What is the score made of?
B represents enough sea sparkle in coastal water. A describes transport and accumulation at the beach. V describes whether darkness, ambient light and wave activity make the glow visible. Reports and media attention do not change this ecological chance.
What did the latest test on past nights show?
In 43 historical comparisons, the current score ranked a reported night higher in 76,0% of cases than its matched comparison night. The previous version reached 72,9% on exactly the same comparisons. For a separate assessment of evening, night, tomorrow and the following day, only 8 positive groups with archived model runs are available.
This is an improvement in ranking, but not yet a complete check of the percentages. Not every comparison night has been proven free of sea sparkle, and some historical wave data were calculated only afterwards.
What makes this approach different?
- A separate landward orientation for each coastal segment instead of one direction for the whole country.
- Up to 72 hours of preceding current, wind and wave context.
- Block validation by year and coastal region, never a random row split.
- Open reporting of negative results, uncertainty and data freshness.
- Voluntary observations are used for research only with separate consent.
How do we test new observations?
Every live forecast is stored immutably in advance and then matched with verified sightings and usable non-detections. Information that became available later therefore cannot make the forecast appear better unnoticed.
- The primary outcome is visible blue light per coastal segment and hour.
- Multiple reports from the same beach night form one ecological episode.
- We collect observations at low, medium and high forecast chances across different coastal regions.
- Years, coastal regions and episodes remain separate during a test.
- We call the model calibrated only after it passes the calibration test described in advance.