Weather observing network
A Socratic walk-through of the weather observing network — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why can no single instrument replace the mix of balloons, buoys and satellites behind a forecast?
Satellites see the whole planet, continuously, from above. Given that, why does anyone still launch a balloon twice a day from eight hundred-odd stations, or push four thousand robotic floats around the ocean? It looks like a museum being kept running alongside a modern factory.
The instinct behind the question is that measurement is measurement, so more coverage should simply win. But ask what each instrument actually returns, and the answer stops being about coverage at all.
Reasoning it through
REASONING #Begin with what a forecast model needs. It does not need a picture; it needs a state — temperature, humidity, pressure and wind at every point of a three-dimensional grid, at one moment, everywhere, including over oceans and at every altitude. No instrument delivers that. Every observation is a partial, biased, unevenly spaced sample of it, and the forecast begins by reconciling all of them into a best estimate. That reconciliation step — data assimilation — is the reason the network can be heterogeneous without being incoherent.
Now put each instrument to a simple test: what does it measure directly, and where?
A radiosonde, carried by a balloon, is a thermometer, a hygrometer and a pressure sensor physically inside the air, reporting as it climbs through the troposphere and into the stratosphere. Its vertical detail is excellent. Its coverage is terrible — a few hundred launch sites, mostly on land in the northern hemisphere, mostly at 00 and 12 UTC.
A satellite radiometer measures nothing of the sort. It measures radiance: how much energy arrives at the sensor in particular wavelength bands. Temperature and humidity are then inferred from that radiance, and the inference is a weighted average over a deep layer of atmosphere, not a reading at a level. So a satellite buys global coverage and pays in vertical sharpness — and it needs an independent instrument in the air to check that its inference has not drifted.
Do you see the shape forming? Each instrument's strength is purchased with a specific weakness, and the weaknesses are not the same weakness. That is what makes them worth keeping together.
The ocean makes the point sharply. An infrared satellite sees the top skin of the sea and nothing beneath it. Argo floats — roughly four thousand of them at any time — drift at depth, rise every ten days while profiling temperature and salinity to about two thousand metres, and transmit by satellite before sinking again. Nothing else gives that interior profile at that scale. Moored buoys in the tropical Pacific, by contrast, give something Argo cannot: a fixed point sampled continuously, which is what you need to catch an El Nino developing rather than sampling it in scattered snapshots.
There is a further reason beyond complementarity, and it is quieter. Satellite instruments drift, and they are replaced by successors with different characteristics. What anchors that long record are the in-situ measurements that measure the quantity itself. Radiosondes and ships and buoys are not just extra data; they are the reference against which the inferred data is calibrated. Remove them and the satellites still report — with slowly degrading confidence that nobody can quantify.
One honest caveat: modern forecasts are dominated by satellite data in sheer volume, and the great majority of observations assimilated by a global model are satellite-derived. Specialisation does not mean equal contribution. It means the small in-situ contribution is doing work no volume of radiance can do.
The analogy
THE ANALOGY #Think of a hospital admitting a patient. There is a scan that shows the whole body at once but resolves soft structures poorly; a blood test that reveals chemistry but says nothing about location; a thermometer that measures one number continuously at the bedside. Nobody argues about which one is best, because they answer different questions, and the diagnosis is assembled from all of them.
A clinician reads each result separately and reasons from them, whereas a forecast system fuses everything into one physically consistent estimate first — so an observation there does not merely add evidence, it corrects the model's guess at places no instrument looked at all.
Clarifying the model
THE MODEL #Three refinements sharpen the picture.
First, "redundancy" is the wrong word for what is going on. The instruments are not spare copies of each other; they are non-substitutable. Losing radiosondes would not shrink the network proportionally, it would remove a kind of information — and the effect was visible when the pandemic grounded much of the commercial fleet that carries automated aircraft sensors, measurably degrading forecast skill in some regions.
Second, direct and inferred measurements are genuinely different objects. It is easy to say "the satellite measured the temperature". It did not. It measured radiance and a retrieval turned that into temperature, using assumptions. Modern systems often skip the retrieval entirely and assimilate the radiances themselves, letting the model predict what the satellite should have seen and correcting from the mismatch.
Third, the network is not a fixed design but a negotiated one, sustained by international agreement precisely because the atmosphere does not respect borders. A gap over one ocean degrades forecasts for the continent downwind of it days later.
A picture of it
THE PICTURE #How to readRead left to right for how much of the planet an instrument covers, and bottom to top for how much of the vertical column it resolves. The striking thing is the empty top-right corner: no instrument sits there, and none is coming. The forecast's job is to synthesise a top-right answer out of four instruments that each occupy a different corner, which is exactly why removing any one of them cannot be compensated by more of another.
What became clearer
WHAT CLEARED #The mix is not historical clutter awaiting consolidation. It exists because coverage, vertical resolution, continuity at a point, and directness of measurement are four different properties, and no physical instrument maximises more than two at once. The satellites supply the breadth; the balloons and floats supply the depth and the calibration anchor; the fixed buoys supply the continuous time series. A forecast is the act of reconciling them, and it is only as good as the diversity of what it has to reconcile.
Where to go next
ONWARD #- How data assimilation actually weighs a sparse accurate observation against a dense inferred one.
- Why observing-system experiments are used to price a proposed new instrument before it is built.
- What commercial constellations and radio occultation have changed about the balance.
Key terms
TERMS #| Term | What it means |
|---|---|
| Radiosonde | a balloon-borne instrument package measuring temperature, humidity and pressure directly as it ascends, with wind derived from its tracked position. |
| Data assimilation | the process of combining diverse observations with a short model forecast to produce the best estimate of the atmosphere's current state. |
| Radiance | the energy per wavelength band arriving at a satellite sensor; the raw quantity from which temperature and humidity are inferred. |
| Argo | an international array of about four thousand drifting floats that profile ocean temperature and salinity to roughly 2,000 metres every ten days. |
| In-situ observation | a measurement made by an instrument physically within the medium, as opposed to inferred remotely. |
Every term the collection defines is gathered in the glossary.