Cyclone API

Models, runs and products

Discover which models, runs and fields are available, and how each field is encoded.

These catalog endpoints tell you what you can request. Check them before building data URLs; don’t hard-code runs or product names.

Models

GET/models

Returns an array with one entry per model or observation source. No account is needed.

[
  {
    "name": "rdps",
    "default": false,
    "prefix": "/models/rdps",
    "forecast_hours": 85,
    "forecast_hour_axis": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12],
    "products": 74
  }
]
namestring

Model ID, used in every other URL.

prefixstring

Path prefix for this model’s endpoints.

forecast_hour_axisinteger[]

What each hour index means: forecast hours for models, minutes since midnight for observation sources. See Runs and hours. The example above is truncated.

forecast_hoursinteger

Length of forecast_hour_axis.

productsinteger

Number of products. List them with /products.

Coverage

GET/models/coverage

Returns a GeoJSON FeatureCollection with each model’s footprint as a polygon, so you can tell which models cover a location. Each feature’s id is the model ID. Its properties are the model’s name, default, prefix and forecast_hours (as in /models), the latest run, the native grid type (rotated_lat_lon, lambert_conformal, polar_stereographic or geostationary), and the grid size as ny × nx. boundary says how the footprint was drawn: grid is the edge of the model’s grid, and footprint is traced around the cells that actually hold data. clip_ring is a simplified copy of the same outline, with fewer points. A model with no runs has a null geometry.

Archive

GET/models/archive

Lists the UTC days each model has runs for. Use it to find past days before asking /runs for a specific one. No account is needed. The response has one entry per model, in the same order as /models:

{
  "models": [
    {
      "name": "hrrr",
      "days": [
        { "day": "2022-05-21", "runs": 1, "pinned": true },
        { "day": "2022-05-22", "runs": 1, "pinned": true },
        { "day": "2026-11-20", "runs": 2, "pinned": false }
      ]
    },
    { "name": "era5", "days": [] }
  ]
}
daysobject[]

Every UTC day that at least one available run covers, oldest first. A run covers each day between its first and last valid time, so a 48-hour forecast appears on three days and future days are included. A model with no runs has an empty list. The example above is truncated.

runsinteger

How many runs cover that day. A run counts once on every day it covers, so adding these up over-counts runs.

pinnedboolean

true when the day is kept indefinitely, usually because it’s part of an archived severe weather event. Days that aren’t pinned may be removed as newer runs arrive.

Runs

GET/models/{model}/runs

Returns the model’s available runs, newest first:

{ "runs": ["20261118T060000Z", "20261118T000000Z", "20261117T180000Z", "20261117T120000Z"] }

Most forecast models keep between two and four recent runs, plus any pinned runs from past events. Some observation sources keep much more: MRMS hail, 30-minute lightning and reflectivity go back to October 2020.

Products

GET/models/{model}/products

Returns every field the model provides, with its unit and tile encoding:

{
  "products": [
    {
      "name": "T2m",
      "range": [250.0, 310.0],
      "units": "K",
      "encoding": {
        "dtype": "int16",
        "littleEndian": true,
        "width": 256,
        "height": 256,
        "scale": 0.0146484375,
        "offset": 280.0,
        "nodata": -32768
      },
      "levels_hpa": null
    },
    {
      "name": "ABSV_ISBL",
      "range": [-0.0016, 0.0016],
      "units": "s^-1",
      "encoding": { "dtype": "int16", "littleEndian": true, "width": 256, "height": 256, "scale": 7.8125e-7, "offset": 0.0, "nodata": -32768 },
      "levels_hpa": [850.0, 700.0, 500.0, 250.0]
    }
  ]
}
namestring

Product name. Names are case-sensitive: T2m, not t2m.

unitsstring | null

Unit of decoded values. Most products use SI units (K, Pa, m/s, J/kg), but not all; always check. An empty string means the value has no unit.

range[number, number] | null

Range of values the tile encoding can represent. Values outside it are clamped to the nearest end.

encodingobject

How to decode this product’s data tiles. Missing when range is null; those products can’t be served as tiles, only through point queries.

levels_hpanumber[] | null

For products on pressure levels, the available levels in hPa. Choose one with the level parameter, which is an index into this array. null for single-level products.

familyobject

Ensemble products only. How the product relates to its deterministic base field. See Ensemble products.

hoursinteger[]

Present when the product exists at only some time steps. The hour indices where it has data; requests for other hours return 418.

Ensemble products

Ensemble products, such as REFS, are ordinary products with a family object that describes them. The kind is one of:

kind Meaning Extra fields
mean, weighted_mean Ensemble mean
spread, stddev Spread between members
pmm, lpmm Probability-matched mean, and its local variant
member One ensemble member member (1-based), members (ensemble size)
percentile A percentile of the ensemble percentile
probability Chance of exceeding a threshold, in % op (gt, lt or between), threshold, thresholdUpper, and optionally method (point, neighborhood or eas)

Thresholds are in the base field’s units. This REFS product (with encoding left out) is the chance of more than 127 mm of precipitation in 12 hours, and exists only every third hour:

{
  "name": "APCP_12h_eas_gt127",
  "range": [0.0, 100.0],
  "units": "%",
  "levels_hpa": null,
  "family": { "base": "APCP_12h", "kind": "probability", "op": "gt", "threshold": 127.0, "thresholdUpper": null, "method": "eas" },
  "hours": [11, 14, 17, 20, 23, 26, 29, 32, 35, 38, 41, 44, 47, 50, 53, 56, 59]
}

Means, members and percentiles use the base field’s units and range. Probabilities are always 0–100 %.

Naming

Product names follow the source model’s conventions, so they vary between models. Some patterns:

Pattern Meaning Examples
…2m, …10m Near-surface fields T2m, Td2m, RH2m, U10m, V10m
…_ISBL Pressure levels (use level) T_ISBL, HGT_ISBL, U_ISBL, ABSV_ISBL
Lowercase Computed by CSSL from model output sb_cape, ml_cin, srh_0_1km, bulk_shear_0_6km, stp_index, theta_e
…_mean, …_spread, …_pmm, …_m01 Ensemble statistics and members APCP_1h_mean, APCP_1h_spread, APCP_1h_m01
…_gt…, …_eas_gt… Ensemble probabilities; p stands for a decimal point APCP_1h_gt12p7 (more than 12.7 mm)

Wind comes as separate U and V components. On some models, including HRDPS and RDPS, these are relative to the model grid rather than true north. See Grid-relative winds.