Wind
Winds come from multi-level NWP output, with Monin–Obukhov log-law shear and bulk-Richardson stability. Power curves are rescaled for air density under IEC 61400-12-1, wakes use the Bastankhah–Porté-Agel Gaussian model, and a storm taper with shutdown hysteresis is anchored to each curve’s cut-out.
Solar
Sun position from NREL’s SPA, Perez anisotropic transposition, single-axis tracking with closed-form backtracking, wind-cooled cell temperature, snow coverage and DC:AC clipping.
Horizons and bands
The ensembles are NOAA’s GEFS (31 members) and ECMWF’s ENS (51 members). The deterministic forecast runs 384 hours in every region. Validated blends cover the front of the horizon and clearly labelled tail members the back. Probabilistic bands run to 360 hours everywhere (published horizon, 30 Sep 2026).
Metered GB and European wind farms that pass a weekly held-out test get P10–P90 bands recalibrated per farm by conformal prediction. In the weekly run of 27 September 2026, 54 farms were accepted, and on held-out days their P10–P90 range caught the metered output 79% of the time, up from 69%, against an 80% target. The accepted set changes weekly and contains no solar farms. On paid plans the calibrated bands appear in the farm panel from Analyst, and in the per-farm API and bulk downloads on Enterprise.
Ground truth
Public scorecards give skill by region, fuel and lead day, and negative skill is published with the rest. Over the 14 days to 30 September 2026, forecasts made within 48 hours of each hour beat same-horizon persistence (a forecast that assumes output stays where it is now) for solar in 28 of 30 scored regions and for wind in 26 of 33.
Site-fitted power curves are built from each farm’s metered output by taking the median power in 0.5 m/s wind bins and then enforcing a monotone curve with weighted isotonic regression (pool-adjacent-violators). Besides the GB Accuracy Proof above, an EU per-unit scorecard checks 3 wind farms (1,363 MW) against ENTSO-E metered output. There, site-fitted curves cut capacity-weighted error by 58% against a generic reference curve on held-out days (computed 27 Sep 2026, negative-price hours screened). With three farms, this is not a fleet-wide claim.
Resource Atlas
Each site’s forecast can be read against its own climatology: the resource capacity factor from our physics run over the native ERA5 record back to 1940, with the CEAtlas reference case beside it. US wind and solar farms also get a labelled comparison of delivered energy against EIA-923 meters. GB follows once complete years of settlement-meter data are available; everything else is served on a resource basis and labelled that way. Coverage is the operating fleet of 10 MW and up, as labelled in the fleet, with monthly matrices, P10–P90 bands and downloads. Explore shows each farm’s long-run capacity factor and monthly seasonality free; the matrices, P10–P90 spread and downloads start at Analyst.
GB settlement data and the API
On the Enterprise API, mapped GB wind farms carry their Elexon BM unit IDs and NESO generator names, with half-hourly output on the settlement-period grid. The GB BMU reference map (open to any signed-in account, refreshed 29 Sep 2026) covers 169 wind farms whose BM units account for 92.7% of the wind capacity listed in NESO’s 14-day BM-unit forecast (23.5 of 25.3 GW).
Enterprise includes the full REST API (10k calls a day), your own farms with data-quality badges, per-tenant isolation, and single sign-on shared with CEGridSight.
The GB Accuracy Proof, the regional skill scorecards and data freshness are public, with no login or card, and a public data-sources page credits every upstream provider and its licence. The map shows the operating fleet by default; one switch adds projects in construction or planning.
