Takeaway: the MSC GeoMet (eccc-*) cloud agrees noticeably more closely with HRDPS
across mean wind, gusts and wind direction than either Weather Underground PWS or Tempest do —
tighter scatter around the 1:1 line, pooled fit slopes closer to 1, and direction roses with much higher
≤30° |Δ| share. MSC SWOB stations are professionally sited (open exposure, standard 10 m mast,
regular calibration, away from buildings and trees), so they sample the same air the model resolves on its
grid. WU and Tempest devices are mostly rooftop or backyard installs at varying heights, often sheltered or
channelled by local obstacles, which both attenuates measured speeds and rotates the apparent direction —
so the wider, more biased clouds for those two networks probably reflect siting and sensor
quality at least as much as any real model error. For paragliding spot decisions: when an MSC station is
nearby, trust it; when only PWS/Tempest are available, expect under-reads and direction wander unless the
station is known to be on an exposed mast. To check any single station yourself — see how much it under- or
over-reads versus HRDPS, look at its direction agreement, and decide whether to rely on it — open its
windmeter quality page (from the
All stations list above, or by clicking
“How good is this windmeter?” in the map’s live-wind panel).
Each block below pools every paired hour from every exported station of one type
(MSC GeoMet, Weather Underground, or Tempest). The coloured
line is one combined origin fit measured ≈ m × forecast on the pooled pairs
(zeros excluded from the slope fit).
Networks cover different sites and climates — this is a descriptive comparison, not a strict A/B test.
Hourly UTC bins; newest HRDPS run wins when archives overlap. Mean/gust axes: 0–50 and 0–70 km/h. Direction roses: measured speed mapped to radius with outer ring at 70 km/h. Scatter and rose dot opacity increases from calm toward √(2)×axis (or √(2)×70 km/h on roses) using √(measured²+forecast²).