nasa_power

NASA POWER — public REST, no auth, hourly point data. Rate-limited to ~1 req/s. Emits ``power_t2m`` (°C), ``power_wind_speed_50m`` (m/s, at 50 m — POWER's ``WS50M``; there is no 100 m parameter) and ``power_ssrd`` (all-sky surface shortwave, W/m², ``ALLSKY_SFC_SW_DWN``). **Timestamps.** The hourly key ``YYYYMMDDHH`` (``time-standard=UTC``) labels the hour that STARTS then, which is the platform's ``interval_start``: no shift. Checked 2026-10-08 against solar geometry at Athens on a clear day (2026-06-21): clear-sky irradiance is symmetric about the 10:00 key (09:00 972.33 W/m², 11:00 972.42) while solar noon is ~10:28 UTC, i.e. the 10:00 value is the mean over 10:00-11:00. Stamped at the hour's end, the symmetry would sit about 11:00. **Radiation lags ~3 months.** POWER's hourly radiation comes only from CERES SYN1deg (the response header lists ``SYN1DEG``), published about a quarter behind; temperature and wind come from MERRA-2/GEOS and are current to a day or two. Probed 2026-10-08: ``ALLSKY_SFC_SW_DWN`` is filled to 2026-06-30 23:00 and -999 after, for every community (RE, SB, AG) — the near-real-time FLASHFlux radiation exists only in the DAILY API. So ``power_ssrd`` stopping three months back is upstream latency, not a wrong parameter; the missing values are counted and logged, not stored. **``dataset="daily"``: current radiation.** The DAILY endpoint serves FLASHFlux all-sky shortwave about five days behind (live 2026-10-09: filled to 2026-10-04). Stored as its own variable, ``power_ssrd_daily``, at granularity ``P1D`` - never mixed into the hourly ``power_ssrd``. Units: POWER answers kWh/m²/day; stored as the day's MEAN irradiance in W/m² (x 1000 / 24), the same physical unit as the hourly series. The day is POWER's local-solar-time day (the daily API has no UTC option; the header says ``time_standard: LST``), stamped at 00:00 UTC of that date - off from the UTC day by longitude / 15 hours (Athens ~1.6 h). Failures raise (there is no credential, so there is no synthetic path for a failed call): HTTP 429/5xx and transport faults are retried by ingest, any other status is a :class:`ConnectorError`.