1    

           

A  comparison  of  seismicity  rates  and  fluid  injection  operations  in  Oklahoma  and  California:  Implications  for  crustal  stresses  

   

Thomas  Goebel  California  Institute  of  Technology  

                                                   _______________  Seismological  Laboratory,  California  Institute  of  Technology  1200E.  California  Blvd.,  Pasadena  California  91125,  USA.  tgoebel@gps.caltech.edu      

2    

1.  Abstract  1    2  Fluid  injection  into  deep  wellbores  can  increase  pore  pressure,  reduce  effective  stress,  and  3  trigger  earthquakes.  The  extent  of   the  seismogenic  response   to   injection  provides   insight  4  into  how  close  faults  are  to  failure  within  the  injection-­‐affected  area.  This  study  examines  5  the  seismogenic  response  to   injection  operations   in  hydrocarbon  basins   in  California  and  6  Oklahoma.   I   test   whether   there   are   significant   spatial   and   temporal   seismicity   rate  7  variations  and  determine  the  corresponding  timing  and   location  based  on  nonparametric  8  modeling  of  background  seismicity  rates.  Oklahoma  experienced  a  recent  surge  of  seismic  9  events,  which  exceeded  the  95%  confidence  limit  of  Poissonian  background  rates  in  ~2010.  10  Annual  injection  volumes  in  OK  increased  systematically  between  1998  and  2013  and  have  11  been  connected  to  several  earthquake  sequences  [e.g.  Keranen,  et  al.  2014].  In  CA,  injection  12  volumes   increased   monotonically   between   2001   and   2009;   however,   the   seismogenic  13  response  was  very  limited,  and  devoid  of  large-­‐scale  background  rate  increase.  I  perform  a  14  detailed   comparison   of   injection   parameters   in   OK   and   CA,   focusing   on   well   density,  15  wellhead   pressures,   peak   and   cumulative   rates   as   well   as   injection   depths.   I   find   no  16  detectable  difference  that  could  readily  explain  the  observed  seismicity  rate  changes  in  OK  17  and   lack   thereof   in   CA.   A   strongly   different   seismogenic   response   to   similar   pressure  18  perturbations   indicates   that   the   injection   parameters   considered   are   only   of   secondary  19  influence   on   the   resulting   earthquake   activity.   I   suggest   that   the   primary   controls   on  20  injection-­‐induced  earthquakes  may  be  the  specific  geologic  setting  and  the  stress  state  on  21  nearby  faults.  22  

  2.  Introduction:  Injection  induced  earthquakes  in  California  and  the  central  23  and  eastern  U.S.  24    25  

The   injection   of   waste   fluids   into   deep   disposal   wells   and   its   environmental  26  consequences  are  a  growing  concern  in  the  central  and  eastern  U.S.  Such  injection  activities  27  can   increase   pore   pressures   and   poroelastic   stresses,  which  may   trigger   earthquakes   on  28  faults  close  to  failure  [e.g.  Kim,  2013;  Ellsworth,  2013].  Several  regions  within  the  central  29  and   eastern   U.S.   exhibited   a   pronounced   increase   in   seismic   activity   coincident   with  30  injection   operations   in   near-­‐by  wastewater   disposal  wells.   The   corresponding   seismicity  31  sequences  include,  the  2011  Mw4.7  Guy,  Arkansas  [Horton,  2012];  the  Mw3.9  Youngstown,  32  Ohio  [Kim,  2013];  and  the  Mw5.7  Prague,  Oklahoma  (OK)  earthquake  sequence  [Keranen  et  33  al.,  2013].  Many  of  these  sequences  were  associated  with  nearby  wastewater  injection  into  34  high-­‐permeability  aquifers  overlying  igneous  basement.  These  basement  layers,  which  are  35  hydraulically   connected   to   the   above   reservoirs,   host   the   majority   of   the   induced  36  earthquakes  including  the  largest  magnitude  events.    37    38  Previous  studies  examined  isolated  cases  of  likely  injection  induced  seismicity;  however,  a  39  synoptic  identification  of  induced  seismicity  and  its  underlying  causes  is  still  missing  in  OK  40  and  the  central  U.S.   In  central  CA,  a  comprehensive,  regional  study  revealed  that   induced  41  seismicity   is   rare   considering   the   extensive   injection   activity   that   has   occurs   in   close  42  proximity   to   active   faults   [Goebel,   et   al.   2015].   The   authors   identified   three   induced  43  

3    

seismicity   sequences   with   magnitudes   up   to   Mw4.7,   based   on   a   rigorous   statistical  44  assessment  of  correlations  between  injection  and  seismic  activity.  45    46  This   study   focuses  on  a   large-­‐scale   assessment  of  differences   in   injection  operations   and  47  possibly  induced  seismicity  in  CA  and  OK.  These  two  regions  exhibit  strong  differences  in  48  tectonic  deformation  and  seismic  activity.  While  OK  experienced  generally   low  seismicity  49  rates  until  2009  [Ellsworth,  2013],  seismicity  rates  in  CA  are  high,  as  a  result  of  pervasive  50  tectonic  deformation  along  many  active  faults.  To  isolate  the  possible  influence  of  injection  51  activity  from  tectonic  earthquakes  in  CA,  I  limit  the  analysis  to  seismicity  that  occurred  in  52  the  major  hydrocarbon  basins.  53    54  The   article   is   structured   as   follows.   First,   I   determine   the   cumulative   distribution   of  55  earthquake  rates  and  annual  injection  rates  in  CA  and  OK  between  1980  and  2014.  I  then  56  test   for   statistically   significant   increases   in   background   seismicity   rates,   and   determine  57  when  and  where  they  occur.  Finally,   I  examine  whether  differences   in  seismicity  rate  can  58  be  traced  back  to  differences  in  injection  parameters  between  CA  and  OK.    59  

  3.  Seismicity  and  injection  data  in  CA  and  OK  60    61  This   work   concentrates   on   the   most   widely   available   and   homogeneous   seismicity   and  62  injection   data-­‐sets,   including   earthquake   catalogs,   fluid-­‐injection   volumes,   well-­‐head  63  pressures  and   injection  depths.   Injection  data  have  been  archived  by   the  OK  Corporation  64  Commission   (OCC)   since  ~1975   and   the   CA  Department   of   Conservation,  Division   of  Oil,  65  Gas  and  Geothermal  Resources  (DOGGR)  since  ~1977.  The  seismicity  record  is  archived  by  66  the  USGS  Advanced  National  Seismic  System  in  CA  and  is  available  from  the  OK  Geological  67  Survey.  68    69  Much  of  the  seismicity  in  CA  is  tectonically  driven  and  localized  along  major  fault  traces.  I  70  exclude   earthquakes   along   major   faults   and   solely   select   seismicity   within   large  71  hydrocarbon  basins.   For   this   purpose,   I   compute   the   largest   convex  hull   of  well   location  72  vertices,  using  a  Delaunay  triangulation  algorithm.  The  subset  of  seismicity  within  the  hull  73  corresponds  to  12%  of   the  total  seismicity   in  CA  (Figure  1).  For  OK  essentially  the  entire  74  (99%)  earthquake  record  is  used.    75  

I   evaluate   changes   in   network   recording   quality   as   a   function   of   time   based   on  76  variations  in  the  magnitude  of  completeness  (Mc).  The  latter  is  computed  by  minimizing  the  77  misfit   between   the   observed   frequency-­‐magnitude-­‐distribution   and   the   modeled   fit  78  assuming  power  law  distributed  data  above  Mc  [Clauset,  et  al.  2009].  In  CA,  Mc  is  generally  79  close  to  2  after  1995  but  also  shows  short-­‐period  fluctuations,  e.g.,  connected  to  the  1994,  80  Mw6.7   Northridge   earthquake   (Figure   1b).   In   OK,   Mc   varies   between   1.5   and   2.5.  81  Consequently,  the  seismicity  record  in  both  regions  is  cut  below  Mc  =  2.5  to  ensure  catalog  82  completeness.  The  following  analyses  are  based  on  the  cut  catalogs  within  the  convex  hulls.  83    84  

[Figure  1  here]  85    86  

4    

In  addition  to  the  seismicity  records,  I  examine  large-­‐scale  injection  data  to  identify  87  possible  temporal  and  regional  variations.  The  overall  well  density  in  OK  is  ~0.05  km-­‐2  (i.e.  88  about  one  well  every  20  km2)  which  is  significantly  lower  than  in  CA  basins  with  a  density  89  of  0.7  km-­‐2  (i.e.  one  well  every  1.4  km2).  In  CA,  injection  data  is  recorded  monthly  for  each  90  well   and   categorized   according   to   the   type   of   injection   activity,   i.e.,  wastewater   disposal  91  (WD),  and  enhanced  oil   recovery  (EOR)  wells.  The   latter  are   further  classified   into  water  92  flooding  (WF),  steam  flooding,  cyclic  steam  injection,  and  pressure  maintenance.  Injection  93  into  WD  and  WF  wells  represent  the  largest  contribution  to  the  total  injected  volumes.  In  94  OK,   insufficient   information   about   WD   vs.   EOR   is   available   so   that   both   well   types   are  95  treated  jointly.  The  overall  contribution  of  WD  wells  to  annual  injection  volumes  is  ~50%  96  in  OK   [Murray,   2014],   and   only  ~20   to   30%   in   CA.   In   CA,   cumulative   injection   rates   are  97  published   for   each   year   by   the   DOGGR,   however,   with   a   significant   time-­‐lag,   so   that  98  presently  2009   is   the  most  recent  year  with  a  complete  record.   Injection  rates  peaked   in  99  the  mid  1980s,  and  in  2000  and  have  been  continuously  increasing  after  2001  (Figure  2a).  100  Moreover,   some   regions   in   CA   such   as   the  major   oilfields   in  Kern   county1   experienced   a  101  continuous   increase   in   injection   rates   over   even   longer   periods,   e.g.   since   ~1995,  102  highlighting  that  injection  rate  variations  are  not  homogeneous  in  space.  This  is  explored  in  103  more   detail   below   through   a   spatial   correlation   of   large-­‐volume   injection   wells   and  104  seismicity  rate  variations.    105  

In  OK,  the  cumulative  annual  injection  volumes  were  determined  by  summing  over  106  all  individual  well  records  from  the  OCC  database  and  comparing  them  to  published  values  107  between  2009  and  2013  [Murray,  2014].  Murray,  [2014]  performed  a  detailed  data  quality  108  assessment   and   removed   obvious   outliers.   Consequently,   the   annual   injection   rates   are  109  lower   compared   to   the   OCC   database,   especially   after   2010.   Nevertheless,   both   datasets  110  indicate   a   systematic   increase   in   annual   injection   rates   between   2008   and   2013   (Figure  111  2b).  The  OCC   injection  data  prior   to  2006   is   incomplete  but   suggest   a   largely  monotonic  112  increase  between  1998  and  2013,   assuming   that   relative   injection  variations  are   reliably  113  depicted   even   in   the   incomplete   dataset.   The   annual   injection   volumes   in   OK   are   only  114  comparable  to  CA  in  2006  with  2.0  Gbbl/yr  and  2.3  Gbbl/yr  respectively,  and  are  generally  115  lower  than  in  CA  during  all  other  periods.  Moreover,  the  injection  in  OK  occurs  over  a  wider  116  area  so  that  the  difference  in  injection  volume  per  area  is  even  higher  in  CA  compared  to  117  OK.  In  addition  to  differences  in  fluid  injection,  net-­‐fluid  production  rates  and  volumes  may  118  influence   the   seismogenic   response   in   a   region.   However,   the   assessment   of   net-­‐fluid  119  production   is   complicated   by   geologic   and   reservoir   complexity   as   well   as   fluid   density  120  variations  and  thus  not  considered  here.    121    122  

[Figure  2  here]  123    124  

 125    126  

                                                                                                                         1  The  data  for  Kern  county  oilfields  was  attained  from  the  DOGGR  online  data  base:  www.conservation.ca.gov/dog/  last  accessed:  12/2014  

5    

4.  Spatial  and  temporal  changes  in  background  seismicity  rates  127    128  Seismicity  rate  changes  are  commonly  dominated  by  aftershock  clustering  (Figure  2),  and  129  thus  provide  only   limited   insight   into  changes   in  external  driving  forces  such  as  pressure  130  increases.  Such  additional   forces  are  more  readily  assessed  using  the  rate  of   independent  131  mainshocks,  i.e.,  the  background  seismicity  rate  (λ0)  [Hainzl,  et  al.  2006].  To  determine  λ0,  I  132  employ  two  different  methods:    133  

1)  I  remove  all  events  within  a  specific  space/time  window  after  a  mainshock,  here  134  the   largest   magnitude   event   within   a   sequence.   This   method   takes   advantage   of   the  135  strongly  localized  occurrence  of  aftershocks  in  space  and  time.    The  size  of  this  window  is  a  136  function  of  mainshock  magnitude  [Gardner  &  Knophoff,  1974].  As  expected,  the  aftershock  137  removal  demotes  abrupt  seismicity  rate  changes  and  the  large  rate  increases,  for  example,  138  connected  to  the  Northridge  earthquake  sequence,  disappear.    139  

2)   To   avoid   inherent   biases   of   the   aftershock   window   selection,   I   also   use   a  140  nonparametric  method   to   determine  mainshock   fractions   and   rates   [Hainzl,   et   al.   2006].  141  The  method  uses  a  Gamma-­‐distribution   to   fit   the  observed   inter-­‐event   time  distributions  142  (ITDs),  i.e.  the  distribution  of  time  intervals  between  consecutive  earthquakes  [e.g.  Hainzl  143  et  al.  2006]:  144  

  𝑝 𝜏 = 𝐶 ∙ 𝜏!!!𝑒!!! ,  

(1)    

where  C  is  a  scaling  constant,  𝜏  is  the  inter-­‐event  time  normalized  by  earthquake  rate,  and  𝛾  145  and  𝛽   are  parameters  describing   the   shape   and   scale   of   the  underlying  distribution.  The  146  background  seismicity  rate  can  be  computed  from  the  parameter  𝛽  and  the  observed  rate,  147  𝜆,  of  the  clustered  catalog:  148     𝜆! =

1𝛽 ∙ 𝜆.  

(2)    

The   Gamma-­‐distribution   parameters   were   estimated   using   a   maximum   likelihood   fit   of  149  ITDs  within  sliding  time  windows  between  1980  and  2014.    150     Applying   the   non-­‐parametric   method   to   the   CA   seismicity   dataset,   I   find   that   λ0  151  decreased  systematically   from  12  to  6  events  per  year  between  1983  and  2011,  however  152  this  variation  falls  well  within  the  95%  confidence  interval  of  Poissonian  distributed  data  153  with  a  mean  of  9  events  per  year  (Figure  3a).  In  OK,  λ0  shows  little  variation  between  1983  154  and  2009,  but  increased  rapidly  thereafter  up  to  23  events  in  2011  and  58  events  in  2013.  155  These   values   exceed   the   95%   confidence   interval   of   Poissonian   distributed   data   with   a  156  mean  of  4  in  ~2010.  I  tested  the  sensitivity  of  these  results  to  the  selected  time  window  by  157  varying   its   length   from   4   to   15   years.   The   results   are   independent   of   particular   time  158  window  selections  such   that   the  systematic   increase   in  λ0   is  observed  consistently   for  all  159  time  windows  as  well  as  for  the  declustered  catalog  using  method  (1).  160    161  

 [Figure  3  here]  162    163  In   addition   to   changes   in   λ0   over   time,   I   examine   spatial   variations   in   numbers   of  164  mainshocks  and  correlate  them  with  locations  of  large-­‐volume  injection  wells  in  OK  and  CA.  165  Spatial  variations  are  computed  within  a  grid  of  0.1  degree  spacing  in  CA  and  0.4  spacing  in  166  OK.   The   rates   at   each   node   are   then   smoothed   using   an   isotropic,   Gaussian   kernel  167  

6    

preserving   the   initial   rates   and   normalized   by   the   corresponding   time   interval   and   grid  168  spacing.  I  compare  the  smoothed  rates  in  CA  before  and  after  the  most  recent,  pronounced  169  increase   in   fluid   injection   rates   in   2001   (see   Figure   2a),   and   highlight   regions   that  170  experienced   an   increase   in   λ0   above   the   95%   confidence   limit   of   a   Poissonian   process  171  (Figure  4).  CA  exhibits  only  localized  areas  with  significant  rate  increase,  connected  to  the  172  2013,   Mw4.8   Isla   Vista   event   [-­‐119.9°N,   34.4°W]   and   the   2008,   Mw5.4   Chino   Hills  173  earthquakes  [-­‐117.8°N,  34.0°W].  174    175  

[Figure  4  here]  176    177  In  OK,  on  the  other  hand,  the  number  of  mainshocks  increased  significantly  after  2009.  This  178  increase   was   concentrated   in   central   and   northern   OK   (Figure5).   Moreover,   the   newly  179  emerging  mainshock   activity   exceeds   the  95%  confidence   limit   of   Poissonian  distributed  180  data  in  a  large  region  (red  contours  in  Figure  5b),  which  is  in  agreement  with  Llenos  et  al.,  181  [2014].  The  significant   increase   in  λ0  occurred  in  the  neighborhood  of  some  large-­‐volume  182  injectors  with  maximum  monthly   rates   of  more   than   1  Mbbl/mo   (i.e.   million   barrel   per  183  month).  The  median  distance  between  individual  earthquakes  and  the  closest  high-­‐volume  184  injection  wells   decreased   significantly   from   39   to   20   kilometers   after   2009   (Figure   5b).  185  This   decrease   in   distance   is   observed   for   wells   with   both   intermediate   and   high   peak  186  injection  rates  and  primarily  a  result  of  a  northward  seismicity  migration.  The  extensive,  187  newly  appearing  mainshock  activity   in  OK  after  2009,  represents  a  strong  contrast  to  the  188  lack  of   observable   changes   in   seismic   activity   in  CA.   If  much  of   this   activity   is   induced   it  189  may  also   indicate  a  substantial  difference   in   injection  parameters  between  OK  and  CA.   In  190  the  following,  I  perform  a  detailed  comparison  of  injection  operations  between  CA  and  OK  191  to  identify  possible  systematic  differences.  192    193  

[Figure  5  here]  194  

  5.  Comparison  between  injection  operations  and  focal  depths  in  CA  and  OK  195    196  Injection  rates  and  pressures  may  significantly  influence  the  seismogenic  potential  of  fluid  197  injection.     I   determine   the   maximum   wellhead   pressures   and   injection   rates   from   the  198  DOGGR  and  OCC  databases   for   individual,   active  wells   to  determine  whether   statistically  199  significant  differences  in  injection  parameters  can  be  identified.  In  CA,  I  examine  both  WD  200  and  WF  wells,  the  latter  as  a  proxy  for  typical  EOR  wells.  In  OK,  both  well  types  are  treated  201  jointly.   As   expected,   much   of   the   injection   activity   is   conducted   at   comparably   low  202  pressures   and   rates.     The   two   study   regions   show   strong   similarities   in   median   peak-­‐203  pressures,   i.e.   4.3±0.2   MPa   in   OK   and   5.5±0.2   MPa   in   CA,   and   peak   injection   rates,   i.e.,  204  2900±200   m3/mo   in   OK   and   4700±200   m3/mo   in   CA   (Figure   6).   Both   peak   rates   and  205  pressures   are  higher   for  WD  compared   to  WF  wells   in  CA,   but  no   systematic  differences  206  between   CA   and   OK   could   be   identified   that   could   be   responsible   for   the   observed  207  difference  in  seismic  activity.  208    209  

[Figure  6  here]  210    211  

7    

Besides  rates  and  pressures,   injection  depth  may  control  induced  seismicity  potentials.  In  212  OK,   the  packer  depth  of   injection  wells,  which   is   a  minimum  value   for   injection  depth,   is  213  readily   available   from   the   OCC   website.   In   CA,   on   the   other   hand,   detailed   depth  214  information   is  only  available   in   the   form  of  non-­‐searchable  pdf   files.   I  used   these   files   to  215  extract   the   effective   depth,   which   provides   a   maximum   value   for   injection   depth,   using  216  open   source   optical   character   recognition   software.   The   quality   of   the   pdf   files   requires  217  visible  inspection  of  each  value,  so  that  I  limited  the  analysis  to  300  randomly  sampled  WD  218  and  300  randomly  sampled  WF  wells.  In  agreement  with  Jordan  &  Gillespie,  [2014],  I  find  219  that  WD   injection  occurs  on   average   at   shallower  depth   than  WF   injection   in  CA   (Figure  220  7a).   The  median   depth   of   the   former   is   ~0.9   km  which   is   comparable   with   the  median  221  depth  of  1  km  of  injection  activity  in  OK  (Figure  7b).  The  median  depth  of  joint  WD  and  WF  222  depths  in  CA  increases  to  ~1.5  km,  in  agreement  with  values  reported  by  the  DOGGR.  This  223  depth   is   substantially   deeper   than   the   value   reported   for   OK.   In   summary,   I   find   no  224  evidence  that  injection  activity  is  deeper  in  OK,  on  the  contrary  fluid  injection  in  CA  occurs  225  on  average  ~0.5  km  deeper  than  in  OK.  226    227  

 [Figure  7  here]  228    229  While   injection  depths   in  CA  and  OK  were  generally   limited   to   the  upper  4  km,   the   focal  230  depths  of  seismic  events  in  both  regions  are  significantly  deeper.    To  compare  focal  depths  231  in  CA   and  OK,   I   use   a  waveform-­‐relocated   catalog   [Hauksson,   et   al.   2012]   and   the   single  232  event  depth  within  the  OGS  catalog.  In  CA,  much  of  the  seismic  activity  occurs  distributed  233  within  the  upper  15  km  with  a  median  of  10  km  (Figure  8).  Events  between  M4  and  M5  and  234  above   M5   show   a   wide   depth-­‐range,   from   ~1-­‐27   km   between   1980-­‐2001.   Much   of   the  235  shallow   earthquakes   are   aftershocks   of   the   1994   Northridge   earthquake,   whereas   deep  236  events   occur   close   to   Ventura   basin.   The   focal   depths   in  OK   did   not   change   significantly  237  between  1980-­‐2001  and  2002-­‐2014,  exhibiting   the   same  mean  depth  of  5  km.  The  more  238  abundant  earthquakes  above  M4  in  OK  between  2002  and  2014  (Figure  8),  occurred  within  239  a   relatively   localized   depth   layer   extending   from   ~2   to   7   km.   While   substantial  240  uncertainties  in  focal  depth  are  expected  especially  is  regions  with  limited  station  coverage  241  and   poorly   constrained   velocity  models   such   as  OK,   the   observed   systematic   differences  242  between   CA   and   OK   seem   to   be   a   robust   feature   within   the   data.   Focal   depths   are  243  significantly  deeper  in  CA  than  in  OK.    244    245  

[Figure  8  here]  246    247  

6.  Discussion    248    249  

In   this   study,   I   compared   fluid   injection   operations   and   examined   temporal   and  250  spatial  seismicity  variations  in  CA  and  OK.  CA  showed  no  large-­‐scale  correlation  between  a  251  recent  increase  in  injection  volumes  and  seismic  activity.  Seismicity  is  generally  deeper  in  252  CA   and   surficial   injection   operations   likely   have   only   limited   influence   on   earthquake  253  activity.   This   is   expected   considering   the   low   upper   crustal   stresses   and   frictional  254  properties   of   shallow   basin   faults,   which   inhibit   seismic   activity   and   large   earthquake  255  

8    

ruptures  [e.g.  Goebel  et  al.,  2015].  OK,  on  the  other  hand,  showed  a  significant  increase  in  256  background   seismicity   rates   starting   in   2009.   This   newly   appearing   seismic   activity  257  encompasses  a  region  that  also  hosts  large-­‐volume  injection  wells  in  central  and  northern  258  OK.  The  spatial-­‐temporal  correlation  between   injection  and  seismicity   indicate   that   fluid-­‐259  injection   may   contribute   to   seismic   activity   at   a   large-­‐scale,   as   pointed   out   by   several  260  previous  studies  [Keranen,  et  al.  2013,  Ellsworth,  2013,  Llenos  &  Michael,  2013,  Keranen,  et  261  al.  2014].    262  

The   injection   operations   in  OK   and  CA   show  many   remarkable   similarities.   These  263  similarities  include  the  overall  well  count,   injection  depths,  well  head  pressures  and  peak  264  injection   rates.   Some   of   the   differences   in   injection   operations   include:   1)   The   overall  265  density   of   injection   wells;   and   2)   the   generally   larger   annual   injection   volumes   in   CA;  266  however,  none  of  these  differences  is  likely  to  cause  the  significantly  stronger  seismogenic  267  response  to  fluid  injection  in  OK.  268  

Several  mechanisms  may  explain  the  strong  seismogenic  response  in  OK.  First,  the  269  observed  seismicity   rate   changes   in  OK  may  be  part  of  natural   rate  variations   that  occur  270  over   long   time-­‐scales.   The   large   recurrence   times   of   intra-­‐plate   earthquakes   and   the  271  limited   temporal   extent   of   the   corresponding   seismicity   record   support   this   possibility.  272  However,  the  strong  correlation  between  injection  and  seismic  activity  in  several  areas  in  273  OK  make   this   scenario   less   likely.   Second,   large-­‐scale,   average   injection   parameters  may  274  not  be  representative  of  the  seismogenic  potential  of  fluid-­‐injection  in  an  entire  region.  The  275  seismogenic  potential  may  be  controlled  by  individual  wells  that  strongly  deviate  from  the  276  average   operational   parameters.   Nonetheless,   the   reported   cases   of   likely   induced  277  seismicity  in  OK  and  CA  were  associated  with  wells  of  average  injection  activity  and  do  not  278  seem  to  support  this  hypothesis  [Keranen,  et  al.  2013,  Goebel,  et  al.  2015].  Third,  perhaps  279  the  considered  set  of  operational  parameters  is  not  complete  and  additional  factors  such  as  280  net  production  rate,   reservoir  pressure,  or   the  rate  of  change  of   injection  rate  may  affect  281  poro-­‐elastic  stresses  resulting  in  seismic  activity  within  a  region.    282  

Finally,  the  difference  in  seismogenic  response  to  fluid  injection  in  CA  and  OK  may  283  be   driven   by   an   overall   difference   in   geologic   setting   and   crustal   stresses   between   plate  284  boundary   and   intraplate   regions.   The   existence   of   higher   crustal   stresses   at   shallower  285  depth   in   OK   is   supported   by   shallower   focal   depths,   which   are   consistent   over   time.  286  Moreover,   several   studies   suggest   that   the   upper   crust   in   OK   is   close   to   failure.   For  287  instance,   the  passage  of  seismic  waves  from  large  teleseismic  earthquakes  and  connected  288  small  dynamic  strains  are  observed  to  result  in  an  uncharacteristically  strong  seismogenic  289  response  in  the  form  of  triggered  earthquakes  [VanDerElst,  et  al.  2013].  Furthermore,  some  290  areas   in   Oklahoma   may   produce   induced   earthquake   sequences   as   a   result   of   pressure  291  perturbations  as  small  as  0.07  MPa  [Keranen,  et  al.  2014].    292  

Besides   the  difference   in  crustal  stresses,   the   larger  scale  geologic  homogeneity   in  293  OK   likely   results   in   extensive   hydraulic   connectivity   of   the   upper   crust.   Keranen   et   al.,  294  [2014]  suggested  that  this  hydraulic  connectivity  may  result  in  lateral  diffusion  of  pressure  295  perturbations   up   to   20-­‐35   km.   In   CA,   the   upper   crust   is   subject   to   constant   tectonic  296  deformation,   so   that   crustal   heterogeneity   and   especially   mature   fault   zones   may   limit  297  lateral  migration  of  pressure  perturbations.  298  

In   conclusion,   my   results   suggest   that   operational   parameters   of   surficial   fluid  299  injection   are   likely   of   secondary   importance   for   the   resulting   seismogenic   response.   The  300  primary   controls   on   injection-­‐induced   seismicity   are   the   specific   geologic   setting,   e.g.  301  

9    

hydraulic   connectivity,   and   the   stress   state   on   near-­‐by   faults.   The   view   that   injection  302  induced   earthquakes   have   successfully   been   avoided   in   CA   in   the   past   because   of   less  303  invasive   injection  operations   is   likely  erroneous.  The  scarcity  of   induced  seismicity   in  CA  304  may   simply   be   an   expression   of   lower   stresses   at   injection  depth   and   lack   of   large-­‐scale  305  hydraulic   connectivity   within   hydrocarbon   basins.   Although   less   probable,   earthquakes  306  may  be   induced   in   CA   through   injection   in   areas   of   active   faulting   as   shown  by   a   recent  307  study   [Goebel   et   al,   2015].   The   largely   similar   injection   operations   in   CA   and   OK   and  308  absence  of  noticeable  seismogenic  response  in  CA  indicate  a  fundamental  difference  in  the  309  state  of   stress  between   the   two  study  areas.  The   specific   geologic   conditions   responsible  310  for  individual,  induced  earthquake  sequences  remain  to  be  understood.    311  

10    

  7.  Acknowledgements  312    313  I  would  like  to  thank  Andrea  Llenos,  Sebastian  Hainzl  and  the  editors,  Robert  Habiger  and  314  Gregory   Beroza   for   their   detailed   reviews   of   the   initial   manuscript.   The   present   work  315  benefitted   from  discussions  with  Preston   Jordan  and  Kyle  Murray.   I   thank   the  Oklahoma  316  Corporation  Commission  and  the  California  Division  of  Oil,  Gas  and  Geothermal  Resources  317  for   making   the   injection   databases   publically   available.   The   earthquake   data   sets   were  318  provided   through   the   USGS   Advanced   National   Seismic   System   and   by   the   Oklahoma  319  Geological  Survey.   I  am  grateful   to  Whenzeng  Yang,  Egill  Hauksson  and  Peter  Shearer   for  320  creating   high-­‐quality   earthquake   catalogs,   and   for   Jeremy   Zechar   who   made   his  321  declustering   algorithm   available   online.   Finally,   I   would   like   to   thank   the   Statistical  322  Seismology  Community  for  the  many  useful  online  resources  on  corssa.org.    323  

   324  

11    

8.  Figures  325    326  

   327  

12    

 328  Figure  1:  Seismicity  (red  dots,  squares,  star;  see  legend  for  magnitudes)  and  injection  well  329  locations  (blue  triangles)  in  CA  (a)  and  OK  (d).  Gray  dots  are  earthquakes  that  were  not  330  used  within  the  following  analysis.  b)  Mc  variations  for  moving  windows  of  200  events  (thin  331  lines)  and  after  applying  a  25-­‐point  median  filter  to  remove  high-­‐frequency  variations  332  (thick  lines).  Major  active  faults  in  CA  are  highlighted  by  black  lines  in  a).    333  

13    

 334  Figure  2:  Annual  fluid  injection  (vertical  bars)  and  number  of  earthquakes  (thick  curves)  in  335  CA  (a)  and  OK  (b)  between  1980  and  2014.    Earthquakes  above  M4  are  shown  by  red  336  squares.  Actual  and  inflation-­‐adjusted  oil  prices  are  highlighted  by  dashed  and  solid  green  337  lines  in  a).  The  light  blue  bars  in  a)  are  annual  injection  rates  for  the  largest  oil  fields  in  338  Kern  County,  CA.  The  dark  and  light  blue  bars  in  b)  are  injection  volumes  from  Murray,  339  [2014]  and  the  OCC  database.    Mbbl/yr  refers  to  million  barrels  per  year.  340  

14    

 341  Figure  3:  Background  seismicity  rates  (large  dots)  and  fraction  of  mainshocks  (squares)  for  342  CA  (a)  and  OK  (b).  Median  values  and  95%  confidence  interval  assuming  a  stationary  343  Poisson  process  are  shown  by  solid  and  dashed  lines.  I  explored  the  influence  of  time  bins  344  by  varying  the  binsize  between  4  to  15  years  (small  dots  and  gray  error)  and  averaged  over  345  these  different  intervals  using  a  2-­‐year  time  step  (large  dots).  Vertical  error  bars  show  the  346  corresponding  standard  deviations  from  bootstrap  resampling.    347  

15    

 348  Figure  4:  Smoothed,  spatial  variation  in  background  seismicity  rates  in  CA  basins  between  349  1980-­‐2001  (a)  and  2002-­‐2014  (b).  Earthquake  locations  are  shown  by  black  dots,  location  350  of  high-­‐volume  injection  wells  (>40,000  m3/mo)  by  blue  triangles  and  areas  of  significant  351  rate  increase  by  red  contours.  The  insets  show  the  distance  between  earthquakes  and  352  closest  high-­‐volume  injection  well.  353  

16    

 354  Figure  5:  Smoothed,  spatial  variation  in  background  seismicity  rates  in  OK  between  1980-­‐355  2008  (a)  and  2009-­‐2014  (b).  Earthquake  locations  are  shown  by  black  dots,  location  of  356  high-­‐volume  (>50,000  m3/mo)  injection  wells  by  blue  triangles  and  areas  of  significant  rate  357  increase  by  red  contours.  Histograms  show  changes  in  distance  between  earthquakes  and  358  closest  high-­‐volume  injection  well  between  the  two  time  periods.  359    360      361  

17    

 362  Figure  6:  Maximum  injection  pressure  (a)  and  injection  rate  (b)  for  all  wells  in  OK  (green)  363  and  for  WD  (orange)  and  WD+WF  wells  (red)  in  CA.  Median  values  and  95th  percentiles  are  364  shown  by  solid  and  dashed  lines.  365      366  

18    

 367    368  Figure  7:  Injection  depth  in  OK  (green)  and  CA  (orange  for  WD,  red  for  WD+WF  wells).  369  Median  depths  and  95%  confidence  intervals  are  shown  by  solid  and  dashed  lines.  Note  370  that  water  flooding  occurs  on  average  deeper  than  water  disposal  in  CA.  Depth  in  OK  is  371  reported  as  packer  depth  whereeas  depth  in  CA  is  reported  as  effective  surface  depth.  372      373  

19    

 374  Figure  8:  Comparison  of  focal  depth  between  1980-­‐2001  (a)  and  2002-­‐2014  (b)  in  CA  (red)  375  and  OK  (green).  Mean  values  and  95%  confidence  intervals  are  highlighted  by  solid  and  376  dashed  lines.  The  depth  of  M4-­‐5  events  is  shown  by  colored  squares  and  for  M>5  events  by  377  colored  stars.  Seismicity  rates  in  OK  did  not  increase  significantly  until  ~2009,  however,  378  the  focal  depth  in  b)  is  still  dominated  by  the  more  abundant  events  after  2009.  379    380      381  

20    

  9.  References  382    383  Clauset,  A.,  Shalizi,  C.R.  &  Newmann,  M.E.J.,  (2009).  Power-­‐law  distributions  in  empirical  384  data,  SIAM  Rev.,  51(4),  661–703.    385  

Ellsworth,  W.  L.  (2013).  Injection-­‐induced  earthquakes.  Science,  341(6142).  386    387  Gardner,  J.  K.,  &  Knopoff,  L.  (1974).  Is  the  sequence  of  earthquakes  in  southern  California,  388  with  aftershocks  removed,  Poissonian?  Bull.  Seismol.  Soc.  Am.,  64(5),  1363-­‐1367.  389    390  Goebel,  T.H.W.,  E.  Hauksson,  F.  Aminzadeh,  &  J.-­‐P.  Ampuero  (2015).  An  objective  method  391  for  the  assessment  of  possibly  fluid-­‐injection  induced  seismicity  in  tectonically  active  392  region  in  central  California,  J.  Geophys.  Res.,  doi:10.1029/2015,  (in  revision).  393    394  Hainzl,  S.,  Scherbaum,  F.,  &  Beauval,  C.  (2006).  Estimating  background  activity  based  on  395  interevent-­‐time  distribution.  Bull.  Seismol.  Soc.  Am.,  96(1),  313-­‐320.  396    397  Hauksson,  E.,  Yang,  W.,  &  Shearer,  P.  M.  (2012).  Waveform  relocated  earthquake  catalog  for  398  southern  California  (1981  to  June  2011).  Bulletin  of  the  Seismological  Society  of  America,  399  102(5),  2239-­‐2244.  400    401  Horton,  S.  (2012).  Disposal  of  hydrofracking  waste  fluid  by  injection  into  subsurface  402  aquifers  triggers  earthquake  swarm  in  central  Arkansas  with  potential  for  damaging  403  earthquake.  Seismol.  Res.  Lett.,  83(2),  250-­‐260.  404    405  Keranen,  K.  M.,  Savage,  H.  M.,  Abers,  G.  A.,  &  Cochran,  E.  S.  (2013).  Potentially  induced  406  earthquakes  in  Oklahoma,  USA:  Links  between  wastewater  injection  and  the  2011  Mw  5.7  407  earthquake  sequence.  Geology,  41(6),  699-­‐702.  408    409  Keranen,  K.  M.,  Weingarten,  M.,  Abers,  G.  A.,  Bekins,  B.  A.,  &  Ge,  S.  (2014).  Sharp  increase  in  410  central  Oklahoma  seismicity  since  2008  induced  by  massive  wastewater  injection.  Science,  411  345(6195),  448-­‐451.  412    413  Kim,  W.  Y.  (2013).  Induced  seismicity  associated  with  fluid  injection  into  a  deep  well  in  414  Youngstown,  Ohio.  J.  Geophys.  Res.,  118(7),  3506-­‐3518.  415    416  Jordan,  P.,  &  Gillespie,  J.  (2013),  Potential  Impacts  of  Future  Geological  Storage  of  CO2  on  417  the  Groundwater  Resources  in  California’s  Central  Valley:  Southern  San  Joaquin  Basin  Oil  418  and  Gas  Production  Analog  for  Geologic  Carbon  Storage.  California  Energy  Commission.  419  Publication  number:  CEC-­‐5002014-­‐029  420    421  Llenos,  A.  L.,  &  Michael,  A.  J.  (2013).  Modeling  earthquake  rate  changes  in  Oklahoma  and  422  Arkansas:  possible  signatures  of  induced  seismicity.  Bull.  Seismol.  Soc.  Am.,  103(5),  2850-­‐423  2861.  424    425    426  

21    

Llenos,  A.  L.,  J.  L.  Rubinstein,  W.  L.  Ellsworth,  C.  S.  Mueller,  A.  J.  Michael,  A.  McGarr,  M.  D.  427  Petersen,  M.  Weingarten,  and  A.  A.  Holland,  (2014),  Increased  earthquake  rates  in  the  428  central  and  eastern  US  portend  higher  earthquake  hazards,  AGU  Fall  Meeting,  Abstract  429  U34A-­‐02.  430    431  Murray  K.E.,  (2014),  Class  II  Underground  Injection  Control  Well  Data  for  2010–2013  by  432  Geologic  Zones  of  Completion,  Oklahoma.  Oklahoma  Geological  Survey  Open-­‐File  Report  433  (OF1-­‐2014).  Norman,  OK.  pp.  32.  434    435  Van  der  Elst,  N.  J.,  Savage,  H.  M.,  Keranen,  K.  M.,  &  Abers,  G.  A.  (2013).  Enhanced  remote  436  earthquake  triggering  at  fluid-­‐injection  sites  in  the  Midwestern  United  States.  Science,  437  341(6142),  164-­‐167.  438    439    440  

 441  

 442