Instrument networks
Weather stations: the oldest sensor network on Earth
Weather observation is the oldest continuously operating sensor network in the world, and it is still expanding. The instruments have barely changed in principle since the nineteenth century — what changed is that they now report every few minutes, automatically, to anyone who asks.
What is actually on the mast
A standard automatic weather station is a small cluster of independent instruments sharing one mounting pole and one radio. Each measures a single physical quantity, and each has a failure mode worth knowing about.
- Cup anemometer — three hemispherical cups on a vertical spindle. Rotation rate is linear with wind speed across the usable range, which is why the design has survived since Robinson proposed it in 1846. It under-reads in gusts, because the cups take time to spin up.
- Wind vane — reports direction, usually as a compass bearing. Many networks transmit only 16-point rose values rather than degrees, which quietly caps resolution at 22.5°.
- Tipping-bucket rain gauge — the funnel feeds a tiny see-saw that tips at a fixed volume, typically 0.2 mm. Each tip is one pulse. It systematically under-reads in heavy rain, because water arriving mid-tip is lost.
- Radiation shield — the stack of white louvred plates. Not decoration: it holds the temperature and humidity probe in shade with free airflow. An unshielded thermometer in sunlight can read several degrees high.
- Barometer — usually inside the housing, since pressure passes through the enclosure. Reported either as station pressure or reduced to sea level, and confusing the two is a common source of nonsense readings.
Three tiers of network
Public weather data comes from three very different kinds of operator, and the distinction matters more than the hardware does.
National meteorological services
Every country runs a synoptic network to World Meteorological Organization standards — calibrated instruments, defined siting rules, documented maintenance. Japan's AMeDAS network shows the density this can reach: StarFlash indexes 1,286 AMeDAS stations, spaced roughly 17 km apart, reporting every ten minutes.
These are the observations that feed numerical weather prediction. When a forecast model is initialised, it is initialised largely on this tier.
Research and agency mesonets
Below the national tier sit regional and sector-specific networks: agricultural mesonets, hydrological services, transport authorities running road-weather stations. Professionally maintained, but tuned to a purpose — a road-weather station cares about surface temperature and freezing point in a way a synoptic station does not.
Personal and community stations
The largest tier by count is amateur. Consumer stations costing a few hundred dollars now match professional instruments closely enough for many uses, and their owners publish the output. What they lack is siting discipline — a station on a sun-facing wall or above a driveway reads warm, consistently, and nothing in the data says so.
What the data is used for
Weather observation is unusual among sensor networks in that its economic value is enormous and almost entirely indirect.
- Aviation — METAR reports are generated at airports worldwide on a fixed half-hourly cadence in a terse coded format that has barely changed in decades. Every commercial flight plan depends on them.
- Agriculture — irrigation scheduling, frost warning, spray timing and disease models all run on local temperature and humidity. A frost warning two hours early can save a season's fruit.
- Energy — wind and solar output are forecast from meteorological data, and grid operators trade power on those forecasts. Observation error becomes trading error.
- Insurance — hail, wind and flood claims are validated against nearby observations. Parametric policies pay out automatically on a measured threshold, making the reading itself the contract trigger.
- Hydrology — rainfall drives river models. StarFlash indexes 729 German river gauges from PEGELONLINE and 3,016 French hydrometric stations from Hub'Eau, only useful alongside the rain that fed them.
The siting problem nobody solves
The hardest problem in weather observation is not the instrument, it is where you put it. A thermometer is accurate to a tenth of a degree; a thermometer two metres from a brick wall is accurate to a tenth of a degree about the wall.
This is why professional networks specify exposure — height above ground, distance from obstructions, ground surface — and why amateur data should be treated as a dense but biased field rather than a sparse unbiased one. Used carefully, the density is worth the bias. Used carelessly, you measure car parks.
A reading is only as good as its surroundings. The instrument is the easy part.
Where it is going
Two trends are reshaping the tier structure. Cheap connectivity has pushed station counts up faster than any national programme could, producing sub-kilometre spacing in some cities. At the same time, machine-learning forecast models trained on historical observations have started matching physics-based models on some measures — which raises the value of long, clean, well-documented observation records.
Both point the same way: more stations, reporting more often, with provenance mattering more than it used to.