Reading environmental data
Measured or modelled?
A measured number came from an instrument pointed at the thing. A modelled number came from a calculation about the thing. Both can be right and both can be wrong, but they fail in different ways and they cannot be compared as though they were the same kind of claim.
How each one fails
The distinction is useful because it tells you what to be suspicious of.
A measurement fails at the instrument. It can be miscalibrated, saturated, obstructed, or reporting its own floor rather than a real value. It describes one point in space at one moment, and nothing else. When CPCB's dissolved-oxygen figure reads 0.3 mg/L at four Delhi stations, that is the floor the method reports — the true value is at or below it, and printing "0.0" would be an invention.
A model fails at its assumptions. A source-apportionment split, an emissions inventory, a death toll attributed to a pollutant and a satellite product that infers a quantity from radiance are all model outputs. They can cover a whole country and a whole year, which no instrument can, and they inherit every uncertainty in their inputs.
Neither is superior. A national figure for anything is almost always modelled, because measuring a nation is not possible. A street-level figure is almost always measured, because modelling a street is not reliable. Asking which one you are holding tells you what it can be used for.
- Counted or measured
- Produced by an instrument or a direct count.
- Modelled
- Produced by calculation from other quantities.
- Reanalysis
- A model that assimilates historical observations to produce a consistent gridded record. Not a measurement, and not a free invention either — the honest description is modelled.
Which of ours are which
Five readings, and what produced each one.
Air readings — CPCB station monitors
Measured
Instruments at named stations
River readings — CPCB grab samples
Measured
Laboratory analysis of collected samples
Fire detections — VIIRS
Counted
Hot pixels on a satellite pass
Heat and rainfall series — ERA5
Modelled
ECMWF reanalysis, not station observation
Source apportionment shares
Modelled
Receptor or dispersion modelling
There is no limit on this page to break. The rule here is editorial, not legal: if you quote a number from this site, quote the kind with it. A modelled figure presented as a measurement is the most common way an honest number becomes a false claim.
Applying the test
Four questions that separate the two in almost every case.
What produced this number? An instrument, a count, or a calculation. If a source does not say, treat it as modelled until it does.
What would have to be true for it to be wrong? For a measurement: a broken sensor, a bad sample, a saturated reading. For a model: a wrong assumption applied uniformly, which is far harder to notice.
What is its coverage? Measurements are sparse and precise. Models are complete and uncertain. A figure that is both complete and precise is worth a second look.
Is it being compared with the other kind? A modelled national estimate set against a measured city reading is not a comparison, even when both are correct.
What it does not measure
- The distinction cannot tell you which number is closer to the truth. It tells you how each one is likely to be wrong.
- It cannot be applied to a figure whose provenance is not published — which is itself the finding.
- It does not settle disagreements between two models, which is a separate problem: see source apportionment.
- It says nothing about whether the thing measured was the right thing to measure.
Where this comes from
Every figure above comes from one of these.
- National Water Quality Monitoring Programme — river data Central Pollution Control Board · A measured series with a stated reporting floor
- ERA5 reanalysis archive ECMWF, served by Open-Meteo · A modelled series covering thirty-five years and fourteen stations
- FIRMS active fire data NASA · A counted series whose unit is a detection, not a fire
Reuse freely — CC BY 4.0. Each source above keeps its own terms.