Weather · Observations

Surface observations

What the weather actually did: temperature, dew point, wind, pressure and visibility from 25 airport stations across the eight ERCOT weather zones. Filed hourly and again whenever conditions change. Collected from the NWS API, reconciled against live responses.

noaa-station-observationshourly, plus specials · 1h rows25 entities
KAUS
26.0°C
2026-09-18 04:50 UTC
KDFW
31.0°C
2026-09-18 04:50 UTC
KIAH
25.0°C
2026-09-18 04:50 UTC
Read from the Dryos API and cached for an hour; the stream itself updates hourly, plus specials.

What is in it

Updates
hourly, plus specials · 1h rows
Entities
25 entities — KAUS, KDFW, KIAH…
Source timezone
UTC — every timestamp Dryos serves is UTC
Coverage
31 August 2026 to now
Rows held
79,358
Primary key
station, interval_start_utc
On the map
25 of 25 placed

What it measures

Every value, its unit, and what the source means by it.

  • Temperature°C

    Air temperature at the station.

    temperature_c
  • Dew point°C

    Dew point — how much moisture the air is holding.

    dewpoint_c
  • Relative humidity%

    Relative humidity, derived from temperature and dew point.

    relative_humidity_pct
  • Wind speedm/s

    Wind speed at the standard 10 m observing height.

    wind_speed_ms
  • PressurehPa

    Barometric pressure at station level.

    pressure_hpa

Columns

Exactly what a request answers with, read from the API rather than written here.

ColumnTypeDescription
interval_start_utcrequiredtimestampStart of the hour the values describe, UTC.e.g. 2026-08-31T12:00:00Z
stationrequiredstringNWS station identifier — the ICAO code for an airport station.e.g. KAUS
zonerequiredstringERCOT weather zone the station sits in, in the load reports' spelling. An attribute of the station, not the entity — assigned by Dryos.e.g. SOUTH_C
temperature_crequirednumberAir temperature, degrees Celsius.e.g. 24.0
dewpoint_cnumberDew point, degrees Celsius.e.g. 23.0
relative_humidity_pctnumberRelative humidity, percent.e.g. 94.1
wind_speed_msnumberWind speed, metres per second. Converted from the km/h the source reports.e.g. 3.6
wind_gust_msnumberPeak gust, metres per second.e.g. 9.3
wind_direction_degnumberWind direction, degrees from north.e.g. 180
pressure_hpanumberBarometric pressure at station level, hectopascals. Converted from pascals.e.g. 1015.9
visibility_mnumberHorizontal visibility, metres.e.g. 16093.4
latrequirednumberLatitude of the reading, degrees north. Published by the source, not inferred: this is where the instrument is or the point that was sampled.e.g. 30.19
lonrequirednumberLongitude of the reading, degrees east of Greenwich.e.g. -97.67
collected_at_utcrequiredtimestampWhen Dryos collected this row, UTC. Stamped by the platform, not the collector.e.g. 2026-08-31T13:02:00Z

Read it

Public, read-only, no key and no sign-up. Relative times stay relative, so a URL keeps working.

The newest rows

shell
curl "https://api.dryos.ai/v1/datasets/noaa-station-observations/query?node=KAUS&limit=12"

The catalogue entry — every column, its type and the source

shell
curl "https://api.dryos.ai/v1/datasets/noaa-station-observations"

Full parameters — bucketing, aggregation, filters and search — are on the API reference.

Ask an agent

The same stream through the Model Context Protocol, so Claude, ChatGPT, Cursor or any MCP client can query it directly.

shell
claude mcp add --transport http dryos https://api.dryos.ai/mcp

Then ask:

What has Temperature been at KAUS over the last 24 hours, and when did it peak?

Setup for every client is on the MCP server page.

Changes to this collector

A change in the collector is a change in the data. Anyone reading a step in a series is owed the day it moved under them.

  1. 2026-09-18

    The gap from 2026-09-04 to 2026-09-18 is a collector fault, not a source one. The run after this heals it back to 2026-09-11 in one batch; 2026-09-04 to 2026-09-11 stays a hole, because the observations endpoint serves about seven days and the rest had aged out before the fault was found.

    The row-count check compared raw batch size against a median taken from twenty-minute runs. This collector reads from the watermark to now, so once a run failed the next batch covered the whole gap — 49,777 rows against a median of 103, which failed the check, which held the watermark, which made the next batch bigger still. The count is normalised per timestamp now, so healing a gap looks like ordinary weather.

How this stream is checked

  • Pulled straight from NOAA · National Weather Service API — no reseller in between — and compared against that file before it is served.
  • Null means the source did not report it — never zero.
  • A reading lands hourly, plus specials; the interval timestamp is the start of the interval, in UTC.
  • Every weather timestamp is UTC at the source and at Dryos.

Related streams