We’ve got another round of enhancements in ForecastWatch. Login now to begin to use them.
Wind observations for SYNOP stations, stored in the correct units again
Wind speed observations for SYNOP-reporting stations in the United Kingdom, Japan, Spain, French Polynesia, Turkey, and Estonia are now stored in the correct units at collection time. The upstream observation feed began delivering these stations’ wind in knots while labeling the values meters per second in late May 2026, which inflated observed wind by a factor of 1.94 and roughly doubled reported wind error for every provider in the affected subregions. Because the issue affected all providers equally, relative rankings were largely unaffected, but absolute wind MAE and bias in those subregions ran high from late May through August. Wind observations and the scores built on them are correct from this release onward; the affected June, July, and August history will be repaired and restated separately. A new nightly check now watches day-1 wind bias across every subregion and alerts us when it sustains a shift, so an observation-side unit or feed problem of this kind is caught within about a week.
Quality alert highlights point at the values that triggered the flag
The Quality Alerts drawer on the ForecastWatch platform now highlights the forecast pair against which a data-quality flag was actually raised. Spread anomalies, zero sentinels, stuck forecasts, and systemic identical values detected on the MOS (7am-7pm) pair are labeled MOS instead of 24h, so the highlighted cells match the offending values in both the drawer and the Data Issues tab. Providers reviewing an invalidation now land directly on the values that tripped the check.
Observation collection restored for SYNOP-verified countries
Observations for stations verified against SYNOP reports (much of Europe, the UK, Japan, and Turkey, among others) are collecting normally again after an upstream metadata change halted parts of hourly collection from September 1 through 3. Because the affected hourly runs stopped mid-batch, some METAR stations sharing those hours were skipped too, including European METAR coverage and US morning batches. The lost hours backfill automatically after the fix deploys, and daily actuals and accuracy scores for August 31 onward regenerate as the backfilled observations process — no gap remains in the record once that completes.
No-data metrics no longer rank at the 0th percentile
A provider with no forecasts for a metric in the current selection is now shown with no percentile, no tier, and no market rank, and is excluded from the “of N” market pool count. Previously a provider whose feed carried no values for a metric was displayed at the 0th percentile with a Developing tier beside a blank score, which read as worst-in-market rather than not-participating. The change applies across Monthly Insights, Long-Term Trends, Market Overview, the PDF snapshots, and the trend rows in data exports. Percentiles, tiers, and gaps are unchanged for any provider that has data.
Also in this release: the precipitation methodology pages now state precisely how an observed precipitation day is determined outside the US — a day is wet if either the measured hourly amount registers (roughly 0.13mm or more in a single hour) or any present-weather code that hour indicates precipitation, with the two signals combined rather than prioritized. The pages previously described a fallback hierarchy with a 0.1mm threshold scoped to the US and Canada.
