Weather Data for Installations

This skill gives an agent live and historical weather, wind and sea-state data for any installation, onshore or offshore, from open APIs that need no account, subscription or API key:

It is built especially to work well on the Norwegian Continental Shelf (MET Norway's 1 km Nordic model plus the Sodir facility register let an agent resolve "Troll A" or "Åsgard B" by name and cross-check two independent forecast sources), while every function also works anywhere in the world given a name (Open-Meteo geocoding) or a latitude/longitude.

It is a data-retrieval skill: it returns measured/modelled weather series and simple empirical summaries, not a certified met-ocean design study. Design wind, wave and temperature return periods, extreme-value (Gumbel/Weibull) statistics and site-specific met-ocean criteria require a qualified met-ocean study; this skill only gives a screening-level, order-of-magnitude picture to scope one.

When to Use

• A task needs the current or forecast ambient temperature, wind speed/ direction/gust, humidity or pressure at an installation, by name or by coordinates.
• A task needs historical weather ("what has the weather been like here") for a past incident window, a fouling/corrosion season, or a design-basis screening statistic (min/mean/max temperature, wind speed percentile).
• A task needs a forward-looking forecast (hours to ~16 days) to plan a lift, a flaring event, a marine operation, or a fire-water/dispersion screening.
• A task needs wave height, wave period or sea-surface temperature for an offshore installation or subsea cooldown screening.
• An agent needs a Pasquill-Gifford stability class for neqsim-gas-dispersion-distance-screening from live wind speed, time of day and cloud cover, instead of an assumed class.
• An agent needs ambient temperature and elevation for neqsim-gas-turbine-performance-screening, or a wind speed for neqsim-firewater-deluge-design's monitor/wind-drift screening, without the user supplying them by hand.
• A root-cause or operational investigation needs the weather at an equipment's location for a past trip/anomaly window, to test whether an ambient-temperature, wind, or sea-state condition was a candidate driver (feeds neqsim-root-cause-analysis's "Weather as Evidence" method and neqsim-autonomous-investigation's lead-lag relationship discovery).

Inputs

• Location, resolved in this order (first available wins):
  1. latitude / longitude (and optional elevation_m) supplied directly - always correct, works anywhere.
  2. name matched against the open Sodir facility register (Norwegian Continental Shelf only) - best-effort, live, open data.
  3. name matched against Open-Meteo's worldwide geocoding API.
• hourly: list of Open-Meteo variable names to request (defaults cover temperature, humidity, pressure, wind speed/direction/gust, cloud cover, precipitation).
• start_date / end_date (YYYY-MM-DD) for historical queries.
• days / past_days for forecast queries (up to 16 / 92 respectively).
• Every HTTP client takes an injectable fetch(url, timeout) -> bytes, which makes it testable offline, mirroring neqsim-ncs-infrastructure-network.

Outputs

• WeatherSeries: normalised hourly data (hourly_time, hourly dict of variable -> list of floats, hourly_units, optional current), the resolved latitude/longitude/elevation_m, the source URL and a retrieval timestamp - the same shape from every source, so callers do not need source-specific parsing.
• SiteConditionsSummary: min/mean/max temperature, mean/max/p99 wind speed, max wind gust, mean relative humidity, mean surface pressure, the period covered, and warnings when the sample is short or a variable is missing.
• estimate_pasquill_stability_class(...): a screening A-F stability class from wind speed, day/night and cloud cover.
• compare_forecast_sources(...): mean absolute temperature/wind deltas between Open-Meteo and MET Norway for the same point and window, useful on the NCS to sanity-check either source.

Engineering Method

1. Location resolution. Explicit coordinates are used as-is. A name is first tried against the Sodir facility layer (ArcGIS REST, outSR=4326, returnGeometry=true); every string attribute is scanned for the search term because Sodir field names are not guaranteed stable across releases. If nothing matches (or the installation is outside Norway), Open-Meteo's geocoding API resolves the name worldwide.
2. Forecast. GET api.open-meteo.com/v1/forecast with wind_speed_unit=ms and temperature_unit=celsius so results already match NeqSim/screening conventions; up to 16 forecast days and 92 past days in one call.
3. Historical. GET archive-api.open-meteo.com/v1/archive reads the ERA5 / ERA5-Land / IFS reanalysis for any start_date/end_date back to 1940, worldwide, at ~9-25 km resolution.
4. Marine. GET marine-api.open-meteo.com/v1/marine (cell_selection=sea) returns wave height/period/direction and sea-surface temperature from MeteoFrance/ECMWF/NOAA wave models.
5. MET Norway cross-check. GET api.met.no/weatherapi/locationforecast/2.0/ complete returns the MET Nordic model (~1 km) forecast for Norway, Sweden and Denmark. MET Norway requires a descriptive User-Agent header (no key); a missing or generic one (e.g. okhttp, Java) gets HTTP 403.
6. Site-condition summary. summarize_site_conditions reduces an hourly series to plain empirical statistics (min/mean/max, a 99th-percentile wind speed as a crude gust proxy) with pure-stdlib arithmetic; it is not a Gumbel/Weibull extreme-value fit and says so in its warnings.
7. Stability class. estimate_pasquill_stability_class follows a simplified Turner (1964) day/night, wind-speed/cloud-cover lookup. It ignores solar elevation, season and terrain and is meant as a stand-in only when no certified meteorological classification exists.

Python Usage Pattern

from weather_data import WeatherDataService

service = WeatherDataService()

# By coordinates - works anywhere in the world
troll_a = service.resolve_location(latitude=60.643, longitude=3.724, name="Troll A")

# By name - tries the Norwegian Sodir facility register first, then
# Open-Meteo's worldwide geocoding
ekofisk = service.resolve_location(name="Ekofisk")

forecast = service.get_forecast(troll_a, days=5)
print(forecast.hourly["wind_speed_10m"][:6], forecast.hourly_units["wind_speed_10m"])

history = service.design_site_conditions(troll_a, years_back=5)
print(history.max_wind_speed_ms, history.min_temperature_c, history.warnings)

waves = service.get_marine(troll_a, days=5)
print(waves.hourly.get("wave_height", [])[:6])

cross_check = service.compare_forecast_sources(troll_a, days=2)
print(cross_check["mean_abs_wind_speed_delta_ms"])

from weather_data import estimate_pasquill_stability_class

stability_class = estimate_pasquill_stability_class(
    wind_speed_ms=4.5, is_daytime=True, cloud_cover_fraction=0.2,
)
# feed directly into neqsim-gas-dispersion-distance-screening's stability_class input

Related NeqSim Functionality

• neqsim-gas-dispersion-distance-screening: wind_speed and stability_class can come from get_forecast/get_historical and estimate_pasquill_stability_class instead of an assumed value.
• neqsim-gas-turbine-performance-screening: ambient_temperature_k and site_elevation_m can be read from get_forecast/design_site_conditions.
• neqsim-firewater-deluge-design: wind_speed_m_s for the fire-monitor wind-drift screening.
• neqsim-surf-cooldown-screening: seabed_temperature/ambient temperature can be informed by get_marine's sea_surface_temperature or design_site_conditions.
• neqsim-root-cause-analysis / neqsim-autonomous-investigation (core equinor/neqsim repo): historical weather for an equipment's location and event window, used as circumstantial evidence or as an extra time series in RelationshipGraph lead-lag discovery, never as a confirmed cause on its own.
• neqsim-ncs-infrastructure-network: the open Sodir DataService pattern this skill's facility resolver follows; use that skill for the wider NCS connectivity graph.
• The installation-weather-agent community agent wraps this skill for conversational use and points to the consuming agents above.

Validation Checklist

• The resolved Location.source is recorded (explicit-coordinates, sodir-facility-register, or open-meteo-geocoding) so provenance is traceable.
• Any coordinate resolved by name (Sodir or geocoding) is verified against an authoritative source (STID, P&ID, operator records) before any safety-critical use.
• SiteConditionsSummary.warnings is checked before quoting min/mean/max or percentile values as a design basis.
• MET Norway calls carry a descriptive User-Agent; a 403 means the header was rejected, not that the location has no data.
• Marine results are only used for genuinely offshore/coastal locations; an inland point returns the nearest sea grid cell, which can be far away.

Common Mistakes

Limitations

• Open-Meteo's non-commercial terms cap usage at under 10,000 API calls/day; self-hosting or a commercial key is available for heavier use (see https://open-meteo.com/en/terms).
• The Sodir facility resolver is best-effort name matching, not an authoritative asset register; never use it to place equipment.
• Historical reanalysis (ERA5/ERA5-Land) has ~9-25 km spatial resolution and is not a substitute for a site-specific met-ocean measurement campaign.
• SiteConditionsSummary statistics are plain empirical min/mean/max/ percentiles over the retrieved window, not a return-period extreme-value analysis.
• estimate_pasquill_stability_class is a simplified screening lookup; use a certified meteorological classification for real consequence studies.
• MET Norway Locationforecast covers the Nordic region at high resolution; outside Norway/Sweden/Denmark, rely on the Open-Meteo forecast only.

References

• Open-Meteo API documentation: https://open-meteo.com/en/docs, https://open-meteo.com/en/docs/historical-weather-api, https://open-meteo.com/en/docs/marine-weather-api, https://open-meteo.com/en/docs/geocoding-api
• Open-Meteo terms and attribution: https://open-meteo.com/en/terms
• MET Norway Locationforecast documentation: https://api.met.no/weatherapi/locationforecast/2.0/documentation
• Sodir (Norwegian Offshore Directorate) open data: https://factpages.sodir.no, NLOD 2.0 licence
• Turner, D.B. (1964). "A Diffusion Model for an Urban Area." Journal of Applied Meteorology, for the simplified Pasquill-Gifford stability lookup.
• NeqSim repository: https://github.com/equinor/neqsim
• NeqSim Skills Guide: https://github.com/equinor/neqsim/blob/master/docs/integration/skills_guide.md