Every observation has to earn trust.
A physical sensor network faces bad siting, dirt, component drift, connectivity gaps and genuine microclimate variation — often at the same station, on the same day. WeatherXM does not treat every received packet as equally useful. It combines automated checks, deployment assessment and published scientific investigation.
Where did this observation come from?
Device identity and signing, where the hardware supports it, establish which station produced a reading and whether it was altered on the way to us. Location mechanisms establish where that station actually is.
That is provenance, and it is the foundation everything else sits on. It says nothing about whether the station is well sited or reading true. Origin is not quality, and we keep the two words apart everywhere on this site.
What we check, and why.
A station can fail in more ways than most people expect, and almost none of them announce themselves. A sensor drifts a degree a year. A rain funnel silts up over a summer. A pole leans after a storm. A perfectly good instrument gets mounted against a south-facing wall. Every one of those produces data that looks fine in isolation.
So we check five things, continuously, across the whole network:
Is the reading physically possible?
Values are tested against what the sensor can actually produce. A temperature outside the sensor's own rated range is not a cold snap; it is a fault.
Does the series behave like weather?
Real weather moves in particular ways. We look for readings that sit unnaturally still, jump further in a minute than the atmosphere does, or stop arriving altogether — each of which points at a different failure.
Is the station really where and how it should be?
Patterns in the data itself reveal a station that is indoors, enclosed, or on a leaning pole, even when it is registered as a normal outdoor installation.
What does the site look like?
Owners send photographs of the installation from each direction, and we assess them against recognized siting guidance — separately for temperature, wind, rain and solar, because a position that is fine for one is often wrong for another.
Is anything in the way?
Obstructions cast a signature. A wall or a tree changes how solar irradiance rises and falls through the day, and it changes how wind behaves from particular directions. Both are detectable in the measurements, and both are compared against nearby stations and against certified reference networks where one is close enough to be meaningful.
Results feed a station's quality assessment, which is visible to the owner and carried alongside the data — and which also affects what that station earns from the network. An owner with a poorly sited station finds out, and has a reason to fix it.
We publish the research, including the unflattering parts.
Automated checks are one kind of evidence. Controlled comparison is another, and we publish it: WeatherXM stations set against reference networks, an experiment on what a nearby wall does to temperature and wind, and an investigation into a temperature offset across one of our own station fleets.
A published comparison is evidence from a particular study under particular conditions. It is not the same thing as the checks running in production, and we do not present it as such.
Side-by-Side Reference Testing
Stations undergo parallel testing against certified meteorological reference instruments to establish empirical error distributions, response times during frontal passages, and aspiration shield efficiency across intense solar loads.
Read the quality signal in context.
Use the available quality information alongside station metadata, data freshness and record length. A quality score summarizes an assessment — it is not a statement of sensor accuracy and it is not a certification. Our methods draw on established observing guidance and on the measured behavior of our own network.
Know the limits of the measurement.
- ▪ Wind — measurement height and nearby obstructions change both speed and direction.
- ▪ Rain — a compact collector's exposure, cleanliness and the weather itself all affect the record.
- ▪ Temperature — shielding, ventilation and nearby radiating surfaces matter more than most people expect.
- ▪ Pressure — measurement uncertainty and the reduction method both affect the published value.
- ▪ Cold — icing and power conditions interrupt measurement.
- ▪ Coverage — a sparse region gives you less redundancy to cross-check against.
For a professional application, agree the variables, period, availability and acceptance criteria up front. That converts a general quality claim into an assessment of the data you are actually going to use.