PhronesisThe cloud
Start where the seed is already established, because most of it is.
The theory came out of the 1906 San Francisco earthquake. Harry Fielding Reid, writing the mechanics volume of the state investigation (Reid, 1910), read fifty years of triangulation surveys along the San Andreas: the land on either side of the fault had drifted about 3.2 meters out of line before the quake. The rock bent. It stored the bending as elastic strain — slowly, invisibly, over decades — until the fault gave way and the two sides sprang back. Elastic rebound. An earthquake releases energy already stored; the day of the quake adds nothing to the account.
That store is the cloud. It is invisible, distributed, fed from far away, and it thickens for decades before anything falls. Its technical name is the strain field. A quake is a sudden local discharge of it — the cloud eroding at one place.
And the erosion redistributes; it does more than dissipate. A rupture spends the local strain and shifts part of the load onto the neighbors. Coulomb stress transfer, computed for the 1992 Landers earthquake by King, Stein & Lin (1994) and reviewed by Stein (1999), maps where a rupture raised the stress on surrounding faults and where it dropped it: seismicity rates climb in the raised lobes and fall in the shadows. Aftershocks and triggered quakes are the cloud re-settling.
Material placements
The seed's strongest half. Placements of mass and fluid have set off earthquakes, on record, for sixty years.
Water impounded. Koyna, India, is the textbook case (Gupta, 2002). The Shivajisagar reservoir filled through the 1960s; on 10 December 1967 came an earthquake of about M6.3 (Mw 6.5–6.6 in some catalogs). Roughly 180–200 people died. The region has stayed seismically active for decades since.
Fluid injected. Denver made the mechanism plain: waste fluid pumped down a deep well at the Rocky Mountain Arsenal, and earthquakes that tracked the injection pressure (Healy et al., 1968). Oklahoma then ran the experiment at state scale — about two M3+ earthquakes a year before 2009, then 579 in 2014 and 903 in 2015 as wastewater disposal peaked (USGS / Oklahoma Geological Survey; Keranen et al., 2014). Across the central and eastern United States as a whole, Ellsworth (2013) counted an average of about 21 M3+ quakes a year through 2000 and more than 300 in 2010–2012. When injection came down — regulation plus the oil-price fall after 2014 — the rate followed: 624 in 2016, 304 in 2017, 194 in 2018.
Mass removed. Extraction is a placement with the sign flipped. Before the 2011 Lorca, Spain earthquake, decades of groundwater pumping had lowered the water table by roughly 250 meters; González et al. (2012) found the quake's slip distribution matched the Coulomb stress change from that unloading. The pumping shaped where the fault slipped and plausibly how much, on a fault already near failure — an influence on the rupture and a candidate trigger, on one well-studied case, and a step short of root cause.
Placements no one made
Nature runs the same experiment on a yearly clock. Seasonal water loading flexes California measurably: small earthquakes come slightly more often in the phases of the annual hydrologic cycle that favor rupture (Johnson, Fu & Bürgmann, 2017) — a modest modulation of timing, and only of the small events. The monsoon does the same to the Himalaya: seasonal water storage swings stress by ~2–4 kPa, and Nepal's seismicity runs roughly twice as high in winter as in summer (Bettinelli et al., 2008). Even erosion is a placement — a negative one. Modeling puts erosion-driven Coulomb stress changes on nearby thrust faults at ~0.1–10 bar over a seismic cycle (Steer et al., 2014); and after Typhoon Morakot dropped about three meters of rain on Taiwan in 2009, shallow earthquake frequency stepped up for over two and a half years where the landsliding was most intense (Steer et al., 2020) — an emerging result, suggestive rather than settled.
The last kilopascal
Why can a reservoir tip a fault that plates loaded for a century? Because the crust is nearly ready almost everywhere.
Deep boreholes and induced seismicity together indicate that the brittle upper crust sits close to frictional failure on well-oriented faults — critically stressed, as a general characterization rather than a universal law (Townend & Zoback, 2000). Earthquake sizes follow the Gutenberg–Richter power law (Gutenberg & Richter, 1944): no preferred size — consistent with a system near criticality. Sandpile models make the analogy formal — self-organized criticality (Bak, Tang & Wiesenfeld, 1987) and its earthquake variant (Olami, Feder & Christensen, 1992) recover Gutenberg–Richter-like statistics from slow loading and threshold release — though whether the crust is strictly critical remains a strong hypothesis, not a theorem.
Near-critical systems answer to small numbers. Stress changes as small as ~0.1 bar (~10 kPa, a tenth of atmospheric pressure) have been associated with detectable changes in seismicity rate (Reasenberg & Simpson, 1992; King, Stein & Lin, 1994), and the off-fault increases that reorganize aftershocks are “rarely more than a few bars” (Stein, 1999). In Oklahoma, modeled pore-pressure changes of ~0.07 MPa tracked the earthquakes out to ~35 km from the wells (Keranen et al., 2014). The figures are rules of thumb, not constants — whether any hard lower threshold exists is itself debated.
The placement supplies the last kilopascal. The plates supplied the budget.
Three corrections
The seed over-reaches in three places. Corrected plainly, it keeps its shape.
The energy is tectonic. “Caused by material placements” hands the placement the whole quake; the record hands it the timing and the address, and sometimes a say in the size. A trigger sets the date. The load sets the scale. Corrected: loaded by plates, tipped — sometimes — by placements.
The cloud is strain, and only strain. Atmospheric clouds neither cause nor predict earthquakes. “Earthquake clouds” and their folklore relatives have no validated support, and the wall stands wider than clouds: reviewing the whole precursor literature after L'Aquila, the International Commission on Earthquake Forecasting concluded that “the search for diagnostic precursors has not yet produced a successful short-term prediction scheme” (Jordan et al., 2011). No reliable short-term earthquake precursor of any kind has been demonstrated. This paper's cloud is the strain field. It predicts nothing.
Induced earthquakes are mostly small. Oklahoma's surge was M3s with a few M5s; the giant events remain tectonic through and through. Humanity tips faults measurably. It leaves the planet's seismic budget where it always was.
The symmetry
The metaphor closes on itself: earthquakes erode in the plain geomorphic sense too. The 2008 Wenchuan quake set off more than 56,000 landslides, and Parker et al. (2011) estimated the rock they mobilized exceeded the quake's own tectonic uplift — a single event removing more mountain than it raised. The estimate is contested: landslide volumes rest on scaling relations with wide error bars, and most of the debris stays parked in the range for decades rather than leaving it, so whether erosion outruns uplift over the full earthquake cycle is open. What stands is the plain fact: the quake erodes the strain cloud and the range in the same minute.
The corrected claim
An earthquake is the erosion of a strain cloud — loaded by plates, near-critical almost everywhere, tipped at the margin by where mass sits. Reservoirs, injection wells, drained aquifers, monsoons, typhoon-stripped hillslopes: placement of mass now shows up measurably in seismicity. That is the true content of the seed, and it is enough.
A live version sits at /field/sandpile: slow grains in, sudden slides out, the same shape.
Kin to Anxiety as the Signature of Displacement — the framework's other slow-load, sudden-discharge essay — and Coherence and Irreducibility, with Sandpile for the criticality live and Surround for the second term on screen.
Rests on: Reid, The Mechanics of the Earthquake, Vol. 2 of the State Earthquake Investigation Commission report, Carnegie Institution Publication 87 (1910) — elastic rebound; King, Stein & Lin, “Static stress changes and the triggering of earthquakes,” BSSA 84(3), 935–953 (1994); Stein, “The role of stress transfer in earthquake occurrence,” Nature 402, 605–609 (1999); Reasenberg & Simpson, Science 255, 1687–1690 (1992); Gupta, “A review of recent studies of triggered earthquakes by artificial water reservoirs...,” Earth-Science Reviews 58, 279–310 (2002) — Koyna; Healy, Rubey, Griggs & Raleigh, “The Denver Earthquakes,” Science 161, 1301–1310 (1968); Ellsworth, “Injection-Induced Earthquakes,” Science 341, 1225942 (2013); Keranen, Weingarten, Abers, Bekins & Ge, Science 345, 448–451 (2014); USGS / Oklahoma Geological Survey catalogs for the Oklahoma M3+ counts; González, Tiampo, Palano, Cannavó & Fernández, “The 2011 Lorca earthquake slip distribution controlled by groundwater crustal unloading,” Nature Geoscience 5, 821–825 (2012) — the “controlled” is scoped to the slip distribution, and the trigger reading is contested; Johnson, Fu & Bürgmann, “Seasonal water storage, stress modulation, and California seismicity,” Science 356, 1161–1164 (2017); Bettinelli et al., “Seasonal variations of seismicity and geodetic strain in the Himalaya induced by surface hydrology,” EPSL 266, 332–344 (2008); Steer, Simoes, Cattin & Shyu, “Erosion influences the seismicity of active thrust faults,” Nature Communications 5, 5564 (2014) — the modeling; Steer et al., “Earthquake statistics changed by typhoon-driven erosion,” Scientific Reports 10, 10899 (2020) — the Morakot signal, emerging; Gutenberg & Richter, “Frequency of earthquakes in California,” BSSA 34(4), 185–188 (1944); Bak, Tang & Wiesenfeld, PRL 59, 381–384 (1987); Olami, Feder & Christensen, PRL 68, 1244–1247 (1992); Townend & Zoback, “How faulting keeps the crust strong,” Geology 28(5), 399–402 (2000); Parker et al., “Mass wasting triggered by the 2008 Wenchuan earthquake is greater than orogenic growth,” Nature Geoscience 4, 449–452 (2011) — contested on sediment export; Jordan et al., “Operational Earthquake Forecasting: State of Knowledge and Guidelines for Utilization,” Annals of Geophysics 54(4), 315–391 (2011). All of these results are established prior art and cited as such. What is proposed is only the reading of them as one picture — the strain field as a cloud, eroded by quakes, tipped at the margin by where mass sits — offered to be argued with.