← Research

Forecast model feed reference

Every builder supplies nine surface fields, five fields on each pressure level, and model terrain elevation to the shared derivation. Access, resolution, cadence, horizon, retention, and field semantics vary by model.

Measurements use live service responses checked through 2026-08-08. When provider documentation and live capabilities disagree, use the live response and update the verification date.

HRDPS continental 2.5 km — supported

Environment and Climate Change Canada's High Resolution Deterministic Prediction System: 2.5 km grid over nearly all of Canada, 48 hourly steps, four runs a day (00/06/12/18Z).

The pipeline reads it through GeoMet's WCS endpoint — not the Datamart. Every request asks for a tiny GeoTIFF crop around the catalogued launches at the model's 0.0225° spacing, and each crop comes back at roughly 2 KB. Measured whole-build transfer costs are in How the static forecast pipeline works.

The trade-off is request count: WCS has no bulk mode, so the build is many small HTTP calls. The client keeps concurrency at five, retries 429/5xx with jittered backoff, honours Retry-After, and reports its request count in the published manifest so the cost stays visible.

HRDPS West 1 km — supported, experimental

ECCC also runs an experimental 1 km HRDPS over BC and western Alberta, published only as whole-domain GRIB2 on the alpha Datamart — no WCS, no subsetting. The grid is a rotated lat-lon, 1330 × 1180 points; ecCodes handles the rotation so the sampler only ever speaks geographic coordinates. Runs are 00Z and 12Z only, files are retained for roughly 24 hours, and the feed goes dark occasionally — the workflow treats a 1 km failure as a warning, never a blocked 2.5 km publish.

The builder downloads each whole-domain file before sampling the catalogue. One launch elevation is 1,485 m; model terrain elevation is 1,311 m at 1 km and 1,072 m at 2.5 km. The 1 km grid reduces the terrain error from 413 m to 174 m. That 239 m shift propagates into cloud base, boundary-layer top, and which pressure levels lie above model terrain.

RDPS 10 km — supported

The Regional Deterministic Prediction System: 10 km, 84 hourly steps, four runs a day. It extends the windgram from HRDPS's two days to three and a half.

Verified facts:

• Fully available on GeoMet WCS under the new-style RDPS_10km_* layer names (the older RDPS.ETA_* convention is on its way out).
• All nine windgram pressure levels (925–600 hPa) exist, and both sensible and latent heat fluxes are published — the full derivation runs with no substitutions.
• WCS sampling requirement: GeoMet stretches whatever resolution you request to fit the subset bounding box, so a native-spacing request over a small box lands the grid cells in the wrong places and returns physically plausible wrong values. Sample finely — 0.0225° reproduced GetFeatureInfo exactly at every site for both RDPS and GDPS — and verify crop values against GetFeatureInfo at a few points. See Seven forecast-data failures that passed parsing.

GDPS 15 km — supported

The Global Deterministic Prediction System: 15 km, 00Z and 12Z, through 240 hours. ECCC's current system description (checked 2026-08-08) identifies it as a coupled atmosphere, ocean, and sea-ice forecast whose large-scale temperature and wind are spectrally nudged toward its GEML data-driven weather model. That is a material change in lineage even though the open-data contract remains familiar. A ten-day windgram still has published sensible and latent heat fluxes rather than proxy surface heating.

Verified facts:

• The old GDPS.ETA_* GeoMet layer names are unavailable; current layers use the GDPS_15km_* prefix.
• All nine pressure levels and both heat fluxes are available — but six of the nine levels (900, 875, 800, 750, 650, 600 hPa) thin to 6-hourly after hour 168; only 925, 850, 700 and the surface fields stay 3-hourly to the end. The builder tolerates a level being absent at a valid time, so day 8–10 columns ride on the three levels that remain.
• Precipitation must be differenced. No fixed-window precipitation layer spans the full horizon; the reliable quantity is the run-total accumulation, so per-step precipitation is accum(h) − accum(h_prev) between consecutive published steps.

HRRR 3 km — supported

NOAA's High-Resolution Rapid Refresh: 3 km over the continental United States, hourly runs, with the synoptic runs (00/06/12/18Z) extending to 48 hours, published to public cloud buckets roughly T+107 minutes after the reference time.

Verified facts:

• Domain geometry: The oft-quoted 47.8°N northern limit is the eastern corner of the Lambert conformal grid. At −117.7°W the boundary is at 51.24°N — the southern-BC founding sites sit about 215 km inside the domain. Check the corner nearest your longitude before assuming HRRR excludes you.
• All 54 records the windgram needs live in a single wrfprs file per hour, and its .idx sidecar lets a client fetch exactly those records by byte range. See How the static forecast pipeline works.
• Heat fluxes are instantaneous values.
• Winds are grid-relative. U/V components on the native Lambert projection point along grid axes, not geographic east/north. Skip the rotation and wind directions bias by 10–15° at these latitudes.

GFS 0.25° — supported

NOAA's Global Forecast System at 0.25°: the long-horizon leg, out to 384 hours. Since the 2021 upgrades the output is 3-hourly all the way to f384 — the old 12-hourly tail beyond day 10 is gone.

Verified facts:

• 875 hPa does not exist in the pgrb2.0p25 files. The windgram's nine levels become eight, or 875 is interpolated; either way, don't loop over the HRDPS level list and expect nine hits.

• Pressure-level dew point isn't published; dew point depression must be computed from temperature and relative humidity per level.

• Flux time semantics: SHTFL/LHTFL exist only as averages over a growing window that resets every 6 hours: f001 averages hour 0–1, f002 averages 0–2, … f006 averages 0–6, then f007 starts again. Difference the window-hours to recover the mean flux over any sub-interval:

  mean(h1→h2) = ((h2−h0)·A(h2) − (h1−h0)·A(h1)) / (h2−h1)
  

  where h0 is the window start and A(h) the published average at hour h. The builder reconstructs each interval before deriving w*.

REPS 10 km ensemble — supported

ECCC's Regional Ensemble Prediction System: 21 members (a control plus 20 perturbed), rotated lat-lon at 0.09° (~10 km), four runs a day (00/06/12/18Z), 3-hourly to 72 hours — no hourly output. Its output here is not a windgram: every member is derived as its own atmosphere and the published file carries the percentile spread of the derived quantities. The first published run is 2026-08-08T00:00:00Z. Read the product semantics in What ensemble spread can—and cannot—tell you.

Verified facts (live, 2026-08-08):

• Both surface heat fluxes exist per member. GeoMet WCS publishes REPS.MEM.ETA_FC.NN / ETA_FV.NN (instantaneous W/m²) and on Datamart as SHTFL_SFC / LHTFL_SFC all-members GRIB2. A test crop of one member's sensible flux at 3 pm MDT returned 45–279 W/m² — real afternoon convection, per member.
• GeoMet serves raw REPS members over WCS through the same ~2 KB crops and SUBSET=x/y + SUBSETTINGCRS=EPSG:4326 mechanics as the deterministic builds, including the same silent zero-fill failure if you subset with the advertised long/lat axis names. Member numbering is off by one across transports: WCS .01 is the control that Datamart GRIB encodes as perturbationNumber 0. The builder pins the mapping in one function, with a test.
• The one WCS gap is wind. No per-member U/V or direction exists on any level, so winds come from Datamart's all-members files. REPS's rotated grid makes the components grid-relative; skipping the rotation produces a ~24° direction error at these latitudes. See Seven forecast-data failures that passed parsing.
• Three of the nine windgram levels exist (925/850/700), and 925 typically sits below REPS model terrain elevation near 1,600 m at one catalogued launch against 925 hPa's ~780 m. Members are fed a 925/850/700/500 column: 500 hPa is not a display level, but it lifts the parcel search's ceiling from ~3,150 m to ~5,700 m so strong days differentiate instead of clamping at the column top. See What ensemble spread can—and cannot—tell you.
• No dew point in any form — derive it from temperature and relative humidity. Precipitation is run-total accumulation, differenced at 3 h (the GDPS pattern). Fluxes and accumulations are absent at hour 000.
• The hybrid build moves ~1.5 GiB per run for the display schedule: kilobytes of WCS crops plus ~10–18 MB per un-croppable Datamart wind file.
• Datamart retention is ~30 days (dd.weather.gc.ca), ~52 on the hpfx.collab.science.gc.ca mirror; GeoMet keeps only ~2 days of reference times, which permits 15-minute polling but not backfill. The legacy /ensemble/reps/ tree is dead (404); everything lives under the date-based YYYYMMDD/WXO-DD/ensemble/reps/10km/grib2/HH/hhh/ tree.

GEPS 0.5° ensemble — evaluated, not adopted

ECCC's Global Ensemble Prediction System: 21 members, regular 0.5° global grid, 00Z and 12Z, 3-hourly to 192 hours then 6-hourly to 384 — sixteen days — with the Thursday 00Z run extending to 936 hours. Verified live on 2026-08-07 and not yet adopted:

• Per-member heat fluxes exist to the full horizon. Unlike REPS's instantaneous values, GEPS fluxes are interval accumulations that must be differenced back to W/m², and the accumulation convention beyond the first step is unverified.
• Datamart only. GeoMet has no per-member GEPS layers — statistical products alone, with no fluxes among them. The fetch unit is the all-members GRIB2 file, 102 MB per forecast step, **9.6 GB per run**. This transfer cost is the reason the feed was not adopted.
• Three of the nine windgram pressure levels exist (925/850/700). Dew point is absent, and run-total precipitation requires 6-hourly differencing beyond hour 192.
• Verify every download against Content-Length. One GEPS fetch returned another model's bytes, correct on retry; the Datamart client now length-checks every download because of it. See Seven forecast-data failures that passed parsing.
• Retention matches REPS; the legacy /ensemble/geps/ tree is likewise dead.

ECMWF IFS open data — evaluated, not adopted

ECMWF's open-data IFS offers a 15-day horizon (360 h at 00/12Z since cycle 50r1) under CC-BY-4.0. Two missing inputs ruled it out:

• No surface heat fluxes are published. w* has no direct input. The physics would need a radiation-proxy flux: net solar radiation partitioned by an assumed Bowen ratio. That method produces a less defensible forecast than models that publish surface fluxes and requires a different product label.
• Only four of our nine pressure levels exist in the open-data set (925/850/700/600) — and at the founding mountain sites, 925 hPa is below the model terrain. Three usable levels do not define a sounding; interpolation would dominate the boundary-layer and stability estimates.

The 15-day horizon could support a labelled proxy-physics outlook. It cannot support the same windgram contract as models that publish surface fluxes and a deeper pressure column.

Licensing and attribution

• ECCC (HRDPS, RDPS, GDPS): used under the Environment and Climate Change Canada Data Server End-use Licence. Derived profiles carry the attribution requirement.
• NOAA (HRRR, GFS): U.S. government open data. Attribution is requested, and use implies no NOAA endorsement of this project.
• ECMWF open data is CC-BY-4.0 — relevant only if an IFS-based leg is adopted later.

Provider overviews: HRDPS, RDPS, GDPS, REPS, HRRR, and GFS.