Main

The 2019 Ridgecrest earthquakes (Mw = 6.4 and Mw = 7.1) triggered a sequence of intense aftershocks that extended well north of the Coso volcanic area3 (Fig. 1a). This seismicity coincided approximately with a northwest-trending lobe of increased Coulomb stress due to the Mw = 7.1 mainshock5 (Fig. 1b). However, no aftershocks were detected within the geothermal field area, which lies within the lobe of increased Coulomb stress4 (Fig. 1). This gap in aftershocks at Coso is unexpected because geothermal and volcanic areas are prone to remote triggering6 and because geothermal operations can trigger intense seismicity7,8. Here we demonstrate that this feature is real and use numerical simulations to show that it resulted from destressing due to the geothermal operation.

Fig. 1: Ridgecrest aftershock and prior seismicity in the Coso area.
figure 1

a, Relocated seismicity10 (Mw > 1) before (since 2010; grey circles) and after (until the end of 2019; blue circles) the 2019 Mw = 7.1 mainshock. The black lines indicate the surface ruptures of the Mw = 7.1 and Mw = 6.4 earthquakes of 20193. Inset, close-up view of the Coso geothermal field. Solid lines represent strike–slip faults parallel to surface ruptures of Ridgecrest earthquake (red) and normal (blue) faults26,27; triangles indicate the locations of geothermal wells (https://maps.conservation.ca.gov/doggr). The red symbol in the inset represents the orientations and relative magnitude of the principal stresses (lines, horizontal; circle, vertical)26,27. b, Changes in static Coulomb stress due to the Mw = 7.1 mainshock, calculated for a right-lateral fault parallel to the main faults ruptured in these events (red faults in inset in a). We used source models derived from remote sensing and high-rate GPS data5, assuming a coefficient of friction 0.6. Black circles indicate Ridgecrest aftershocks. c, Depth distribution of earthquakes in section XX′ in a (all events in the orange box in a), along with the coseismic slip distribution5 (red colour scale). d, Yearly rate of Mw > 1 earthquakes (grey circles in a and c) in the main field (black, left axis; Extended Data Fig. 1c–e) and in the entire Coso area (main field and east flank; grey, left axis), and simulated change in Coulomb stress at the centre of the reservoir (red, right axis; Fig. 3a). See Extended Data Figs. 1–3 for detailed seismicity history.

Source data

Full size image

The lack of aftershocks at Coso could be interpreted to reflect the locally shallower seismogenic depth range. Seismicity cuts off at a shallow depth of around 4 km within the Coso field area (Fig. 1c), probably owing to a shallow brittle–ductile transition due to the large temperature gradient9. Away from Coso, the seismicity extends to the typical 10–15 km depth of the seismogenic zone in California10. According to this explanation, we would expect a rate of aftershocks about three to four times lower at Coso than elsewhere. However, even if we assume that the after-Ridgecrest seismicity is entirely aftershock (that is, no contribution from the geothermal operation), the catalogue shows an aftershock rate about 20 times lower than in the areas immediately northwest or southeast of the geothermal field (Extended Data Fig. 1). We therefore conclude that the lack of aftershocks truly reflects a lower sensitivity of the geothermal field to static triggering. A recent study11 based on a local seismic network reports more aftershocks at Coso than observed by the regional network. These data, once declustered, show that areas around the geothermal field experienced an increased rate of seismicity after the Ridgecrest earthquake. However, although the reservoir was included in the analysis, it could not be independently resolved owing to the declustering. Another recent study, with non-declustered local seismic network data, shows that the Ridgecrest earthquake did not affect the total seismicity rate within the reservoir12.

Geothermal power production at Coso started in 1987 with an electrical power capacity of 230 MW13. This large-scale operation induced substantial ground deformation and seismicity. Subsidence measured using interferometric synthetic aperture radar (InSAR) exceeded 14 cm over the injection area between September 1993 and June 199814 and was most probably driven by thermal contraction and pressure depletion, as commonly observed over other geothermal fields15. There may also be natural tectonic or magmatic sources of vertical deformation in the Coso area16, but it seems unlikely that these sources could explain the deformation signal measured from InSAR, owing to the strong correlation between the signal and the geothermal field operation. Recent InSAR observations show continued subsidence, but at a lower rate (Fig. 2f).

Fig. 2: Production, power generation and surface subsidence for our reference simulation.
figure 2

a, Reported injection (blue) and production (red) flow rates from the Coso field (thin lines), with simulation results superimposed (bold lines). The onset of production is set to 1 January 1989 in the simulation. b, Predicted change in temperature since the onset of production at all 25 producing wells. c, Time evolution of earthquake (Mw > 1) depth in the Coso main field. The migration towards greater depths is consistent with the cooling of the reservoir. d, Cumulative line-of-sight displacement over the Coso area, recovered from InSAR measurements between May 1996 and June 1998. Image adapted with permission from ref. 14, American Geophysical Union. e, Predicted cumulative line-of-sight displacement after 30 years of production. The arrow shows the line-of-sight unit vector (0.38, −0.09, 0.92)14 surface deformation (see Extended Data Fig. 7 for separate x, y and z components). f, Time evolution of the maximum line-of-sight displacement (solid black line) and observed line-of-sight displacement rates (solid blue solid)14,33,34, with their extrapolations (blue dashed lines). The black dashed line indicates the maximum displacement rate at years 5–7.

Source data

Full size image

The paucity of aftershocks during the 2019 Ridgecrest earthquake sequence is surprising because the Coso reservoir has been seismically active since the beginning of geothermal field development in 198110 (Fig. 1a, c, grey circles). We consider the time evolution of seismicity within the Coso reservoir before 2019. Seismicity ramped up substantially when geothermal power production started in 1987, peaking after around 10 years of production (Fig. 1d). We identify spatiotemporal evolution of seismicity that reflects the details of the geothermal field development (Extended Data Fig. 2). For example, an early transient peak in 1984 probably coincides with a stimulation or testing phase. A more recent peak is due to the development of the East Flank area after 2000. The overall evolution is dominated by the seismicity within the main field, which gradually increased, peaked about 10 years after the beginning of geothermal production and decreased substantially afterwards.

Reservoir pressure has been monotonically decreasing during production17, a feature consistent with production exceeding injection (Fig. 2a). The pressure drop must have induced a gradual decrease in Coulomb stress (increase in effective normal stress), so it cannot explain the sustained seismicity observed in the reservoir. Hence, another effect that increases Coulomb stress, such as thermal destressing, is required. Changes in thermoporoelastic stress may be substantial, owing to elastic coupling18, and may trigger seismicity at relatively large distances from the boreholes18,19. An initial flow test in the Coso area showed a gradual increase in permeability associated with a decrease in injection pressure by around 0.5 MPa over 40 days20. This observation also points to a dominating thermal effect. Fracturing and faulting induced by thermal contraction tend to enhance permeability21. By contrast, the pressure reduction should have prevented faulting and allowed for fracture closing, resulting in a decrease in permeability.

Thermal effects evolve slowly and may considerably modify the state of stress21,22,23 and affect the depth of the brittle–ductile transition. The seismicity in the reservoir area clearly migrated to a greater depth during the field operation, as would be expected from reservoir cooling (Fig. 2c). There is also a hint in the time evolution of focal mechanisms within the Coso area24 that the stress field was substantially altered during geothermal production. The decrease in seismicity after the peak in the 1990s coincides with an increase in the diversity of focal mechanisms and in the proportion of normal faulting events (Extended Data Fig. 3). We therefore hypothesize that the cumulative stress changes induced by geothermal heat production from Coso since 1987 impeded earthquake triggering during the 2019 Ridgecrest earthquake sequence.

We model the geomechanical effect of the geothermal operation using the thermohydromechanical simulator TOUGH–FLAC25 (Methods). The simulations were designed using public information on the geothermal field operation (https://openei.org/, https://maps.conservation.ca.gov/doggr/) and previous studies26,27,28. The Coso geothermal field consists of more than 100 wells, which were developed sequentially over the 30 years of production, in an area with multiple strike–slip (dominant in the main field area) and normal (dominant in the east flank area) faults (Fig. 1a, inset). The first successful production well was completed in 1981, but large-scale production did not start until 1987, once the development of the main field was completed. The development of the east flank area started in the early 2000s27,28. We consider a simplified setting consistent with the size of the developed area and constrained by reported flow rates and energy production. Our model consists of 50 wells (25 injectors and 25 producers at depths of 1,800 m and 1,300 m) in a 4 km × 4 km × 3 km reservoir, which is embedded in a 30 km × 30 km × 18 km domain (Extended Data Fig. 4). The reservoir and cap are assigned a bulk Mohr–Coulomb rheology, with a cohesion of 2 MPa and a friction coefficient of 0.6 (or 0.3), with the medium outside the reservoir and cap considered fully elastic. All elements are assigned a volumetric thermal contraction coefficient of 6.0 × 10−5 K−1, consistent with laboratory measurements at 250 °C29. Vertical stress is controlled by gravitational loading; horizontal stresses are applied on the boundary, consistent with a previous study of the local stress field26,27,28 (Methods).

Reservoir cooling is driven mainly by fluid advection associated with the cold-water injection and therefore depends primarily on the flow rate. Hence, the flow rate is a dominating factor in defining the change in reservoir temperature. The data reported by the operator show production rates of nearly double the injection rates (Fig. 2a). The excess production must have been balanced by either a reduction in the pore volume or an influx of groundwater in the reservoir. The cumulated excess volume of around 5.3 × 108 m3 (assuming 1 ton ≈ 1 m3) would require an unrealistic pressure drop of around 400 MPa (assuming a roughly 2 km × 2 km × 4 km reservoir with a bulk modulus of 13 GPa). The excess production was therefore probably compensated by substantial groundwater supply. This inference is consistent with geochemical evidence of recharge from a shallow cold aquifer and from regional groundwater30. It also implies that the reservoir pore space is partly filled with steam due to vaporization. Because our simulation assumes a hydraulically closed domain boundary and single-phase flow, fitting both the production and the injection data would require an unreasonably large drop in pore pressure. We therefore carried out simulations targeted to fit either the production (Fig. 2a) or the injection rates. We choose the first scenario as a reference as it is probably more realistic because the excess production probably comes from colder surrounding areas or groundwater. We consider the second scenario to provide a lower bound on thermal effects (Extended Data Fig. 5). Even our reference scenario may underestimate thermal contraction because it ignores the cooling effect of evaporation implied by the continuous increase of the steam fraction in the produced flow over the operation period31.

Our simulations account for the continuous decrease in pressure by controlling the injection and production pressure (Extended Data Fig. 6). The resulting pressure drop after 30 years is approximately 5.5 MPa in the reservoir and decreases gradually beyond it (Extended Data Fig. 6b). The predicted rate of decrease in pressure (around 0.18 MPa yr−1) matches the reported rate (around 0.17 MPa yr−1 during 2012–2014)17. Our simulations show that, even if heat extraction is driven by only the injection flow rate, the effect of the drop in pore pressure on stress changes is only a fraction of the thermal stresses.

In our reference simulation, the production temperature decreases from 250 °C (year 0) to between 150 °C and 210 °C after 30 years (Fig. 2b). This simulation initially yields electricity generation of about 170 MW, which declines over time to about 50 MW, assuming an average efficiency of the geothermal power plant of 12%32. This estimate is smaller than the reported power output capacity (230 MW)13, but gets closer if we assume 16% efficiency, which is the upper bound for double-flash power plants (the technology used at Coso)32. This comparison shows that the heat extraction predicted in our simulation may be similar to, or slightly smaller than, that in reality.

Our reference simulation predicts cumulative surface line-of-sight displacement with a pattern and amplitude (about 65 cm over 30 years) that are generally consistent with the InSAR data14,33,34 (Fig. 2d–f). The peak deformation rate is about 3.0 cm yr−1 in our simulation (Fig. 2f), comparable to the observed value14 of about 3.5 cm yr−1. In general, our reference model provides a reasonable estimate of the strain change within the reservoir. We conducted additional simulations without thermal stress (Extended Data Fig. 8) and found that the pressure depletion could, in principle, contribute about 57% of total subsidence. Because the pressure drop takes place over a wider area than does thermal cooling, the surface subsidence predicted by the isothermal simulation is less localized and a poorer fit to the satellite observations (Extended Data Fig. 8). This conclusion conflicts with the claim35 that the subsidence in the Coso area is mostly driven by pressure. If the pressure effect was dominant, it would not explain the sustained seismicity during the geothermal field operation at Coso. Thermal effects help to reconcile surface deformation with seismicity.

Thermal contraction results in a decrease in the compressive normal stresses, shifting the Mohr circles progressively towards the Mohr–Coulomb failure envelope (Fig. 3c). As a result, some reservoir areas fail, resulting in a gradual decrease in the differential stress. Failure limits the decrease in the minimum principal stress so that the Mohr circle shrinks in diameter, leading to a very large decrease in shear stress in the reservoir (Fig. 3). We also conducted simulations assuming a purely elastic reservoir. In this case, there is no reduction in shear stress and tensile stress at the centre of the reservoir becomes unrealistically large (18 MPa; Extended Data Fig. 10). Such large tensile stresses are not possible in reality, because of the limited strength of the rocks that comprise the reservoir.

Fig. 3: Simulation of stress changes due to thermal contraction, pore pressure changes and anelastic failure of the reservoir.
figure 3

a, Change in Coulomb stress at the end of the simulation (year 30), calculated on faults parallel to the Mw = 7.1 rupture on a cross-section through the reservoir area (see inset). b, Shear (xy) stress at the end of the simulation (year 30) for the same cross-section as in a. c, Mohr circle representation of the evolution of stress at the centre of the reservoir during the simulation (stresses are averaged along the yellow line in a and b). The Mohr circles do not touch the failure envelope because failure occurs only in a fraction of the reservoir volume, close to the injection wellbore. The grey dashed line indicates the input failure criteria in this simulation: tensile cut-off, 1 MPa; cohesion, 2 MPa; friction, 0.6. d, Change in shear (xy) stress at the centre of the reservoir (averaged along the yellow line in a and b).

Source data

Full size image

Our model predicts that the shear stress decreases from an initial value of about 9 MPa to about 2.9 MPa after 30 years of operation (Fig. 3d). Because friction tends to be reduced at higher temperatures, we also tested a case with lower internal friction (0.3 instead of 0.6 for the reference model). We found that the final shear stress is even smaller with lower friction (Extended Data Fig. 9). The decrease in shear stress explains the gap in large aftershocks (Mw > 1) detected by the regional seismic network in the Coso area in 2019. According to our simulation, the differential stress (the difference between the maximum and minimum principal stresses, which are both horizontal) decreases from about 30 MPa to less than 10 MPa (Extended Data Fig. 3d–f). At the injector depth, the maximum horizontal stress decreases gradually to the vertical stress and eventually even lower. Such evolution should, in principle, favour normal faulting events. Above the reservoir, at a depth shallower than about 1 km, the vertical stress becomes smaller than the minimum horizontal stress, so thrust events should eventually be favoured. This evolution of the stress field is qualitatively consistent with the increasing diversity of focal mechanisms observed during the geothermal field operation (Extended Data Fig. 3).

The predicted stress evolution is also consistent with the evolution of seismicity. According to our model, the Coulomb stress on faults parallel to the Ridgecrest rupture initially increased by as much as 6.7 MPa, owing to thermal contraction of the reservoir (Fig. 3a). This stress increased rapidly, by around 0.6–0.8 MPa yr−1, during the first 3 years before slowly decreasing to less than 0.1 MPa yr−1 at the end of the simulation (Fig. 1d). Thus, the observed seismicity rate approximately follows the simulated Coulomb stress rate, as would be expected from a standard Coulomb failure model with an instantaneous drop in coseismic stress36. In reality, the response of the seismicity should be damped because earthquake nucleation is a time-dependent process37. The effect of nucleation can probably be neglected on a multiyear timescale36.

Although poor sensitivity to dynamic triggering in the Coso geothermal field has been reported38, this observation is questioned by recent studies11,39. The suggested mechanism38, which involves unclogging of fluid pathways and subsequent pore pressure equalization, is unlikely to explain the paucity of aftershocks at Coso in 2019. A homogenous pore pressure should not inhibit static triggering of earthquakes by the increase in coseismic Coulomb stress.

Although thermal contraction of the reservoir induced considerable seismicity1,40, it eventually depleted the stress available to drive the aftershocks during the Ridgecrest sequence. This destressing results from anelastic but mostly aseismic deformation. The total released seismic moment calculated by summing the scalar moments of the events10 in the main field area is 2.1 × 1015 N m. Using the Kostrov approach41, this seismicity accounts for a shear strain of at most 1.3 × 10−5, a value estimated by considering a 2 km × 2 km × 2 km volume of rock with a shear modulus of 10 GPa and assuming that all events occurred on parallel fault planes. This strain is about two orders of magnitude smaller than the maximum anelastic shear strain predicted by our simulation (about 1.1 × 10−3; Extended Data Fig. 11), which implies that the deformation of the Coso reservoir was mostly aseismic. This result is in agreement with the theoretical considerations and with observational evidence from fluid injection experiments and geothermal operations42,43,44,45, which all suggest that faults tend to creep aseismically at low normal effective stresses.

We conclude that seismic and aseismic anelastic deformation induced by the geothermal operations at Coso probably substantially released the shear stress initially available to drive earthquakes. The thermal destressing of the Coso area reduced aftershock productivity. Such destressing could, in principle, form a barrier to the propagation of a large earthquake. However, the shallow brittle–ductile transition beneath the broader Coso volcanic area is a more likely cause of the arrest of the rupture in 2019, given that the rupture stopped about 10 km away from the geothermal field. The observed reduction in aftershock productivity may provide a general model of the early-time potential for induced earthquake mitigation, if project-terminating, triggered seismicity in deep geothermal projects46,47,48 can be avoided and long-term seismicity49 is reduced to acceptable levels.

Methods

We used TOUGH–FLAC25 to simulate thermohydromechanical processes during the geothermal operation. Stress changes within the reservoir depend on the mechanical response of the surrounding medium21,50. We therefore consider a simulation domain (30 km × 30 km × 18 km) that is substantially larger than the geothermal reservoir (4 km × 4 km × 3 km). The reservoir is similar in size to the currently developed area of the Coso geothermal field.

Extended Data Fig. 4 shows the geometry and initial stress field. The domain is meshed into 13,312 blocks, which are divided into either reservoir or host blocks. All blocks are assigned a volumetric thermal contraction coefficient of 6.0 × 10−5 K−1, a bulk modulus of 13 GPa and a Poisson’s ratio of 0.2. The magnitude of thermal expansion is consistent with an experimental result with quartz-rich rock at a temperature of 250 °C29. Reservoir blocks (blue) are embedded at depths between 1 km and 4 km. Reservoir (blue) and upper host (dark green) are assumed to fail according to the Mohr–Coulomb criteria, with a friction coefficient of 0.6 and cohesion of 2 MPa. Lower host rock blocks (light green) are fully elastic. To achieve a stable flow rate and electricity generation, we assumed constant permeability over the entire domain. Reservoir elements have high permeabilities (16 md and 10 md for higher and lower flow rates, respectively; Extended Data Fig. 5). The host (lower and upper) block has a much smaller permeability of 0.05 md. The permeabilities were tuned to yield a production rate comparable to that of the Coso field (https://openei.org). See Extended Data Table 1 for model parameters.

The stress field accounts for gravity and tectonic loading. Gravitational body forces are calculated assuming an effective density (rock density minus water density) of 1,400 kg m−3. The stress field is assumed to be initially homogeneous, with maximum and minimum horizontal stresses of 150% and 50% of the vertical stress, respectively, on the basis of a previous study of the local stress field26,27. The maximum principal stress initially strikes N 20° E, which is oriented at 65° from the right-lateral strike–slip faults (Extended Data Fig. 4, inset). Rollers with shear stress are applied at the domain boundary, except at the ground surface, which is assumed to be traction-free. We tested roller and constant-stress boundary conditions at the domain boundary and found no substantial differences, confirming that the model domain is large enough. To reduce the computational cost, we assumed an initial uniform temperature of 250 °C over the domain. Also, because water density and viscosity do not depend strongly on pressure, the pressure here represents the overpressure from the hydraulic pressure (gravitational flow is ignored). Accordingly, the stresses in our simulation represent effective stresses (stress minus fluid pressure).

The Coso geothermal field consists of more than 100 wells, developed sequentially over the 30 years of production, in an area with multiple strike–slip faults. We simplified the field into 50 wellbores. The wells include 25 injectors and 25 producers, which form a five-spot pattern that accesses two depths layers (1,300 m and 1,800 m; Extended Data Fig. 4). The distance between the injectors and producers is about 500 m, slightly more closely spaced at the centre of the reservoir (Extended Data Fig. 4). We used a Peaceman well-block pressure model51, with a virtual wellbore radius of 10 cm and a Skin factor of −4 located in the well block (Extended Data Fig. 4). The initial wellbore pressure is set to 5 MPa overpressure and −1 MPa underpressure at the injector and producer wells, respectively (Extended Data Fig. 6a), and decreases with time at a different rate to achieve the target flow rate (Fig. 2a) and pressure drop (Extended Data Fig. 6).

We acknowledge that our simulation is a simplification of reality. A more realistic simulation should consider the distribution of initial pressure and temperature. In reality, the geothermal gradient is much smaller, even in the vicinity of the Coso area17. This temperature distribution affects the evolution of the reservoir temperature though fluid influx from the surrounding area. The influx from the deeper, hot zone increases the reservoir temperature, whereas the influx from all other surroundings cools the reservoir. The cooling effect is probably much larger than the heating effect, because the cooler zone has larger contact with the reservoir, and cold water is continuously supplied from precipitation and aquifer flow. Although smaller than reality, our simulation includes considerable fluid influx (overproduction) from the surrounding area (Fig. 2a). The temperature of the influx in our simulation is probably higher than reality because we assume a uniform, high temperature (250 °C) even for the surrounding cooler area. Accordingly, if we were to use a more realistic initial temperature distribution, larger thermal destressing is expected, owing to the cold-water influx from the surroundings. Furthermore, the application of a realistic pressure gradient would induce the endothermic effect of evaporation in and above the reservoir and therefore also increase the cooling effect. Overall, we expect the application of realistic pressure and temperature gradients to enhance the thermal destressing effect relative to our simulation.

The nearly uniform thermal depletion in our simulation is not expected in a real, highly fractured reservoir. Also, a substantial change in the permeability would be expected21,52 with production. In reality, thermal depletion is dominant in the vicinity of existing fractures, owing to the high permeability and thermal stimulation21. Therefore, the thermal depletion of the fractures may become wider than predicted in our idealized depletion model. Given that the subsidence predicted by our model is comparable in extent and rate with the available observations of surface deformation, we believe that, even with these simplifications, our model provides a reasonable first-order estimate of the stress changes imparted by geothermal operations at Coso.