Journal of Seismology

, Volume 17, Issue 1, pp 109–135

Mechanical and statistical evidence of the causality of human-made mass shifts on the Earth’s upper crust and the occurrence of earthquakes

Authors

    • Think Geohazards
    • NorthWest Research Associates
Original Article

DOI: 10.1007/s10950-012-9321-8

Cite this article as:
Klose, C.D. J Seismol (2013) 17: 109. doi:10.1007/s10950-012-9321-8

Abstract

A global catalog of small- to large-sized earthquakes was systematically analyzed to identify causality and correlatives between human-made mass shifts in the upper Earth’s crust and the occurrence of earthquakes. The mass shifts, ranging between 1 kt and 1 Tt, result from large-scale geoengineering operations, including mining, water reservoirs, hydrocarbon production, fluid injection/extractions, deep geothermal energy production and coastal management. This article shows evidence that geomechanical relationships exist with statistical significance between (a) seismic moment magnitudes M of observed earthquakes, (b) lateral distances of the earthquake hypocenters to the geoengineering “operation points” and (c) mass removals or accumulations on the Earth’s crust. Statistical findings depend on uncertainties, in particular, of source parameter estimations of seismic events before instrumental recoding. Statistical observations, however, indicate that every second, seismic event tends to occur after a decade. The chance of an earthquake to nucleate after 2 or 20 years near an area with a significant mass shift is 25 or 75 %, respectively. Moreover, causative effects of seismic activities highly depend on the tectonic stress regime in which the operations take place (i.e., extensive, transverse or compressive). Results are summarized as follows: First, seismic moment magnitudes increase the more mass is locally shifted on the Earth’s crust. Second, seismic moment magnitudes increase the larger the area in the crust is geomechanically polluted. Third, reverse faults tend to be more trigger-sensitive than normal faults due to a stronger alteration of the minimum vertical principal stress component. Pure strike-slip faults seem to rupture randomly and independently from the magnitude of the mass changes. Finally, mainly due to high estimation uncertainties of source parameters and, in particular, of shallow seismic events (<10 km), it remains still very difficult to discriminate between induced and triggered earthquakes with respect to the data catalog of this study. However, first analyses indicate that small- to medium-sized earthquakes (<M6) seem to be induced and large-sized events (>M6) seem to be triggered. The rupture propagation of triggered events might be dominated by pre-existing tectonic stress conditions.

Keywords

Mass shiftsGeoengineeringHuman-madeEarthquakesHuman-causedTriggeredInducedShallow eventsHazard

1 Introduction

For decades, it has been disputed whether man-made mass redistributions and, in particular, large-scale geoengineering constructions can induce or trigger earthquakes. Since the first half of the twentieth century, numerous studies about earthquakes that are thought to be caused by geoengineering operations have been documented and published, including artificial water reservoirs, underground and open-pit mining, coastal management, hydrocarbon production, and fluid injections/extractions. In general, it is known that geological and tectonophysical processes control crustal deformation rates, which, in turn, tend to control seismic cycles of faults (Scholz 2002). These cycles seem to determine the recurrence frequency of earthquakes (Steacy et al. 2005). Thus, any additional triggering stress perturbation (e.g., due to human activities) on top of the natural stress accumulation (e.g., due to tectonic deformation, volcanism, or isostasy) can enhance seismic activities.

This paper provides an overview of global observations of mass shifts of geoengineering and earthquakes that occurred in their vicinity. The objective of this first study is to quantify correlatives among descriptive parameters (e.g., seismic moment magnitude versus mass change). Results also provide statistical significance and mechanical causalities of the relationships and their physical interpretations. Section 2 gives an overview about the data and the utilized statistical methods to identify relationships that are physically reasonable. Finally, results are discussed in Section 3, and conclusions are provided in Section 4.

1.1 Earthquakes associated with geomechanical pollution

Two factors may determine the likelihood of seismic activities in the vicinity of large-scale geoengineering operations: (a) conditions of the nature system (e.g., geological and tectonic conditions of the Earth’s crust) and (b) conditions of the human system (e.g., dimension of the construction/excavation) (Simpsom 1976; Klose 2010). Processes to induce or trigger seismic activities by geoengineering include the following:
  • mass removal or mass accumulation, e.g., movement of water,

  • volumetric changes (e.g., contraction of underground excavations or carbon dioxide storage sites),

  • fluid pressure changes with/without fluid flow in rock fractures/pores (e.g., watering and flooding of mines, reservoir impoundings),

  • thermal stress changes due to temperature gradients (e.g., around injected carbon dioxide storage sites, geothermal reservoirs, underground excavations).

Based on the concept of a human-nature coupled system (Klose 2007b), geoengineering activities (human system) alter stresses in the interior of the Earth’s crust (nature system) that can bring faults to failure. Moreover, earthquakes associated with geoengineering activities can be induced or triggered as new events (Carder 1945; Simpsom 1976) or clock-advanced as part of a natural seismic cycle of a fault (McGarr et al. 2002; Klose 2011). The phenomenon of causing earthquakes due to geoengineering activities in the upper part of the Earth’s crust can be understood as geomechanical pollution of the crust. The terminology “geomechanical pollution” was inspired by C.D. Klose and L. Seeber (Klose 2007b; Seeber 2002). The following definition is used throughout this paper:

Seismic activity is induced by geoengineering when its intensity of stress perturbation in the Earth’s crust is above daily tidal changes and spatially large enough to cause a mainshock by covering the entire area of an affected fault that is going to rupture. The mainshock would not happen without geoengineering. On the other hand, seismic activity is triggered by geoengineering when its intensity of stress perturbation in the Earth’s crust is above tidal changes and can set off seismic events that are likely to happen in the future without geoengineering. Moreover, the geomechanically polluted area is spatially not large enough to cover the entire fault. The rupture propagation of a triggered event tends to be dominated by pre-existing tectonic and geologic conditions on one or many affected fault zones with respect to their seismic cycle.

1.2 Triggering stress perturbations

Stress perturbations induced by geomechanical pollution cause earthquakes when they reach or exceed natural stress levels on faults ranging at 1–10 kPa (about a 1/10th of an atmosphere), for example, due to daily tidal strains of the crust by the Sun and the Moon (Evans 1989; Klose 2011). Thus, such seismic events result from the mechanical response of the continental crust (e.g., elastostatic, poroelastic) due to very small induced stress perturbations. A theoretical derivation of this empirical 10-kPa threshold for tidal stress perturbations on an oblique oriented fault is outlined by Klose (2011). Maximum shear stresses τ and normal stresses σn that are induced by the Sun and Moon and that need to be exceeded by any additional triggering stress (e.g., due to surface loading) are
$$ \begin{array}{rll} &&{\kern-6pt}{\rm by~the~Moon} \tau = 5.03~{\rm kPa}~{\rm and}~\sigma_n = 9.07~{\rm kPa},\\ &&{\kern-6pt}{\rm by~the~Sun} \tau = 2.32~{\rm kPa}~{\rm and}~\sigma_n = 4.18~{\rm kPa}. \end{array} $$
The tidal stress changes (a) are daily induced by elongations of the Earth due to the Sun (Moon) and (b) seem to trigger seismicity occasionally on favorably oriented faults (Tanaka et al. 2002; Cochran et al. 2004). Perturbations due to geoengineering that are below this threshold are unlikely to cause seismic activity. Moreover, it can be anticipated that the influence of these stress perturbations tends to be small in lower crustal parts, distant from the geoengineering “operation points.” An operation point can be seen as the location where engineering activities cause a large or abnormal mass change due to impounding of an artificial water reservoir or injecting fluids deep underground (see Fig. 2). Therefore, it is often hypothesized, that geomechanical pollution brings only shallow faults to failure in the upper part of the crust (<10 km). This study, however, shows how deep earthquakes can occur underground underneath geoengineering operation points.

1.3 Types of earthquake triggering

Triggering processes, induced by geoengineering activities, are complex. In general, two types of triggering mechanisms are known: (a) static stress perturbations (e.g., tectonic forces, natural erosion/sedimentation, water accumulations in artificial reservoirs) and (b) transient or dynamic stress perturbations (e.g., seismic loads from other earthquakes, seasonal water level changes in water reservoirs or lakes). Both types of triggering tend to have different effects on the clock advancement of a mainshock occurring at the end of an earthquake cycle (Steacy et al. 2005). Static changes applied early in a cycle advance the clock of the mainshock much more than that applied late in the cycle (Steacy et al. 2005). In contrast, transient stress perturbations applied late in an earthquake cycle advance the clock of the mainshock much more than that applied early in the cycle. In particular, long periodic waves (e.g., annual water level changes) advance the clock more than short-periodic waves (e.g., seismic waves within a few seconds) (Beeler and Lockner 2003).

Some processes can bring faults close to failure, while simultaneously other processes bring the same fault away from failure. Stress alterations that destabilize a fault would bring the same fault away from failure if the geologic/tectonic conditions or if the involved processes were reverse. Dewatering of deep mines of the Newcastle coal field in Southwest Australia is a good example to illustrate how mass removal can bring faults close to failure (Klose 2007a, b). In contrast, seismic investigations at the Tarbela water reservoir in Pakistan showed evidence that mass accumulation by the water reservoir impounding in 1974 locked a seismically active part of the crust in up to 70 km depth beneath the artificial lake (Jacob et al. 1979). This example can be considered as a true negative case.

2 Data and methods

2.1 Global statistics

The number of human-caused earthquakes (Fig. 1), for example, increases on average (per decade) since the twentieth century (Klose 2007b, 2010). This may result from an increasing awareness, where more attention is paid to human-caused earthquakes, or maybe due to the influence of global population increase, with higher energy and water demand. However, it should be noted that the presented earthquake catalog is not complete. Many earthquakes were not analyzed and classified as candidates of being caused by geoengineering.
https://static-content.springer.com/image/art%3A10.1007%2Fs10950-012-9321-8/MediaObjects/10950_2012_9321_Fig1_HTML.gif
Fig. 1

Number of human-caused earthquakes versus time. The earthquake catalog is based on events with moment magnitudes Mw > 4.5. (After (Klose 2010))

The majority of human-caused earthquakes is observed in stable continental regions (>75 %) (Klose 2007b). These regions show a very low nature-triggered seismic activity (<5 %) (Johnston and Kanter 1990; Schulte and Mooney 2005), where (a) earthquakes tend to rupture very shallow (0–10 km) close to the surface (Klose and Seeber 2007) and (b) tectonic conditions are stable over very long periods (e.g., thousands, ten-thousands or millions of years). In contrast, active continental regions (e.g., California, Japan or Turkey) show a high nature-triggered seismic activity (>95 %) but a low frequency of human-caused events (<25 %). The low human-caused seismicity rate in active continental regions might be also due to the fact that less attention is paid to such seismic events in these regions and that most earthquakes are assumed to be of natural origin.

Natural tectonic stresses in stable continental regions can build up over long periods of time. These stresses load faults close to failure without overcoming the strength of the rock in the faults. Thus, faults in stable continental regions and, in particular, in regions with high tectonic stress concentrations can be very earthquake trigger-sensitive due to their long earthquake cycles of major events (e.g., >10,000 years) (Hough et al. 2003; Townend and Zoback 2000). These regions, in particular, are sensitive to any static or transient stress perturbation. Overall, human-caused earthquakes tend to
  • occur in regions close to the geoengineering activities, e.g., rural and urbanized areas,

  • occur in naturally stable continental regions,

  • nucleate in very shallow depths (<10 km), while damaging seismic energy can reach the Earth’s surface (Klose and Seeber 2007), and

  • create lasting seismic activities in areas that were initially stable (e.g., Australia, Europe, America) (Ahmad and Smith 1988; Klose 2007a; Kraft et al. 2009).

Earthquakes like Australia’s M5.6 Newcastle event of 1989, for example, have shown how large-scale geoengineering activities (e.g., deep coal mining) can have major effects on human security. The Newcastle event caused US$ 5 billion damage (value of 2011) and 13 fatalities (Klose 2010).

2.2 The earthquake catalog

In this article, a catalog of 92 earthquakes is made available to the public with seismic moment magnitudes M<7.9. The data set contains physical properties that were collected and initially summarized in an unpublished earthquake catalog presented at a meeting of the Seismological Society of America in 2006 (Klose 2006). The seismic events result from a larger set of more than 500 scientific research papers, conference proceedings, and abstracts, which were published since the first half of the twentieth century. However, the half out of more than 200 earthquakes and their associated geoengineering operations are seen to be useful for further analyses due to data completeness and publication quality (e.g., geographic map with scale, production time line). A complete data set of high quality includes physical properties and their uncertainties that are discussed in the next paragraph “Selected properties and uncertainties.” A data set is incomplete if more than one parameter is missing. The seismic events and their properties are summarized in Table 1. All seismic events were reported to be the largest events (maximum magnitude values) associated with geoengineering activities, including the following:
  • artificial water reservoir (e.g., impoundment and/or operation stage),

  • open pit and underground mining (e.g., black coal, copper),

  • hydrocarbon production and enhanced recovery (e.g., crude oil, natural gas, steam injections),

  • waste water injection deep into underground (e.g., radioactive, toxic or other hazardous liquids),

  • injection of carbon dioxide deep underground (to reduce green house gases into atmosphere),

  • deep geothermal energy production (> 1,000 m) and

  • coastal management (e.g., formation of artificial coast land).

Table 1

Catalog of human-caused earthquakes, including the properties of the seismic event and the associated type of geoengineering operation

M

Time (year)

Depth (m)

± (m)

Δm (kg)

± (kg)

Type of geoengineering specification

Name

Start (year)

Duration (months)

Depth (m)

± (m)

Lat. distance (m)

± (m)

Country

Tectonic

References

7.9w

2008

13,000

5,000

1.10 1012

NA

Reservoir

Dam

Zipingpu

2006

30

−156

NA

10,000

5,000

PRC

R

Klose (2011)

7.2S

1984

15,000

5,000

6.00 1011

NA

Extraction

Gas

Gazli

1962

168

1,500

100

10,000

2,500

Usbek

R

Adushkin and Yudin (2000), Bossu et al. (1996), Simpson and Leith (1976)

7.1S

1959

20,000

5,000

9.94 1011

NA

Reservoir

Dam

Lake Hebgen

1915

-

−19

NA

7,000

3,000

USA

N

Tocher (1962), Witkind et al. (1962), Klein (1976)

7.0S

1976

15,000

5,000

6.00 1011

NA

Extraction

Gas

Gazli

1962

168

1,500

100

7,500

2,500

Usbek

R

Adushkin and Yudin (2000), Bossu et al. (1996), Simpson and Leith (1976)

6.8S

1976

15,000

5,000

6.00 1011

NA

Extraction

Gas

Gazli

1962

168

1,500

100

7,000

2,500

Usbek

R

Adushkin and Yudin (2000), Bossu et al. (1996), Simpson and Leith (1976)

6.5w

1983

10,000

100

3.24 1011

NA

Extraction

Oil

Coalinga

1952

108

2,000

10

2,000

100

USA

R

McGarr (1991), Grasso (1992)

6.3S

1967

7,000

3,000

2.78 1012

NA

Reservoir

Dam

Koyna

1962

60

−103

NA

10,000

1,000

India

N

Guha et al. (1974), Gupta and Rastogi (1974), Rajendran and Harish (2000), Rastogi et al. (1997b), Gupta (1985), Seeber et al. (1996), Gupta et al. (1997), Bozovic (1974), Talwani (1995, 1997)

6.1S

1962

4,700

1,000

1.15 1013

NA

Reservoir

Dam

Xinfengjiang

1959

24

−105

NA

4,000

1,000

PRC

SS

Gupta (2002), Shen et al. (1974), Shen and Chang (1995)

6.1w

1993

2,600

1,000

1.25 1011

NA

Reservoir

Dam

Killari

1990

36

−20

NA

12,000

2,000

India

R

Seeber et al. (1996), Kayal and Mukhopadhyay (2002)

6.1w

1985

11,400

100

1.48 1011

NA

Extraction

Oil

Kettleman

1967

216

1,500

10

2,500

100

USA

R

McGarr (1991), Grasso (1992)

5.9L

1983

8,000

3,000

1.18 1013

NA

Reservoir

Dam

Srinagarind

1977

-

−140

NA

1,000

1,000

Thail

R

Chung and Liu (1992)

5.9L

1987

14,600

100

1.62 1011

NA

Extraction

Oil

Whittier-Narrows

1924

757

1,500

10

1,600

100

USA

R

McGarr (1991), Grasso (1992)

5.8b

1962

1,000

1,000

1.75 1014

NA

Reservoir

Dam

Kariba

1959

32

−128

NA

25,000

1,000

Zambia

N

Gough and Gough (1970a, b), Rastogi et al. (1995), Gupta and Rastogi (1976), Meade (1991)

5.8L

1975

8,800

1,000

4.40 1012

NA

Reservoir

Dam

Oroville

1967

24

−210

NA

10,500

1,000

USA

N

Rajendran and Gupta (1986), Rajendran et al. (1993), Ryall et al. (1976), Bufe et al. (1976), Savage et al. (1976)

5.7w

1981

19,000

1,000

1.64 1014

NA

Reservoir

Dam

Aswan

1964

84

−91.5

NA

10,000

10,000

Egypt

N

Gupta (1985), Meade (1991), Hassoup (2002), Awad and Mizoue (1995a, b)

5.7L

1967

4,500

1,000

2.73 1014

NA

Extraction

Gas

Arette (Lacq)

1950

204

4,200

700

20,000

2,000

FR

N

Segall and Grasso (1994), Grasso and Freignier (1990), Grasso (1992), Maury et al. (1992), Jenner and Dienesh (1965), Winnock and Pontalier (1968), Grasso and Wittlinger (1990)

5.6S

1966

12,000

1,000

4.75 1012

NA

Reservoir

Dam

Kremasta

1965

12

−160

NA

25,000

1,000

GR

N

Therianos (1974), Stein et al. (1982)

5.6L

1989

800

200

2.25 1010

2.00 109

Mining

Potash

Volkerhausen

1925

768

700

50

1,000

100

DE

N

Knoll (1990), Leydecker (1998)

5.6L

1989

9,100

1,000

2.97 1011

8.59 1010

Mining

Coal

Newcastle

1801

2,256

250

250

1,250

1,250

AUS

R

Klose (2007a)

5.5L

1951

1,500

1,000

12.0 1012

NA

Extraction

Gas

Caviaga

1946

60

500

500

5,000

5,000

IT

N

Maury et al. (1992)

5.3L

1962

5,000

1,000

2.75 1011

NA

Reservoir

Dam

Monteynard

1962

7

−130

NA

10,000

NA

FR

N

Gupta and Rastogi (1976), Rothe (1970, 1973), Bozovic (1974)

5.3L

1977

3,000

1,000

3.00 1012

NA

Reservoir

Dam

Charvak

1972

-

−167

NA

18,000

5,000

UZB

N

Kalinin and Kuzin (1982), Plotnikova et al. (1992)

5.3L

1868

4,000

4,000

6.39 1010

5.34 1010

Mining

Coal

Newcastle

1801

804

150

150

1,250

1,250

AUS

R

Klose (2007a)

5.3L

1925

4,000

4,000

3.46 1010

2.45 1010

Mining

Coal

Newcastle

1801

1,488

250

250

1,250

1,250

AUS

R

Klose (2007a)

5.3L

1994

1,300

1,300

1.36 1011

7.85 1010

Mining

Coal

Ellalong

1801

2,316

250

250

1,250

1,250

AUS

R

Klose (2007a)

5.3L

1967

5,000

1,000

5.85 108

5.00 106

Injection

Fluid

Denver

1962

48

3,671

100

5,000

1,000

USA

SS

Healy et al. (1968)

5.2L

1975

800

500

1.40 1011

1.00 109

Mining

Potash

Suenna

1925

600

800

50

3,000

100

DE

N

Leydecker (1998)

5.2L

1974

5,800

2,000

5.40 1011

NA

Reservoir

Dam

Shenwo

1972

24

−97

NA

3,000

1,000

PRC

SS

Zhong et al. (1997)

5.1L

1980

6,400

1,000

2.03 1014

NA

Extraction

Gas

Lacq

1950

360

4,000

500

500

500

FR

N

Grasso (1992), Maury et al. (1992)

5.0L

1966

1,200

1,000

2.04 1012

NA

Reservoir

Dam

Benmore

1964

-

−110

NA

17,000

5,000

NZ

N

Adams (1974)

5.0L

1939

4,000

1,000

3.60 1013

NA

Reservoir

Dam

Hoover

1935

20

−140

NA

2,000

2,000

USA

N

Carder (1945)

5.0w

2001

3,000

1,000

2.86 1013

NA

Injection

Oil

Ekofisk

1981

240

300

100

500

500

NOR

N

Zoback and Zinke (2002), Selby et al. (2005), Cesca et al. (2007)

5.0L

1996

12,000

1,000

1.10 1012

 

Reservoir

Dam

Thomson

1983

-

−165

NA

3,000

500

AUS

R

Allen et al. (2000)

5.1w

2010

10,000

1,000

3.90 1012

NA

Mining

Gold

Kalgoorlie

1992

216

640

50

1,000

500

AUS

R

Cranswick (2011)

4.9L

1997

4,500

100

5.68 1011

NA

Extraction

Oil

Alabama

1975

288

2,100

10

4,000

100

USA

N

Gomberg and Wolf (1999)

4.9L

1983

5,000

1,000

9.57 1011

NA

Reservoir

Dam

Bhatsa

1983

3

−88

NA

4,000

1,000

IN

SS

Rastogi et al. (1986a)

4.9S

1961

100

100

1.49 1011

NA

Reservoir

Dam

Kurobe

1960

12

−180

NA

10,000

1,000

JP

SS

Hagiwara and Ohtake (1972)

4.9L

1967

5,000

1,000

3.40 1011

NA

Reservoir

Dam

Baji-Basta

1966

12

−89

NA

7,000

1,000

YU

SS

Bozovic (1974), Rastogi et al. (1995)

4.9w

1986

3,250

1,000

1.33 109

NA

Injection

Fluid

Perry

1975

96

1,851

10

12,000

1,000

USA

SS

Ahmad and Smith (1988)

4.9L

1841

4,000

4,000

2.09 1010

1.08 1010

Mining

Coal

Newcastle

1801

1,488

50

50

1,250

1,250

AUS

R

Klose (2007a)

4.7L

1973

11,800

1,000

2.50 1013

NA

Reservoir

Dam

McNaughton

1973

1

−191

NA

15,000

5,000

CND

N

Ellis et al. (1976)

4.7L

1973

4,800

1,000

1.11 1011

NA

Reservoir

Dam

Anderson

1950

-

−72

NA

10,000

2,500

USA

SS

Bufe (1975)

4.7L

1973

7,000

1,000

1.60 1013

NA

Reservoir

Dam

Danjiangkou

1967

16

−100

NA

100

1,000

PRC

SS

Hu (1992)

4.6L

1969

10,000

5,000

1.00 1012

NA

Reservoir

Dam

Kastraki

1969

1

−96

NA

1,000

1,000

GR

N

Gupta (2002)

4.6L

2001

4,850

1,250

3.99 1012

NA

Injection

Fluid

Trinidad

1988

158

1,800

300

2,750

1,250

USA

N

Meremonte et al. (2002)

4.6S

1972

5,000

1,000

3.00 1012

NA

Reservoir

Dam

Nurek

1971

23

−315

NA

5,000

1,000

Usbek

SS

Soboleva and Mamadaliev (1976), Simpson and Negmatullaev (1981), Rastogi et al. (1995)

4.6L

1977

27,000

5,000

1.15 1013

NA

Reservoir

Dam

Toktogul

1973

78

−215

NA

5,000

1,000

Kirgh

SS

Simpson et al. (1981), Yunga et al. (1996)

4.6b

1994

2,000

1,000

4.00 109

NA

Mining

Quarry

Cacoosing

1992

24

50

NA

1,000

100

USA

R

Seeber et al. (1998)

4.5L

2005

2520

1,000

1.48 1012

NA

Extraction

Steam

Geysers

1973

384

2,520

100

5,040

100

USA

N

Majer and Peterson (2007), Hamilton and Muffler (1972)

4.5L

1988

4,900

1,000

2.00 1012

NA

Reservoir

Dam

Idukki

1975

18

−169

NA

10,000

1,000

India

SS

Rastogi et al. (1995)

4.3L

1974

1,500

1,000

2.50 1012

NA

Reservoir

Dam

Clark Hill

1952

1

−67

NA

3,000

1,000

USA

N

Talwani (1976a)

4.3L

1978

8,000

1,000

4.50 1010

NA

Injection

Oil

Snyders Cogdell

1948

264

2,100

100

2,000

1,000

USA

N

Dumas (1989), Davis and Pennington (1989)

4.3L

1993

3,500

1,000

1.60 1012

NA

Extraction

Gas

Fashing

1958

312

3400

200

2,000

500

USA

N

Davis et al. (1995), Pennington et al. (1986)

4.3L

2006

15,000

2,000

2.94 1012

NA

Reservoir

Dam

Karun

2004

20

−205

NA

10,000

1,000

IRI

SS

Kangi and Heidari (2008)

4.3b

2001

3,750

1,000

3.40 108

NA

Injection

Fluid

Ashtabula

1986

84

1,851

10

7,000

3,000

USA

SS

Ahmad and Smith (1988), Seeber et al. (2004)

4.2L

1986

7,000

1,000

4.00 1012

NA

Reservoir

Dam

Fierza-Komani

1978

30

−167

NA

3000

2000

ALB

N

Muco (1991a, b)

4.2L

2007

1,000

1,000

2.80 109

NA

Coastal

Coast

Folkestone

1806

2412

−5

5

1,000

1,000

UK

N

Klose (2007)

4.1L

1989

220

100

2.50 1012

NA

Mining

Copper

Khibiny

1929

720

220

NA

1,200

600

RU

N

Kozyrev et al. (2005), Kremenestskaya and Trjapitsin (1995), Melnikov et al. (1996)

4.1L

1985

14,000

1,000

1.30 1010

NA

Reservoir

Dam

Ridracoli

1984

10

−103

NA

6,000

1,000

IT

R

Piccinelli et al. (1995)

4.0L

1986

1,750

1,000

2.56 1012

NA

Injection

Oil

Romashkino

1948

456

1,550

50

1,000

1,000

RUS

N

Adushkin and Yudin (2000)

4.0L

2008

1,000

1,000

1.00 1010

1.00 109

Mining

Coal

Primsmulde

1945

756

500

NA

2,000

1,000

D

R

Fritschen (2010)

3.9L

1984

2,400

1,000

1.91 1011

NA

Extraction

Oil

Imogene

1945

336

2,200

100

3,500

100

USA

N

Pennington et al. (1986)

3.8L

1994

1,500

1,000

2.85 1011

NA

Reservoir

Dam

Dhamni

1983

60

−59

NA

650

350

IN

N

Rastogi et al. (1997a)

3.7L

1983

8,000

1,000

3.00 1013

NA

Reservoir

Dam

LG-3

1981

17

−80

NA

100

5,000

CND

N

Anglin and Buchbinder (1985)

3.7L

1962

5,000

1,000

1.60 109

NA

Extraction

Oil

Inglewood

1954

36

500

200

500

500

USA

N

Hamilton and Meehan (1971)

3.6L

1991

3,400

1,000

8.16 109

NA

Extraction

Oil

Falls City

1948

-

1550

250

3,500

100

USA

N

Davis et al. (1995), Pennington et al. (1986)

3.5L

1982

1,500

1,000

1.17 1011

NA

Reservoir

Dam

Osmansagar

1920

9

−36

NA

3,000

1,000

India

N

Rastogi et al. (1986b)

3.5L

1974

5,000

1,000

2.25 1011

NA

Reservoir

Dam

Emmosson

1973

17

−180

NA

4,000

1,000

CH

SS

Bock and Mayer-Rosa (1981)

3.4L

1985

500

500

2.10 1012

NA

Reservoir

Dam

Wujiangdu

1980

24

−165

NA

50

50

PRC

N

Hu et al. (1996)

3.4L

2006

4,600

300

8.80 106

NA

Injection

Geoth

Basel

2006

7

4,300

10

200

500

CH

SS

Kraft et al. (2009)

3.3L

1,955

472

100

7.56 1013

NA

Extraction

Oil

Wilmington

1937

216

1,000

100

100

100

USA

SS

Kovach (1974), Yerkes and Castle (1976), Kosloff et al. (1980)

3.3L

1961

518

100

9.96 1013

NA

Extraction

Oil

Wilmington

1937

218

1,000

100

100

100

USA

SS

Kovach (1974), Yerkes and Castle (1976), Kosloff et al. (1980)

3.3b

1974

1,000

500

2.50 1010

NA

Mining

Quarry

Wappinger Falls

1952

264

50

NA

100

100

USA

R

Pomeroy et al. (1976)

3.2L

1972

5,167

1,000

7.17 1012

NA

Reservoir

Dam

Zhelin

1972

17

−64

NA

100

1,000

PRC

N

Hu (1992)

3.2L

1975

2,000

500

1.43 109

NA

Reservoir

Dam

Jocassee

1971

48

−107

NA

2,500

1,000

USA

N

Talwani (1976b)

3.2L

1982

5,000

100

1.90 1010

NA

Injection

Geoth

Larderello-Travale

1978

48

1300

600

5,000

NA

IT

N

Kravanja et al. (2000), Batini et al. (1978, 1985, 2003)

3.1L

1970

2,750

1,000

5.70 108

NA

Injection

Fluid

Rangely

1969

24

2,200

NA

1,750

250

USA

R

Raleigh et al. (1976)

3.0L

1960

400

100

1.50 1011

NA

Reservoir

Dam

Vajont

1960

1

−262

NA

1,000

1,000

IT

N

Scheidegger (1973), Chowdhury (1978), Voight and Faust (1992)

3.0L

1991

1,500

100

2.00 107

NA

Injection

Geoth

Ahuachapan

1991

2

1,250

150

3,000

NA

ElSal

N

Fabriol and Beauce (1997)

3.0L

1975

3,350

1,000

1.56 1010

NA

Extraction

Oil

CBP War Wink

1967

96

5,000

1,000

2,000

100

USA

N

Dose et al. (1991)

2.9

1978

2,000

1,000

4.95 1011

NA

Reservoir

Dam

Monticello

1977

757

−48

NA

1,000

500

USA

R

Secor et al. (1982), Fletcher (1982), Zoback and Hickman (1982), Rajendran and Talwani (1992), Rastogi et al. (1995), Chen and Talwani (2001)

2.7

1982

3,200

200

1.14 1010

NA

Reservoir

Dam

Takase

1979

12

−176

NA

500

100

JP

N

Okamoto et al. (1985)

2.3

2001

1,500

1,000

x1.25 1013

NA

Reservoir

Dam

Toulnustouc

1998

35

−171

NA

10

1,000

CND

N

Lamontagne et al. (2008)

1.9

1997

3,200

100

1.00 108

NA

Injection

Geoth

Soultz

1997

4

2,950

950

1,350

150

FR

N

Bourouis and Bernard (2007), Evans et al. (2004), Phillips (2000)

1.5

1997

1,350

100

4.50 108

NA

Injection

Geoth

Olkaria

1997

4

1350

50

1,000

NA

Kenya

N

Simiyu (1999), Omenda (1998)

1.3

1983

3,372

75

2.16 107

NA

Injection

Geoth

Fenton-Hill

1983

1

3,463

10

300

200

USA

SS

Cash et al. (1983), Fehler (1989)

1.2

1994

8,800

NA

2.20 105

NA

Injection

Fluid

KTB

1994

2

9,100

10

75

25

DE

SS

Zoback and Harjes (1997)

1.0

1997

3,000

1,000

2.58 105

NA

Injection

Fluid

Nojima

1997

1

1,580

1,000

3,000

300

JP

R

Tadokoro et al. (2000), Ando (2001)

0.4

2000

8,500

NA

4.00 106

NA

Injection

Fluid

KTB

2000

2

9,100

10

500

200

DE

SS

Baisch and Harjes (2003), Baisch et al. (2002)

Seismic magnitudes are differentiated between moment magnitudes (w), local magnitudes (L), body wave magnitudes (b) and surface wave magnitudes (S). Depths are given in positive values. The negative depth values indicate the height of the dam constructions of artificial reservoirs

The earthquakes of the presented catalog ruptured on pre-existing or new faults near operation points of the engineering activities. For simplification purposes, an operation point (see Fig. 2) is considered as the center of the geographical location of a geoengineering operation in the Earth’s crust (subsurface) or on top of the Earth’s crust (surface). The lateral distance of the operation point to the earthquake is defined as distance d. The direct distance \(\overline{d}\) is the Euclidean distance to an earthquake \(\overline{d} = \sqrt{z^2 + d^2}\), where z is the depth of an earthquake beneath an operation point. The engineering operation lasted for a certain time, called perturbation duration tp, while the earthquake occurred with a time delay t after beginning of the operation. Moreover, all parameters provided in Table 1 are also characterized by their uncertainties.
https://static-content.springer.com/image/art%3A10.1007%2Fs10950-012-9321-8/MediaObjects/10950_2012_9321_Fig2_HTML.gif
Fig. 2

Schematic sketch to illustrate the properties mass change Δm, lateral distance d, and depth z of the earthquake from the operation point of the geoengineering activity and time delay of the earthquake Δt

The observed earthquakes showed evidence to be “triggered” or “induced” by man-made mass shifts, because of the causal relations in space and time that were outlined in scientific publications at different times, by different researchers, and for different events, different geoengineering activities, and different geological regions worldwide. Thus, the conclusions of the original authors to consider an earthquake as “triggered” or “induced” are based on independent observations, methodologies, and interpretations. Therefore, each seismic event and its associated geoengineering activity are independent realizations. This fact is crucial for further statistical analyses, such as t tests, correlation tests, or regression tests, in order to reduce the effect of systematic sampling errors. Such an error would occur if a subjectively biased author analyzed not one event but multiple events simultaneously; let us assume the 1939 M5.0 Hoover dam earthquake and the 1974 M5.2 Shenwo earthquake. In summary, although not yet complete, the earthquake catalog discussed in this article can be considered as a statistical representative sample due to the outlined unbiased sampling criteria that was applied during the screening process of more than 500 original research papers.

Earthquake selection criteria:

  • A scientific peer-reviewed article, a conference proceedings paper or abstract must exist.

  • The earthquake that is claimed to be “induced” or “triggered” must be characterized by (a) time of nucleation, (b) dominant focal mechanism, (c) geographic location ± uncertainty of the, (d) depth ± uncertainty, and (e) maximum observed seismic magnitude ± uncertainty.

  • The geoengineering operation that is claimed to be the cause of the earthquake must be characterized by (a) start time of operation, (b) geographic location ± uncertainty, (c) depth in Earth’s crust, (d) mass change of the responsible man-made mass redistribution, and (e) if possible, an end time of operation.

Selected properties and uncertainties

Six physical properties and their uncertainties were extracted from the original catalog, which include the following:
  • Observed seismic moment magnitude of the earthquake \(M_{\textrm{obs}} = M_w\) (see Eqs. 14),

  • Lateral distance d±δd of the operation point from the earthquake nucleation point,

  • Depth z±δz of the earthquake nucleation point beneath the operation point,

  • Time delay Δt±Δt of the earthquake nucleation after start of the geoengineering operation,

  • Perturbation duration Δtp±Δtp of geoengineering operation,

  • Mass change Δm±Δm caused by the geoengineering operation.

The time delay is Δt = teq − ts with the starting time of the geoengineering operation ts and the time of the mainshock teq. It is emphasized that Δt does not consider the ending time of the operation te, while the perturbation duration Δtp = te − ts does considers te. Some operations have a very high mass shift rate with very small perturbation durations, such as water reservoir impoundments (e.g., weeks to few months), while others have very long durations with very low shift rates, such as mining.

Interpretations of the relationships of the earthquake catalog data have limitations due to (a) comparability of different magnitudes and (b) uncertainties of single properties. Equations 24 indicate that only earthquakes with magnitudes >M3.0 can be corrected and expressed in moment magnitudes Mw. Thus, the observed magnitude \(M_{\textrm{obs}}\) is determined by correcting all given seismic magnitudes ML, MS, or mb to use a standard magnitude Mw that allows to compare seismic events with each other (Kanamori 1983; Scordilis 2006).
$$ M_S = 1.27\,(M_L-1) - 0.016\,M_L^2 \label{eq_ML} $$
(1)
$$ \begin{array}{rll} M_w &=& 0.67\,(\pm0.005)\,M_S \\ &&+\, 2.07(\pm0.03)\;\;\;3.0 \leq M_S \leq 6.1 \end{array} $$
(2)
$$ \begin{array}{rll} M_w &=& 0.99\,(\pm0.02)\,M_S \\ &&+\, 0.08(\pm0.13)\;\;\;6.2 \leq M_S \leq 8.2 \end{array} $$
(3)
$$ \begin{array}{rll} M_w &=& 0.85\,(\pm0.04)\,m_b \\ &&+ 1.03(\pm0.23)\;\;\;3.5 \leq m_b \leq 6.2 \end{array} $$
(4)
Moreover, uncertainties of single properties may result from the measurement or estimation errors as a result of limited accuracies and precisions. A high (low) accuracy means that a true and unknown value that needs to be determined, such as a true focal depth of 10 km, can (not) be exactly estimated or measured. Data points are distributed near (away from) the true value with 10.2 ± 0.5 km (15 ± 0.5 km). A high (low) precision means that a true and unknown value that needs to be determined can be a measured or estimated with low (high) statistical errors. Data points are distributed with a small (large) standard deviation around their mean value of 10.2 ± 0.5 km (10.2 ± 5 km).

Moreover, time delay and perturbation duration have relatively low uncertainties because the beginning and ending of the geoengineering operation and the occurrence time of the earthquake activity can be accurately and precisely measured (e.g., \(\frac{{\delta \Delta {\rm{t}}}}{{\Delta {\rm{t}}}} = \frac{{\pm 1{\rm{month}}}}{{10{\rm{years}}}} = 1\) %). The uncertainty of a mass change is also low because injection volumes, produced material of coal or water levels in water reservoirs can be also measured relatively accurately and precisely (e.g., \(\frac{{\delta {\rm{m}}}}{{\rm{m}}} = \frac{{\pm {{10}^{10}}{\rm{Mt}}}}{{{{10}^{11}}{\rm{Mt}}}} = 10 \) % water in a reservoir). Higher interpretation uncertainties, however, may result from seismological data such as from distance values \(\frac{{\delta {\rm{d}}}}{{\rm{d}}} = \frac{{\pm 0.2\,{\rm{km}}}}{{1\,{\rm{km}}}} = 20 \) % and from the seismic magnitudes Δm which are about ±0.3 for medium- to large-sized earthquakes on a global scale (Rhoades 1996). Both uncertainties still tend to be small in comparison to the uncertainty resulting from depth values of shallow earthquakes (e.g., \(\frac{{\delta {\rm{z}}}}{{\rm{z}}} = \frac{{\pm 5\,{\rm{km}}}}{{5\,{\rm{km}}}} = 100 \) %).

Finally, with respect to the focal mechanisms, the earthquake catalog considers only reverse, normal or strike-slip faults to keep interpretations as simple as possible. A focal mechanism was classified based on the dominant fault plain of the two potential failure planes that define a fault-plane solution. A dominant fault plain, for example, might be in conjunction with pre-existing tectonic conditions.

2.3 Correlations, regressions, and physical and statistical significances

The given properties were analyzed with respect to their physical meaningful relationships, proportionalities, and statistical correlations and significance. The following property combinations were taken into consideration and are discussed in the next section: M ∝ Δm, M ∝ d, M ∝ z, M ∝ Δt, M ∝ Δtp, and \(\overline d \propto \Delta t\).

It is hypothesized that these property combinations illuminate the physical most meaningful and significant relationships between geoengineering and earthquake activities. First, the expected seismic magnitude M should increase with an increasing man-made mass shift Δm because the more material is accumulated or removed, the larger is the un/loading of or the bending effect on the continental plate with respect to the pre-existing tectonic conditions (Pomeroy et al. 1976; Simpsom 1976; Klose 2011). Second, the direct distance of an earthquake to the operation point \(\overline d\) should increase with time t for fluid injections and reservoir loadings because triggering stress perturbations migrate away for the operation point due to pore pressure diffusion (McGarr et al. 2002).

For each property set, regression and correlation analyses were conducted by utilizing Kendall’s correlation test, Pearson’s test, and Spearman’s test. The three different correlation tests provide non-parametric measures (known as Kendall’s τ, Pearson’s product-moment correlation coefficient, and Spearman’s ρ) to quantify the extent of statistical dependence between pairs of observations or variables. The most common test is the Pearson’s test. The tests include an estimation for statistical significance. However, each test may not always provide good results in terms of statistical significance, and therefore all three tests were utilized to increase the interpretation quality of statistical dependencies between two variables, e.g., M and Δm. Results of these tests are presented in Tables 2 and 3.
Table 2

Results of regression and correlation analyses (see Table 1), including statistic significance testing for property combinations and with respect to different tectonic regimes

Magnitude \(M_{\textrm{obs}}\) versus

Tectonic regime

Regression

Kendall’s correlation test

Pearson’s test

Spearman’s test

Conclusion

\(M_{\textrm{obs}}=\hat \beta\,\log_{10}x\)\(\hat \beta\)

Significance p value

Coeff. corr

Significance p value

Coeff. τ

Significance p value

Coeff. ρ

Significance p value

Corr.

Significance

Mass change Δm

N, SS, R

0.415 ± 0.010

< <0.01

0.29

< <0.01

0.59

< <0.01

0.41

< <0.01

Medium

Yes

N

0.391 ± 0.011

< <0.01

0.41

< <0.01

0.62

< <0.01

0.58

< <0.01

Strong

Yes

SS

0.393 ± 0.023

< <0.01

0.09

0.29

0.54

< <0.01

0.14

0.27

Medium

No

R

0.488 ± 0.020

< <0.01

0.55

< <0.01

0.83

< <0.01

0.68

< <0.01

Strong

Yes

Distance d

N, SS, R

1.394 ± 0.040

< <0.01

0.38

< <0.01

0.47

< <0.01

0.53

< <0.01

Medium

Yes

N

1.377 ± 0.036

< <0.01

0.46

< <0.01

0.61

< <0.01

0.62

< <0.01

Strong

Yes

SS

1.289 ± 0.103

< <0.01

0.27

0.06

0.38

0.05

0.44

0.03

Medium

No

R

1.575 ± 0.083

< <0.01

0.44

< <0.01

0.60

< <0.01

0.56

< <0.01

Strong

Yes

Depth z

N, SS, R

1.315 ± 0.068

< <0.01

0.25

< <0.01

0.15

0.09

0.36

< <0.01

Medium

Yes

N

1.198 ± 0.129

< <0.01

0.13

0.11

0.02

0.46

0.20

0.11

Small

No

SS

1.344 ± 0.074

< <0.01

−0.30

0.93

−0.33

0.89

−0.40

0.93

Negative

No

R

1.456 ± 0.066

< <0.01

0.46

< <0.01

0.62

< <0.01

0.62

< <0.01

Strong

Yes

Time delay t

N, SS, R

2.229 ± 0.095

< <0.01

0.23

< <0.01

0.46

< <0.01

0.33

< <0.01

Medium

Yes

N

2.131 ± 0.127

< <0.01

0.08

0.24

−0.08

0.13

0.14

0.19

None

No

SS

2.572 ± 0.246

< <0.01

0.02

0.45

0.47

0.02

0.06

0.41

Small

No

R

2.227 ± 0.177

< <0.01

0.23

0.06

0.52

< <0.01

0.33

0.06

Medium

Yes

Table 3

Results of regression and correlation analyses (see Table 1), including statistic significance testing for property combinations and with respect to different geoengineering operations

 

Regression

Kendall’s correlation test

Pearson’s test

Spearman’s test

Conclusion

\(M_{\textrm{obs}}=\hat \beta\,\log_{10}x\)\(\hat \beta\)

Significance p value

Coeff. corr

Significance p value

Coeff. τ

Significance p value

Coeff. ρ

Significance p value

Corr.

Significance

Magnitude \(M_{\textrm{obs}}\) ∝ mass change Δm

  Mining

0.455 ± 0.017

< <0.01

0.20

0.18

0.03

0.46

0.25

0.21

Medium

No

  Extraction

0.420 ± 0.025

< <0.01

0.20

0.12

0.24

0.16

0.33

0.09

None

No

  Reservoir

0.401 ± 0.014

< <0.01

0.23

0.02

0.33

0.02

0.32

0.02

None

No

  Injection

0.385 ± 0.033

< <0.01

0.59

< <0.01

0.70

< <0.01

0.78

< <0.01

Strong

Yes

Magnitude \(M_{\textrm{obs}}\) ∝ perturbation duration tp

  Mining

1.770 ± 0.087

< <0.01

0.20

0.18

0.03

0.46

0.25

0.21

Medium

No

  Extraction

2.226 ± 0.145

< <0.01

NA

NA

0.18

0.24

0.07

0.39

None

No

  Reservoir

3.154 ± 0.285

< <0.01

0.19

0.13

0.19

0.13

0.24

0.08

Medium

No

  Injection

2.192 ± 0.249

< <0.01

0.83

< <0.01

0.84

< <0.01

0.83

< <0.01

Strong

Yes

Direct distance \(\overline{d}\) ∝ delay time t

  Mining

0.806 ± 0.046

< <0.01

−0.01

0.52

0.07

0.41

−0.003

0.50

None

No

  Extraction

0.634 ± 0.026

< <0.01

−0.11

0.74

−0.02

0.54

−0.19

0.78

None

No

  Reservoir

0.404 ± 0.028

< <0.01

−0.03

< <0.01

−0.09

0.72

−0.05

0.62

None

No

  Injection

0.424 ± 0.056

< <0.01

0.49

< <0.01

0.70

< <0.01

0.66

0.01

Strong

Yes

Correlations between catalog properties, which are discussed in the next section, are only illustrated by regression lines when all three correlation tests perform, in average, a strong correlation and strong statistical significance. Otherwise, a relationship is not considered as reliable, and no regression line is shown.

Again, it should be noted that observed magnitudes, that were provided for each seismic event throughout the literature, differ in type. Moreover, seismic moment magnitudes Mw that were converted from the different magnitude types mb, ML, or MS (Eq. 14) show less systematic variations. Empirical moment conversion models, however, indicated that systematic variations between different seismic moments tend to be smaller than the statistical variations observed in this study due to different geological and tectonic regimes (Braunmiller et al. 2005; Bakun 1984; Castellaro and Bormann 2007).

3 Discussion and results

3.1 Proportionality between mass change and seismic moment release

Figure 3 shows proportionalities between man-made mass changes Δm and seismic moment magnitudes \(M_{\textrm{obs}}\) of the observed earthquakes that nucleated in the vicinity of the operation points. The figure shows cross plots with respect to the tectonic regime and the type of geoengineering operation. It can be seen that correlations are only medium when the entire earthquake catalog is considered, independently from any other descriptive feature, such as type of tectonic regime or engineering operation. This could confirm the fact that the proportionality of M ∝ Δm is disputed in an ongoing debate since human-caused earthquakes were discovered. A closer look, however, shows a more differentiated result. The Δm-M-proportionality highly depends on the tectonic regime in which the operation takes place. Seismic magnitudes increase with the magnitude of the mass shift. This is independent whether masses were removed or accumulated in a region. Of less importance is the type of operation which causes the mass change. Here, no significant differences can be observed among the different geoengineering activities (Fig. 3).
https://static-content.springer.com/image/art%3A10.1007%2Fs10950-012-9321-8/MediaObjects/10950_2012_9321_Fig3_HTML.gif
Fig. 3

Seismic moment magnitude M > 3 of observed earthquakes versus associated man-made mass shift Δm in their vicinity. Regression curves are provided for statistically strong and significant correlations, which include normal fault regimes (N) and reverse fault regimes (R) (see Tables 2 and 3). Distributions are shown with respect to a focal mechanisms (N normal fault, SS strike-slip fault, R reverse fault) and b the associated geoengineering operations

In detail, strike-slip fault regimes tend to show no significant correlation with coefficients <0.5 (Table 2). Therefore, no regression line is shown in Fig. 3. Strong correlations with strong statistical significance, on the other hand, can be observed in normal and reverse fault regimes. Regression lines in Fig. 3 illustrate both Δm-M relationships. Proportionalities in normal and reverse fault regimes tend to differ slightly. The linear graphs in Fig. 4 show the best fits of a linear regression curves (\(M = \hat \beta \Delta m\)) for normal and reverse fault regimes. In fact, the proportionality constant \(\hat \beta\) derived from the regression analysis seems higher in reverse fault regimes, which suggests that revere faults tend to be more trigger-sensitive than normal faults. This could be explained by the alteration of the vertical lithostatic stress component in a reverse fault regime which is the minimal principal stress component σ3. This component is dominantly affected by the un/loading σL in the upper part of the Earth’s crust and, in turn, has a larger effect on the failure stress change in comparison to both horizontal lithostatic stress components, following the Mohr–Coulomb failure criteria. Due to the elastic response of the crust, the vertical stress component changes by Δσ3 = ΔσL, while both horizontal components change by a fraction of it \(\frac{\nu}{1 - \nu}\, \Delta \sigma_L \leq \Delta \sigma_{1,2} < \Delta \sigma_L\), depending on the Poisson’s ratio ν. Moreover, it is more likely that \(\Delta \sigma_{1,2} \geq \frac{\nu}{1 - \nu}\, \Delta \sigma_L\), because horizontal principal strains ϵ1,2 are infinitesimally small and can be assumed to be ϵ1,2 = 0 in >10-km depth, where faults tend to be locked (Sibson 1982, 1983). Geodetic surveys, however, tell us otherwise, because this assumed relationship is true only for a very unrealistic boundary condition.
https://static-content.springer.com/image/art%3A10.1007%2Fs10950-012-9321-8/MediaObjects/10950_2012_9321_Fig4_HTML.gif
Fig. 4

Seismic moment magnitude M > 3 of observed earthquakes versus lateral distance d of the earthquakes from nearby operation points of geoengineering activities. Regression curves are provided for statistically strong and significant correlations, which include normal fault regimes (N) and reverse fault regimes (R) (see Tables 2 and 3). Distributions are shown with respect to a focal mechanisms (N normal fault, SS strike-slip fault, R reverse fault) and b the associated geoengineering operations

3.2 Proportionality between distance, depth, and seismic moment

Figure 4 shows proportionalities between the lateral distance d and the seismic moment magnitude \(M_{\textrm{obs}}\) of the earthquakes. Correlations between d and \(M_{\textrm{obs}}\) are strong and statistically significant in reverse and normal fault regimes (Table 2). Seismic magnitudes \(M_{\textrm{obs}}\) tend to increase with the lateral distance d. The linear graphs in Fig. 4 illustrate the best fits of linear regression curves for normal and reverse fault regimes. The regression coefficients and their statistical significance are outlined in Table 2. No significant differences exist among groups of different engineering activities, as outlined in Table 3.

The positive correlation between d and \(M_{\textrm{obs}}\) could be explained by three effects: (a) the moment magnitude M increases with larger rupture size while d represents an approximation of the rupture size, (b) the distance d is a measure for the size of the geomechanically polluted area on the fault that is going to rupture, or (c) the plate bending effect increases on the continental crust the further away the weakest part of the crust is from the engineering operation point, where vertical forces are applied on the crust. The three effects are discussed as follows.

For the first case, it can be anticipated that the shown d-M distribution in Fig. 4 illuminates the moment magnitude M which increases with larger rupture size d. Thus, d might be a realization of the earthquake rupture, where M is in a causal relationship to d. However, this case seems to be less likely because the statistically significant d-M-correlations do only exist for normal and reverse faults but not for strike-slip faults.

The second case seems more likely because the larger the geomechanically polluted volume beneath an operation point is, the larger is the area on the fault that is going to rupture. However, such a fault needs to be favorably oriented in a compressive or extensive tectonic regime, where geomechanical pollution due to applied vertical forces (un/loading) destabilizes oblique faults. The lateral extension of the polluted area can be expressed by the observed distance d which, in turn, can be, to some extent, approximated as a function of the half length l/2 of a fault that is going to rupture (Soliva et al. 2008).

In the third case, d may also have a causal relationship to M. The more distant (measured by d) the weakest part of the crust (e.g., pre-existing fault) is from the operation point, the more may a vertically applied force, given by a mass change Δm, try to bend the continental plate. Such a plate-bending effect would favor the generation/reactivation of oblique reverse and normal faults. Strike-slip faults, on the other hand, are mechanically less likely to develop/rupture because their fault planes tend to be vertically oriented. Thus, only the second and third case could explain why reverse and normal faults show and strike-slip faults do not show statistically significant d-M correlations (see Table 2).

When comparing Figs. 4 and 5, the d-M-proportionality seems stronger with much higher statistical significance for normal and reverse faults than the relationship between M and depth z, as shown in Table 2. This could result from the fact that for older seismic events and, in particular, for earthquakes in stable continental regions, relatively little is known about very shallow seismogenesis. Hypocentral depth values are the least constrained parameters during iterative location procedures and have very high estimation errors because (a) regional seismic networks tend to be sparse, (b) recordings of shallow earthquakes tend to be exclusively via ray paths of narrow-degree angles between source and receiver, and (c) trial depths tend to remain unchanged during the iterations (Klose and Seeber 2007). Despite these issues, the observed depth z could be also approximated as a function of the half fault width w/2 (Soliva et al. 2008).
https://static-content.springer.com/image/art%3A10.1007%2Fs10950-012-9321-8/MediaObjects/10950_2012_9321_Fig5_HTML.gif
Fig. 5

Seismic moment magnitude M > 3 of observed earthquakes versus depth z of the earthquakes beneath nearby operations points of geoengineering activities. A regression curve is provided for statistically strong and significant correlation of reverse fault regimes (R) (see Tables 2 and 3). Distributions are shown with respect to a focal mechanisms (N normal fault, SS strike-slip fault, R reverse fault) and b the associated geoengineering operations

The physical and statistical relationships that are shown in Figs. 4 and 5 might explain that the larger the rock mass volume of the stress perturbation in the Earth’s crust is, the larger are l/2 and w/2 of the perturbed area on a fault that is going to rupture at a certain time in the future. As aforementioned, the size of the perturbed fault area and the expected seismic moment release on that fault, might be an effect of (a) the un/loading, (b) the crustal depth of the polluted fault, and (c) the pre-existing tectonic stress conditions. For small (shallow) earthquakes, l/2 and w/2 are typically on a similar scale and expressed by a rupture radius r; for larger (shallow) events, l/2 is typically larger then w/2, which is in agreement with an average brittle/ductile behavior as a function of crustal depth (Scholz 2002).

The three aforementioned conditions could explain the differences between induced and triggered seismic events. The human-caused stress perturbation on the continental crust and pre-existing tectonic conditions could be responsible for the earthquake nucleation and initial rupture of both types of earthquakes. The entire propagation of a triggered event is, in addition, a combined effect with regional geologic and tectonic conditions. Here, the polluted crustal volume is simply not large enough to bring an entire fault to failure. Given these facts, the 1989 M5.6 Newcastle earthquake, for example, can be classified as an induced event, and the 2007 M7.9 Wenchuan earthquake, which is currently discussed in the seismological community, could be classified as a triggered event.

Australia’s 1989 Mw5.6 Newcastle earthquake nucleated underneath a major 500-m deep coal mining field after 160 years of production (Klose 2007a). Coal and mining water were removed from the mining area, and the induced unload polluted a pre-existing reverse fault in a highly compressive stress regime. The unload brought a relative small part of the fault close to failure with a radius that was estimated r = l/2 = w/2 = 2014±107 m based on a geomechanical model, while exceeding critical failure stresses of 10 kPa. The following Eq. 5 is valid for this relatively shallow medium-sized earthquake to estimate the seismic moment that is expected to be released (Scholz 2002):
$$ M_o = \frac{16}{7}\,\Delta\sigma\;r^3 $$
(5)
and
$$ \log M_o = 1.5\, M_w + 9.1 $$
(6)
with an expected fault rupture radius r, which is equivalent to or ranges between d and z (including asperities), and an average stress drop of Δσ = 10 MPa. Both equations basically determine small- to medium-sized magnitudes for the specific cross-over regime (Scholz 2002). Thus, the resulting seismic moment magnitude is M\(_w = 5.4\frac{+0.2}{-0.3}\) (standard mean ± error). The observed seismic moment was Mw = 5.6 (Klose 2007a).

China’s 2008 M7.9 Wenchuan earthquake nucleated beneath the Zipingpu water reservoir 2.5 years after impounding on a listric reverse fault in a highly compressive stress regime (Klose 2011). A d = l/2 = 10±5-km-long and z = w/2 = 13±5-km-wide Beichuan fault area (see Table 1) was brought closer to failure by exceeding critical failure stresses. Equation 5 determines a seismic moment magnitude M\(_w = 6.9\frac{+0.3}{-0.5}\) (standard mean ± error) with respect to the following cross-over regime of w < l < 10·w (Scholz 2002). In comparison, the observed seismic moment magnitude was Mw = 7.9, and the actually ruptured fault was l = 250±50 km long and w = 13±3 km wide. This might illuminate that the Wenchuan earthquake might have been triggered and not induced because the polluted area covered only 5 % of the entire Beichuan fault, and the estimated seismic magnitude was M > 0.7 lower than the actually observed magnitude. The stress Δσ had been built up during the seismic cycle along the Beichuan fault, and the entire Longmen Shan. Thus, Δσ resulted from the natural tectonic loading of the fault. The water reservoir only provided the trigger and influenced the initial rupture, which was evident in the initial earthquake propagation upwards toward the Minjiang River Valley and directly to the surface loading area of the Zipingpu water reservoir. After the first 20 s, the rupture propagation process changed to a >100 s, lasting right-lateral NNE strike-slip rupture mechanism (Klose 2011).

3.3 Proportionality between perturbation duration, time delay, and seismic moment release

Overall, Figs. 6 and 7 show no significant correlations neither between the observed seismic moment magnitude \(M_{\textrm{obs}}\) and the time delay Δt of the earthquake nucleation after beginning of the engineering operation nor between \(M_{\textrm{obs}}\) and the perturbation duration Δtp. The Δtp-M distribution shows lower correlation coefficients than the Δt-M distribution for mining, reservoir impoundings, and fluid extraction (Table 3). Water reservoirs have the lowest perturbation durations Δtp in comparison to the time delays Δt and all other geoengineering activities. Impoundments of reservoirs occur within 86 ± 63 months. Such a “short” duration is also observed for fluid injections. However, seismic magnitudes of earthquakes caused by fluid injections tend to be significantly correlated with Δt and Δtp. This proportionality could be explained by the permeability which increases due to the overpressure of the injected fluids. Moreover, Fig. 8 shows a positive proportionality between the direct distance \(\overline{d}\) and the time delay Δt of earthquakes that nucleate near fluid injection points. A pressure front caused by fluid injections, for example, may migrate away from an operation point until it reaches the weakest nearby fault while further reducing their frictional strength (Zoback and Harjes 1997; Nicholson and Wesson 1992; Ahmad and Smith 1988). In contrast, Fig. 8 does not show any correlation between \(\overline d\) and the time delay Δt for mining, fluid extractions, and even reservoirs. This would confirm the understanding that the permeability near fluid extraction wells or watering shafts of underground mines decreases due to fracture closing caused by the lithostatic confining pressure (Jaeger et al. 2007).
https://static-content.springer.com/image/art%3A10.1007%2Fs10950-012-9321-8/MediaObjects/10950_2012_9321_Fig6_HTML.gif
Fig. 6

Seismic moment magnitude M > 3 of observed earthquakes versus time delay t of the earthquakes after nearby geoengineering activities started. Distributions are shown with respect to a focal mechanisms (N normal fault, SS strike-slip fault, R reverse fault) and b the associated geoengineering operations

https://static-content.springer.com/image/art%3A10.1007%2Fs10950-012-9321-8/MediaObjects/10950_2012_9321_Fig7_HTML.gif
Fig. 7

Seismic moment magnitude M >3 of observed earthquakes versus perturbation duration tp of nearby geoengineering activities. Distributions are shown with respect to a focal mechanisms (N normal fault, SS strike-slip fault, R reverse fault) and b the associated geoengineering operations

https://static-content.springer.com/image/art%3A10.1007%2Fs10950-012-9321-8/MediaObjects/10950_2012_9321_Fig8_HTML.gif
Fig. 8

Time delay t of human-caused earthquakes versus direct distances \(\overline{d}\) of the earthquakes from the operations points of the geoengineering activities, a as a function of their focal mechanisms (N normal fault, SS strike-slip fault, R reverse fault) and b as a function of the associated geoengineering operations

3.4 Behavior of man-made mass shifts in different tectonic regimes

First, Table 4 indicates, as aforementioned, that fault zones in reverse fault regimes tend to be, in general, much more trigger-sensitive than in normal or strike-slip fault regimes. Reverse faults seem to set off earthquakes due to mass changes that are about 1/10th to 1/20th of the mass needed to cause earthquakes on normal faults. Moreover, reverse faults tend to rupture deeper but geographically closer to an operation point in comparison to normal or strike-slip faults. Mining-caused earthquakes tend to occur in much shallower depths than seismicity triggered by fluid injections/extractions or reservoirs. This would confirm why there is the general notion that earthquakes need to be shallow to be considered as “human-caused.” The research findings show evidence that earthquakes triggered by human-made mass changes can occur in depths greater than 6 km. Finally, every second, human-caused earthquake, independent from the tectonic regime or the type of geoengineering, sets off 8 years (median) after the beginning of the geoengineering operation. On the other hand, every fourth event occurs after 2 years (1st quartile) or 23 years (3rd quartile). This shows that the 2.5-year time delay of the 2008 M7.9 Wenchuan earthquake after the impounding of the Zipingpu water reservoir is not an extreme short-time delay or even an outlier.
Table 4

Results of descriptive statistics (mean  ±  s.d.) on single properties (see Table 1) with respect to different tectonic regimes and different geoengineering operations

 

Seismic magnitude \(M_{\textrm{obs}}\)

Mass change Δm = (t)

Lat. distance \(d=\mbox{(km)}\)

Depth \(z=\mbox{(km)}\)

Time delay \(\Delta t=\mbox{(month)}\)

Perturbation duration \(\Delta t_p=\mbox{(month)}\)

Normal faults

4.4 ± 0.2

2.1 ± 0.8 1010

5.1 ± 1.0

3.6 ± 

257 ± 58

203 ± 63

Strike-slip faults

4.2 ± 0.3

1.1 ± 0.6 1010

4.9 ± 0.9

5.9 ± 

79 ± 21

52 ± 17

Reverse faults

4.9 ± 0.3

8.6 ± 4.8 108

3.2 ± 0.8

6.7 ± 

435 ± 132

477 ± 144

Mining

4.8 ± 0.2

5.7 ± 3.3 108

1.3 ± 0.2

2.6 ± 0.8

881 ± 199

881 ± 199

Extraction

4.9 ± 0.3

3.5 ± 1.8 1010

4.1 ± 1.1

4.5 ± 1.2

285 ± 40

220 ± 40

Reservoir

4.8 ± 0.2

1.2 ± 0.5 1010

6.6 ± 0.9

6.8 ± 1.0

123 ± 58

86 ± 63

Injection

3.1 ± 0.4

2.0 ± 1.6 109

2.8 ± 0.7

1.3 ± 0.4

94 ± 32

84 ± 31

4 Conclusion

Human-caused earthquakes have been studied as an environmental hazard since the twentieth century. This type of geological hazard is associated with large-scale geoengineering operations, such as mining, water reservoir impoundment, hydrocarbon production, fluid injection/extractions, coastal management, and deep geothermal energy production. This article is a first quantitative attempt to discuss systematics of historic earthquakes induced or triggered by human-made mass shifts on the upper Earth’s crust that were previously reported and independently analyzed in different research articles. A catalog of 92 earthquakes was generated based on more than 500 scientific research papers, conference proceedings, or abstracts.

The data specifically show that human-made mass changes can advance the clock of natural seismic cycles and induce or trigger new earthquakes. First, although the data set contains high statistical uncertainties, the analysis shows statistical significant correlations of physically explainable causalities in normal and reverse fault regimes. The uncertainties result mainly from limitations of seismological data estimations and due to different geographic locations, geology, tectonophysical conditions, and type of engineering activity.

The research findings show that human-made mass shifts in the upper part of the Earth’s crust ranging from kilotons to teratons can trigger or induce small- to large-sized earthquakes within a radius of up to 30 km. The expected seismic moment magnitude M increases the more the mass Δm is accumulated or removed and the more distant (measured by d) the most trigger-sensitive fault is from an operation point. The Δm-M and d-M relationships can be explained as causality by the elastostatic response of the Earth’s crust due to applied vertical forces of the man-made mass shifts at the upper part of the crust. These mass shifts induce bending moments on and change lithostatic stresses in the continental crust deep underground. Nearby faults can be brought closer to failure until they rupture. Furthermore, reverse faults tend to be more trigger-sensitive than normal or strike-slip faults.

Human-caused earthquakes can rupture after a decade. Every fourth event, however, nucleates after 2 years and the seismicity can last for decades, as the Koyna reservoir seismicity in India has been showing since the 1960s. A significant positive correlation is observed for fluid injections between seismic moment magnitude M and the time delay t of the nucleation time after engineering operations started. The correlation between the perturbation duration tp and M is even higher with >0.8. This indicates that pressure diffusion plays a major role for injections in comparison to mining, reservoir impoundings, or fluid extractions.

Finally, uncertainties still remain besides the statistical significances shown in this study, in particular, high estimation errors of seismologic data. Findings presented in this article are further investigated in ongoing research initiatives, which include numerical modeling and lab/in-situ experiments. Moreover, robust mechanical models, including hazards mapping tools in space and time are going to be developed, tested, and validated to ensure sustainable geoengineering in the Earth’s crust, in particular, in regions of high population densities.

Acknowledgements

The author expresses his gratitude to Think Geohazards, Lamont-Doherty Earth Observatory, Columbia University, Ritsumeikan University, the German Science Foundation, the Japan Society for the Promotion of Science, and NATO’s Science for Peace and Security Programme for their support of funding and resources from 2005 to 2012. He is grateful to A. McGarr and three anonymous reviewers who reviewed this manuscript. He would also like to thank C.H. Scholz, H. Ogasawara, V.I. Khalturin, K.H. Jacob, P.H. Liotta, L. Seeber, D.W. Simpson, and S.K. Negmatullaev for their general support and their feedback on the earthquake catalog data.

Copyright information

© Springer Science+Business Media B.V. 2012