Free planning tool
How many earthquake sensors does your area actually need?
Place sensors on the map, drop an earthquake on it, and see whether the network would have detected it — or stayed silent. Nothing is saved, and you do not need an account.
Scenarios
Two layouts, same five sensors:
Or replay a real earthquake. Your sensors move there if they are far away, so you can see the result straight away.
Your saved plans:
On the map
Warning time
Place your location above and this fills in.
Drag the blue pin to where you actually are — warning time depends entirely on distance.
Place at least one sensor
Click anywhere on the map. Then set the magnitude and see what the network would do.
Earthquake
Your sensors — 0
No sensors yet.
Finds the arrangement that covers the most ground for M4.0 events — a network is designed around the smallest earthquake you want to catch; bigger ones come free. Turn on Past earthquakes first and it will favour the places that actually shake.
Takes the network you have already placed and finds the single spot that unlocks the most new ground.
can be located by three or more of your sensors — the only area where an earthquake would actually be published. Zoom does not change this number.
What this plan costs
Shipping and taxes included, worldwide.
Assembling them yourself? The bare board is $99.99. Planning more than 500? Talk to us.
Why three sensors, and not one
GeoShake does not publish an earthquake because a single sensor shook. Shaking has many causes — a lorry, a slammed door, someone moving the shelf the sensor sits on. The network only publishes an event when at least three sensors register it within seconds of each other, with arrival times consistent with a P-wave travelling at 6 km/s.
That one rule is why placement matters more than count. Five sensors clustered in a valley behave completely differently from five spread across a region — same hardware, same money, very different networks. Try the two examples above.
What the tool assumes
Detection range grows with magnitude. The tool uses:
- M3.0 — about 16 km
- M4.0 — about 47 km
- M5.0 — about 121 km
- M6.0 — about 305 km
- M6.3 and above — capped at 400 km
Those figures are for a shallow earthquake, around ten kilometres down, which is where most damaging crustal earthquakes happen. A deeper earthquake of the same size is felt less strongly at the surface and the range shrinks: an M5.0 reaches about 116 km at 35 km depth and about 99 km at 70 km. Far below that the shaking never reaches the surface strongly enough for any sensor to separate it from ordinary movement, and the tool says so by giving no range at all.
The cap matters: the bare formula would put an M8.8 at over 2,000 km, which is nonsense. Very distant large earthquakes arrive as long, slow ground motion that a consumer accelerometer cannot separate from ordinary building movement.
The curve began as an engineering estimate. It is now built from real earthquakes. Every catalogued earthquake that happens near one of our sensors is recorded along with what that sensor actually did, and the table below is that record in full. It is not a selection of the good days. On 13 September 2026 the record was large enough to replace the original guess: the ranges above now come from the smallest ground movement our sensors are measured to trigger on, carried through the same attenuation relation that estimates magnitude.
What our sensors have actually recorded
When this section was updated on 2 November 2026, the network had 130 stations, 121 of them reporting in the last day. Between 8 September, when we lowered the trigger threshold across the fleet, and 1 November, a catalogued earthquake landed close enough to a live sensor to be worth scoring 372 times, covering 133 earthquakes from M1.6 to M5.5 and 65 sensors. Only nine of those cases date from 13 September or earlier, when the curve was fitted, so the rest test it rather than shape it. Here is every one of them, counted by magnitude.
| Earthquake | Tool said: in range | Tool said: out of range | ||
|---|---|---|---|---|
| Triggered | Silent | Triggered | Silent | |
| Under M3 | 0 | 12 | 1 | 141 |
| M3 to M3.5 | 5 | 7 | 4 | 70 |
| M3.5 to M4 | 5 | 2 | 14 | 35 |
| M4 to M4.5 | 1 | 0 | 3 | 30 |
| M4.5 and over | 7 | 5 | 5 | 25 |
| All | 18 | 26 | 27 | 301 |
One case is one catalogued earthquake and one live sensor near enough to score, roughly within three times its detection range. Each band includes its lower bound but not its upper one. Green counts are where the tool was right, red where it was wrong. Sensors that were offline at the time are left out (30 cases), and so is everything before 8 September, when the sensors used a less sensitive trigger.
Overall, the tool’s yes or no matched what the sensor did in 319 of 372 cases. Most of those are the easy ones, though: a small or distant earthquake that the tool says no sensor will hear, and none did. The harder test is when the tool says a sensor should register an earthquake. That happened 44 times, and the sensor triggered in 18.
The tool is too optimistic in two places. The first is small earthquakes close to a sensor: 19 of the 26 silent cases were below M3.5. The second is the outer edge of a circle: the five silent cases at M4.5 and over all sat between 73% and 92% of the way out to the edge. In the other direction it is sometimes too cautious: sensors triggered 27 times on earthquakes it placed out of range. Eleven of those were deep, 43 to 109 km down, and most of the rest were within about half a radius of the edge. All but two of those triggers were within 10% of the threshold, so some may be ordinary noise that happened to coincide with the earthquake.
In practice, treat the edge of a circle as uncertain, and plan so that the places you care about sit well inside the circles of several sensors.
The curve itself is not a guess with a slope picked to look sensible. It is the point where the shaking from an earthquake of a given size and depth falls to the smallest movement our sensors actually trigger on, and that last number is measured. It also means the ranges move when we change the trigger, and when they do, this page changes with them.
The model also ignores depth, soil conditions and directionality, all of which matter in reality. It is a planning aid for geometry, not a hazard assessment. And it is deliberately optimistic about warning time: it assumes zero processing and network delay, so treat any figure it gives as a ceiling.
Privacy
Your sensor positions stay in your browser. We do not store them, and there is no account or cookie behind this page. We do count how many people plan a network in each region, rounded to a roughly 10 km grid before anything is written down — enough to know where coverage is wanted, never enough to locate anyone.
If the answer surprised you
Most people find they need more sensors than they expected, arranged more tightly than they expected. That is the honest shape of the problem: density buys warning time, and a thin network buys very little.
If you are planning coverage for a town, a campus or a region and want a second pair of eyes on it, write to us. We will model it properly — what your layout would catch, what it would miss, and whether a different arrangement of the same budget does better. No charge, no obligation.