Three Gorges Probe

“Fill a lake, start an earthquake”: Damming and Reservoir-Induced Seismicity

(April 04, 2000) A paper by Heather Gingerich, a medical geologist who specializes in hydrogeochemistry and the director of the Canadian chapter of the International Medical Geology Association, detailing evidence in support of reservoir-induced seismicity RIS as a distinct geophysical phenomenon causally related to reservoir impoundment.

ABSTRACT: Much evidence exists in support of reservoir-induced seismicity RIS as a distinct geophysical phenomenon that is causally related to reservoir impoundment. Forensically, RIS cases fall into two categories, Rapid and Delayed Response type, based primarily on the time lag between initial reservoir filling and the occurrence of the largest earthquake. Within the scientific community, a dichotomy of opinion exists as to a probable geophysical triggering mechanism for RIS events and hydrogeochemical factors are rarely, if ever, considered.  According to Coulomb criterion, fault failure may be induced through the manipulation of shear stress, normal stress, pore pressure, friction, and cohesion parameters, either individually, or in concert.

A review of the literature reveals that most engineers and the general public subscribe to the “weight of water” RIS triggering mechanism, which is supported by the occurrence of Rapid Response-type seismic events. In this case, RIS is thought to be the result of an elastic response to reservoir loading which induces an increase in vertical compressive stress as well as decreased normal stress due to short-term pore pressure increases in the rock volume directly beneath the reservoir. Not only is this “Poroelastic” model incongruous with the temporal and spatial distribution of Delayed Response type-RIS events, but it has been shown that weight-induced stress changes across a normal or strike-slip fault plane are sufficient to trigger an earthquake only when the preexisting stress regime is in a near critical state, and/or when pore spaces in compressible rock are saturated down to hypocentral depths.

Most cases of RIS fall into the Delayed Response category, whereby the combined effects of water diffusion from the reservoir (i.e. increased pore pressure, decreased normal stress, coefficient of friction, and possibly cohesion due to chemical dissolution) at hypocentral depths are thought to cause fault failure. Although this “Diffusion-based” model of RIS appears more comprehensive than the Poroelastic model, it fails to adequately explain the stabilizing effect of reservoirs in certain tectonic environments and Rapid Response type RIS events wherein the diffusion of water to hypocentral depths would have been limited by the permeability of the reservoir bottom.

In order to perform a rigorous assessment of RIS hazard, the two geophysical models must therefore be considered in tandem. Only by implementing a multidisciplinary approach that also includes rudimentary hydrogeochemical modeling, can a comprehensive and realistic assessment of RIS hazard be performed. 


Review of the scientific literature pertaining to reservoir-induced seismicity (RIS) reveals an alarming lack of consensus in both theory and practice among engineers and politicians (the Dambuilders) and seismologists (the Dambusters). Given the current fashionability of large-scale hydro-engineering projects (Rich, 1994), it has become crucial that this dichotomy of opinion as to the governing principles of RIS be resolved.  Based on a review of the existing body of knowledge, this paper will examine the quantitative and qualitative evidence for the seismologic phenomenon of reservoir-induced earthquakes and will critically evaluate the proposed triggering mechanisms.

“Fill a lake, start an earthquake” (Rothé, 1968)

It has long been suspected that man’s engineering activities may influence the way crustal stresses are released in earthquakes (Gupta, 1992). However, it was not until recently that increased earthquake activity associated with the creation of artificial lakes was recognized as anything more than a series of surprise earthquakes in inconvenient places (Papazachos, 1974).

The reticence on behalf of the scientific community in accepting RIS may be attributed in part to the fact that not all reservoirs appear to trigger earthquakes. Of 126 reservoirs studied by Stuart-Alexander and Mark (prior to 1976) between 90 and 250 meters in depth, only 15 (12%) were associated with any increased seismicity above “normal” background levels (Figure 1). It is noteworthy that the probability of RIS was greatest (26%) in the category of greatest reservoir depth.

Figure 1. Observed frequency percentage of magnitude >3 earthquakes associated with reservoir impoundment

(from Gupta, 1992 after Stuart-Alexander and Mark, 1976)

It remains virtually impossible to identify an RIS-prone environment prior to reservoir impoundment. However, statistical consideration of RIS and certain reservoir characteristics by Baecher and Keeney (1982) revealed significant correlations between RIS and reservoir volume (0.27) and RIS and reservoir depth (0.23). Lesser correlations were found between RIS and both reservoir geology and the preexisting stress state of the reservoir area. Insufficient data was available for statistical evaluation of RIS and faulting in and around the reservoir area.

Distinguishing characteristics of RIS

Prior to assignment of reservoir impoundment as the discrete cause of increased earthquake activity, RIS events must be established as being distinguishable from “normal” earthquake sequences that would have occurred regardless of the reservoir. RIS events typically display four such characteristics. From each, inferences regarding the changes in mechanical properties of crustal layers induced by reservoir impoundment may be drawn (Witherspoon, 1977).

In general, the frequency distribution of earthquakes over an observed range of magnitudes in a particular area can be described by the equation,

Equation 1. log10N = A – bM


N is the number of shocks of magnitude ³ M

M is the lowest magnitude earthquake considered

A is a constant that depends on the period of observation, the size of the region considered, and the level of seismic activity

b is a constant, which describes the relative abundance of small and large earthquakes

Insofar as the study of seismology is concerned, the b parameter from the above equation is of some considerable interest. Given a reasonably low-magnitude threshold (i.e. M £ 3.0), a small b value (<1.0) is usually indicative of a relatively homogeneous medium, a high preexisting stress regime, and that larger magnitude earthquakes are likely to occur. Conversely, a large b value (>1.0) suggests a relatively heterogeneous medium, a low-stress regime, and that small earthquakes occur frequently. The b parameter can also be related to the tectonic environment, with small b values characterizing continental rifts and large b values being associated with mid-ocean ridges (Mogi, 1967).

1) Earthquake frequency-magnitude relation

Extensive study of b values for RIS sequences has been carried out by several researchers (Gupta et al., 1972). Typically, RIS sequences are characterized by numerous small earthquakes before and after the main shock, as is indicated by relatively large b values, especially in relation to the background seismicity of the region. Table 1 lists relevant the b values for the reservoir-induced earthquake of 6.2 RM at Kremasta, Greece (1966).

Table 1. Kremasta (Greece) RIS b values.

Seismicity location and type b value
Kremasta Reservoir foreshocks 1.41
Kremasta Reservoir aftershocks 1.12
Kremasta Region normal earthquakes 0.64
Greece Region shallow normal earthquakes 0.

(after Gupta and Rastogi, 1976)

Kremasta mainshock 6.2 RM (1966)

The weight of water confined within a reservoir induces a state of high stress in the underlying rock volume. Depending on the mechanical properties of the rock and the rate of strain accumulation

(i.e. reservoir filling), fractures may be initiated in hitherto homogenous rock. Increased pore fluid pressure, either due to the diffusion of water or mechanical pore space compression, may cause failure in previously competent fractures (Scholz et al, 1973). Heterogeneity is introduced, thus dividing the original rock volume into smaller volumes that are capable of releasing stored energy when their competence is exceeded, resulting in numerous low-energy earthquakes and a high b value (Gupta, 1992).

1) Largest aftershock to mainshock magnitude relation

Gupta et al. (1972) noted that the largest aftershocks in RIS sequences are similar in magnitude to the mainshocks. Although the sample size (n=4) is too small to be statistically significant, Table 2 shows that the ratio of the largest aftershock magnitude to the mainshock magnitude is near 0.9 for the four largest known RIS events to date.

Table 2. Relationship of mainshock to largest aftershock in known RIS events.

Reservoir Main-shock RM (Mo) Largest after-shock RM (M1) M1/Mo b
Kariba 6.1 6.0 0.98 1.02
Kremasta 6.2 5.5 0.89 1.12
Koyna 6.3 5.1 0.83 1.09
Xinfenkiang 6.1 5.3 0.87 1.04

(after Gupta, 1992)

Heterogeneity within the rock volume may be introduced as a result of the mainshock energy release or inelastic strain accumulation in the rock due to incremental reservoir filling (Gupta, 1983). Increased heterogeneity within the rock implies increased relative permeability, thus allowing the diffusion of water to new hypocentral locations or depths. In effect, the relatively large magnitude aftershocks associated with RIS events may represent mainshock activity of “parallel” earthquake sequences.

1) Decay of aftershocks

The relatively slow aftershock attenuation rate of RIS sequences, as compared with normal earthquake sequences, has been noted by several researches (Gupta et al., 1972; Gupta and Rastogi, 1976). This perhaps a reflection of the presence of water in the earthquake environment, as high b values are associated with both RIS aftershock sequences and mid-ocean ridges. In a “dry” environment, aftershocks represent a “marginal readjustment” of the rock volume following the mainshock energy release that is quickly resolved. The constant supply of water from a reservoir may influence the crust in such a way as to “prolong” the active fault’s re-stabilization process, thus delaying aftershock attenuation.

2) Foreshock-aftershock patterns

Figure 2 represents Mogi’s (1963) classification of foreshock and aftershock patterns. Differences in pattern type are attributable to the structural state of the rock volume and the distribution of the applied stress. With small elastic foreshocks and numerous aftershocks, the 6.3 RM Koyna earthquake of 1967 (Figure 3) exemplifies how most RIS sequences may be categorized as Mogi’s Type II earthquakes. According to Mogi (1963), this implies that the rock volume has a rather heterogeneous structure and/or that the applied stress is not uniform.

Figure 2. Mogi’s 3 types of foreshock-aftershock patterns                        Figure 3. Koyna RIS shock pattern

(from Gupta, 1992 after Mogi 1963)

The characteristic foreshock-aftershock pattern displayed by most reservoir-induced earthquake sequences is consistent with other distinguishing characteristics of RIS. Heterogeneity within the rock medium has already been discussed in previous sections. Mogi’s classification illustrates the importance of reservoir geometry and its location relative to the fault plane in determining the uniformity of the applied stress caused by reservoir loading. Although a shallow and wide reservoir would distribute the water’s weight more evenly than a deep and narrow reservoir, it would still not approach the uniformity of stress applied by “natural” tectonic forces (Gupta, 1992).

Two types of crustal response to reservoir impoundment

Mere visual assessment of Figure 4 reveals that reservoir-induced earthquakes may be roughly categorized into two types. Although other distinguishing characteristics exist, classification of an RIS event as either Rapid or Delayed Response-type is made primarily on the basis of the temporal distribution between initial filling of the reservoir and a significant increase in seismic activity.  Some reservoirs (e.g. Nurek) display both types of temporal response to filling.

Figure 4. Two types of temporal response to reservoir impoundment – Rapid and Delayed RIS events

(from Simpson et al., 1988)

Rapid Response RIS – Manicougan-3, Monticello and Nurek

As the name suggests, Rapid Response RIS immediately follows initial reservoir impoundment. It is further characterized by extremely shallow, low-magnitude, swarm-like earthquakes that are confined to the immediate reservoir area but that usually occur near the periphery of the lake. Furthermore, seismic sequences are closely temporally correlated with incremental changes in water level (Simpson et al., 1988).

Delayed Response RIS– Kariba, Aswan, Koyna, Oroville and Nurek

Except for Nurek, the crustal response to reservoir impoundment is significantly delayed, with increased seismicity occurring only after several cycles of filling and emptying or after peak water levels have been maintained for extended periods of time. Delayed Response type RIS events tend to be moderately shallow and larger magnitude earthquakes. The epicentres may extend significantly beyond the confines of the reservoir and there may be a significant time lag between seismic events and major changes in water level (Simpson et al., 1988).

Proposed Triggering Mechanisms for RIS

For the purpose of evaluating proposed RIS triggering mechanisms, fault stability is determined by modified Coulomb criterion and may be described by the equation,

Equation 2. S = So + mf (snp) – t


S is fault strength (S < 0 = fault failure)

So is degree of cohesion of the medium

mf is the coefficient of friction

p is the pore fluid pressure

sn is the compressive normal stress across the fault plane

t is the shear stress in the direction of faulting

(after Bell and Nur, 1978)

Aside from simply increasing shear stress until the accumulated strain exceeds the competence of the rock volume (i.e. cohesion) and/or effective normal stress, fault failure may be induced by,

2)             decreasing the coefficient of friction by fault lubrication,

3)             increasing pore fluid pressure, thereby effectively decreasing normal stress across the fault plane,

4)              decreasing the cohesion of the medium by water saturation or chemical dissolution.

1)         Poroelastic Model

Most engineers and the general public subscribe to the “weight of water” RIS triggering mechanism (Fearnside, 1988; CYJV, 1988). This model proposes that RIS is the result of an elastic response of the crust to reservoir loading that increases vertical compressive stress in the direction of gravity. Effective normal stress across the fault is reduced due to increasing pore fluid pressures that result from elastic compression of pore spaces (Gough, 1969). This model assumes a closed system, and does not account the diffusion of water from the reservoir above.

Figure 5 emphasizes the importance of reservoir geometry in determining the crust’s elastic response to reservoir loading. As would be expected, maximum compressive deformation occurs near the surface of the rock volume and directly beneath the deepest part of the lake. Maximum pore pressure increases also correspond to the greatest depth of water column and tend to be somewhat shallow within the rock volume.

Figure 5. Compressive deformation and pore pressure relative to depth of water column in reservoir

(from Withers and Nyland, 1976)

In order for effective normal stress across the fault plane to be significantly reduced by reservoir loading, the rock must be compressible and the pore spaces must be saturated. Figure 6 illustrates that this coupled response is immediate and short-term in nature, with deformation of the rock and induced pore pressure increases attenuating with time to near pre-loading levels.

Figure 6. Response of the crust beneath the reservoir to loading under various conditions

(from Simpson et al., 1988)

This model for triggering reservoir-induced earthquakes is supported by certain characteristics of Rapid Response-type RIS. The Poroelastic Model very neatly accounts for the extremely shallow focal depths, the restriction of epicentres within the reservoir, and the tight temporal correlation of seismic events with changing water levels associated with this type of crustal response.

However, the Poroelastic Model fails to explain the occurrence of Delayed Response-type RIS events, and in his consideration of the Oroville earthquakes, Beck (1976) concluded that the weight-induced stress changes in the hypocentral region could not have caused fault failure. Furthermore, the question remains as to why the epicentres of Rapid Response type RIS events are usually located near the periphery of the reservoir where the water is presumably more shallow.

It becomes apparent that the triggering of RIS events by means mechanical loading is fraught with a set of unwieldy pre-conditions.  Not only must the underlying rock volume be compressible, but it must also be impermeable with nonetheless saturated pore spaces. The reservoir geometry must be such that stress is sufficiently concentrated at an appropriate location vis à vis tangential normal or strike-slip faults that are suitably oriented and that are experiencing a near-critical stress regime. In addition, the rate of reservoir filling must be such that the accumulated strain within is not allowed to diffuse with time (Simpson, 1976).

2)         Diffusion Model

Healy et al. (1968) and Evans’ (1966) work on the triggering effects of high-pressure fluid injection at the Rocky Mountain Arsenal, Denver (CO) affirmed that it was possible to initiate earthquakes by the manipulation of pore fluid pressures alone (Figure 7). Without the compressive influence of mechanical loading on pore spaces, increasing pore fluid pressure may be accomplished only by the diffusion of water.

Figure 7. Inducing earthquakes through the manipulation of pore pressures at Rocky Mountain Arsenal

(from Evans, 1966)

Contrary to the Poroelastic Model, the Diffusion Model assumes that the reservoir bottom is somewhat permeable to the water above. The diffusion of water could trigger fault failure by decreasing the effective normal stress across the fault plane by increasing pore fluid pressure, decreasing the coefficient of friction by fault lubrication (Talwani, 1997; Talwani and Acree, 1985), and/or decreasing cohesion of the rock volume by water saturation or chemical dissolution (as per Equation 3 from Chen and Talwani, 1998 for limestone).

Equation 3. CaCO3 + CO2 + H2O = Ca(HCO3)2 = Ca2+ + 2HCO3

The occurrence of Delayed Response type RIS events alludes to a time-dependent triggering mechanism, as is proposed by the Diffusion Model. Depending on the relative permeability of the reservoir bottom

(Table 3), it is reasonable to expect some delay as water from the reservoir diffuses to the hypocentral region after initial impoundment or major changes in water level.

Table 3. Representative values of hydraulic conductivity for various rock types

(from Domenico and Scwartz, 1990)

The Diffusion Model accounts for increased seismicity in regions that are beyond the influence of mechanical loading (i.e. accounts for epicentres that are located beyond the confines of the reservoir and deeper focal points). While this model assumes a certain degree of heterogeneity within the rock volume to allow for the diffusion of water, it does not require heterogeneity at hypocentral depths in order to trigger fault failure. Comparatively larger volumes of rock may be otherwise induced to release stored energy, yielding the larger magnitude earthquakes associated with Delayed Response RIS.

Although the Diffusion Model is more comprehensive, it obviously neglects the occurrence of Rapid Response RIS events during which the rate of water diffusion to hypocentral regions would have been limited by impervious reservoir bottoms. Furthermore, this model fails to account for the stabilizing effects of reservoir loading in certain tectonic environments. Figure 8 suggests the thrust faults are stabilized by reservoir loading. This might explain the notable absence of RIS in the Himalayan foothills noted by Chandra (1978).

Figure 8. Response of various fault types to oscillating reservoir loading

(from Talwani, 1997 after Roeloffs, 1974)


The distinctly different responses of the crust to reservoir impoundment, as characterized by Rapid and Delayed Response types, suggest that different RIS triggering mechanisms prevail under different conditions. Kariba is a classic example of a reservoir that has elicited both crustal responses during its life span (Gough and Gough, 1971), thus increasing the exposure of the dam structure to possible failure. Clearly, rigorous assessment of RIS hazard prior to dam construction is in the best interest of all concerned. A comprehensive model of RIS considers the Poroelastic and Diffusion Models in tandem. Implementation of a multidisciplinary approach, that takes the geophysical, hydrogeological and geochemical aspects of RIS into account, is therefore recommended during the planning and assessment phase of any damming activity.

Heather Gingerich, April 04, 2000


Baecher, B.G., and Keeney, R.L., 1982. Statistical examination of reservoir induced seismicity. Bulletin of the Seismological Society of America, 72: 553-569.

Barber, M., and Ryder, G., (eds.) 1990. Damming the Three Gorges: What Dam Builders Don’t Want You to Know.  Probe International, Toronto, pp. 126-132.

Beck, J.L., 1976. Weight-induced stresses and the recent seismicity at Lake Oroville, California. Bulletin of the Seismological Society of America, 66:1121-1131.

Bell, M.L., and Nur, A., 1978. Strength changes due to reservoir-induced pore pressure and stresses and application to Lake Oroville. Journal of Geophysical Research, 83(B9): 4469-4483.

Canadian International Project Managers Ltd. Yangtze Joint Venture, 1988. Three Gorges Water Control Project Feasibility Study, P.R.C. Canadian International Development Agency (sponsor), Ottawa.

Chandra, U., 1978. Seismicity, earthquake mechanisms and tectonics along Himalayan mountain range and vicinity. Physics of the Earth and Planetary Interiors, 16: 109-131.

Chen, L., and Talwani, P., 1998. Reservoir-induced seismicity in China. Pure and Applied Geophysics, 153:133-149.

Domenico, P.A., and Schwartz, F.W., 1990. Physical and Chemical Hydrogeology. John Wiley and Sons, Inc., New York, pp. 29-41.

Evans, M.D., 1966. Man made earthquakes in Denver. Geotimes, 10: 11-17.

Fearnside, P.M., 1988. China’s Three Gorges Dam: “Fatal” project or step towards moderization? World Development, 16(5): 615-630.

Gough, D.I., 1969. Incremental stress under a two-dimensional artificial lake. Canadian Journal of the Earth Sciences, 6: 1067-1075.

Gough, D.I., 1978. Induced Seismicity. In: The Assessment and Mitigation of Earthquake Risk. UNESCO. Paris, pp. 91-117.

Gough, D.I., and Gough, W.I., 1971. Time dependence and trigger  mechanisms for the Kariba (Rhodesia) earthquakes. Engineering Geology, 10:211-217.

Gupta, H., 1983. Induced seismicity hazard mitigation through water level manipulation: a suggestion. Bulletin of the Seismological Society of America, 73: 679-682.

Gupta, H., 1985. The present status of RIS investigations with special emphasis on Koyna earthquakes. Tectonophysics, 118: 257-279.

Gupta, H., 1992. Reservoir-Induced Earthquakes. Elsevier, Amsterdam.

Gupta, H., Rastogi, B.K., and Narain, H., 1972. Common features of reservoir associated seismic activities. Bulletin of the Seismological Society of America, 62:481492.

Gupta, H., Rastogi, B.K., and Narain, H., 1972. Some discriminatory characteristics of earthquakes near Kariba, Kremasta, Koyna artificial lakes. Bulletin of the Seismological Society of America, 63: 493-507.

Healy, J.H., Rubey, W.W., Griggs, D.T., and Raleigh, C.B., 1968. The Denver earthquakes. Science, 161:1301-1310.

Hu, P., and Hu, Y., 1992. Advances in reservoir-induced seismicity in China. Tectonophysics, 209: 331-337.

Hu, Y., Yang, Q., Chen, X., Hu, P., Ma, W., and Lei., J., 1996. Induced seismicity at Wujiangdu Reservoir, China: A case induced in the karst area. Pure and Applied Geophysics, 147(2): 409-418.

Li, F., and Zhang, B., 1997.  Possibility of reservoir-induced seismicity around Three Gorges Dam on Yangtze River. International Journal of Rock Mechanics and Mining Sciences, 34:3-4, paper no. 076.

Papazachos, B.C., 1974. On the relation between certain artificial lakes and the associated seismic sequences. Engineering Geology, 8:39-48.

Roeloffs, E.A., 1988. Fault stability changes induced beneath a reservoir with cyclic variations in water level. Journal of Geophysical Research, 93(B3): 2107-2124.

Scholz, C.H., 1968. The frequency-magnitude relation of microfracturing in rock and its relation to earthquakes. Bulletin of the Seismological Society of America, 58:399-415.

Scholz, C.H., Sykes, L.R., and Aggarwal, Y.P., 1973. Earthquake prediction: a physical basis. Science, 181:803-810.

Simpson, D.W., 1976. Seismicity changes associated with reservoir loading. Engineering Geology, 10: 123-150.

Simpson, D.W., Leith, W.S., and Scholz, C.H.,1988. Two types of reservoir-induced seismicity. Bulletin of the Seismological Society of America, 78(6): 2025-2040.

Talwani., P., 1997. On the nature of reservoir-induced seismicity. Pure and Applied Geophysics, 150:473-492.

Talwani, P., and Acree, S., 1984/85. Pore pressure diffusion and the mechanism of reservoir-induced seismicity. Pure and Applied Geophysics, 122: 947-965.

Topping, A.R., 1995. Ecological Roulette: Damming the Yangtze. Foreign Affairs, 74(5):132-146.

Wang, M.Y., Yang, M.Y., Hu, Y.L., Li, T.C., Chen, Y.T., Chin, Y., and Feng, J., 1976. Mechanism of reservoir impounding earthquakes at Hsinfengkiang and a preliminary endeavour to discuss their cause. Engineering Geology, 10:331-351.

Withers, R.J., and Nyland, E., 1976. Theory for the rapid solution of ground subsidence near reservoirs on layered porous media. Engineering Geology, 10: 169-185.

Withers, R.J., and Nyland, E., 1978. Time evolution of stress under an artificial lake and its implication for induced seismicity. Canadian Journal of the Earth Sciences, 15: 1526-1534.

Witherspoon, P.A., and Gale, J.E., 1977. Mechanical and hydraulic properties of rocks related to induced seismicity. Engineering Geology, 11:23-55.


1 reply »

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s