as such wherever it appears.

Source of the recurring ECMWF full-res forecast loops and 9-km EC snowfall panels.

Some products now behind Pivotal Plus. Card must not imply free access to paywalled loops.

The "modez" (ECMWF) Australian panels, sub-seasonal anomaly charts and rapid-update animations.

The KB sticky's canonical snow-level links — EC snowline and GFS 0 °C isotherm height.

DUAL ROLE: also the ground-truth cross-check for our own snow-level panel. QA ticket P1-7 sanity-checks our thickness-derived snow level against this product for the same run.

Expert forecaster commentary; season outlooks and weekly forecasts read by the same audience.

Official gridded AU forecast layer (7-day, includes snow) — the authoritative comparison for our own snow-level panel.

MetEye is a heavy JS app — link-out card. ADFD gridded access is registered-only.

Independent DWD global ensemble point spread for long-range alpine comparison.

DWD produces ICON-EPS; this application retrieves only Open-Meteo point-ensemble data and neither downloads nor republishes direct DWD GRIB files. A live alpine probe on 2026-08-03 returned a bare control plus member01–member39 (40 total), approximately 180 hours of usable data, and a null tail in a 15-day request at Open-Meteo's selected approximately 0.25° grid point. Actual coverage is always derived from returned member data; run metadata for this exact ensemble domain is not verified, so no run time is claimed.

An independent deterministic centre in the model-comparison view, publishing the full nine-level pressure profile at half the panel models' cadence.

15 km grid. Verified live 2026-08-07: usable horizon of ~229 h (~9.5 days) at Perisher coordinates, and all nine requested pressure levels (1000/950/925/900/850/800/700/600/500 hPa) non-null. Storage domain is cmc_gem_gdps_15km, not the more obvious cmc_gem_gdps — that domain returns HTTP 200 but is an abandoned copy, last run initialised 2026-05-26. It is a comparison track only, not one of the panel models: it updates 2×/day against their 4×/day.

An independent ensemble system in the long-range comparison — coarser grid, but a genuinely separate model family.

0.5° (~50 km) grid — the coarsest product rendered on either ensemble page, with adjacent resorts routinely sharing one grid cell. 21 members (control + 20). Verified live 2026-08-02: temperature_2m, precipitation, snowfall and snowfall_water_equivalent all present and non-null to hour 384 at Perisher coordinates. Deterministic GEM (gem_global) is out of scope — its storage domain has not initialised since 2026-05-26 and is not rendered anywhere.

Free official EC/AIFS/ENS charts incl. meteograms and anomaly products — the primary-source alternative to third-party chart sites.

Highest-value Tier B source: authoritative, CC-BY, replaces the ToS-murky Pivotal/Meteologix panels for synoptic context. Prefer over all of them.

Raw fields for own-rendered MSLP/precip/thickness maps over an Australian domain.

GRIB decoding in a Worker is impractical — use the standalone scheduled Python/ecCodes producer, emitting a canonical-grid JSON artifact for the static site.

Ensemble-at-a-glance widget; useful benchmark for our own multi-model meteogram.

Cleared for embedding ONLY via meteoblue Plug&Play widgets with a visible METEOBLUE reference close to the data. Scraping/framing the plain page is not covered.

The trans-Tasman flank of the same synoptic systems.

Raw GFS/GEFS GRIB with byte-range subsetting for the Australian domain.

Use the `.idx` byte-range filter — retrieve only the selected full-grid message ranges, then interpolate the Australian domain to the canonical output grid.

Resort-level multi-elevation forecasts; widely used outside the forum.

Cleared for embedding ONLY via their free official widget (snow-forecast.com/pages/configure_feed1) with attribution. Framing the raw resort page is explicitly banned — swap `url` to the widget endpoint before any Phase 5 embed.

Independent UK global ensemble point spread for long-range alpine comparison.

Retrieved only as Open-Meteo point-ensemble data under this project's existing non-commercial Open-Meteo route; the project neither downloads nor republishes direct Met Office files. A live point probe on 2026-08-03 returned control plus 17 perturbed members and approximately ten days of temperature, precipitation, snowfall and snowfall-water-equivalent data at the native approximately 20 km grid point. Actual coverage is always derived from the returned member data.

Alternative multi-model visualiser with a good snow-depth layer.

Cleared for embedding ONLY via embed.ventusky.com with an active link back to ventusky.com. No API exists.

The community's reference model; run-to-run continuity of "EC" is the core currency of the threads.

Two explicitly separate point products are live. The 0.25° feed supplies the pressure-level profile used for snow-level and resort-elevation phase analysis. Native deterministic IFS supplies an approximately 9 km Australian surface precipitation and snowfall-water-equivalent reference. The native surface track is unphased and never substituted for, or silently merged into, the 0.25° profile. Community views IFS mostly through third-party chart sites; we render directly from Open-Meteo point data and ECMWF open GRIB fields.

Second reference model; the over-eager one, used to cross-check EC.

Fully open via NOMADS and AWS Open Data. No auth, no rate paywall.

Where the analysis this site serves actually happens — the primary audience and the reason the hub exists.

Persistent "discuss this on the forum" link in the header. This is a companion to the forum, not a replacement.

Early-signal tiebreaker at long lead — "the EC robot is also beginning to show some interest".

Available in Open-Meteo as a deterministic model and as a separate 51-member AIFS 0.25° ensemble. Ensemble point values retain native-grid semantics, remain separate from the other ensemble systems, and are not described as snowfall at a resort elevation or at 1500 m.

Daily AAO index plot with GFS/ENS forecast tail — the number the threads quote for "negative SAM event".

Best Phase 4 target: plain ASCII index series + public domain = own-rendered chart with our own styling, no ToS question at all.

Free multi-model point API (including ECMWF ensembles, GEFS, GEPS, MOGREPS-G and ICON-EPS) — the data backbone for every own-rendered panel.

MANDATORY ATTRIBUTION: "Weather data by Open-Meteo.com", linked, on every panel carrying its data — not footer-only. CONSTRAINT: free tier is CC BY 4.0 non-commercial, ~10k calls/day; project is committed to staying non-commercial. Migration path if that changes is Open-Meteo's paid endpoint (customer-api.open-meteo.com + API key, same API shape) — NOT raw model ingestion. All calls sit behind the WeatherProvider interface (ADR 0001 §7).

The probabilistic backbone; posters police the norm that beyond ~5 days you read spread, not runs.

Resort-point spread and independent liquid-equivalent precipitation probability are live via Open-Meteo's global ECMWF IFS 0.25° ensemble. Native approximately 9 km ensemble delivery listed for Europe is a geographic forecast domain, not an account or citizenship entitlement, and does not cover Australian resort points. Domain anomaly/map context remains P3-5; UI shows separate spread and probability, never a single member.

Fully open ensemble system — our best shot at own-rendered spread/probability panels.

Open-Meteo exposes GEFS members at points — enough for resort-level probability-of-snow plumes without touching GRIB.

Official SAM commentary — the standard long-range framing device in-thread.

RMM phase diagram and the ECMWF MJO forecast page — long-range context alongside SAM.

The ECMWF-sourced plot on the CPC page is ECMWF's — attribute both.

SH polar vortex and SSW events materially drive SAM and Australian cold outbreaks; the forum runs a dedicated thread.

10 hPa zonal-mean wind + temperature is the card. Quiet 11 months a year — collapse it when nothing is happening.

Narrative synthesis of the model picture; well received in-thread.

Substack exposes a public RSS feed — headline + date + link only, never full text.

Levi Cowan's free site — source of most GFS/EPS/AIFS Australia-domain panels posted in-thread.

Highest-frequency chart source in the threads (20+ uploads). Ships as a deep-linked Tier C card until the audit says otherwise — never hotlink the .gif files.

NOAA's experimental AI global model, retrieved direct from NOMADS as GRIB2 and own-rendered rather than borrowed as someone else's chart.

Reclassified from "not worth own-rendering". That verdict was about the Tropical Tidbits chart route and was correct for it; the direct NOMADS GRIB2 route is a different proposition — public directory, .idx sidecars, byte-range retrieval, our own rendering, no third party's terms involved. Wired into produce_synoptic.py as a third, deterministic system alongside ECMWF ENS and GEFS (P3-5), and included in the twice-daily scheduled refresh's core-anomaly views.

NOAA's hybrid AI-and-physics ensemble, retrieved direct from NOMADS as GRIB2. Carried as the centre's own published ensemble MEAN, which is a real product — the HGEFS equivalent of the GEFS `geavg` this project already prefers over a mean it computes itself.

Publishes `ensstat` only — an `avg` and a `spr`, with no individual members served (verified 2026-08-30). No percentile range can therefore be computed from it, and none is: deriving a P10–P90 from a mean and a standard deviation would assume the members are normal about the mean, which precipitation, bounded at zero and skewed, is not. Carried as one line named as an ensemble mean, with nothing drawn around it. Publishes no snowfall field, and no snow level is derived — a wet-bulb level taken from the ensemble MEAN profile is not any member's snow level.

NOAA's experimental AI ensemble, retrieved direct from NOMADS as GRIB2. Every one of the 31 members is fetched and retained, so percentiles are computed from the members themselves rather than inferred from the published mean and spread.

Publishes no snowfall field — the surface stream carries 10 m winds, 2 m temperature, MSLP and total precipitation only — and no member-level pressure profiles are retrieved, so this source can offer no snow depth and no elevation-specific phase probability. Precipitation distribution only, stated as such wherever it appears.

SH extratropical circulation monitoring; cited directly in the KB sticky.

Source of the recurring ECMWF full-res forecast loops and 9-km EC snowfall panels.

Some products now behind Pivotal Plus. Card must not imply free access to paywalled loops.

The "modez" (ECMWF) Australian panels, sub-seasonal anomaly charts and rapid-update animations.

Classic SH 500 hPa spaghetti; posted occasionally for ensemble agreement at range.

Host is ambiguous in-thread (`spag_f192_shbg.gif`). Audit must identify the actual origin before this ships at all.

Expert forecaster commentary; season outlooks and weekly forecasts read by the same audience.

Dedicated AU snow forecaster; long-range outlooks and season tracking.

Independent DWD global ensemble point spread for long-range alpine comparison.

DWD produces ICON-EPS; this application retrieves only Open-Meteo point-ensemble data and neither downloads nor republishes direct DWD GRIB files. A live alpine probe on 2026-08-03 returned a bare control plus member01–member39 (40 total), approximately 180 hours of usable data, and a null tail in a 15-day request at Open-Meteo's selected approximately 0.25° grid point. Actual coverage is always derived from returned member data; run metadata for this exact ensemble domain is not verified, so no run time is claimed.

An independent ensemble system in the long-range comparison — coarser grid, but a genuinely separate model family.

0.5° (~50 km) grid — the coarsest product rendered on either ensemble page, with adjacent resorts routinely sharing one grid cell. 21 members (control + 20). Verified live 2026-08-02: temperature_2m, precipitation, snowfall and snowfall_water_equivalent all present and non-null to hour 384 at Perisher coordinates. Deterministic GEM (gem_global) is out of scope — its storage domain has not initialised since 2026-05-26 and is not rendered anywhere.

Free official EC/AIFS/ENS charts incl. meteograms and anomaly products — the primary-source alternative to third-party chart sites.

Highest-value Tier B source: authoritative, CC-BY, replaces the ToS-murky Pivotal/Meteologix panels for synoptic context. Prefer over all of them.

Raw fields for own-rendered MSLP/precip/thickness maps over an Australian domain.

GRIB decoding in a Worker is impractical — use the standalone scheduled Python/ecCodes producer, emitting a canonical-grid JSON artifact for the static site.

Raw GFS/GEFS GRIB with byte-range subsetting for the Australian domain.

Use the `.idx` byte-range filter — retrieve only the selected full-grid message ranges, then interpolate the Australian domain to the canonical output grid.

Independent UK global ensemble point spread for long-range alpine comparison.

Retrieved only as Open-Meteo point-ensemble data under this project's existing non-commercial Open-Meteo route; the project neither downloads nor republishes direct Met Office files. A live point probe on 2026-08-03 returned control plus 17 perturbed members and approximately ten days of temperature, precipitation, snowfall and snowfall-water-equivalent data at the native approximately 20 km grid point. Actual coverage is always derived from the returned member data.

Daily AAO index plot with GFS/ENS forecast tail — the number the threads quote for "negative SAM event".

Best Phase 4 target: plain ASCII index series + public domain = own-rendered chart with our own styling, no ToS question at all.

The season-long snow-depth tracking thread built on the Snowy Hydro readings.

The probabilistic backbone; posters police the norm that beyond ~5 days you read spread, not runs.

Resort-point spread and independent liquid-equivalent precipitation probability are live via Open-Meteo's global ECMWF IFS 0.25° ensemble. Native approximately 9 km ensemble delivery listed for Europe is a geographic forecast domain, not an account or citizenship entitlement, and does not cover Australian resort points. Domain anomaly/map context remains P3-5; UI shows separate spread and probability, never a single member.

Official SAM commentary — the standard long-range framing device in-thread.

Official Southern Hemisphere monitoring for ENSO, IOD and SAM context; it replaced the retired Climate Driver Update in December 2024.

THE verification dataset for Australian snow — weekly snow-course readings back to 1954 plus newer near-daily instrumentation.

Ground truth, and the emotional centre of the season for this community ("Spencers Creek 2026" thread). The historical series makes a current-vs-1954–present percentile chart the single best panel on the site — IF the audit clears reuse. Otherwise: link-out card, no exceptions.

Monthly ENSO status and Niño 3.4 outlook — background seasonal framing.

RMM phase diagram and the ECMWF MJO forecast page — long-range context alongside SAM.

The ECMWF-sourced plot on the CPC page is ECMWF's — attribute both.

SH polar vortex and SSW events materially drive SAM and Australian cold outbreaks; the forum runs a dedicated thread.

10 hPa zonal-mean wind + temperature is the card. Quiet 11 months a year — collapse it when nothing is happening.

Narrative synthesis of the model picture; well received in-thread.

Substack exposes a public RSS feed — headline + date + link only, never full text.

Levi Cowan's free site — source of most GFS/EPS/AIFS Australia-domain panels posted in-thread.

Highest-frequency chart source in the threads (20+ uploads). Ships as a deep-linked Tier C card until the audit says otherwise — never hotlink the .gif files.

Peer-reviewed anchors the forum itself cites — projected snowpack decline; SAM/rainfall relationships.

SH extratropical circulation monitoring; cited directly in the KB sticky.

Sea-surface-temperature anomalies underpinning ENSO/IOD discussion.

Publishes an annual Australian snow outlook; same ecosystem as AerisCast.

Second seasonal opinion alongside CFSv2.

Monthly/seasonal temp-anomaly panels posted when discussing winter prospects.

Underlying CFSv2 is US-gov public domain but the rendered panel is Tropical Tidbits' — audit gates the embed.

Expert forecaster commentary; season outlooks and weekly forecasts read by the same audience.

Home-grown NINO3.4/SOI/SAM/IOD prediction threads — colour, not authority, but proof of appetite for exactly this content.

Dedicated AU snow forecaster; long-range outlooks and season tracking.

Official ACCESS-S Southern Hemisphere outlook graphs for Relative Niño3.4, IOD, SAM and sea surface temperature scenarios.

Reanalysis composites and SH teleconnection tools for analog work.

The community's recommended intro course, including the long-/short-wave page.

1000–500 hPa thickness ↔ snow level — the exact rule our own snow-level panel encodes.

Link this from the snow-level panel's "how is this calculated?" disclosure. Methodology transparency is the whole point.

The community's own curriculum — the spine of our reference shelf.

Terminology reference.

Snowflake microphysics — why -12 to -18 °C makes the good powder.

Why 9-km EC ≠ 25-km GFS — the most common misreading in the threads.

Australia-specific snow lore wiki.

Deeper theory — posted in-thread to explain data assimilation.

The reading behind the community's AIFS scepticism-and-curiosity.

Peer-reviewed anchors the forum itself cites — projected snowpack decline; SAM/rainfall relationships.

The trans-Tasman flank of the same synoptic systems.

Reanalysis composites and SH teleconnection tools for analog work.