Study 003 · Climate data quality

The hottest place in Europe was a broken sensor.

One community monitor reported 32,783.8°C. The absurdity is obvious; the quieter errors are not. Before heat data meets age, housing or hospitals, it needs a provenance trail and a quality gate.

Published 13 July 202618,102 observationsSensor.Community

The most dangerous number on a climate map may be the one that looks plausible.

DataVault captured 18,102 observations from Sensor.Community across Europe and beyond. Of 8,034 temperature values, 249 sat outside a deliberately broad -40°C to 60°C range. That is 3.1%: small enough to overlook, large enough to distort a map, average or ranking.

18,102sensor observationssingle snapshot
8,034temperature valuesparsed observations
249outside QA range-40°C to 60°C
3.1%flagged implausiblenot silently deleted

The spectacular errors are the easy ones

The unfiltered capture included 32,783.8°C in Germany, 640.1°C in Serbia and 512.2°C in Poland. No analyst would mistake those for ambient air temperature. But a poorly shielded sensor against a sunlit wall can produce a high reading that remains physically possible and therefore slips through a simple range check.

That distinction matters. A quality gate can flag impossible values; it cannot automatically turn an uneven community network into an official meteorological station network.

A clean number is not automatically a representative number.

After the first quality gate

Keeping values between -40°C and 60°C left a snapshot-wide mean of 29.8°C and a maximum of 58°C. Germany contributed 3,065 accepted temperature values and 76 rejected ones; the Netherlands 1,161 and 59; Poland 556 and 18; Belgium 356 and three.

These counts expose another source of bias: coverage is not uniform. Germany had 7,097 total observations in the capture, while Belgium had 755. A European map coloured without accounting for density would partly be a map of volunteer participation.

The reason to clean the thermometer is the people beside it

Heat becomes a social risk when it is joined to age, health, housing quality, green space, night-time cooling and access to care. The European Environment Agency has shown why vulnerability and exposure must be considered together, especially in urban heat islands.

Those joins are powerful, but only if each layer carries its uncertainty. A community sensor can provide valuable local texture. It should not silently impersonate an official station or a population-weighted regional estimate.

DataVault quality principleKeep the original observation, the quality flag, the source, the collection time and the transform. Analysts should be able to reproduce both the raw and filtered result.

Analytical plate

Coverage and quality are part of the finding

Observation counts reflect this one community-network snapshot, not national weather coverage.

Observations by country

Largest contributors in the capture

Germany
7,097
Netherlands
2,857
Poland
1,260
Belgium
755

Responsible join path

Raw evidence remains available at every step

sensorrange flagstation contextneighbourhoodage / housing / health

Evidence ledger

A snapshot with a fingerprint

Sensor.Community current measurements

18,102 observations; 8.76 MB; raw values retained with country and sensor identifiers.

27547a51…0db

Immutable content reference

The artifact can be checked against its stored capture before any filtering.

bef53114…456d

Method & limits

Flag first. Interpret second.

We selected temperature values from one current snapshot and applied a broad plausibility gate of -40°C to 60°C. Rejected values remain evidence; they are not rewritten. Country counts describe the source network at that time.

What this cannot showThe filtered mean is not Europe’s mean temperature. Sensors are community-operated, unevenly located and may differ in shielding, height or calibration. Even a value inside the range can be unsuitable for climate or health inference.