The only layer that shows nothing observed
What CAMS draws is a computation of how aerosols and trace substances will be distributed in the atmosphere, fed by emission estimates, weather fields and assimilated satellite observations, but not itself measured.
samoEKO therefore places it in a layer group of its own, “forecast”, with the model caveat in its description and in a colour language that cannot be confused with any measurement layer. A forecast that looks like a measurement is a source of error in an operations centre.
The samoEKO emission balance has nothing to do with CAMS: it computes from mapped burnt area, vegetation type and biomass following the IPCC guidelines and is a separate, fully disclosed calculation.
Particulate matter and why the sea misleads
The European particulate matter layer computes near-surface PM2.5 loading every hour. Over the Adriatic it regularly shows high values. That is physically correct and, in health terms, something entirely different: it is mostly natural sea salt aerosol from spray, together with the absence of deposition over water and, seasonally, Saharan dust.
The same numerical value over Kakanj is combustion loading; over open water it is not. The layer carries this caveat in the legend, without it, it would be the best example of a correct number carrying a false statement.
The right quantity for fire smoke
For fire smoke the narrower quantity sits beside it: the aerosol optical depth of biomass burning at 550 nanometres. It contains exactly the biomass burning contribution, without the sea salt and industrial share and reaches 120 hours ahead.
One detail governs how it must be read: the display style starts at 0.1. An empty tile here means “no smoke”, not “no data”. The reverse confusion would be the more dangerous one, which is why it is stated here.
Limits
Roughly forty kilometres of grid width: a valley basin such as Kakanj is a fraction of one pixel within it. Anyone needing a statement about the air at a specific location will not get it from CAMS, that is what ground measurement is for.
According to the service, the smoke layer is published only around 20:00 UTC from the midnight run; between midnight and 20:00 it therefore shows the previous day's run. Irrelevant for a five-day forecast, but the time stamp on the layer has to be read.
Access runs through an open token of the ECMWF service. It is not contractually guaranteed; should it be throttled, a dedicated account with the Atmosphere Data Store is the robust but more laborious route.