Plate Tectonics Explained: How Continents Move and Why Quakes Happen

Continents move because Earth's rigid outer shell, the lithosphere, is broken into plates that ride on a hotter, slowly flowing layer of mantle beneath them. The plates move a few centimetres a year, pulled mainly by old, cold seafloor sinking back into the mantle at subduction zones. Earthquakes happen where plate edges lock together, store elastic strain for decades or centuries, and then slip in seconds; volcanoes form where plates pull apart or where one dives beneath another and melts the mantle above it.

This post goes one level deeper than our explainer on Earth's layers, from crust to inner core, which describes what the planet is made of, and sits alongside our guide to the layers of the atmosphere. Here we follow the evidence for plate tectonics, name the plates and boundaries, work through the numbers for plate speed, magnitude, tsunami speed and warning time, and look at the networks that watch the plates in real time, including the US ShakeAlert early warning system and how its software is protected.

A note on scope: Safeguard is a software supply chain security company and does not do geoscience. The geological facts come from the US Geological Survey (USGS), NOAA, NASA, the EarthScope Consortium and peer reviewed papers; arithmetic we add is labelled as a worked example.

Key takeaways

• The lithosphere (crust plus the rigid top of the mantle) averages at least 80 km thick, says the USGS, and is broken into plates: 14 large and 38 small in the PB2002 model, 25 covering 97 percent of the surface in the MORVEL model.
• A young ocean floor, magnetic reversals recorded in rock, seafloor spreading and earthquake belts along trenches and ridges turned continental drift into plate tectonics in the 1950s and 1960s. GPS later confirmed the motion directly.
• Plates move at roughly 2.5 cm a year at the Mid-Atlantic Ridge and more than 15 cm a year at the fastest part of the East Pacific Rise. A useful rule: 1 mm per year equals 1 km per million years.
• Moment magnitude measures the size of the fault slip. Each whole step is about 32 times more energy, so an M7 releases about 32 times the energy of an M6 and about 1,000 times that of an M5.
• Subduction zones host the great megathrust earthquakes (all four of NOAA's example tsunami earthquakes were magnitude 9.0 to 9.2) and explosive arc volcanoes.
• ShakeAlert detects the fast P waves and races the slower, more damaging S waves; where stations are dense, the USGS says the first alert is routinely out less than 5 seconds after the quake begins.
• Early warning is software as much as sensors: the USGS describes federal security accreditation, encrypted and authenticated alert feeds, configuration checks every 5 minutes and replay testing before code reaches production.

A dated timeline: from drifting continents to real time warning

Dates below come from the USGS publication This Dynamic Earth, the USGS ShakeAlert technical implementation plan, NOAA and the cited papers.

What evidence led from continental drift to plate tectonics?

Wegener's case, and why it stalled

The idea is older than geology. The USGS notes that the Dutch mapmaker Abraham Ortelius suggested in 1596 that the Americas had been torn away from Europe and Africa. The first full scientific theory came in 1912, when Alfred Wegener, then a 32 year old German meteorologist, published two articles arguing that a supercontinent he called Pangaea began breaking apart around 200 million years ago.

His evidence: the matching coastlines of South America and Africa, the same unusual rocks and fossils on both sides of the Atlantic, and coal from tropical plants in Antarctica, which suggested that land had once sat closer to the equator.

What he could not supply was a mechanism. Nobody could explain what force would push continents through solid ocean floor, and the USGS says continental drift was "largely dismissed" for decades after Wegener's death. The revival came from the seafloor, not from the continents.

Mapping the ocean floor

Echo sounders, developed after the First World War, showed the Atlantic floor was far more rugged than assumed. In 1947, seismologists on the US research ship Atlantis found that the sediment on the Atlantic floor was much thinner than expected. If the oceans were billions of years old, the sediment should have been thick. It was not.

Surveys in the 1950s then revealed the global mid-ocean ridge, which the USGS describes as more than 50,000 km long, in places more than 800 km across, rising about 4,500 m above the seafloor and virtually encircling the Earth.

Magnetic striping and the reversal test

Ocean floor basalt contains magnetite, a strongly magnetic mineral, and when lava cools the grains lock in the direction of Earth's magnetic field at that moment. Paleomagnetists such as Bernard Brunhes (1906) and Motonari Matuyama (1920s) had already found rocks with "normal" and "reversed" polarity, so the field must have flipped many times.

In the 1950s, magnetometers adapted from wartime submarine detection were towed over the oceans. They found stripes of normal and reversed magnetism running parallel to the ridges, laid out symmetrically on either side like a barcode. In 1963, Frederick Vine and Drummond Matthews, writing in Nature, and Lawrence Morley working independently, proposed that the stripes record new crust forming at the ridge and spreading away while the field reversed. A team of USGS scientists (Allan Cox, Richard Doell and Brent Dalrymple) used potassium argon dating of continental lava flows to build a dated history of reversals over the past 4 million years. In 1966 that history was matched against the seafloor stripes. The fit was close enough that, as the USGS puts it, the calculated and observed profiles "provided one of the clinching arguments."

Seafloor spreading, drilling and the earthquake belts

Harry Hess, a Princeton geologist, reasoned that if crust is created at ridges it must be destroyed somewhere, or Earth would grow. His conveyor belt carries new crust away from the ridges and down into the trenches that ring the Pacific, which explained the thin sediment and why ocean rocks are much younger than continental ones. Robert Dietz of the US Coast and Geodetic Survey coined the term "seafloor spreading."

Two more lines of evidence closed the case. In 1968 the drilling ship Glomar Challenger cored the seafloor across the Mid-Atlantic Ridge between South America and Africa, and the rock ages increased with distance from the ridge. Meanwhile, the Worldwide Standardized Seismograph Network, set up in the 1960s to monitor the 1963 treaty banning above ground nuclear tests, gave seismologists far better data. It showed earthquakes concentrated along ridges and trenches, and inclined zones of earthquakes, now called Wadati Benioff zones, dipping 40 to 60 degrees beneath trenches and extending several hundred kilometres into the Earth. Those zones trace the sinking slabs Hess had predicted.

What is a plate, physically, and what drives it?

A plate is not the crust alone. The USGS gives oceanic crust about 5 km of thickness and continental crust about 30 km on average. The upper part of the mantle is cooler and more rigid than the mantle below, and together with the crust it forms the lithosphere, which the USGS says averages at least 80 km thick over much of the Earth and is thinnest under the oceans and in volcanically active continental areas. Below it lies the asthenosphere, a narrow zone of hot, semi-solid mantle that can soften and flow over geological time. The rigid lithosphere moves over it.

What moves it? The USGS summary in This Dynamic Earth says scientists generally agree with Hess that the driving force is the slow movement of hot, softened mantle below the plates, a form of convection. It adds that the gravity driven sinking of cold, dense oceanic slabs at subduction zones, called slab pull, which drags the rest of the plate along, is now considered the main driving force, and that some geologists argue magma intruding at ridges adds a smaller "ridge push." How convection organises itself in detail remains, in the USGS's word, a mystery.

What are the main plates and boundary types?

The major plates

There is no single official count. Two peer reviewed models are widely used:

• PB2002 (Peter Bird, Geochemistry, Geophysics, Geosystems, 2003) maps 52 plates: the 14 large plates whose motion was described by the older NUVEL-1A model, plus 38 small plates.
• MORVEL (DeMets, Gordon and Argus, Geophysical Journal International, 2010) describes the current motion of 25 plates that cover 97 percent of Earth's surface. The motions of 19 plates bordered by mid-ocean ridges come mainly from spreading rates and fault directions; six smaller plates are tied in using GPS station velocities.

The 14 large plates in Bird's list are the African, Antarctic, Arabian, Australian, Caribbean, Cocos, Eurasian, Indian, Juan de Fuca, Nazca, North American, Pacific, Philippine Sea and South American plates. The 38 small ones include the Okhotsk, Amur, Sunda, Mariana, Tonga and Easter plates. The large plates are much the same in every model; the small plates are where models differ.

Divergent boundaries: where plates pull apart

At a divergent boundary, plates move apart and magma rises to make new crust. The Mid-Atlantic Ridge is the classic example. The USGS gives its average spreading rate as about 2.5 cm a year. Iceland straddles the ridge, so you can watch spreading on land: around Krafla volcano, rifting episodes between 1975 and 1984 opened the ground by a total of about 7 m.

Divergence also happens inside continents: the USGS describes the Red Sea, where Arabia has been torn from Africa, and the East African Rift, where a new spreading centre may be developing.

Convergent boundaries: where plates collide

Earth's size has not changed much in at least 600 million years, so crust must be destroyed as fast as it is made. That happens at convergent boundaries, in three combinations:

• Oceanic under continental. The oceanic Nazca Plate dives under South America at the Peru Chile trench, lifting the Andes. Deep trenches like this are 8 to 10 km deep.
• Oceanic under oceanic. The fast moving Pacific Plate converges with the slower Philippine Sea Plate at the Mariana Trench. The USGS notes the Challenger Deep there is nearly 11,000 m deep, deeper than Mount Everest is tall. Volcanoes built from the melting above such slabs form curved chains of islands called island arcs, such as the Marianas and the Aleutians.
• Continental against continental. Continental crust is too buoyant to sink far, so the plates crumple. The USGS says India's collision with Asia about 50 million years ago pushed up the Himalaya and the Tibetan Plateau, with most of the growth in the past 10 million years.

Transform boundaries: where plates slide past each other

At a transform boundary, crust is neither made nor destroyed; plates grind sideways. Most transform faults are short offsets on the ocean floor, but the San Andreas fault zone runs on land for about 1,300 km through California. The USGS says the Pacific Plate has been grinding northwest past the North American Plate along it for 10 million years at an average of about 5 cm a year.

Boundary zones and hotspots

The USGS adds a fourth category, plate boundary zones: broad belts where the boundary is not well defined. Some volcanism happens far from any boundary. Hawaii sits above a hotspot. The islands get older to the northwest: the USGS gives rocks on Kauai an age of about 5.5 million years, while the Island of Hawaii is still erupting. The chain continues underwater as the Hawaiian Ridge and Emperor Seamount Chain, a track left by the Pacific Plate moving over the hotspot.

Boundary types compared

How fast do plates move? Worked examples

Plate speeds are quoted in millimetres or centimetres a year; geological time in millions of years. One identity links them:

1 mm per year × 1,000,000 years = 1,000,000 mm = 1 km.

So a speed in millimetres per year is the same number in kilometres per million years.

Worked example 1: the Atlantic

The USGS gives the Mid-Atlantic Ridge an average spreading rate of about 2.5 cm a year.

• 2.5 cm per year = 25 mm per year = 25 km per million years.
• The Atlantic began opening roughly 200 million years ago (the USGS gives a range of 100 to 200 million years for its growth from a narrow inlet).
• 25 km per million years × 200 million years = 5,000 km.

Thousands of kilometres of ocean, from a rate that adds up to about 2 m over an 80 year human life.

Worked example 2: the San Andreas

At the USGS's long term average of about 5 cm a year for 10 million years:

• 5 cm per year = 50 km per million years.
• 50 km × 10 = 500 km of total offset.

The EarthScope Consortium gives a lower figure for the fault itself, roughly 30 mm a year. The two numbers describe different things: the USGS figure is a long term average for the plate boundary, and the boundary in California is a system of faults, not one. The USGS notes, for example, that much of the nearby Calaveras fault creeps at 5 to 6 mm a year.

Worked example 3: measuring Iceland splitting with two GPS stations

NASA's Jet Propulsion Laboratory (JPL) publishes velocities for its GNSS stations. In its table (reference frame IGS14, reference epoch 1 January 2026), two Icelandic stations sit on opposite sides of the spreading ridge:

Subtract one from the other to get relative motion:

• East: 13.896 minus (minus 9.692) = 23.588 mm per year apart.
• North: 14.977 minus 19.834 = minus 4.857 mm per year.
• Combined: square root of (23.588² + 4.857²) = square root of (556.4 + 23.6) = square root of 580.0 = about 24.1 mm per year.

That is about 2.4 cm a year of opening, matching the USGS's 2.5 cm a year from magnetic stripes that record millions of years. The geological and satellite measurements agree to within a millimetre a year. HOFN also shows a large vertical velocity in the table; that vertical signal is not part of the horizontal spreading calculation.

Worked example 4: how fast Hawaii is moving

JPL's station MKEA on Mauna Kea moves 35.203 mm per year north and 62.899 mm per year west in the same frame.

• Horizontal speed = square root of (35.203² + 62.899²) = square root of (1,239.3 + 3,956.3) = square root of 5,195.5 = about 72.1 mm per year, toward the northwest.
• That is 72 km per million years, or roughly 400 km over the 5.5 million years since Kauai's rocks formed.

These are velocities in a global reference frame, not relative to a neighbouring plate.

How do earthquakes release strain?

Elastic rebound in plain terms

Plates move steadily, but faults do not. Friction locks the two sides together while the plates keep moving, so the rock near the fault bends like a stretched spring. When the stress exceeds the strength of the locked patch, the fault slips suddenly and the rock snaps back, sending out seismic waves.

The USGS describes both behaviours on the San Andreas. Some segments creep, slipping steadily with many small to moderate earthquakes that cause little damage. Others lock for long periods and then lurch; the USGS notes that the great 1906 San Francisco earthquake ruptured a previously locked segment about 430 km long, from Cape Mendocino south to San Juan Bautista. Subduction zones behave the same way: the USGS describes the deepest part of the Nazca Plate breaking into pieces that lock for long periods before moving suddenly to generate large earthquakes, sometimes lifting the land by a few metres.

How a rupture grows

A rupture starts at a point called the hypocentre and spreads along the fault. The USGS ShakeAlert implementation plan says the rupture front races along the fault at about 3 km per second (2 miles per second), can progress irregularly as it meets strong and weak patches, and can jump from one fault to another. For warning systems, that means an alert issued in the first seconds describes an earthquake that may still be growing.

Seismic waves

The rupture sends out several kinds of wave. The ShakeAlert fact sheet describes the two that matter for warning: the P wave (primary), a compressional wave that travels fastest and arrives first, and the S wave (secondary), a slower shear wave that usually does more damage. Early warning lives in the gap between P and S arrivals.

What do magnitude scales measure?

Magnitude is not intensity

Magnitude describes the size of the earthquake at its source, one number per earthquake. Intensity describes the shaking at a particular place, so one earthquake has many intensities. ShakeAlert publishes its shaking estimates on the Modified Mercalli Intensity (MMI) scale.

From Richter to moment magnitude

The original Richter scale (local magnitude, ML) and the surface wave magnitude (MS) are based on the amplitude of particular seismic waves. Hanks and Kanamori showed in 1979 that these older scales line up with a moment based scale only over limited ranges, roughly ML 3 to 7 and MS 5 to 7.5. For the largest earthquakes, seismologists use moment magnitude, which is tied directly to the physics of fault slip.

Moment magnitude is built on the seismic moment, M₀. By the standard definition used in seismology, seismic moment equals the rigidity of the rock × the fault area × the average slip. The Hanks and Kanamori formula is:

M = ⅔ log₁₀ M₀ minus 10.7, with M₀ in dyne centimetres (1 newton metre = 10⁷ dyne centimetres).

NOAA's Science On a Sphere team summarises what this means for energy: each whole number step in moment magnitude is about 32 times more energy released, and two steps is about 1,000 times.

Worked example 1: M6 versus M7

Rearranging the Hanks and Kanamori formula, log₁₀ M₀ = 1.5 × (M + 10.7). The 10.7 cancels when you compare two earthquakes, so:

• Ratio of seismic moment = 10^(1.5 × difference in magnitude).
• M7 versus M6: 10^1.5 = 31.6, which is the "about 32 times" figure.
• M8 versus M6: 10^3 = 1,000.
• M9 versus M6: 10^4.5 = about 31,600.

Radiated energy scales in roughly the same proportion, which is why NOAA can say an M7 releases about 32 times the energy of an M6. A useful consequence: an M7 is not "a bit bigger" than an M6. It would take about 32 M6 earthquakes to release the energy of one M7.

Worked example 2: from a fault to a magnitude

Take an illustrative fault patch 100 km long and 30 km deep (down dip) that slips an average of 2 m. Assume a typical crustal rigidity of 3 × 10¹⁰ pascals (our assumption, a common textbook value).

• Area = 100,000 m × 30,000 m = 3.0 × 10⁹ m².
• M₀ = 3 × 10¹⁰ × 3.0 × 10⁹ × 2 = 1.8 × 10²⁰ newton metres = 1.8 × 10²⁷ dyne centimetres.
• log₁₀ (1.8 × 10²⁷) = 27.26.
• M = ⅔ × 27.26 minus 10.7 = 18.17 minus 10.7 = 7.47, so about M7.5.

Double the slip to 4 m and M₀ doubles, but the magnitude rises by only ⅔ × log₁₀ 2 = 0.2, to about M7.7. Reaching magnitude 9 takes a far larger fault area and slip, which is why the M9 events in NOAA's list below are all subduction megathrusts.

Worked example 3: Tohoku 2011 against an M6

The USGS catalogue lists the 11 March 2011 Great Tohoku Earthquake as moment magnitude 9.1 (type "mww"), at a depth of 29 km, origin time 05:46:24 UTC.

• log₁₀ M₀ for M9.1 = 1.5 × (9.1 + 10.7) = 1.5 × 19.8 = 29.7, so M₀ = about 5 × 10²⁹ dyne centimetres, or 5 × 10²² newton metres.
• For M6.0: 1.5 × 16.7 = 25.05, so M₀ = about 1.1 × 10²⁵ dyne centimetres.
• Ratio = 10^(29.7 minus 25.05) = 10^4.65 = about 45,000.

One M9.1 released roughly the seismic moment of 45,000 M6 earthquakes; an M9 versus an M6 is 10^4.5, about 31,600.

How are subduction zones linked to volcanoes and tsunamis?

Why subduction makes volcanoes

The USGS notes that scientists still debate the exact source of subduction zone magma, whether partial melting of the sinking slab or of the mantle above it, but the link to subduction is clear. The volcanoes line up above the sinking slab, parallel to the trench: the Andes above the Nazca Plate, the Cascade Range above the Juan de Fuca Plate, and island arcs such as the Marianas and the Aleutians above oceanic slabs. The USGS calls the rim of the Pacific, ringed by subduction zones, the Ring of Fire, the most seismically and volcanically active zone on Earth.

The USGS explains that subduction zone volcanoes such as Mount St. Helens in Washington and Mount Pinatubo in the Philippines typically erupt explosively, because their magma is too stiff to let volcanic gases escape easily; pressure builds until it is released violently.

Why subduction makes the largest earthquakes

The USGS ShakeAlert plan describes three earthquake types in a subduction zone such as Cascadia: crustal earthquakes in the overriding plate; megathrust events on the plate interface, which can reach M9.0 or larger; and intraslab events inside the sinking plate as it bends and breaks. Intraslab quakes are deeper and usually shake less than comparable interface or crustal quakes, but they can still cause damage under cities; the plan cites the 2001 M6.8 Nisqually earthquake near Seattle as an example.

Subduction quakes can also be extremely deep: the USGS describes a magnitude 8.3 earthquake beneath Bolivia on 9 June 1994 at a depth of 636 km, in the subduction zone between the Nazca and South American plates.

How a megathrust makes a tsunami

When a megathrust slips, the seafloor above it moves suddenly and displaces the water column above it; on land, the USGS notes, such quakes can lift the ground by as much as a few metres. NOAA notes that earthquake generated tsunamis originate mainly in subduction zones, and that landslides, volcanic activity and even atmospheric pressure disturbances (meteotsunamis) can also generate them.

NOAA's tsunami program lists these earthquake examples:

Worked example: how fast a tsunami crosses an ocean

NOAA gives the formula: tsunami speed equals the square root of water depth times the acceleration of gravity.

• Deep ocean, 4,000 m: square root of (9.81 × 4,000) = square root of 39,240 = about 198 m per second, or about 713 km per hour (443 mph).
• NOAA's own example, 15,000 feet: square root of (32.2 × 15,000) = square root of 483,000 = about 695 feet per second, or about 474 mph. NOAA rounds this to about 475 mph and says that at such speeds a tsunami travels from the Aleutian Islands to Hawaii in about five hours.
• Near shore, 10 m: square root of (9.81 × 10) = about 9.9 m per second, or about 36 km per hour (22 mph). NOAA describes tsunamis slowing to "the speed of a car" near land, approximately 20 or 30 mph.

Watching the ocean: DART buoys and warning centres

NOAA's Deep-ocean Assessment and Reporting of Tsunamis (DART) stations pair a seafloor bottom pressure recorder with a moored surface buoy. According to the National Data Buoy Center, the recorder samples every 15 seconds and converts pressure to sea surface height. Normally it reports values at 15 minute intervals; when its onboard software detects an event, it switches to sending 15 second values, then one minute averages, by acoustic link to the buoy and satellite to the warning centres. NOAA completed the original six buoy array in 2001 and expanded it to 39 stations in March 2008.

NOAA says that where seismic networks are dense, the US Tsunami Warning Centers can issue messages within five minutes of an earthquake; where networks are sparser, response time rises to 10 to 15 minutes. For a local tsunami, NOAA's advice is not to wait for an official message: if you feel strong shaking near the coast, move to high ground or inland.

How do seismometer networks and earthquake early warning work?

Seismometers, accelerometers and the station network

A seismic station is a site with sensors, a datalogger, power and telemetry. The USGS plan explains that seismic sensors are inertial: masses and springs. Strong motion sensors (accelerometers) stay on scale in violent shaking; broadband sensors are more sensitive and can detect distant events. The plan says all ShakeAlert stations have strong motion sensors and about 25 percent also have broadband sensors.

In the US, these stations belong to the USGS led Advanced National Seismic System (ANSS) and its regional networks. The 2018 plan called for 1,675 early warning capable stations (1,115 in California and 560 in Oregon and Washington), spaced about 10 km apart in urban areas, 20 km in source areas that threaten population centres and 40 km elsewhere. Free-field stations run on solar power with battery backup.

Inside ShakeAlert: from first P wave to a message

The West Coast production prototype is described by Kohler and others in Seismological Research Letters (2018); the USGS 2018 plan describes three processing layers:

1. Data layer. Raw ground motion from field stations flows into the ANSS Quake Monitoring System (AQMS). Live data streams are forked to ShakeAlert's production layer as Earthworm TraceBuf messages.
2. Production layer. Algorithms detect earthquakes, estimate location and magnitude, and predict shaking.
3. Alert layer. A decision module checks whether the event meets publication thresholds and, if so, publishes a ShakeAlert Message.

ShakeAlert's own algorithm page describes the pieces:

Two details show the care taken over bad alerts. The USGS plan says running several similar detection algorithms in parallel produced more false alerts, so in 2017 they were merged into EPIC. And ShakeAlert's "alert pause" limits alert regions to a 100 km radius for the first 5 seconds of a sequence, because the earliest magnitude estimates come from only a few stations and can be wrong.

Licensed technical partners subscribe to the alert servers and deliver alerts to phones and other channels, or trigger automated actions; the ShakeAlert fact sheet lists slowing trains, closing bridge gates, shutting water valves and opening firehouse doors. Delivery methods do not all use the same thresholds, so not every channel activates for every earthquake.

Worked example: how much warning is there?

This example is illustrative. We assume typical crustal wave speeds of about 6 km per second for P waves and 3.5 km per second for S waves (textbook values, not ShakeAlert parameters), ignore the earthquake's depth, and use the USGS statement that in areas with dense station coverage, ShakeAlert's first alert is routinely available in less than 5 seconds after origin time.

• A city 100 km from the epicentre. S waves arrive after 100 ÷ 3.5 = about 28.6 seconds. If the alert is out at about 5 seconds, the city has roughly 23 seconds, minus delivery delays.
• A town 40 km away. S waves arrive after 40 ÷ 3.5 = about 11.4 seconds, leaving about 6 seconds.
• The break even distance. Warning time reaches zero where distance ÷ 3.5 = 5 seconds, at about 17.5 km. Closer than that, strong shaking can arrive before or with the alert; the USGS plan notes that people near the epicentre may not get an alert before strong shaking.

The USGS plan records real cases: the M5.1 La Habra earthquake of 29 March 2014 was detected 4.2 seconds after origin time, and the M6.0 South Napa earthquake of 24 August 2014 gave beta users in the San Francisco Bay Area 5 to 8 seconds of warning.

A 2019 review of early warning systems worldwide by Richard Allen and Diego Melgar draws the same line the ShakeAlert fact sheet does: early warning is not prediction, because the alert comes only after the earthquake has begun. The fact sheet adds that people may receive an alert before, during or after shaking, depending on distance and delivery path.

Limits that matter

Near the epicentre, warning time is short or zero, a limit set by physics rather than software. Large earthquakes keep growing after the first alert, which is why FinDer models rupture length and GFAST-PGD adds GNSS for M7 and above. And thresholds are a trade-off: the USGS plan describes follow-up messages after a false alert, public acknowledgement after a missed one, and the risk that too low a threshold causes "alert fatigue."

What do space geodesy and GNSS stations show about plate motion?

The techniques

NASA's Space Geodesy Project describes four techniques that together underpin global positioning: Very Long Baseline Interferometry (VLBI); Satellite Laser Ranging (SLR); Global Navigation Satellite Systems (GNSS); and DORIS (Doppler Orbitography and Radiopositioning Integrated by Satellite). NASA operates a network of co-located stations for all four, plus a global GNSS network. Their data define the International Terrestrial Reference Frame (ITRF), the coordinate system against which plate velocities are measured.

From metres to millimetres

The EarthScope Consortium explains the jump in precision: basic GPS receivers are accurate to within a few metres, while geodetic receivers analyse a different aspect of the signal and apply corrections to measure plate velocities of one millimetre a year or less. EarthScope runs the Network of the Americas, permanent GNSS stations from Alaska to the Caribbean used to study earthquakes, volcanoes, weather and sea level.

NASA's JPL says it has analysed data from more than 2,000 receivers with its GipsyX software and publishes position time series and velocity estimates for each site; the table we used for the worked examples above is one of its products. For more on how satellite positioning itself has evolved, see our post on GPS and GNSS evolution and location privacy.

What GNSS has shown

• The plates move as the rocks said they would. The USGS notes that space geodetic rates agree with rates averaged over millions of years. The MORVEL authors report that GPS derived plate rotations fit their geologically based model better than earlier models did.
• Large earthquakes move the ground measurably. JPL publishes break estimates alongside its time series, and ShakeAlert's GFAST-PGD uses GNSS peak ground displacement to size M7 and larger events. The USGS plan budgeted to upgrade about 475 GNSS stations for early warning.

Walkthrough: pull earthquake and plate motion data yourself

The numbers in this post can be checked with public data and a terminal. Steps 1 to 4 are read only queries.

Step 1: count large earthquakes in the USGS catalogue

The USGS FDSN event web service supports query and count methods with parameters such as starttime, endtime and minmagnitude.

# How many M6+ earthquakes did the USGS catalogue record in a year?
curl -s "https://earthquake.usgs.gov/fdsnws/event/1/count?starttime=2025-10-01&endtime=2026-10-01&minmagnitude=6"

When we ran this on 23 October 2026, it returned 136. Catalogue values are revised as events are reviewed, so your count may differ slightly.

Step 2: look up a specific earthquake

# M7+ events on 11 March 2011, as tab separated text (needs jq)
curl -s "https://earthquake.usgs.gov/fdsnws/event/1/query?format=geojson&starttime=2011-03-11&endtime=2011-03-12&minmagnitude=7" \
  | jq -r '.features[] | [((.properties.time/1000|floor)|todate), .properties.mag, .properties.magType, .properties.place, .geometry.coordinates[2]] | @tsv'

The output includes the mainshock, listed as 2011-03-11T05:46:24Z 9.1 mww 2011 Great Tohoku Earthquake, Japan 29, plus two large aftershocks within the first hour. The last column is depth in kilometres.

Step 3: find a seismic station's metadata

# Station metadata for IU.ANMO (Albuquerque) from the EarthScope FDSN station service
curl -s "https://service.earthscope.org/fdsnws/station/1/query?net=IU&sta=ANMO&level=station&format=text" | head -5

Each line gives the network, station code, latitude, longitude, elevation and the time span of that station configuration.

Step 4: compute plate speeds from JPL's GNSS velocities

# Horizontal speed (mm/yr) for three stations from JPL's velocity table
curl -s https://sideshow.jpl.nasa.gov/post/tables/table2.html \
  | sed 's/<[^>]*>//g' \
  | awk '$2=="VEL" && ($1=="REYK" || $1=="HOFN" || $1=="MKEA") {printf "%s N=%s E=%s horiz=%.1f mm/yr\n", $1, $3, $4, sqrt($3^2+$4^2)}'

On 23 October 2026 this printed MKEA at 72.1 mm a year, REYK at 22.1 and HOFN at 20.4. Those are speeds in the global frame. For the spreading rate across Iceland, subtract the two Icelandic velocity vectors as in worked example 3.

Step 5: set up a reproducible Python environment for waveform work

For waveform work, ObsPy is an open source Python library for seismology. Its FDSN client maps the names "EARTHSCOPE" and "IRIS" to https://service.earthscope.org. Install it in a dedicated virtual environment with pinned versions, so the same analysis gives the same result next year.

python3 -m venv .venv && source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install "obspy==1.5.1"
python -m pip freeze > requirements.txt

# fetch_anmo.py: download one hour of vertical broadband data from IU.ANMO
from obspy import UTCDateTime
from obspy.clients.fdsn import Client

client = Client("EARTHSCOPE")
t0 = UTCDateTime("2011-03-11T05:46:24")
st = client.get_waveforms("IU", "ANMO", "00", "BHZ", t0, t0 + 3600)
print(st)
st.plot(outfile="anmo_tohoku.png")

ObsPy 1.5.1 was the current release on PyPI when we checked, and it requires Python 3.8 or later. Check the current version before you pin.

Step 6: check what you just installed

A scientific Python environment pulls in many transitive dependencies. Treat it like any other software you run. With Safeguard's CLI, a dependency scan of the project folder is one command, and results are written as JSON under ./output:

safeguard scan --dir .

If your team builds a container image for a processing pipeline, safeguard scan --image IMAGE_NAME scans the image, and safeguard scan --sbom sbom.json scans an existing SBOM. The scan tells you which components and versions you ship and which known vulnerabilities affect them. It says nothing about whether the seismology is right.

How is earthquake early warning software protected?

An early warning system is a chain: sensors in the field, telemetry links, processing servers, algorithms, a message broker, alert servers, and then partners' apps, sirens and control systems. A failure anywhere in that chain, whether a dead link, a bad configuration, a buggy update or an attacker, can mean a missed alert or a false one. The USGS 2018 implementation plan is unusually specific about how ShakeAlert handles this, and it is a useful model for anyone running public safety or critical infrastructure software.

Accreditation and access control. The plan says ShakeAlert must comply with the federal security assessment and accreditation process under the Federal Information Security Management Act of 2002, and that it received a conditional authority to operate in March 2018, with exceptions tracked in a plan of action and milestones. The reason given: full compliance would have meant running inside government data centres, causing "unacceptable delays" for alerts. Users connecting to the alert layer must authenticate with USGS issued credentials, all interactions use SSL certificates, and data streams are encrypted. Alert layer servers are hosted only on USGS servers and administered only by USGS staff.

Configuration and change control. All servers run identical operating system environments and revision levels (Red Hat Linux 7 at the time of the plan) and are patched regularly. A configuration management tool runs every 5 minutes to verify the operating system, application code and configuration files, and would detect and report any change to the modules. Code is version controlled in a GitLab repository workflow. Changes reach production only through the same configuration management after passing testing and certification.

Testing before deployment. The testing and certification platform runs candidates in real time alongside production, and replays a suite of about 120 historical waveform sets, including 47 historic earthquakes and known problem signals such as station recentring and calibration pulses. Results are compared against the current production baseline, and only changes that improve performance are deployed.

Monitoring. Production servers are monitored for faults, resource use and security issues, logs are archived, and the plan says port scans run monthly.

Resilience by design. The infrastructure is engineered, to the extent possible, to keep working through shaking of up to 2.0 g, a week without commercial power and the collapse of commercial telecommunications. Processing runs on redundant, geographically distributed servers at seismic network centres in Seattle, Menlo Park, Berkeley and Pasadena.

Open source inside the chain. The plan says all communication between layers runs over ActiveMQ, which it describes as open source, enterprise level messaging middleware, and station data reach the production layer as Earthworm TraceBuf messages. Third party components like these have their own release histories and, over time, their own vulnerabilities, which is why the configuration control and patching described above matter.

The same questions of shaking, power loss and broken links apply to anyone siting a data centre or a cable landing station near an active fault or a tsunami coast, and many network paths between regions run under the sea. Our posts on how undersea internet cables work, how often undersea cables break and the evidence on cable cuts and internet resilience cover that side of resilience.

Common mistakes

• Treating magnitude as linear. An M7 is about 32 times more energy than an M6, not 17 percent more. An M8 is about 1,000 times an M6.
• Mixing up magnitude and intensity. One earthquake has one magnitude but many intensities. The USGS plan notes that for a given magnitude, the shaking a user feels depends on their distance from the fault rupture.
• Saying plates "float on molten rock." The asthenosphere is hot, semi-solid rock that flows over geological time. The outer core is liquid; the mantle is not a sea of magma.
• Calling early warning a prediction. ShakeAlert detects an earthquake that has already started.
• Expecting warning everywhere. Close to the epicentre there may be little or no warning. Protective action on feeling shaking still matters.
• Comparing GNSS velocities without the reference frame. A station's velocity in a global frame such as IGS14 is not the same as its motion relative to a neighbouring plate. Subtract vectors from stations on each side.
• Leaving research environments unpinned. An analysis that installs "latest" every time is neither reproducible nor easy to audit. Pin versions, record them and scan what you install.

FAQ

How fast do tectonic plates move?

A few centimetres a year. The USGS gives about 2.5 cm a year for the Mid-Atlantic Ridge, under 2.5 cm a year for the Arctic Ridge and more than 15 cm a year for the fastest part of the East Pacific Rise. JPL's GNSS station on Mauna Kea moves about 72 mm a year in a global reference frame.

How many tectonic plates are there?

It depends on the model. The PB2002 model maps 52 plates, 14 large and 38 small. The MORVEL model describes 25 plates covering 97 percent of Earth's surface. Small plates and microplates are where models disagree; the large plates are much the same in all of them.

What causes tectonic plates to move?

Heat escaping from Earth's interior drives slow convection in the mantle. The USGS says the gravity driven sinking of cold, dense oceanic plates at subduction zones, called slab pull, is now considered the main force, with "ridge push" at spreading centres adding to it.

What is the difference between the Richter scale and moment magnitude?

The Richter scale measures the amplitude of seismic waves and matches a moment based scale only over a limited range. Moment magnitude, defined by Hanks and Kanamori in 1979, comes from the seismic moment (rigidity times fault area times slip), so it describes the largest earthquakes properly. The USGS catalogue lists the 2011 Tohoku earthquake with a moment magnitude of 9.1.

How much stronger is a magnitude 7 than a magnitude 6?

About 32 times more energy and seismic moment. Two whole steps, such as M6 to M8, is about 1,000 times.

Can earthquakes be predicted?

Not in the sense of naming a time, place and size in advance. The USGS describes that question as one that has "challenged and frustrated" scientists. Early warning is different: it detects a quake that has already started.

How does ShakeAlert know an earthquake is coming?

It detects the fast P waves at nearby stations, estimates the location and magnitude within seconds, predicts the shaking intensity across the region, and sends alerts ahead of the slower, more damaging S waves. In areas with dense stations, the USGS says the first alert is routinely available less than 5 seconds after the quake begins.

Does GPS really measure continents moving?

Yes. Geodetic GNSS stations measure positions to millimetres over years. Two stations on opposite sides of Iceland's rift separate by about 2.4 cm a year in JPL's data, matching the 2.5 cm a year the USGS infers from magnetic stripes on the seafloor.

Where Safeguard fits, honestly

Safeguard does not do geoscience, does not measure earthquakes, and does not run or secure ShakeAlert. Its fit is indirect and narrow: the software that monitoring, alerting and research teams build. Safeguard scans repositories, container images and SBOMs for open source components and known vulnerabilities, shows which projects contain a given component and version, ranks findings by reachability, enforces policies and gates, and can open fix pull requests (autonomous remediation ships switched off by default). That helps a team that maintains a processing pipeline, a message client or a research environment know what it ships and keep it patched. It does not replace accreditation, change control, network security, physical resilience or the science.

Earthquake monitoring, tsunami warning and geoscience research all run on software assembled from open source parts, from message brokers and data formats to the Python libraries a single install command pulls in. The geology is not a software problem, but the systems that carry it are, so the steps below apply to their open source dependencies and container images (fork and patch is in early access).

Fix it without waiting for upstream

Whatever the vulnerability, most of the risk sits in two gaps. The first is the time between a fix being published and that fix actually running in your production builds. The second, and worse, is the time before any fix exists at all, while a maintainer works out a patch and a release schedule. Waiting through both gaps is a habit, not a requirement. Safeguard gives teams three ways to close them.

1. Autonomous remediation: pull requests, not tickets

When a fixed version exists, Safeguard's Griffin models open a tested, policy gated pull request that moves the affected dependency to the safe version. You can run it on one repository or in bulk across every project that carries the vulnerable version. Remediation covers eight categories, including dependencies, licenses, security posture and findings. Reachability analysis ranks the vulnerable code paths your application actually calls first, so the first pull requests you merge are the ones that matter most. Autonomous remediation ships switched off, with every category set to Manual, so your team decides how much runs on its own and what still needs a human approval.

2. Gold Open Source: the safe version in seconds, free

gold.safeguard.sh is free and needs no login. Search any CVE or package to see affected and fixed versions, CVSS, EPSS, CISA KEV status and a recommended safe version, or use the "Am I affected?" checker. The same data is available through a free read API, README status badges and RSS threat feeds. The safeguard-sh/gold-scan-action GitHub Action can fail a build on a known exploited vulnerability, on any known vulnerability, or never, so a version you have just removed cannot quietly come back in next week's merge.

3. Gold Registry: start from components that are already fixed

registry.safeguard.sh provides hardened, drop in replacements for upstream packages and container images. Every Gold artifact is held continuously to zero exploitable critical or high CVEs, zero malware and a completed license check, with SLSA L3 or higher provenance, a Sigstore signature and a public evidence page you can hand to an auditor. An artifact loses its Gold badge if it regresses. Teams can point their own package directory at the registry so hardened equivalents resolve without changing how developers install anything.

When upstream has no fix yet, or will not ship one, Safeguard first works to get the fix accepted upstream. If the issue is severe or upstream will not move, Safeguard builds a fork, fixes the vulnerability there and publishes a hardened Gold build, so you are not stuck on a maintainer's release schedule. You can also request a hardened, compatibility tested build of a specific package. This fork and patch flow is in early access today, so ask which packages it covers before you plan around it.

One honest limit: when the vulnerable code lives in a vendor appliance, a hosted service or an operating system runtime you do not build yourself, the fix still has to come from that vendor or distribution. The steps above apply to the open source dependencies and container images you build and ship, which is where most teams carry the largest and least tracked share of their exposure.

See how Safeguard would help

The first demo is a real CLI session: install, run safeguard to open the interactive terminal, and ask it to "fix the lodash CVE in this repo." It reads the finding and advisory, asks before each command, edits package.json to lodash 4.17.21, runs npm install and npm test with approval, prints a summary and pushes nothing. The same flow applies to bumping a vulnerable library in a Python processing pipeline.

Click through it at your own pace, or open it full screen.

The second demo sets up the Safeguard runner on your own machine, approves it with its device code and then scans from the terminal. That suits research groups and operations teams that keep code and data on their own servers and want scans to run where the work runs.

Click through it at your own pace, or open it full screen.

The third demo connects Claude to the Safeguard MCP server with OAuth and shows how to choose which tools Claude can use. A team that already uses Claude for analysis can ask about the components in its own projects without giving the assistant more access than it needs.

Click through it at your own pace, or open it full screen.

To try it on your own repositories, download Safeguard and follow the CLI documentation.

Quick check

Sources

• USGS, This Dynamic Earth: Historical perspective
• USGS, This Dynamic Earth: Developing the theory
• USGS, This Dynamic Earth: Magnetic stripes and isotopic clocks
• USGS, This Dynamic Earth: Understanding plate motions
• USGS, This Dynamic Earth: Inside the Earth
• USGS, This Dynamic Earth: Hotspots
• USGS, This Dynamic Earth: Some unanswered questions (what drives the plates)
• USGS, This Dynamic Earth: Plate tectonics and people
• USGS Open File Report 2018-1155, Revised Technical Implementation Plan for the ShakeAlert System (PDF)
• ShakeAlert, System Algorithms
• ShakeAlert, Alert Delivery Thresholds
• ShakeAlert, GitLab Portal for technical partners
• ShakeAlert, Earthquake Early Warning in the US and Canada fact sheet (PDF)
• NOAA and National Weather Service, Tsunami frequently asked questions
• NOAA National Data Buoy Center, DART description
• NOAA Science On a Sphere, Earthquake Magnitude Perspective
• USGS, FDSN Event Web Service documentation
• USGS, M 9.1 2011 Great Tohoku Earthquake event page
• NASA JPL, GNSS Time Series
• NASA JPL, GNSS station positions and velocities table
• NASA, Space Geodesy Project
• EarthScope Consortium, What is GPS?
• ObsPy documentation, FDSN client
• DeMets, Gordon and Argus (2010), Geologically current plate motions, Geophysical Journal International 181, 1 to 80
• Hanks and Kanamori (1979), A moment magnitude scale, Journal of Geophysical Research 84(B5), 2348 to 2350
• Vine and Matthews (1963), Magnetic anomalies over oceanic ridges, Nature 199, 947 to 949
• Bird (2003), An updated digital model of plate boundaries, Geochemistry, Geophysics, Geosystems 4(3)
• Allen and Melgar (2019), Earthquake early warning: advances, scientific challenges, and societal needs, Annual Review of Earth and Planetary Sciences 47, 361 to 388
• Kohler and others (2018), Earthquake Early Warning ShakeAlert System: West Coast Wide Production Prototype, Seismological Research Letters 89(1), 99 to 107

Last reviewed: 23 October 2026.