REVIEW Streamflow and Water Well Responses to Earthquakes David R. Montgomery1 and Michael Manga2,3 groundwater levels and stream flow (21). Earthquake-induced crustal deformation and ground shaking can alter stream flow and Liquefaction and consolidation also can oc- water levels in wells through consolidation of surficial deposits, fracturing of solid rocks, cur in the nearfield to the intermediate field aquifer deformation, and the clearing of fracture-filling material. Although local condi- (22), and far-field effects most likely result tions affect the type and amplitude of response, a compilation of reported observations from the interaction of aquifer properties and of hydrological response to earthquakes indicates that the maximum distance to which transient strain during the passage of seismic changes in stream flow and water levels in wells have been reported is related to waves. Here, we review observations from earthquake magnitude. Detectable streamflow changes occur in areas within tens to studies of hydrologic responses to earth- hundreds of kilometers of the epicenter, whereas changes in groundwater levels in wells quakes and compile observations on the rela- can occur hundreds to thousands of kilometers from earthquake epicenters. tions between surface and subsurface re- sponse, distance from the epicenter, and ffects of earthquakes on the direction, at scales ranging from pores to continents. earthquake magnitude. quantity, and rate of surface and sub- The great variety of reported hydrological Esurface water flow have been noted for responses to earthquakes may be system- Groundwater Response centuries. Over the past several decades, atized by considering near-field versus far- The development of seismographs and automat- measurements of hydrological response to field, transient versus sustained, and rapid ed water level monitors in the early 20th century numerous earthquakes have quantified changes versus delayed responses. More specifically, led to the now common observation that seis- in both surface water (streamflow) and ground- we suggest that it is helpful to distinguish mically induced oscillation in water levels in water levels in wells (1, 2). As a result of these between near-field effects within about one wells provides a form of hydroseismograph. observations, a variety of mechanisms have rupture length of the fault, intermediate-field Both near-field and far-field well responses can been proposed to explain hydrological respons- effects from 1 to 10 fault-rupture lengths include high-frequency oscillations (23–27) and es to earthquakes (Fig. 1). Changes in stream- away, and far-field effects at greater distanc- step responses that may be either transient or flow and water levels in wells have been es. It is also useful to distinguish the transient sustained (28, 29). Analytical solutions have attributed to expulsion of fluids from the seis- response to passage of seismic waves and the been developed for both the high-frequency mogenic zone (3), pore-pressure diffusion after effects of liquefaction from more sustained or (ranging from seconds to minutes) oscillation of strain occurs in the upper groundwater levels and pore- crust (1, 4, 5), compression pressure fluctuations in re- of shallow aquifers (1, 6–8), sponse to the passage of seis- increased permeability of mic waves (2, 30, 31), as surficial materials resulting well as the strain amplifica- from either shaking of near- tion that leads to well level surface deposits (9–11)or oscillations. Such a response opening of bedrock fractures depends on the interaction (12–14), and decreased per- between the flow in the well meability resulting from and the flow into and out of consolidation of surficial de- Fig. 1. Interactions between earthquakes and hydrological processes. the aquifer (32). The particu- posits (15–17). Although lar amplitude of transient often relegated to the status of strange or permanent responses arising from aquifer high-frequency well response at any given site is curious phenomena, pore-pressure responses compression or dilation. Transient response a result of the site-specific effect of the interac- to distant earthquakes may trigger local earth- ranges from rapid oscillations in water levels tion of the aquifer properties and the volumetric quakes (18), and chemical and thermal changes in wells to increased stream flows that last for strain associated with propagating seismic in hydrological systems have been suggested weeks, whereas sustained response can per- waves (33). as having potential use in earthquake pre- sist for months or reflect permanent changes Sustained changes in water levels in wells diction (19, 20). Moreover, interactions in aquifer properties. Rapid response initiates depend on both the structure of the aquifer and between seismological and hydrological during ground shaking (coseismic), whereas the direction and magnitude of local seismically processes offer the potential for new in- delayed response arises after ground shaking induced strain (2). A good example of such sights into the temporal and spatial variabil- ceases (postseismic). These different styles of effects is provided by postseismic water level ity of hydrological properties and processes response reflect different mechanisms, as decreases in wells situated on bedrock ridges, well as proximity to the epicenter and geo- changes that have been attributed to water table 1Department of Earth and Space Sciences, University logical context. lowering resulting from the opening of bedrock of Washington, Seattle, WA 98195, USA. 2Depart- Different mechanisms likely characterize fractures (13, 14, 34). Changes in water levels ment of Earth and Planetary Science, University of responses at various distances from epicen- have also been attributed to local strain, such as California, Berkeley, CA 94720, USA. 3Earth Sciences Division, Lawrence Berkeley National Laboratory, ters. In the nearfield, changes in properties of in cases in which water levels in wells rise in Berkeley, CA 94720, USA. the fault zone itself can influence both zones of compression on the hanging walls of

www.sciencemag.org SCIENCE VOL 300 27 JUNE 2003 2047 R EVIEW The causes of delayed far- mented from earthquake epicenters as a func- field changes in water levels in tion of earthquake magnitude (40). We com- wells hundreds to thousands of piled data for 221 gauging stations where kilometers from an epicenter streamflow response was reported for more are not well understood. Brod- than 16 earthquakes, and we show in Fig. 3 sky et al.(29) proposed that that the limiting distance to which liquefac- local ground shaking can loosen tion has been reported also defines the upper fracture-blocking colloidal ma- limit to the envelope of reported streamflow terial, which in turn results in responses to earthquakes. Although this com- changes in water pressure. pilation suggests that the maximum extent of streamflow responses is limited by the ability Streamflow Response of liquefaction to occur, this does not mean Streamflow changes observed af- that liquefaction is the exclusive cause of ter earthquakes include near-field increased flow. Whereas other mechanisms and intermediate-field transient likely affect near-field streamflow response, Fig. 2. Distance from epicenter (or hypocenter if reported) versus responses and sustained changes we suggest that liquefaction generally may be earthquake magnitude (moment magnitude Mw, if reported) for locations reported to have exhibited seismically induced changes that altered the low discharges the limiting control on the distance to which (high-frequency to persistent) in well levels or groundwater- that define base flows in the peri- such response occurs. controlled springs. The line represents the theoretical limit to ods between storm events. Most In some cases, near-field–to–intermediate- Ͼ Ϫ8 coseismic strain 10 derived by Dobrovolsky et al.(37). studies reporting streamflow field streamflow responses appear more close- Compiled data were reported directly in previous publications or changes record increased dis- ly related to structural changes. In the 1983 measured off of maps presented in the original sources (5, 10, 13, 26, 28, 34, 41–55), with the exception of observations from the charge, and three types of mech- Borah Peak earthquake, increases in stream 1964 Alaska earthquake (23), which were corrected for the chord anisms have been proposed to ex- flow occurred on the down-dropped hanging length from the point of observation to the earthquake plain changes in stream flow after wall of the fault, whereas no response was epicenter, as were data for the 2002 Denali earthquake earthquakes: (i) transient stream- observed on the footwall. Streamflow re- ϭ (29). Data for the 1906 San Francisco earthquake (Mw flow changes resulting from aqui- sponses to the 1999 Chi Chi earthquake in 7.9) are based on accounts reported in Lawson (56). Each fer deformation, (ii) sustained Taiwan showed a pattern of increased and data point represents the response of a single well; re- sponse to a single earthquake may include many wells changes in stream discharge re- decreased stream flow, corresponding to areas of sulting from changes in hydraulic compressional and extensional deformation (35). (such as that from the Mw 9.2 1964 Alaska earthquake). conductivity or the opening or Areas with the greatest increase in stream flow in closing of near-surface fractures, response to the 2001 Nisqually earthquake were normal faults (7, 35). Water levels in wells and (iii) transient changes resulting from con- located in the Seattle Basin on the down-dropped drilled in unconsolidated valley-bottom depos- solidation of surficial deposits. In some cases, side of the deep seismogenic normal fault (17). its have exhibited seismically triggered rises the absence of evidence for changes in the rate Hence, streamflow response to earthquakes can that have been attributed to aquifer compaction of change of stream discharge during periods reflect a range of mechanisms that give rise to (35). A sustained change in well level reflects between storm events has been used to argue sustained versus transient and rapid versus nonrecoverable deformation, such as from that the properties of aquifers feeding surface delayed response. Consequently, a key chal- changes in aquifer storage capacity (36). Dif- flows did not systematically change in response lenge for studies of hydrological response to ferent styles of sustained changes in near-field to seismic events (15–17). earthquakes is to elucidate the circumstances groundwater levels can arise from the interac- Liquefaction of valley-bottom deposits that favor each mechanism. tion of seismic deformation, site-specific geo- provides a mechanism for rapid logic structure, and the topography. increases in stream flow, be- Dobrovolsky et al.(37) derived theoreti- cause compaction or settling of cal relations between earthquake magnitude, saturated near-surface deposits distance from the epicenter, and volumetric could yield large volumes of strain and reported that, in general, detectable runoff as a result of the high seismically induced strain exceeds 10Ϫ8 (38). specific storage that is typical In a typical aquifer, compressive strain of of shallow aquifers. Moreover, 10Ϫ6 could produce a pressure rise of1mof liquefaction of valley-bottom water (2), whereas strain of 10Ϫ8 would pro- deposits occurred in areas that duce a 1-cm rise barely detectable with high- experienced increased stream precision instrumentation. We compiled data flow because of the 1983 Borah from the response of 912 wells to more than Peak (6) and the 2001 Nis- 44 earthquakes and show in Fig. 2 that the qually (17) earthquakes. Al- distance to which earthquake-induced chang- though consolidation of surfi- es in water levels in wells have been reported cial deposits could produce Fig. 3. Distance from epicenter versus earthquake magnitude, increases systematically with earthquake rapid runoff response, in some Mw, for gauging stations that exhibited seismically induced magnitude. Moreover, the limit to which wa- cases it may be difficult to ex- streamflow response. The line represents the empirical limit to ter well response has been reported corre- plain large volumes of sus- the distance from the epicenter, beyond which liquefaction sponds to the distance at which a strain of tained streamflow increases by has been observed (39). Data represent increases in stream 10Ϫ8 would be expected. Hence, the coseis- this mechanism (16). Papado- flow reported for the Nisqually earthquake (17), observations mic response of wells to earthquakes extends poulos and Lefkopoulos (39) compiled by Manga (15), and additional data from published reports or taken from published maps (9–12, 14, 16, 50, 51, 53, as far from earthquake epicenters as we are reported an empirical limit to ϭ 54, 57). Data for the 1906 San Francisco earthquake (Mw able to reasonably detect such changes in the maximum distance to which 7.9) are based on accounts of changes in stream flow and typical near-surface aquifers. liquefaction has been docu- spring flow reported in Lawson (56).

2048 27 JUNE 2003 VOL 300 SCIENCE www.sciencemag.org R EVIEW

Time Scale of Response ize the maximum extent of areas over which where rw is the radius of the well, T is the transmissivity (units of length2/time), ␶ is the period of the seismic The time scale over which a change in water susceptible sites respond, but the magnitude of ␣ ␻ wave, w is a dimensionless constant given by rw( S/ levels in wells occurs can be used to estimate hydrologic response to earthquakes is inherent- T)0.5, ␻ is the angular frequency of the seismic wave the maximum depth at which the subsurface ly site-specific because of local geological con- (radians), S is a dimensionless aquifer storage coeffi- structure or fluid pressure has been altered by ditions. The variety of interactions between hy- cient, He is the effective height of the water column, and Kei and Ker are Kelvin functions (30). an earthquake. Roeloffs (2) derived a relation drological and seismological processes pro- 33. Large response can reflect either high transmissivity to estimate the time scale of vertical pore- vides opportunities for insight into controls (the product of hydraulic conductivity and aquifer pressure diffusion to the water table in near- on the development of crustal-scale perme- thickness) or high specific storage, Ss, which is the surface aquifers as a function of hydraulic ability and aquifer properties, fault behavior volume of water released by a unit volume of an aquifer in response to a unit decline in hydraulic diffusivity D and depth below the water table and strength, and the nature and distribution ϭ␳ ␣ϩ ␤ ␳ head, and is given by Ss g ( n ) where g is that a hydrologic response originated. For of seismic hazards. Moreover, the different the unit weight of water, ␣ is the matrix compress- ␤ typical hydraulic diffusivities of unconsoli- spatial scales over which surface and subsur- ibility, n is porosity, and is the fluid compressibility. Ϸ 2 2 Ϫ1 34. G. M. Fleeger et al., U.S. Geol. Surv. Water Resources dated sands (D 1to10 m s ), seismi- face hydrologic responses to earthquakes oc- Inventory Rep. WRIR 99-4170 (1999). cally triggered pore-pressure response originat- cur suggest different general mechanisms for 35. M. Lee, T.-K. Ma, Y.-M. Chang, Geophys. Res. Lett. 29, ing within 100 m of the water table would reach near-surface sources of stream flow and 10.1029/2002GL015116 (2002). 36. Storativity is the product of S (33) and aquifer the water table within hours to days of an deeper groundwater response. s thickness, and it characterizes the volume of water earthquake. For a hydraulic diffusivity more released by an aquifer from storage per unit surface Ϫ2 2 typical of fractured rock (10 to 10 m References and Notes area of the aquifer in response to a unit decline in the sϪ1), the same pore-pressure signal would 1. R. Muir-Wood, G. C. P. King, J. Geophys. Res. 98, hydraulic head normal to the surface. take from hours to a year to reach the water 22035 (1993). 37. I. P. Dobrovolsky, S. I. Zubkov, V. I. Miachkin, Pure 2. E. Roeloffs, Adv. Geophys. 37, 135 (1996). Appl. Geophys. 117, 1025 (1979). table. Consequently, near-instantaneous 3. R. H. Sibson, in Earthquake Prediction, American Geo- 38. Specifically, Dobrovolsky et al.(37) showed that the high-frequency water level changes in a physical Union Maurice Ewing Series, D. W. Simpson, distance corresponding to a strain of 10Ϫ8 could be ϭ 0.43M well must result from stress or changes in P. G. Richards, Eds. (American Geophysical Union, modeled by D 10 , where D is in kilometers Washington, DC, 1981), vol. 4, pp. 593–603. and M is earthquake magnitude. The volumetric the vicinity of the well, whereas delayed or 4. A. Nur, Geology 2, 217 (1974). strain ε can be related to pore-pressure change p in kk ϭ ϩ␯ ␯ ε sustained changes may involve deeper or 5. H. Wakita, Science 189, 553 (1975). a well by p –(2GB/3)[(1 u)/(1 – 2 u)] kk, where 6. C. J. Waag, T. G. Lane, Earthquake Spectra 2, 151 G is the shear modulus of the material, B is Skemp- more distant sources. ␯ (1985). ton’s coefficient, and u is the “undrained” Poisson Different mechanisms of hydrologic re- 7. S. H. Wood et al., Earthquake Spectra 2, 127 (1985). ratio (2). Typical values of the coefficient on the sponse to earthquakes should have different 8. B. F. Atwater, J. Geophys. Res. 97, 1901 (1992). right-hand side of the equation are 5 to 50 Gpa, Ϫ ranges of characteristic response times, and 9. R. C. Briggs, H. C. Troxell, California Division of Mines indicating that compressive strain of 10 6 could and Geology Bulletin 171, 81 (1955). produce a pressure rise of 1 m of water (2), whereas Ϫ one also should expect to see a wide range of 10. R. Waller, U.S. Geol. Surv. Prof. Pap. 544-A (1966). strain of 10 8 would produce a 1-cm rise. time scales for hydrologic response to earth- 11. C.-Y. King et al., Appl. Geochem. 9, 83 (1994). 39. G. A. Papadopoulos, G. Lefkopoulos, Bull. Seismol. quakes, depending on the local geological 12. R. O. Briggs, Water Resour. Bull. 27, 991 (1991). Soc. Am. 83, 925 (1993). and geomorphological contexts. Streamflow 13. S. Rojstaczer, S. Wolf, Geology 20, 211 (1992). 40. Specifically, Papadopoulos and Lefkopoulos (39) re- 14. S. Rojstaczer et al., Nature 373, 237 (1995). ported that the maximum distance to which lique- response to the Nisqually earthquake, for ex- 15. M. Manga, Geophys. Res. Lett. 28, 2133 (2001). faction had been reported was a function of earth- ample, occurred within 12 hours, indicating a 16. M. Manga, E. E. Brodsky, M. Boone, Geophys. Res. quake magnitude in which M ϭ –0.44 ϩ 3 ϫ 10Ϫ8 Lett. 30, 10.1029/2002GL016618 (2003). ϩ shallow source, and therefore precluding ex- De 0.98logDe where the distance to the epicenter 17. D. R. Montgomery, H. M. Greenberg, D. T. Smith, (De) is given in centimeters. pulsion of overpressured fluids in the seismo- Earth Planet. Sci. Lett. 201, 19 (2003). 41. Y. Chia et al., Bull. Seismol. Soc. Am. 91, 1062 (2001). genic zone and pore-pressure diffusion after 18. E. E. Brodsky, V. Karakostas, H. Kanamori, Geophys. 42. G. Grecksch et al., Geophys. J. Int. 138, 470 (1999). coseismic strain in the upper crust as mech- Res. Lett. 27, 2741 (2000). 43. V. Le´onardi et al., Geodinamica Acta 11, 85 (1998). anisms for the observed streamflow changes 19. P. G. Silver, H. Wakita, Science 273, 77 (1996). 44. I. Kawabe et al., Geophys. Res. Lett. 15, 1235 (1988). 20. C.-Y. King et al., Geophys. J. Int. 143, 469 (2000). 45. T. Kunugi et al., J. Geophys. Res. 105, 7805 (2000). associated with the earthquake (17). Sus- 21. A. Gudmundsson, Geophys. Res. Lett. 27, 2993 46. G. Igarashi, H. Wakita, J. Geophys. Res. 96, 4269 (2000). tained streamflow response, such as the (1991). 22. Liquefaction resulting from earthquakes generally oc- greatly increased stream flow that persisted 47. T. L. Masterlark et al., Bull. Seismol. Soc. Am. 89, 1439 curs in loose sandy soils that densify during ground (1999). for many months after the Loma Prieta earth- shaking, reducing the pore volume. The reduction in quake (13, 14), indicates a deeper source, pore space increases pore water pressures and reduc- 48. N. Matsumoto, Geophys. Res. Lett. 19, 1193 (1992). 49. E. G. Quilty, E. A. Roeloffs, Bull. Seismol. Soc. Am. 87, whereas rapid transient streamflow response es the soil strength by reducing the effective normal stress. Liquefaction occurs when the shear strength 310 (1997). to earthquakes may reflect the expulsion of of the soil is reduced enough that it no longer can 50. E. G. Quilty et al., U.S. Geol. Surv. Open-File Rep. water from surficial deposits in response to resist failure and the soil behaves as a viscous liquid. 95-813 (1995). consolidation during ground shaking. Distin- Increased pore water pressures that result from liq- 51. E. Roeloffs et al., U.S. Geol. Surv. Open-File Rep. uefaction force water upward toward the water table. 95-42 (1995). guishing among the potential mechanisms of 23. R. C. Vorhis, U.S. Geol. Surv. Prof. Pap. 544-C (1967). 52. E. Roeloffs, E. Quilty, Pure Appl. Geophys. 149,21 hydrological response to earthquakes should 24. D. R. Bower, K. C. Heaton, Can. J. Earth Sci. 15, 331 (1997). incorporate consideration of time scales, vol- (1978). 53. T. Sato et al., Geophys. Res. Lett. 27, 1219 (2000). umes of water, spatial patterns of response 25. R. L. Whitehead, R. W. Harper, H. G. Sisco, Pure Appl. 54. R. Waller, U.S. Geol. Surv. Prof. Pap. 544-B (1966). Geophys. 122, 280 (1985). 55. D. Woodcock, E. Roeloffs, Oregon Geol. 58,27 and their relation to the distribution of surfi- 26. M. Ohno et al., Geophys. Res. Lett. 24, 691 (1997). (1996). cial deposits, and patterns of seismically in- 27. C.-Y. King et al., J. Geophys. Res. 104, 13073 (1999). 56. A. C. Lawson, The California Earthquake of April 18, duced strain. 28. E. A. Roeloffs, J. Geophys. Res. 103, 869 (1998). 1906 (Report of the State Earthquake Investigation 29. E. E. Brodsky et al., J. Geophys. Res., in press. Commission, Carnegie Institution of Washington, 30. H. H. Cooper et al., J. Geophys. Res. 70, 3915 (1965). Washington, DC, 1908), vol. 1. Conclusions 31. G. Bodvarsson, J. Geophys. Res. 75, 2711 (1970). 57. R. C. McPherson, L. A. Dengler, California Geol. 45,31 Our comparison of the maximum distance 32. Specifically, a response function derived for the am- (1992). plification A of seismic waves in water wells by from earthquake epicenters for stream flow 58. This review was partially supported by the director, assuming both radial flow into the well and that Office of Science, of the U.S. Department of Energy and well response highlights differences in pore-pressure changes in the aquifer are proportional under contract no. DE-AC03-76SF00098. Figure 1 is the spatial extent of surface versus subsur- to dilational strain associated with passage of the modified from an illustration by D. Mays, and H. face hydrologic response to earthquakes. seismic wave predicts Greenberg determined the chord distances for the ϭ ␲ 2 ␶ ␣ ␲2 ␶2 2 ϩ A {[1–( rw /T ) Kei w –(4 He/ g)] response to the 1964 Alaska earthquake used in Fig. Models for the spatial limits of liquefaction 3. The comments of three anonymous reviewers are ␲ 2 ␶ ␣ 2 –0.5 and detectable response to strain character- ( rw /T ) Ker w) } greatly appreciated.

www.sciencemag.org SCIENCE VOL 300 27 JUNE 2003 2049