Skip to content

dmi-radar.md

Latest commit

 

History

History
233 lines (174 loc) · 18.9 KB

dmi-radar.md

File metadata and controls

233 lines (174 loc) · 18.9 KB
Raw

DMI Weather Radar Network (Denmark)

What this is

The Danish Meteorological Institute (DMI) operates a five-radar C-band network covering Denmark and its surrounding seas, and publishes the full observational output — national composite, per-radar pseudo-CAPPI, and raw per-radar polarimetric volumes — as ODIM HDF5 through an open, unauthenticated STAC / OGC API-Features service. All three product families are genuine gridded/polar binary data, not rendered imagery.

This is a single-operator national network. The distinguishing features are the depth of the open archive (a rolling 180 days, unusually generous for a real-time radar feed — most European national feeds retain days) and the fact that raw, unfiltered volume data including the full dual-polarisation moment set is published operationally with no account, no key, and no approval gate. Denmark also runs two interleaved scan strategies — a long-range reflectivity scan and a short-range dedicated Doppler scan, alternating every 5 minutes — and both are exposed separately through a scanType query parameter.

Who operates it

  • Operator / coordinating programme: Danmarks Meteorologiske Institut (DMI), Ministry of Climate, Energy and Utilities
  • Country / region: Denmark (domain extends over the North Sea, Baltic, southern Norway/Sweden and northern Germany)
  • Data distributor (if different): DMI Open Data — Radar Data API (opendataapi.dmi.dk/v1/radardata)
  • Contributing networks / members: n/a (single operator)

Network composition

Five Vaisala WRM200 C-band radars (wavelength 5.33 cm, beamwidth 0.95°, dual-polarisation). Site metadata read from the ODIM what/source and where groups of live volume files (29 July 2026):

Site API stationId ODIM what/source Lat (°N) Lon (°E) Antenna height Active from
Stevns 06177 WMO:06177,RAD:DN41,PLC:Stevns,NOD:dkste 55.32619 12.44928 53 m 2001-04-01
Rømø / Juvre 60960 WMO:06096,RAD:DN42,PLC:Romo,NOD:dkrom 55.17311 8.55200 15 m 1992-03-01
Sindal 06036 WMO:06036,RAD:DN43,PLC:Sindal,NOD:dksin 57.48931 10.13647 109 m 1994-07-01
Bornholm 06194 WMO:06194,RAD:DN44,PLC:Bornholm,NOD:dkbor 55.11275 14.88752 171 m 2002-03-05
Samsø 06133 WMO:06133,RAD:DN46,PLC:Samso,NOD:dksam 55.81190 10.58530 48 m 2024-10-11

Notes on the table:

  • Coordinates above are the in-file (antenna) values. DMI's published station table gives values differing by roughly 50–150 m (e.g. Stevns 55.32562 / 12.44817 vs 55.32619 / 12.44928). Prefer the in-file where/lat,where/lon for georeferencing.
  • RAD:DN45 is absent from the live network. DMI's station table lists a sixth radar, Virring/Skanderborg (06103, 56.02387 / 10.02517, active from 2008-11-01), with status Inactive; Samsø came online 2024-10-11. Whether DN45 was Virring's code is not stated in any source checked — TBD.
  • No non-radar sensors are blended into any of these products (no gauge adjustment, no lightning, no NWP). This is radar-only.

Scan strategy (from how and dataset*/where, live 29 July 2026)

Two interleaved volume scans, each 10 sweeps × 360 rays, 500 m range gates, nominal elevations 0.5°–15° (measured set ≈ 0.49, 0.65, 0.96, 1.47, 2.37, 4.80, 8.40, 9.97, 12.99, 15.00°):

scanType=fullRange scanType=doppler
Minutes past hour :00, :10, … :50 :05, :15, … :55
how/task REFLECTI VELOCITY
Range gates 474–475 (varies by site) 239
Max range 237.0–238.5 km (documented as 240 km) 119.5 km (documented as 120 km)
how/prf 625 Hz, prffac 1 1180 Hz, prffac 2 (dual-PRF)
Nyquist velocity (how/NI) ±8.3 m s⁻¹ ±47.2 m s⁻¹

Products

1. Composite (composite)

National maximum-value reflectivity composite across all five radars.

  • Filename: dk.com.<YYYYMMDDHHMM>.500_max.h5
  • Format: ODIM HDF5 v2.0, what/object = COMP, single dataset, quantity = DBZH, uint8
  • Grid: 1728 × 1984 (ny × nx) at 500 m = 864 × 992 km
  • Projection: polar stereographic — +proj=stere +ellps=WGS84 +lat_0=56 +lon_0=10.5666 +lat_ts=56
  • Cadence: 5 minutes (288 files/day, verified 28 July 2026)
  • Compositing rule: maximum value in areas of overlap
  • Both scan types are written to the same grid; the Doppler-based composite simply covers less of it (a sample pair showed 46.2 % nodata for fullRange vs 78.8 % for doppler, consistent with the 240 km / 120 km range difference)
  • Clutter-filtered (image-processing chain applied)

2. Pseudo-CAPPI (pseudoCappi)

Per-radar near-surface reflectivity, composed across the sweeps of a single volume, at a minimum height of 500 m.

  • Filename: <prefix><YYYYMMDD>_<HHMM>.Z_nn_S1.ps.500.wrk.h5 — prefixes observed: ekxs (Stevns), eksn (Sindal), eksm (Samsø), ekrn (Bornholm)
  • Format: ODIM HDF5 v2.2, what/object = IMAGE, product = PCAPPI, quantity = DBZH, uint8
  • Grid: 960 × 960 at 500 m = 480 × 480 km, one grid per radar
  • Projection: gnomonic, centred on each radar — +proj=gnom +ellps=WGS84 +lon_0=<site lon> +lat_0=<site lat> (note: a different projection family from the composite)
  • Cadence: 10 minutes (≈575 files/day across four sites)
  • Only four of the five radars are published — Rømø is absent (see Notes)
  • Clutter-filtered

3. Volume (volume)

Raw, unfiltered per-radar polarimetric volumes — the most basic product.

  • Filename: dk<site>_<YYYYMMDDHHMM>.vol.h5 (dkste, dkrom, dksin, dkbor, dksam)

  • Format: ODIM HDF5 v2.0 (what/version = H5rad 2.0), what/object = PVOL, 10 sweeps (dataset1…dataset10), uint8

  • Geometry: 360 rays × 239 or 474/475 gates, rscale 500 m, rstart 0.5 km, a1gate 0

  • Moments (8, identical on every sweep):

    quantity Meaning
    DBZH Horizontal reflectivity (corrected)
    TH Horizontal total (uncorrected) reflectivity
    VRAD Radial velocity
    WRAD Spectrum width
    ZDR Differential reflectivity
    RHOHV Co-polar correlation coefficient
    PHIDP Differential phase
    LDR Linear depolarisation ratio

    KDP is not distributed — derive it from PHIDP if needed.

  • Cadence: 5 minutes per site (≈1438 files/day across five sites)

  • No clutter filtering — clutter is present as recorded

Value scaling (all products)

value = raw × gain + offset, with nodata and undetect sentinels read from the same what group.

  • DBZH / TH: gain = 0.5, offset = −32.0 → dBZ = 0.5 × raw − 32.0, spanning −31.5 … +95.0 dBZ over raw 1…254
  • nodata = 255 (outside coverage), undetect = 0 (in coverage, no echo). Because undetect is 0, raw 0 must not be read as −32.0 dBZ; the usable scale starts at raw 1.
  • VRAD gain/offset vary by scan type and by file — fullRange samples read gain = 0.065576, offset = −8.328125; doppler samples read gain = 0.371421, offset = −47.1705. Always read the per-dataset what attributes rather than hardcoding.
  • Z–R relation: how/zr-a = 200.0, how/zr-b = 1.6 (Marshall–Palmer)
  • No clipping observed: composites from the convective afternoon of 17 July 2026 reached 68.5 dBZ with a smooth distribution up to the maximum, and volume DBZH reached 67.5 dBZ.

Data availability

  • Is the data free? Yes — no account, no registration, no API key. (api-key is an accepted query parameter on every collection but is not required; all requests below were made anonymously.)

  • License: CC BY 4.0. DMI holds the IP rights and licenses its open data under Creative Commons Attribution 4.0; attribution to DMI and a link to the licence are required, commercial use is permitted, and no share-alike obligation applies. The /collections response carries a machine-readable rel="license" link to DMI's terms-of-use page, which is the direct confirmation that the licence covers the radar service specifically. Licence terms last changed 30 November 2023.

  • Is the data downloadable? Yes.

  • Access tier: Open (no account).

  • Data formats: ODIM HDF5 (application/x-hdf5) — v2.0 for composite and volume, v2.2 for pseudo-CAPPI.

  • Update cadence: composite 5 min; pseudo-CAPPI 10 min; volume 5 min per site.

  • Latency: ~8–13 minutes measured. Files are published in pairs on a 10-minute batch: the :00 fullRange and :05 doppler products both appeared at ≈:13.

  • Primary access — Radar Data API (STAC API-Features, base https://opendataapi.dmi.dk/v1/radardata):

    Endpoint Purpose
    /collections List the three collections; carries the rel="license" link
    /collections/{composite|pseudoCappi|volume}/items Query features (one per file)
    /collections/{collection}/items/{id} Single feature
    /download/{file-name} Retrieve the HDF5 file
    /conformance, /api Conformance classes; OpenAPI 3.0 definition

  • Query parameters (support differs per collection — verified by sending a bogus parameter and reading the 400 response):

    Parameter composite pseudoCappi volume
    datetime, bbox, bbox-crs, limit, offset, sortorder, api-key ✅ ✅ ✅
    scanType (doppler | fullRange) ✅ — (rejected) ✅
    stationId — (rejected, 400) ⚠️ accepted but silently ignored ✅ works

    limit defaults to 1000 and caps at 300000; offset caps at 1000000. sortorder accepts only datetime,DESC — any other value, including datetime,ASC, returns 400.

  • New-data notifications: none. No SNS, webhook, or feed — poll the items endpoint. DMI's fair-use guidance asks users to poll at the product's actual update cycle rather than continuously, to filter server-side, and to cache; no numeric rate limit is published, but DMI reserves the right to throttle per-IP.

  • Other mirrors: none known for the radar products. DMI's terms reference "Bulk Download Services" alongside the APIs, but no bulk radar channel is documented — TBD.

  • Archive depth: Rolling 180 days, matching DMI's documentation and confirmed empirically with a sharp cutoff. Date-window probes on 29 July 2026 returned zero features for 2026-01-29 (181 days back) and non-zero for 2026-01-30 (180 days back) in all three collections. No deeper archive is offered through this service.

  • Approximate volume: ~1.5 GB/day (volumes ≈1.0–1.15 MB each dominate; composites 37–166 kB, pseudo-CAPPI ≈20 kB), so roughly 270 GB for the full 180-day window.

Example — fetch the latest composite

BASE=https://opendataapi.dmi.dk/v1/radardata

# Newest full-range composite
ID=$(curl -s "$BASE/collections/composite/items?limit=1&scanType=fullRange&sortorder=datetime,DESC" \
     | python3 -c 'import json,sys; print(json.load(sys.stdin)["features"][0]["id"])')

curl -sL "$BASE/download/$ID" -o "$ID"
import h5py, numpy as np

with h5py.File(ID, "r") as f:
    raw = f["dataset1/data1/data"][:]          # (1728, 1984) uint8
    w   = f["what"].attrs                      # composite: gain/offset live in root /what
    gain, offset = float(w["gain"]), float(w["offset"])
    nodata, undetect = float(w["nodata"]), float(w["undetect"])

dbz = np.where((raw == nodata) | (raw == undetect), np.nan, raw * gain + offset)

Note the group layout differs between products: in the composite the scaling attributes sit in the root /what and quantity is an attribute directly on the dataset1/data1 group; in pseudo-CAPPI they are in dataset1/what; in volumes they are in dataset1/dataN/what. A generic reader must handle all three.

Scope note

Clean fit — no tension to flag. Open licence (CC BY 4.0), open access with no account or key, gridded/polar binary containers, permanent documented operational channel, and a real-time feed rather than an archive-only dataset.

Notes

Observation, not forecast. All three product families are observational. No nowcast or extrapolation products are carried in this API, so — unlike the DWD and ČHMÚ trees — there is nothing here to split off to nowcasting_models/.

Rømø's stationId is not its WMO number. The API indexes Rømø as 60960, which is DMI's own station-registry number and matches DMI's published station table. The ODIM what/source inside the file records WMO:06096. Querying stationId=06096 (or 6096) returns zero features; only 60960 works. The other four sites' stationId values coincide with their WMO numbers, so this one site is an easy trap when building a station map from file metadata.

stationId is silently ignored on pseudoCappi. The parameter is accepted without error but has no effect — stationId=06194 returns all four stations interleaved. It filters correctly on volume, and is rejected outright (400) on composite. Filter pseudo-CAPPI client-side on properties.stationId, the filename prefix, or how/nodes.

Rømø publishes no pseudo-CAPPI. Only four sites appear in the pseudoCappi collection (06036, 06133, 06177, 06194). Spot checks at 2026-02-01, 03-15, 05-01, 06-15 and 07-20 returned the same four throughout, so this is a persistent gap across the whole retention window, not an outage. Rømø is present in volume and contributes to the composite. Cause not documented — TBD.

Only descending sort is available. sortorder=datetime,DESC is the sole accepted value. To reach the oldest end of the archive, page with datetime intervals rather than sorting ascending.

Composite georeferencing needs care — the corner metadata is partly unreliable.

  • where omits xsize/ysize (required by ODIM). Take the shape from the dataset itself: (1728, 1984). Pseudo-CAPPI does carry both.
  • The four corner lat/lons are pixel centres, not cell edges: the projected span between LL and UR is (nx−1) × 500 and (ny−1) × 500, not nx × 500.
  • UL is given as exactly (3.0, 60.0) — suspiciously round, and inconsistent with the grid, which implies ≈(2.99900, 60.00445).
  • LR_lat is bit-identical to LL_lat (52.29427206432812), which is geometrically impossible for a rectangular grid in a stereographic projection. The grid implies LR ≈ (18.90052, 52.15864), so the published LR_lat is wrong by ≈0.136° ≈ 15 km.
  • LL, UL and UR are mutually consistent to ~0.01°. Georeference from LL + projdef + array shape, and do not trust LR.

The API's STAC bbox understates the domain. Features report bbox = [4.379, 52.294, 20.735, 59.828], computed from LL and UR only. The true north-west extent reaches lon ≈2.999 and lat ≈60.004 at the UL corner, so bbox filtering may behave unexpectedly near the north-west of the composite.

STAC conformance is partial. /conformance declares OGC API-Features 1.0 (core, oas30, geojson) plus api.stacspec.org/v1.0.0-rc.2/core, but: Collection objects omit extent entirely (hence the empirical retention measurement above), items expose the download link under asset (singular) rather than STAC's standard assets, and sortorder/stationId are custom parameters in place of STAC's sortby. Generic STAC clients will not resolve the download hrefs without a shim.

ODIM deviations — DMI documents this explicitly ("our radar data files to some extent differs from the ODIM HDF5 standard in relation to metadata"). Observed:

  • Composite what/source = DMI-RADARGROUP — not valid ODIM source syntax (no WMO:/RAD:/NOD: token).
  • Pseudo-CAPPI what/source = CTY:611 — country only. Identify the radar from how/nodes (dkste/dkrom/dksin/dkbor/dksam) or the filename prefix.
  • Pseudo-CAPPI omits prodpar, so the CAPPI height is absent from the file; DMI's documentation gives the minimum height as 500 m.
  • Volume files are properly formed, with full WMO:/RAD:/PLC:/NOD: source strings.

Radial velocity from fullRange is close to useless. Its ±8.3 m s⁻¹ Nyquist (single-PRF, 625 Hz) means pervasive folding. The dedicated doppler scan exists precisely for this and gives ±47.2 m s⁻¹ via dual-PRF. VRAD is nonetheless present in both.

Clutter handling differs by product. Volumes are raw as recorded — land and sea surface, buildings, wind turbines, birds, WiFi and solar interference all included. Composite and pseudo-CAPPI have passed a clutter-removal chain, which DMI notes is imperfect in both directions (residual clutter, and occasional removal of real precipitation).

Reflectivity interpretation. DMI's rule of thumb: ≈20 dBZ light precipitation (<1 mm/h), ≈35 dBZ moderate (≈5 mm/h), >45 dBZ heavy (>20 mm/h) or hail.

Feed completeness. Near-complete but not perfect: 28 July 2026 returned 288/288 composites, 575/576 pseudo-CAPPI, and 1438/1440 volumes.

Data feed vs viewer. DMI's public radar animation at dmi.dk/radar is rendered imagery and is not this feed. The API above is the data.

Relationship to OPERA. Denmark is a EUMETNET OPERA member and these radars contribute to the pan-European OPERA composite; this DMI feed is the national-resolution counterpart, and the only route to the raw Danish volumes.

Relationship to NWP. DMI runs a HARMONIE-AROME configuration (see harmonie-dmi). Whether these radars are assimilated into that suite, and in what form, is not stated in the radar documentation — TBD.

Reading the files. Standard ODIM HDF5 tooling works: h5py, wradlib, xradar, Py-ART. Expect to handle the group-layout and source deviations noted above.

Official documentation

Entry verified against live API responses and downloaded ODIM HDF5 files on 29 July 2026.