etas fore 3 3 · 2011. 2. 24. · ! 6! 102!...

25
1 The Spatial Density of Foreshocks 1 Emily E. Brodsky 2 Dept. of Earth & Planetary Sciences 3 UC Santa Cruz; [email protected] 4 Abstract 5 Aftershocks follow a welldefined spatial decay pattern in the intermediate field. Here I 6 investigate the same pattern for foreshocks. Foreshock linear density decays as r 1.5+/0.1 over 7 distances r of 0.130 km for 15 minutes after magnitude 34 mainshocks. This trend is the same 8 as that of the aftershocks within the error of the measurement. This consistency of spatial 9 decays can be explained by the clustering inherent in earthquake interactions. No additional 10 preparatory process beyond earthquake triggering is necessary to explain the spatial decay. 11 Introduction 12 Earthquake triggering is a possible route into studying earthquake initiation and, by 13 extension, predictability. Earthquake interactions are generally dissected by looking at 14 relationships between earthquakes that are close in either time or space. For instance, 15 Felzer and Brodsky [2006] examined the spatial pattern of smaller earthquakes following 16 small magnitude mainshocks within 5 minutes. The study found that the linear density ρ of 17 aftershocks followed a welldefined pattern of ρ r γ where r is the distance from the 18 mainshock. The persistence of this spatial trend to large distances was interpreted as 19 indicative of a consistent triggering process across the entire range of distances. Since 20

Upload: others

Post on 01-Oct-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  1  

The  Spatial  Density  of  Foreshocks  1  

Emily  E.  Brodsky  2  

Dept.  of  Earth  &  Planetary  Sciences  3  

UC  Santa  Cruz;  [email protected]  4  

Abstract  5  

Aftershocks  follow  a  well-­‐defined  spatial  decay  pattern  in  the  intermediate  field.  Here  I  6  

investigate  the  same  pattern  for  foreshocks.  Foreshock  linear  density  decays  as  r-­‐1.5+/-­‐0.1  over  7  

distances  r  of  0.1-­‐30  km  for  15  minutes  after  magnitude  3-­‐4  mainshocks.  This  trend  is  the  same  8  

as  that  of  the  aftershocks  within  the  error  of  the  measurement.    This  consistency  of  spatial  9  

decays  can  be  explained  by  the  clustering  inherent  in  earthquake  interactions.  No  additional  10  

preparatory  process  beyond  earthquake  triggering  is  necessary  to  explain  the  spatial  decay.      11  

Introduction  12  

Earthquake  triggering  is  a  possible  route  into  studying  earthquake  initiation  and,  by  13  

extension,  predictability.  Earthquake  interactions  are  generally  dissected  by  looking  at  14  

relationships  between  earthquakes  that  are  close  in  either  time  or  space.  For  instance,  15  

Felzer  and  Brodsky  [2006]  examined  the  spatial  pattern  of  smaller  earthquakes  following  16  

small  magnitude  mainshocks  within  5  minutes.  The  study  found  that  the  linear  density  ρ  of  17  

aftershocks  followed  a  well-­‐defined  pattern  of  ρ ∝r-­‐γ  where  r  is  the  distance  from  the  18  

mainshock.  The  persistence  of  this  spatial  trend  to  large  distances  was  interpreted  as  19  

indicative  of  a  consistent  triggering  process  across  the  entire  range  of  distances.    Since  20  

Page 2: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  2  

seismic  waves  are  thought  to  be  significant  triggerers  at  great  distances,  the  inference  was  21  

those  seismic  waves  were  an  important  part  of  triggering  at  all  distances  [Hill  et  al.,  1993;  22  

Brodsky  et  al.,  2000;  West  et  al.,  2005;  Hill  and  Prejean,  2007;  Van  der  Elst  and  Brodsky,  23  

2010].    24  

Foreshocks  are  the  best-­‐known  predictor  of  earthquakes.  Around  ~50%  of  earthquakes  are  25  

preceded  by  observable  foreshocks  [Abercrombie  and  Mori,  1996].    Early  thinking  26  

attributed  foreshocks  to  a  stress  accumulation  process  that  ultimately  culminated  in  a  large  27  

earthquake  [e.g.,  Jones  and  Molnar,  1979].    In  this  scenario,  the  spatial  distribution  of  28  

foreshocks  is  controlled  by  the  stress  distribution  and  fatigue  on  the  fault  plane.    More  29  

recent  work  has  entertained  other  possibilities  such  as  the  foreshocks  being  symptomatic  30  

of  an  earthquake  cascade.  Earthquakes  are  constantly  triggering  each  other.    When  31  

seismicity  is  tightly  clustered,  the  likelihood  of  a  large  earthquake  increases  and,  after  the  32  

fact,  the  seismicity  cluster  is  interpreted  as  foreshocks  [Helmstetter,    et  al.,  2003;  Felzer  et  33  

al.,  2004].  This  type  of  clustering  results  in  distinct  patterns,  like  the  Inverse  Omori’s  Law  34  

[Helmstetter  et  al.,  2003].    Still  other  work  has  suggested  that  both  kinds  of  foreshocks  35  

exist  in  earthquake  catalogs  [McGuire  et  al,  2005;  Zhuang  et  al.,  2008].    36  

This  paper  starts  by  establishing  that  the  foreshock  spatial  trend  follows  a  similar  37  

relationship  to  the  aftershock  one.  At  first,  this  observation  is  disconcerting.  If  the  spatial  38  

decay  of  r-­‐γ  is  generated  by  aftershocks,  then  why  is  it  present  before  the  mainshock?    I  will  39  

follow  up  the  observation  by  showing    that  a  statistical  seismicity  model  in  which  40  

aftershocks  follow  the  spatial  decay  law  of  r-­‐γ    automatically  generates  foreshocks  with  the  41  

Page 3: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  3  

same  spatial  decay.  The  trend  can  be  a  natural  consequence  of  the  clustering  provided  by  42  

the  mutual  triggering  of  the  earthquake  cascade.  43  

Observation  44  

I  begin  by  using  the  well-­‐located  LSH  catalog  of    Southern  California  earthquakes  1981-­‐45  

2005  to  compare  the  spatial  distribution  of  aftershocks  and  foreshocks  [Lin  et  al.,  2007].  46  

Only  magnitudes  >2  are  included  to  ensure  completeness.  For  the  purpose  of  comparison,  I  47  

use  a  similar  windowing  criteria  as  Felzer  and  Brodsky  [2006]  to  isolate  sequences.  48  

Mainshocks  are  defined  as  earthquakes  that  are  at  least  4  days  after  and  0.5  day  before  any  49  

larger  earthquake  at  any  distance  in  the  catalog  from  any  larger  earthquake.  Aftershocks  50  

are  defined  to  be  earthquakes  that  follow  a  larger  identified  mainshock  within  the  specified  51  

time  period  Δt.  Foreshocks  are  earthquakes  that  are  followed  by  a  larger  identified  52  

mainshock  within  a  specified  time  period  Δt.      When  comparing  foreshocks  and  aftershocks  53  

for  a  particular  sequence,  the  same  value  of  Δt  is  used  to  define  both  types  of  seismicity.    54  

These  windowing  criteria    are  not  meant  to  imply  a  physical  limit  to  interaction.    The  55  

insensitivity  of  the  aftershock  spatial  decay  to  the  windowing  details  was  established  in  56  

previous  work  [Felzer  and  Brodsky,  2006].  The  point  of  the  present  study  is  simply  to  57  

compare  foreshock  decays  to  aftershock  decays  given  the  same  conditioning  on  the  data.    58  

To  measure  density,  the  mainshocks  (and  their  accompanying  sequences)  for  the  entire  59  

catalog  are  combined  to  improve  the  statistical  sampling.  A  single,  ordered  vector  of  60  

aftershock-­‐mainshock  hypocentral  distances  is  created,   r .  The  linear  density  of  61  

Page 4: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  4  

aftershocks  (and  foreshocks)  is  estimated  at  the  midpoints  between  each  element  of  the  62  

combined  aftershock  distance  vector   r by    63  

ρ ri + ri+12

⎛⎝⎜

⎞⎠⎟=

1ri+1 − ri

    (1)  64  

For  mainshocks  with  magnitude  <4,  the  fault  rupture  length  is  expected  to  be  less  than  the  65  

spatial  precision  of  the  data  (~0.1  km)  and  therefore  the  point  approximation  embedded  in  66  

eq.  1  is  appropriate.  Larger  magnitude  mainshocks  would  require  a  better  geometrical  67  

model  for  earthquake  density  estimation.  68  

The  aftershocks   following   the  magnitude  3-­‐4  mainshocks  yields  a   fit  of  γ=1.5+/-­‐  0.08   for  69  

data   from  distances  of  0.1-­‐30  km   for  aftershocks   for  15  minute   (Figure  1).   (Error  ranges  70  

here  and  throughout  the  paper  are  standard  deviations  of  1000  bootstrap  trials  on  the  fit).  71  

This  observation  is  similar  to  Richard-­‐Dinger  and  Stein  [2010].    This  value  of  γ    is  defined  72  

slightly  differently   than  Felzer  and  Brodsky  [2006]  because   it   is  based  on  epicenters,  not  73  

hypocenters.    This  study  uses  epicenters  to  facilitate  comparison  with  the  2-­‐D  ETAS  model  74  

below.  For  Δt=  5  minutes,  a  similar   trend  appears,  but   far   fewer   foreshocks  are  recorded  75  

(Figure  1b).  76  

For  close  distances,  the  largest  mainshocks  in  this  range  can  no  longer  be  treated  as  point  77  

sources  and  so  the  distance  r  is  likely  inaccurate.  For  large  distances,  the  number  of  78  

triggered  events  is  sufficiently  small  that  unrelated  seismicity  overwhelms  the  signal.  The  79  

background  level  can  be  seen  clearly  by  shuffling  the  event  times  randomly  for  the  same  80  

earthquake  locations  and  re-­‐measuring  the  density.  The  thin  light  lines  in  Figure  1  show  81  

Page 5: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  5  

100  such  reshufflings  and  the  thick  light  lines  show  the  mode  values  of  the  time-­‐82  

randomized  set.  As  an  unusually  closely  spaced  pair  of  events  can  easily  dominate  the  83  

median  or  mean  of  the  density,  the  mode  is  the  best  representative  of  the  ordinary  84  

background  level  of  seismicity.  This  background  level  demonstrated  that  indeed  the  85  

background  interferes  with  recording  the  aftershock  signal  at  large  distances,  but  an  86  

aftershock  signal  is  visible  up  to  at  least  30  km  distance  from  the  mainshock.    87  

Interestingly,   the   foreshocks   follow   almost   the   same   trend   (Figure   1).   For   15   minutes  88  

before,   the   decay   of   the   composite   foreshock   sequence   from   the   future   mainshock   site  89  

follows   r-­‐1.5+/-­‐0.1   and   for   the   5  minutes   foreshocks   the   density   is   r1.6+-­‐/-­‐0.2.     The   foreshock  90  

decay  is  also  truncated  by  the  background  seismicity  at  large  distances.  91  

 92  

Clustering  Model  93  

In  order  to  recreate  the  observed  foreshock  trend,  I  will  employ  an  Epidemic  Triggering  94  

Aftershock  Sequence  (ETAS)  model  that  combines  empirical  statistical  laws  of  seismicity  to  95  

generate  self-­‐consistent  synthetic  earthquake  catalogs  [Ogata,  1999].  An  earthquake  96  

triggers  successive  earthquakes  with  a  probability  determined  by  the  magnitude  of  the  97  

mainshock.  Omori’s  Law  and  a  spatial  decay  relationship  determine  the  time  and  distance  98  

separating  the  aftershock  from  the  mainshock.  Successive  earthquake  magnitudes  are  then  99  

determined  by  Gutenberg-­‐Richter  statistics  and  then  each  earthquake  generates  its  own  100  

aftershocks.    101  

Page 6: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  6  

The  ETAS  formulation  has  been  explored  analytically  and  numerically  in  the  literature  102  

[Helmstetter  and  Sornette,  2002;  Felzer  et  al.,  2004;  McGuire  et  al.,  2005;  Zhuang  et  al.,  103  

2007].  Windowing  and  selection  criteria  often  result  in  behaviors  that  are  challenging  to  104  

evaluate  analytically.  I  therefore  use  a  numerical  model  directly  for  this  work.    The  105  

parameters  used  here  are  meant  to  illustrate  the  relationship  between  the  foreshock  and  106  

aftershock  spatial  decay.  They  are  certainly  not  exhaustive,  but  are  realistic  and  suffice  to  107  

elucidate  the  role  of  clustering  in  generating  a  foreshock  spatial  pattern.  Variations  on  the  108  

basic  ETAS  model  are  implemented  by  utilizing  different  space  or  time  modifications  of  the  109  

observational  laws.  As  the  permutations  of  the  model  are  important  and  the  practicalities  110  

of  implementation  are  sometimes  difficult,  I  will  go  into  some  detail  on  the  exact  forms  111  

used  in  the  numerical  calculations  done  here.  112  

ETAS  Implementation  113  

For  this  particular  implementation,  background  seismicity  is  imposed  as  a  Poissonian  114  

process  with  rate  λ  is  over  the  duration  of  the  observed  catalog  (24  years)  with  the  115  

magnitude  of  each  earthquake  determined  by  Gutenberg-­‐Richter.  The  background  rate  is  116  

chosen  so  that  the  total  number  of  synthesized  events  (including  aftershocks)  is  117  

approximately  equal  to  the  observed  number.  Each  of  these  background  earthquakes  118  

generates  aftershocks,  which  in  turn  spawn  their  own  sequences.  The  cascade  is  produced  119  

by  the  combination  of  four  empirical  equations  cast  in  terms  of  probability  distributions:  120  

I) Gutenberg-­‐Richter  121  

The  number  of  aftershocks  with  magnitudes  greater  than  or  equal  to  M  is  governed  by    122  

Log  N(M)=  -­‐bM+a                 (2)  123  

Page 7: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  7  

where  a  and  b  are  constants.    124  

 125  

To  obtain  a  cumulative  density  function  (CDF),  N(M)  is  normalized  by  the  total  number  of  126  

earthquakes  in  the  catalog,  i.e.,  N(Mmin)  where  Mmin  the  minimum  magnitude  of  the  127  

simulated  catalog.    A  random  number  p1  is  then  selected  between  0  and  1  for  each  128  

aftershock  and  inverted  using  the  CDF  into  a  magnitude,  i.e.,    129  

M=-­‐log  (p1(10-­‐b  Mmin  -­‐10-­‐b  Mmax))/b           (3)  130  

where  p1  is  a  random  variable  between  0  and  1  and  Mmax  is  the  maximum  magnitude  size  to  131  

be  generated.    The  extra  term  in  brackets  is  a  normalization  that  accounts  for  discarding  132  

values  of  p1  that  result  in  M>Mmax  [Sornette  and  Werner,  2005].    133  

 134  

II) Omori’s  Law  135  

The  rate  of  seismicity  as  a  function  of  time  t  from  a  mainshock  is    136  

dNAS/dt  =K/(t+Comori)β (4)  137  

where  β  and  Comori  are  constants  and  K  is  a  function  of  mainshock  magnitude  to  be  138  

discussed  below.  The  constant  Comori  has  proven  particularly  problematic  to  measure  as  it  is  139  

often  biased  by  completeness  problems  at  short  times  after  an  earthquake.  Recent  work  140  

has  suggested  that  Comori  is  significantly  shorter  than  previously  imagined  [Peng  et  al.,  141  

2007].    142  

In  probabilistic  terms,  143  

t=((tmax  +  COmori)1-­‐β  p2  +  (1-­‐p2)  COmori(1-­‐β))1/(1-­‐β)    -­‐  COmori     (5)  144  

where  p2  is  a  random  variable  between  0  and  1.  145  

146  

Page 8: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  8  

III)     Aftershock  Productivity  147  

The  numerator  of  Omori’s  Law  determines  the  number  of  aftershocks  from  a  mainshock  of  148  

magnitude  M  149  

K  =  C10α(M-­‐Mmin)               (6)  150  

where  C  and  α are constants. 151  

Studies agree that α=1 for Southern California seismicity [Felzer et al., 2004; Helmstetter et al., 152  

2005]. The productivity constant is much more poorly determined, in part because it depends on 153  

the minimum active magnitude [Hardebeck et al., 2008]. Furthermore, the parameter C above is 154  

defined for the instantaneous rate in Omori’s Law. In order to use eq. 6, the total number of 155  

aftershocks must be computed for a sequence and so in practice what is required is the integral of 156  

eq. 4 over the finite time tmax, that the aftershock sequence will be simulated. The practical 157  

equation is therefore, 158  

NAS = C’  10α(M-­‐Mmin) 159  

where  C’=C((tmax  +  COmori)1-­‐β    -­‐  COmori  (1-­‐β))/(1-­‐β)       (7)  160  

Eq. 7 is implemented deterministically based on aftershock magnitudes. In reality, aftershock 161  

productivity probably varies in time and space, but those deviations are not sufficiently well 162  

mapped out to justify a more sophisticated implementation at this time. In the simulations, I 163  

apply a value of C’ that fits the observed productivity in the LSH data set with the aftershock 164  

identification criteria used here (Figure 2). 165  

For non-integer value of NAS, the fractional part is interpreted as a probability. For each 166  

sequence, a uniform random variable p is generated between 0 and 1 and if the value is less than 167  

the non-integer fraction, an extra aftershock is added. 168  

Page 9: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  9  

169  

IV) Distance  Decay    170  

The  distance  fall-­‐off  is  implemented  by  normalizing  the  distribution  to  form  a  CDF  171  

assuming  that  the  aftershocks  are  distributed  a  distances  greater  than  dmin  from  the  172  

mainshock.  Since  γ>1,  no  upper  bound  to  the  distance  fall-­‐off  is  necessary.    The  173  

probabilistic  form  for    determining  r  for  each  aftershock  is    174  

r  =    dmin  (1-­‐p3)  1/(1-­‐γ)                 (8)  175  

where  p3  is  a  random  variable  between  0  and  1.  As  there  are  no  constraints  on  the  angular  176  

distribution  an  angle  θ  is  selected  with  uniform  probability  from  0  to  2π.  177  

Combining  eqs.  3,  5,  7  and  8  yields  an  ETAS  catalog.  In  the  implementation  of  ETAS,  it  is  178  

helpful  to  note  the  order  of  calculation.    First  the  total  number  of  aftershocks  must  be  179  

calculated  for  each  extent  earthquake  using  eq.    7,  then  each  aftershock  is  assigned  a  time,  180  

magnitude  and  location  using  eqs.  3,  5  and  8.    Then  the  aftershocks  are  allowed  to  generate  181  

their  own  aftershocks  by  the  same  procedure  until  the  cascade  comes  to  an  end.  182  

Simulation  Results  183  

The  ETAS  catalog  was  made  to  mimic  the  catalog  of  Figure  1  as  closely  as  possible.  The  184  

parameters  of  Table  1  were  used  and  the  same  windowing  criteria  were  used  to  isolate  185  

mainshocks  as  in  Figure  1.  The  same  stacking  procedure  and  the  same  windows  for  186  

identifying  15  and  5-­‐minute  aftershock  and  foreshocks  were  used.  187  

Page 10: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  10  

The  overall  trends  of  the  Figure  1  are  reproduced  in  the  simulated  catalog  (Figure  3).  Both  188  

the  aftershock  and  foreshock  densities  have  a  best-­‐fit  spatial  decay  consistent  with  the  189  

imposed  exponent.    This  consistency  in  foreshock  and  aftershock  behavior  exists  even  190  

though  there  was  no  specific  foreshock  preparatory  process  in  the  ETAS  model.  The  utility  191  

of  the  statistical  simulation  is  that  it  elucidates  that  non-­‐intuitive  behavior,  like  Figure  1,  192  

emerges  naturally  from  the  earthquake  sequences.    Like  the    Inverse  Omori’s  Law,  the  193  

spatial  decay  of  the  foreshocks  is  symptomatic  of  the  type  of  increased  seismicity  that  is  194  

likely  to  trigger  a  large  earthquake.  Figure  3  demonstrates  that  the  observed  parallel  195  

between  the  foreshock  and  aftershock  decay  can  arise  naturally  from  a  clustering  model.  196  

At  very  large  distances,  the  simulated  catalog  has  larger  clusters  due  to  unisolated  ongoing  197  

sequences.  The  isolation  distance  of  100  km  based  on  the  SCSN  catalog  is  insufficient  for  198  

the  simulation  because  of  the  more  homogeneous  distribution  of  epicenters,  i.e.,  there  are  199  

no  simulated  faults.      200  

There  is  one  conspicuous  difference  between  the  simulations  and  the  real  catalog.  The  ratio  201  

of  identified  aftershocks  to  foreshocks  in  the  real  catalog  is  much  lower  than  in  the  202  

simulated  catalog.  In  the  actual  catalog,  the  ratio  is  2.2  for  Δt=5  min  and  2.8  for  Δt=15  min.,  203  

while  the  simulations  generally  result  in  ratios  between  7  and  8  for  the  parameters  used  204  

here.  Aftershock-­‐foreshock  ratios  have  been  analyzed  extensively  elsewhere  [e.g.,  McGuire  205  

et  al.,  2005]  and  are  not  the  focus  of  this  study.  Nonetheless,  some  assessment  of  the  source  206  

of  the  discrepancy  is  in  order.    207  

One  possibility  is  that  the  observed  catalog  artificially  depresses  the  aftershock/foreshock  208  

ratio  because  of  completeness  issues.  It  is  generally  more  difficult  to  detect  aftershocks  209  

Page 11: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  11  

than  foreshocks  because  the  mainshock  and  temporally  high  seismicity  rate  can  obscure  210  

individual  events.  Therefore,  the  earthquakes  that  are  above  the  magnitude  of  catalog  211  

completeness  for  aftershocks  can  be  significantly  higher  than  normal  [e.g.,  Peng  et  al.,  212  

2007].  213  

I  assess  the  effects  of  catalog  incompleteness  following  mainshocks  issue  in  two  different  214  

ways.  First,  I  restrict  the  data  to  a  higher  completeness  level  consistent  of  M=2.5  (Figure  4).  215  

This  restriction  decreases  the  simulated  aftershock/foreshock  ratio  as  the  ratio  is  sensitive  216  

to  the  range  of  magnitudes  between  the  mainshock  and  Mth  [McGuire  et  al.,  2005].    For  217  

Δt=15  minute,  the  simulated  ratio  in  the  realization  of  Figure  4c  is  3.8.    In  contrast,  the  218  

observed  aftershock  to  foreshock  ratio  for  the  Δt=  15  minute  data  increases  slightly  to  2.9.  219  

This  improved  consistency  between  simulation  and  observation  is  expected  based  on  220  

previous  work  on  foreshock  rates  in  the  region    [Felzer  et  al.,  2004]  and  indicates  that  the  221  

lack  of  aftershock  detection  is  an  important  issue  in  the  observations.  However,  the  222  

simulated  ratios  here  still  do  not  exactly  match  the  observations.    It  is  possible  that  the  223  

completeness  threshold  needs  to  be  raised  still  further  to  ensure  aftershock  detection,  but  224  

any  further  restrictions  limit  the  dataset  to  an  unreasonable  small  size.  225  

As  an  alternative  method  of  probing  time-­‐variable  completeness,  I  exclude  a  short  window  226  

immediately  before  and  following  the  mainshock  in  measuring  the  aftershock  and  227  

foreshock  densities  (Figure  5).  This  strategy  has  the  advantage  of  preserving  the  number  of  228  

events  at  larger  times  when  the  completeness  threshold  drops  to  the  overall  catalog  value.  229  

Excluding  a  window  of  1  minute  on  either  side  of  the  mainshock  increases  the  230  

aftershock/foreshock  ratio  to  3.3  for  Δt=  15  min.  This  ratio  is  still  smaller  than  the  231  

Page 12: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  12  

simulated  ratio,  but  it  does  again  indicate  that  early-­‐time  aftershock  detection  is  likely  a  232  

problem  in  the  observations.      233  

The  aftershock-­‐foreshock  ratio  can  be  adjusted  in  the  simulations  reducing  the  minimum  234  

magnitude  of  simulation  Mmin  or  increasing  the  productivity  constant  C’,  both  of  which  have  235  

the  effect  of  increasing  the  average  number  n  of  aftershocks  per  mainshock  [Sornette  and  236  

Werner,  2005].  As  n  increases,  the  system  approaches  a  critical  state  and  the  foreshock  rate  237  

should  increase  relative  to  the  aftershock  rate  [McGuire  et  al.,  2005].  The  calculations  also  238  

become  prohibitively  expensive.  Thus  far,  the  observed  foreshock  to  aftershock  ratio  has  239  

not  been  simulated.    One  possible  explanation  is  that  there  is  a  genuine  difference  in  the  240  

ETAS  process  from  the  observations.  If  so,  the  aftershock/foreshock  ratio  is  the  only  241  

evidence  of  such  a  process.  Another  explanation  is  that  aftershock  completeness  is  still  the  242  

overwhelming  observational  problem.    243  

More  importantly  for  this  study,  the  spatial  decay  of  the  foreshock  density  in  both  Figures  4  244  

and  5  remains  consistent  with  the  aftershock  decay.  This  consistency  of  Figure  5  is  245  

particularly  useful  in  light  of  the  commentary  of  Richards-­‐Dinger  et  al.  [2010]  that  the  246  

observed  aftershock  density  decay  may  be  controlled  by  events  that  occurred  before  the  247  

arrival  of  the  seismic  waves.  No  such  events  are  included  in  the  fits  of  Figure  5  as  P  waves  248  

travelling  at  6  km/s  will  have  passed  through  360  km  in  the  first  minute.      249  

Conclusions  250  

Earthquakes  are  preceded  by  a  halo  of  earthquakes  that  decays  with  distance  much  like  251  

aftershocks.  These  foreshocks  are  an  expected  consequence  of  earthquake  interaction  and  252  

Page 13: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  13  

clustering.  Their  existence  and  location  near  the  eventual  rupture  does  not  in  itself  indicate  253  

a  preparatory  strain  accumulation  process  beyond  ordinary  earthquake  triggering.    254  

A  foreshock  preparatory  process  could  exist  in  the  Earth  exists  beyond  the  clustering  255  

process,  but  it  is  not  required  by  the  observations.  The  only  potential  evidence  seen  here  256  

for  such  a  preparatory  process  is  in  the  ratio  of  total  number  of  foreshocks  to  aftershocks.  257  

However,  the  data  indicate  that  catalog  completeness  is  a  serious  problem  that  could  258  

artificially  depress  the  aftershock/foreshock  ratio  and  therefore  this  ratio  is  not  the  most  259  

reliable  indicator  of  the  underlying  physics.  The  spatial  decay  of  both  the  aftershocks  and  260  

foreshocks  is  well  explained  by  the  clustering  model.  261  

 262  

Acknowledgements  263  

Many  thanks  to  Karen  Felzer,  Keith  Richards-­‐Dinger,  Peter  Shearer,  Ross  Stein  and  Agnes  264  

Helmstetter  for  many  helpful  conversations  and  ideas.  This  work  was  funded  in  part  by  the  265  

Southern  California  Earthquake  Center.    266  

267  

Page 14: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  14  

a267  

b  268  

Figure  1.  Aftershocks  and  foreshocks  separated  from  the  mainshock  by  time  (a)    Δt=  15  minutes  and  (b)  Δt  =5  269  

minutes   for   mainshock   magnitudes   3-­4.   A   least-­squares   fit   of   the   0.1-­30   km   data   yields   γ=1.5+/-­   0.08   for  270  

aftershocks   and  1.5+/-­0.1   for   foreshocks.     For   5  minute   data,   γ   =1.5+/-­0.1   and  1.6   +/-­   0.2   for   aftershocks   and  271  

foreshocks,  respectively.  Time-­randomized  catalog  super-­imposed  to  illustrate  background  level  of  density.    The  272  

Page 15: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  15  

thin,   light   lines   illustrate   each   of   100   time-­randomized   catalogs   for   the   same   aftershock   (blue)   and   foreshock  273  

(red)  selection  criteria  as  used  in  the  original  data.  The  thick   lines  are  the  mode  values   for   logarithmic  bins  of  274  

these  100  realizations.    275  

276  

Page 16: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  16  

 276  

 277  Figure  2.  Simulate  and  observed  identified  aftershocks  as  a  function  of  magnitude  using  the  parameters  in  Table  278  

1.   The   fit   between   the   simulation   and   observations   for   the  magnitudes   in   the   study   range   (M   3-­4)   is   used   to  279  

calibrate   the   simulation   aftershock   productivity   C’.     Solid   lines   indicate   best   fit   trends   to   the   simulations   and  280  

observations  for  M  3-­4  mainshocks.  The  apparent  depletion  of  aftershocks  for  mainshocks  with  magnitude>  4  is  281  

likely  a  completeness  issue  as  documented  elsewhere  (Felzer  et  al.,  2004;  Helmstetter  et  al.,  2005].  282  

 283  

284  

Page 17: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  17  

a284  

b  285  

Figure  3  Simulated  aftershock  and  foreshock  decay  from  the  ETAS  parameters  in  Table  1  with  (a)    Δt=15  minutes  286  

and   (b)   Δt=5  minutes.   The   best-­fit   decay   exponent   is   γ=1.5/-­0.04   for   the   aftershocks   and   1.4+/-­0.09     for   the  287  

foreshocks   for   Δt=15   minutes.   For   Δt=5   minutes,   γ=1.5+/-­0.05   for   the   aftershocks   and   1.5   +/-­0.1   for   the  288  

foreshocks.    289  

Page 18: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  18  

 290  

a291  

b  292  

Page 19: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  19  

c293  

d  294  

Figure   4.   Observed   and   simulated   aftershock   and   foreshock   sequences   with   a   completeness   magnitude   of  295  

Mth=2.5.  All  other  simulation  parameters  are  the  same  as  Table  1.    (a)  Observed  earthquake  densities  for    Δt=15  296  

minutes  and  (b)  Δt=5  minutes.  (c)  Simulated  earthquake  densities  for    Δt=15  minutes  and  (d)  Δt=5  minutes.    297  

298  

Page 20: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  20  

a298  

b299  

Page 21: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  21  

c300  

d  301  

Figure  5.  Observed  and  simulated  aftershock  and  foreshock  sequences  excluding  a  1  minute  interval  both  before  302  

and   after   the  mainshock.   All   other   simulation   parameters   are   the   same   as   Table   1.     (a)   Observed   earthquake  303  

densities  for  Δt=15  minutes  and  (b)  Δt=5  minutes.  (c)  Simulated  earthquake  densities  for  Δt=15  minutes  and  (d)  304  

Δt=5  minutes.    305  

306  

Page 22: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  22  

 306  

 307  

 308  

 309  

 310  

Parameter   Values   Rationale  

Completeness  threshold  Mth   2   Catalog  completeness  

tmax   24  years   Catalog  duration  

Mmin   0   Computationally  limited  

Spatial  Decay  exponent  γ 1.5   Aftershock  density  spatial  fit  (Figure  1)  

Productivity  Constant  C’   0.03  Earthquakes/day   Aftershock  productivity  fit  (Figure  3)  

Omori  Exponent  β 1.34     Hardebeck  et  al.  [2008]  

Background   earthquake   rate   λ for  

M>Mmin  

300  Earthquakes/day   Matches   observed   catalog   rate   with  aftershocks  are  included  

Minimum  aftershock  distance  dmin   100  m   Minimum  location  accuracy  

Omori  Delay  Comori   130s     Peng  et  al.  [2007]  

b     1   Standard  value  

α 1   Felzer  et  al.  [2004];  Helmstetter  et  al.  [2005]  

Table  1.  ETAS  Parameters  for  simulations  in  Figures  2-­5.    311  

312  

Page 23: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  23  

References  312  

Abercrombie,  R.E.  and  J.  Mori  (1996),  Occurrence  patterns  of  foreshocks  to  large  313  

earthquakes  in  the  western  United  States,  Nature,  381,  303-­‐307.  314  

Brodsky,  E.  E.,  V.  Karakostas,  and  H.  Kanamori  (2000),  A  New  Observation  of  Dynamically  315  

Triggered  Regional  Seismicity:  Earthquakes  in  Greece  Following  the  August,  1999  Izmit,  316  

Turkey  Earthquake,  Geophys.  Res.  Let.,  27,  2741-­‐2744.  317  

Felzer,  K.  R.,  R.  E.  Abercrombie,  and  G.  Ekstrom  (2004),  A  common  origin  for  aftershocks,  318  

foreshocks,  and  multiplets,  Bull.  Seism.  Soc.  Am.,  94,  88-­‐98.  319  

Felzer,  K.  R.  and  E.  E.  Brodsky  (2006),  Decay  of  aftershock  density  with  distance  indicates  320  

triggering  by  dynamic  stress,  Nature  ,  441,  735-­‐738.  321  

Helmstetter,  A.  and  D.  Sornette  (2002)  Subcritical  and  supercritical  regimes  in  epidemic  322  

models  of  earthquake  aftershocks,  J.  Geophys.  Res.,  107,  B10,    2237,    doi:  323  

10.1029/2001JB0001580.  324  

Helmstetter,  A.,  D.  Sornette,  and  J.-­‐R.  Grasso  (2003),  Mainshocks  are  aftershocks  of  325  

conditional  foreshocks:  How  do  foreshock  statistical  properties  emerge  from  aftershock  326  

laws,  J.  Geophys.  Res.,  108(B1),  2046,  doi:10.1029/2002JB001991.  327  

Helmstetter,  A.,  Y.  Y.  Kagan,  and  D.  D.  Jackson  (2005),  Importance  of  small  earthquakes  for  328  

stress  transfers  and  earthquake  triggering,  J.  Geophys.  Res.,  110,  B05S08.  329  

Hill,  D.  P.,  et  al.  (1993),  Seismicity  remotely  triggered  by  the  magnitude  7.3  Landers,  330  

California,  earthquake,  Science,  260(5114),  1617-­‐1623.  331  

Page 24: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  24  

Hill,  D.  P.,  and  S.  G.  Prejean  (2007).  Dynamic  triggering,  in  Earthquake  Seismology,  Treatise  332  

on  Geophysics,  H.  Kanamori  (Editor),  Elsevier,  Amsterdam.  333  

Jones,  L.  M.,  and  P.  Molnar  (1979),  Some  characteristics  of  foreshocks  and  their  possible  334  

relationship  to  earthquake  prediction  and  premonitory  slip  on  faults,  J.  Geophys.  Res.,  335  

84(B7),  3596–3608.  336  

Lin,  G.,  P.  M.  Shearer,  and  E.  Hauksson  (2007),    Applying  a  three-­‐dimensional  velocity  337  

model,  waveform  cross  correlation,  and  cluster  analysis  to  locate  southern  California  338  

seismicity  from  1981  to  2005,  J.  Geophys.  Res.,  112,  B12309,  doi:10.1029/2007JB004986.  339  

McGuire,  J.J.,  M.S.  Boettcher  and  T.H.  Jordan  (2005),  Foreshock  sequences  and  short-­‐term  340  

earthquake  predictability  on  East  Pacific  Rise  transform  faults,  Nature  434,  457-­‐461.  341  

Ogata,  Y.  (1999),  Seismicity  Analysis  through  Point-­‐process  Modeling:  A  Review,  Pure  Appl.  342  

Geophys.  155,  471-­‐507.    343  

Peng,  Z.,  J.  E.  Vidale,  M.  Ishii,  and  A.  Helmstetter  (2007),  Seismicity  rate  immediately  before  344  

and  after  main  shock  rupture  from  high-­‐frequency  waveforms  in  Japan,  J.  Geophys.  Res.,  345  

112,  B03306,  doi:10.1029/2006JB004386.  346  

Richards-­‐Dinger,  K.B.,  R.  Stein,  and  S.  Toda,  Decay  of  aftershock  density  with  distance  does  347  

not  indicate  triggering  by  dynamic  stress.  Nature  .  467,  583-­‐586,  2010.  348  

Sornette,  D.  and  M.J.  Werner  (2005),  Apparent  clustering  and  apparent  background  349  

earthquakes  biased  by  undetected  seismicity,  J.  Geophys.  Res.  110,  B09303  350  

doi:10.1029/2005JB003621.    351  

Page 25: ETAS fore 3 3 · 2011. 2. 24. · ! 6! 102! The!ETAS!formulation!has!beenexplored!analytically!and!numerically!in!the!literature! 103! [Helmstetter!and!Sornette,!2002;!Felzer!et!al.,!2004;!McGuire!et

  25  

Van  Der  Elst,  N.J.  and  E.  E.Brodsky  (2010),  Connecting  near-­‐field  and  far-­‐field  earthquake  352  

triggering  to  dynamic  strain.  Journal  of  Geophysical  Research  115,  B07311,  2010.  353  

West,  M.,  J.  J.  Sanchez,  and  S.  McNutt  (2005),  Periodically  Triggered  Seismicity  at  Mount  354  

Wrangell,  Alaska,  After  the  Sumatra  Earthquake,  Science,  308,  1144-­‐1146.  355  

Zhuang,  J.,  A.  Christophersen,  M.  K.  Savage,  D.  Vere-­‐Jones,  Y.  Ogata,  and  D.  D.  Jackson  (2008),  356  

Differences  between  spontaneous  and  triggered  earthquakes:  Their  influences  on  357  

foreshock  probabilities,  J.  Geophys.  Res.,  113,  B11302,  doi:10.1029/2008JB005579.  358  

 359