Streamflow 
  Projections 
  for 
  the 
  upper 
  Gila 
  River 
  

David 
  S. 
  Gutzler 
  
Department 
  of 
  Earth 
  & 
  Planetary 
  Sciences 
  
University 
  of 
  New 
  Mexico 
  
gutzler@unm.edu 
  
 
  
Submitted 
  to 
  the 
  New 
  Mexico 
  Interstate 
  Stream 
  Commission 
  
for 
  Deliverables 
  2 
  and 
  3 
  
 
  
UNM 
  Contract 
  No. 
  37675 
  
 
  
 
  
final 
  version 
  
December 
  10, 
  2013 
  
 
  

 Contents 
  
 
  
1. 
   
  Introduction 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

3 
  

2. 
   
  Historical 
  Streamflow 
  at 
  the 
  Gila 
  River 
  Gila 
  gage 
   
   
   
   
   
  

. 
  

. 
  

. 
  

. 
  

. 
  

5 
  

. 
   
   
  

. 
  

. 
  

10 
  

. 
  

. 
  

. 
  

17 
  

 
  

 
  
3. 
   
  Dynamical 
  Projection 
  of 
  Streamflow 
  in 
  the 
  upper 
  Gila 
  River 
   
   
  . 
   
   
  
 
  
4. 
   
  Statistical 
  Projection 
  of 
  Streamflow 
  in 
  the 
  upper 
  Gila 
  River 
   
   
   
  . 
  
 
  
5. 
   
  Conclusions 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

22 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

. 
  

25 
  

 
  
6. 
   
  References 
   
  

 
  

2 
  

 1. 
   
  Introduction 
  
This 
  paper 
  discusses 
  two 
  approaches 
  to 
  estimating 
  projected 
  streamflows 
  in 
  the 
  
upper 
  Gila 
  River 
  over 
  the 
  next 
  few 
  decades. 
  One 
  approach 
  employs 
  dynamical 
  models 
  of 
  
the 
  climate 
  system 
  and 
  surface 
  hydrologic 
  system 
  to 
  calculate 
  projected 
  changes 
  in 
  
streamflows. 
  The 
  second 
  approach 
  uses 
  statistics 
  of 
  current 
  observed 
  climate 
  variability 
  to 
  
develop 
  an 
  empirical 
  model 
  for 
  use 
  in 
  projecting 
  future 
  flows. 
  This 
  paper 
  follows 
  an 
  earlier 
  
document, 
  entitled 
  "Observed 
  and 
  Projected 
  Climate 
  Changes 
  and 
  Their 
  Effects 
  on 
  Snow-­‐
fed 
  Rivers 
  in 
  Southwestern 
  North 
  America" 
  (hereafter 
  referred 
  to 
  as 
  the 
  Review 
  Paper), 
  
that 
  was 
  submitted 
  to 
  the 
  New 
  Mexico 
  Interstate 
  Stream 
  Commission 
  (ISC) 
  in 
  March 
  2013. 
   
  
Taken 
  together 
  these 
  papers 
  have 
  been 
  drafted 
  for 
  the 
  purpose 
  of 
  providing 
  guidance 
  to 
  
the 
  ISC 
  on 
  projected 
  future 
  flows 
  in 
  the 
  Gila 
  River, 
  for 
  use 
  regarding 
  decisions 
  on 
  water 
  
allocations 
  within 
  New 
  Mexico 
  related 
  to 
  the 
  2004 
  Arizona 
  Water 
  Settlements 
  Act. 
   
   
  
The 
  Gila 
  River 
  has 
  its 
  headwaters 
  in 
  the 
  Gila 
  Wilderness 
  of 
  southwestern 
  New 
  Mexico 
  
(Figure 
  1) 
  and 
  flows 
  in 
  a 
  general 
  westward 
  direction 
  into 
  southern 
  Arizona 
  as 
  a 
  tributary 
  of 
  
the 
  Lower 
  Colorado 
  River 
  (Thomson 
  2012). 
   
  The 
  Gila 
  River 
  is 
  fed 
  by 
  winter 
  snowpack 
  at 
  
high 
  elevations 
  near 
  its 
  headwaters. 
  Summer 
  flows 
  are 
  augmented 
  by 
  warm 
  season 
  season 
  
rainfall 
  associated 
  with 
  the 
  North 
  American 
  Monsoon 
  circulation. 
   
  The 
  Gila 
  River 
  is 
  
arguably 
  the 
  southernmost 
  snow-­‐fed 
  river 
  in 
  North 
  America. 
  Future 
  flows 
  on 
  the 
  upper 
  
Gila 
  River 
  will 
  be 
  controlled 
  by 
  climatic 
  variability 
  and 
  change 
  in 
  the 
  surface 
  water 
  budget 
  
with 
  regard 
  to 
  precipitation 
  (P) 
  in 
  winter 
  and 
  summer, 
  and 
  by 
  temperature-­‐driven 
  
variability 
  and 
  change 
  in 
  water 
  losses 
  due 
  to 
  evapotranspiration 
  off 
  the 
  surface 
  (E). 
  On 
  a 
  
large 
  scale, 
  climate 
  models 
  project 
  that 
  the 
  surface 
  water 
  budget 
  P-­‐E 
  across 
  the 
  Southwest 
  
will 
  tend 
  toward 
  drier 
  conditions 
  (a 
  negative 
  change 
  in 
  P-­‐E) 
  during 
  the 
  21st 
  Century 
  
(Seager 
  et 
  al. 
  2008), 
  leading 
  to 
  projections 
  of 
  diminished 
  future 
  streamflows 
  in 
  
southwestern 
  rivers 
  as 
  described 
  in 
  the 
  Review 
  Paper. 
   
  
The 
  Review 
  Paper 
  summarized 
  recent 
  research 
  on 
  future 
  flows 
  in 
  the 
  Colorado 
  and 
  
Rio 
  Grande 
  basins. 
   
  Several 
  studies 
  have 
  indicated 
  that 
  annual 
  flows 
  in 
  major 
  southwestern 
  
snow-­‐fed 
  rivers 
  could 
  decline 
  by 
  approximately 
  20% 
  by 
  the 
  end 
  of 
  the 
  21st 
  Century. 
  
Substantial 
  quantitative 
  uncertainties 
  are 
  inherent 
  in 
  this 
  general 
  projection, 
  although 
  
there 
  is 
  a 
  very 
  strong 
  consensus 
  that 
  flows 
  are 
  expected 
  to 
  decrease 
  as 
  the 
  result 
  of 
  the 
  
significantly 
  warmer 
  temperatures 
  projected 
  to 
  occur 
  in 
  the 
  middle 
  of 
  the 
  North 
  American 
  

 
  

3 
  

 continent 
  this 
  century. 
   
  The 
  projected 
  temperature 
  increase 
  is 
  a 
  very 
  robust 
  projection 
  that 
  
is 
  generated 
  by 
  all 
  modern 
  climate 
  models 
  in 
  response 
  to 
  much-­‐increased 
  concentrations 
  
of 
  greenhouse 
  gases 
  that 
  are 
  currently 
  observed 
  and 
  will 
  inevitably 
  continue 
  to 
  increase 
  for 
  
the 
  foreseeable 
  future 
  (IPCC 
  2007; 
  Karl 
  et 
  al. 
  2009). 
  Precipitation 
  projections 
  across 
  the 
  
Southwest 
  are 
  considerably 
  less 
  robust 
  (Gutzler 
  and 
  Robbins 
  2011; 
  Cook 
  and 
  Seager 
  2013), 
  
but 
  the 
  possibility 
  of 
  diminished 
  precipitation 
  plays 
  a 
  secondary 
  role 
  to 
  the 
  effect 
  of 
  
increasing 
  temperature 
  on 
  projections 
  of 
  diminished 
  future 
  streamflow 
  (Hurd 
  and 
  Coonrod 
  
2008, 
  2012). 
   
  Additional 
  uncertainty 
  due 
  to 
  the 
  interannual 
  and 
  decadal 
  variability 
  of 
  
precipitation, 
  which 
  will 
  be 
  emphasize 
  in 
  this 
  report, 
  has 
  received 
  very 
  little 
  attention 
  to 
  
date 
  in 
  studies 
  of 
  century-­‐scale 
  climate 
  change. 
   
  
Section 
  2 
  of 
  this 
  paper 
  reviews 
  historical 
  flows 
  on 
  the 
  upper 
  Gila 
  River, 
  including 
  
means, 
  variability 
  and 
  trends 
  in 
  the 
  data 
  record. 
  We 
  have 
  chosen 
  to 
  focus 
  on 
  the 
  flow 
  
record 
  at 
  the 
  Gila 
  gage 
  (USGS 
  gage 
  number 
  09430500). 
  We 
  also 
  review 
  historical 
  climatic 
  
variability 
  of 
  temperature 
  and 
  precipitation 
  in 
  southwestern 
  New 
  Mexico. 
   
  
Downscaled 
  climate 
  change 
  projections 
  generated 
  by 
  dynamical 
  models 
  have 
  recently 
  
been 
  applied 
  to 
  the 
  watershed 
  of 
  the 
  upper 
  Gila 
  River. 
  Section 
  3 
  assesses 
  a 
  new 
  set 
  of 
  such 
  
streamflow 
  projections 
  generated 
  by 
  the 
  Bureau 
  of 
  Reclamation 
  as 
  part 
  of 
  its 
  West 
  Wide 
  
Water 
  Assessment 
  projection 
  (Reclamation 
  2011a, 
  2011b). 
  We 
  analyze 
  runoff 
  at 
  the 
  
location 
  of 
  the 
  Gila 
  gage 
  yielded 
  by 
  these 
  projections, 
  and 
  assess 
  the 
  quality 
  and 
  possible 
  
limitations 
  of 
  these 
  dynamical 
  projections. 
  We 
  consider 
  the 
  Reclamation 
  projections 
  to 
  
represent 
  the 
  current 
  state 
  of 
  the 
  art 
  in 
  dynamical 
  streamflow 
  projections 
  of 
  western 
  
rivers. 
   
  
Section 
  4 
  then 
  develops 
  an 
  empirical 
  projection 
  of 
  flow 
  at 
  the 
  Gila 
  gage. 
  Average 
  
temperature 
  in 
  the 
  upper 
  Gila 
  basin 
  is 
  projected 
  to 
  increase 
  significantly 
  in 
  coming 
  
decades, 
  and 
  may 
  depart 
  altogether 
  from 
  the 
  historical 
  envelope 
  of 
  variability 
  in 
  the 
  
instrumental 
  record 
  by 
  mid-­‐century. 
  The 
  magnitude 
  of 
  streamflow 
  change 
  driven 
  largely 
  
by 
  this 
  projected 
  temperature 
  change 
  can 
  be 
  estimated 
  by 
  statistical 
  analysis 
  of 
  the 
  effects 
  
of 
  temperature 
  changes 
  on 
  streamflow 
  in 
  the 
  current 
  climate, 
  taken 
  together 
  with 
  
projections 
  of 
  precipitation 
  variability 
  and 
  change. 
   
  

 
  

4 
  

 The 
  results 
  in 
  Sections 
  3 
  and 
  4 
  represent 
  distinct 
  and 
  complementary 
  ways 
  to 
  
estimate 
  projected 
  streamflow 
  change 
  in 
  the 
  upper 
  Gila 
  River. 
  Each 
  approach 
  -­‐-­‐ 
  one 
  
derived 
  from 
  dynamical 
  models, 
  the 
  other 
  from 
  statistical 
  extrapolation 
  using 
  observed 
  
data 
  -­‐-­‐ 
  includes 
  considerably 
  uncertainty. 
   
  However 
  we 
  will 
  show 
  that 
  these 
  two 
  
approaches 
  yield 
  similar 
  estimates 
  of 
  change 
  for 
  the 
  period 
  2021-­‐2050 
  in 
  the 
  upper 
  Gila 
  
River, 
  which 
  bolsters 
  confidence 
  that 
  either 
  approach 
  yields 
  reasonable 
  and 
  actionable 
  
projections. 
   
  
Principal 
  conclusions 
  are 
  then 
  summarized 
  in 
  Section 
  5. 
   
  We 
  include 
  both 
  an 
  estimate 
  
of 
  average 
  future 
  flows, 
  and 
  -­‐-­‐ 
  importantly 
  -­‐-­‐ 
  place 
  the 
  projected 
  change 
  into 
  context 
  of 
  
natural 
  variability. 
  The 
  projection 
  of 
  estimated 
  change, 
  and 
  its 
  magnitude 
  relative 
  to 
  
decadal 
  and 
  interannual 
  variability, 
  are 
  offered 
  as 
  an 
  element 
  of 
  water 
  management 
  
planning 
  for 
  the 
  upper 
  Gila 
  River 
  in 
  New 
  Mexico. 
   
  
2. 
   
  Historical 
  streamflow 
  at 
  the 
  Gila 
  River 
  Gila 
  gage 
  
To 
  set 
  projected 
  flows 
  into 
  historical 
  context, 
  we 
  first 
  review 
  the 
  observed 
  seasonal 
  
and 
  interannual 
  record 
  of 
  streamflow 
  on 
  the 
  upper 
  Gila 
  River. 
  This 
  paper 
  will 
  focus 
  on 
  
observed 
  and 
  projected 
  flows 
  at 
  the 
  Gila 
  gage 
  (USGS 
  gage 
  09430500) 
  on 
  the 
  upper 
  Gila 
  
River. 
  The 
  location 
  of 
  the 
  Gila 
  gage 
  is 
  shown 
  by 
  a 
  red 
  square 
  in 
  Figure 
  1. 
  Annual 
  average 
  
(mean) 
  flow 
  at 
  the 
  Gila 
  gage 
  for 
  the 
  entire 
  period 
  of 
  record, 
  Water 
  Years 
  1929-­‐2012, 
  is 
  
155.6 
  cubic 
  feet 
  per 
  second 
  (cfs), 
  or 
  113×103 
  acre-­‐feet 
  per 
  year 
  (113 
  kaf/yr). 
   
  
The 
  82-­‐year 
  average 
  (median) 
  flow 
  for 
  each 
  calendar 
  day 
  in 
  this 
  record 
  is 
  shown 
  as 
  
the 
  thick 
  black 
  line 
  in 
  Figure 
  2. 
  Flows 
  average 
  about 
  60 
  cfs 
  from 
  October 
  through 
  
December, 
  then 
  increase 
  to 
  about 
  100 
  cfs 
  by 
  the 
  end 
  of 
  February. 
  The 
  Spring 
  season 
  
snowmelt 
  runoff 
  peak 
  occurs 
  in 
  March 
  and 
  April, 
  when 
  the 
  median 
  flow 
  peaks 
  near 
  200 
  cfs. 
  
Flows 
  then 
  steadily 
  decrease 
  to 
  less 
  than 
  50 
  cfs 
  by 
  the 
  end 
  of 
  June. 
  A 
  secondary 
  peak 
  in 
  the 
  
seasonal 
  hydrograph 
  of 
  median 
  flows, 
  about 
  80 
  cfs, 
  occurs 
  near 
  the 
  end 
  of 
  the 
  Water 
  Year 
  
in 
  August-­‐September, 
  as 
  rainfall 
  associated 
  with 
  the 
  summer 
  monsoon 
  replenishes 
  flow 
  in 
  
the 
  upper 
  Gila 
  River. 
   
  
The 
  shape 
  of 
  the 
  seasonal 
  hydrograph 
  at 
  the 
  Gila 
  gage 
  is 
  quite 
  robust 
  in 
  the 
  historical 
  
record. 
  Median 
  flows 
  based 
  on 
  just 
  the 
  first 
  half 
  (WY 
  1929-­‐1970) 
  and 
  second 
  half 
  (WY 
  

 
  

5 
  

 1971-­‐2012) 
  of 
  the 
  observed 
  record 
  are 
  shown 
  in 
  Figure 
  2 
  as 
  thin 
  green 
  and 
  red 
  lines, 
  
respectively. 
  Low 
  flow 
  periods 
  in 
  the 
  hydrograph, 
  in 
  October-­‐November 
  and 
  June-­‐early 
  
July, 
  are 
  nearly 
  the 
  same 
  in 
  both 
  halves 
  of 
  the 
  data 
  record. 
  However 
  the 
  median 
  flows 
  
during 
  higher 
  flow 
  periods 
  (December-­‐March, 
  April-­‐May, 
  and 
  late 
  July-­‐August) 
  are 
  greater 
  
in 
  the 
  latter 
  half 
  of 
  the 
  record. 
  Annual 
  average 
  (mean) 
  flow 
  for 
  the 
  1929-­‐1970 
  period 
  is 
  128 
  
cfs, 
  and 
  for 
  the 
  following 
  1971-­‐2012 
  period 
  is 
  183 
  cfs, 
  a 
  fluctuation 
  of 
  ±18% 
  about 
  the 
  82-­‐
year 
  average. 
  It 
  is 
  worth 
  noting 
  that 
  this 
  change 
  could 
  be 
  considered 
  a 
  statistically 
  
significant 
  long-­‐term 
  increasing 
  trend, 
  if 
  we 
  had 
  no 
  other 
  information 
  to 
  suggest 
  that 
  flows 
  
have 
  exhibited 
  multi-­‐decadal 
  natural 
  variability 
  associated 
  with 
  precipitation 
  fluctuations 
  
for 
  a 
  much 
  longer 
  period 
  of 
  time. 
   
  
In 
  addition 
  to 
  seasonal 
  variability, 
  the 
  Gila 
  River 
  exhibits 
  extensive 
  interannual 
  and 
  
decadal 
  variability. 
  The 
  time 
  series 
  of 
  the 
  entire 
  observed 
  record 
  of 
  flow 
  at 
  the 
  Gila 
  gage, 
  
with 
  annual 
  values 
  calculated 
  over 
  Water 
  Years 
  (Oct-­‐Sept), 
  is 
  shown 
  in 
  the 
  top 
  panel 
  of 
  
Figure 
  3. 
   
  As 
  indicated 
  in 
  Figure 
  2, 
  the 
  second 
  half 
  of 
  the 
  data 
  record 
  has 
  a 
  considerably 
  
higher 
  average 
  flow 
  than 
  the 
  first 
  half. 
  Notable 
  decade-­‐scale 
  low-­‐flow 
  periods 
  include 
  the 
  
1950s-­‐1960s, 
  and 
  the 
  years 
  around 
  the 
  turn 
  of 
  the 
  21st 
  Century. 
  The 
  interannual 
  and 
  
decadal 
  fluctuations 
  are 
  most 
  evident 
  in 
  terms 
  of 
  high 
  flow 
  years 
  (the 
  peaks 
  in 
  the 
  time 
  
series 
  in 
  the 
  top 
  panel 
  of 
  Figure 
  3). 
  These 
  are 
  higher 
  during 
  decades 
  of 
  high 
  flow, 
  e.g. 
  the 
  
cluster 
  of 
  years 
  with 
  annual 
  flow 
  exceeding 
  250 
  kaf 
  in 
  the 
  1980s 
  and 
  1990s. 
  Low 
  flow 
  years 
  
exhibit 
  little 
  trend: 
   
  low 
  flow 
  years 
  in 
  the 
  annual 
  average 
  time 
  series 
  tend 
  to 
  reach 
  a 
  
minimum 
  value 
  of 
  about 
  50 
  kaf/yr 
  throughout 
  the 
  data 
  record. 
   
  
Most 
  of 
  the 
  interannual 
  and 
  decadal 
  variability 
  in 
  flow 
  at 
  the 
  Gila 
  gage 
  is 
  contained 
  
within 
  the 
  winter 
  and 
  spring 
  months. 
  The 
  peaks 
  in 
  annual 
  flow 
  (high 
  flow 
  years) 
  
correspond 
  to 
  peaks 
  in 
  Dec-­‐Jun 
  flow, 
  shown 
  in 
  the 
  middle 
  panel 
  of 
  Fig. 
  3. 
   
  The 
  spring 
  
seasonal 
  maximum 
  in 
  streamflow 
  occurs 
  despite 
  the 
  observation 
  that 
  local 
  precipitation 
  
peaks 
  in 
  the 
  summer, 
  not 
  the 
  winter, 
  in 
  southwestern 
  New 
  Mexico 
  (Gutzler 
  2012). 
   
  
In 
  Sections 
  3 
  and 
  4, 
  we 
  will 
  develop 
  projections 
  for 
  future 
  flows 
  at 
  the 
  Gila 
  gage 
  that 
  
focus 
  on 
  Dec-­‐Jun 
  condtions, 
  and 
  emphasize 
  changes 
  in 
  median 
  flows. 
  The 
  statistics 
  of 
  
median 
  flows 
  are 
  less 
  affected 
  by 
  relatively 
  infrequent 
  very 
  high-­‐flow 
  years, 
  and 
  represent 
  
a 
  more 
  stable 
  metric 
  of 
  average 
  flow 
  than 
  the 
  mean. 
  Median 
  observed 
  flow 
  at 
  the 
  Gila 
  gage 
  
for 
  1951-­‐2012, 
  averaged 
  from 
  December 
  through 
  June, 
  is 
  56 
  kaf/yr. 
   
  

 
  

6 
  

 Summer 
  season 
  flows 
  (bottom 
  panel) 
  are 
  much 
  smaller 
  than 
  cold 
  season 
  flows 
  in 
  all 
  
but 
  a 
  few 
  years, 
  consistent 
  with 
  the 
  average 
  hydrographs 
  in 
  Fig. 
  2. 
  Recent 
  years 
  have 
  
exhibited 
  higher 
  summer 
  flow 
  variability, 
  and 
  several 
  of 
  the 
  major 
  peaks 
  in 
  summer 
  flow 
  
(1988, 
  2006) 
  follow 
  low 
  flow 
  seasons 
  in 
  the 
  Dec-­‐Jun 
  time 
  series 
  (Gutzler 
  2000). 
   
  
A 
  considerable 
  fraction 
  of 
  interannual 
  variance 
  in 
  streamflow 
  is 
  associated 
  with 
  shifts 
  
in 
  the 
  winter 
  large-­‐scale 
  atmospheric 
  storm 
  track, 
  forced 
  by 
  fluctuations 
  of 
  equatorial 
  
Pacific 
  Ocean 
  temperature 
  (Molles 
  and 
  Dahm 
  1990; 
  Seager 
  et 
  al. 
  2005). 
  Winters 
  in 
  which 
  
equatorial 
  Pacific 
  Ocean 
  temperature 
  is 
  anomalously 
  warm 
  (El 
  Niño) 
  tend 
  to 
  feature 
  a 
  
southward-­‐displaced 
  storm 
  track, 
  yielding 
  higher 
  than 
  average 
  snowpack 
  and 
  high 
  March-­‐
April 
  streamflow 
  when 
  the 
  snow 
  melts. 
  Winters 
  in 
  which 
  equatorial 
  Pacific 
  temperature 
  is 
  
cold 
  (La 
  Niña) 
  tend 
  to 
  lead 
  to 
  negative 
  snowpack 
  and 
  streamflow 
  anomalies. 
   
  
Recent 
  research 
  indicates 
  that 
  ocean 
  temperature 
  anomalies 
  in 
  both 
  the 
  Pacific 
  
(Gershunov 
  and 
  Barnett 
  1998; 
  Gutzler 
  et 
  al. 
  2002) 
  and 
  the 
  Atlantic 
  (McCabe 
  et 
  al. 
  2004) 
  are 
  
largely 
  responsible 
  for 
  the 
  pronounced 
  decadal 
  fluctuations 
  of 
  cold 
  season 
  precipitation 
  
variability 
  observed 
  in 
  data 
  from 
  southwestern 
  North 
  America. 
  The 
  Pacific 
  was 
  cold 
  and 
  
the 
  Atlantic 
  was 
  warm 
  during 
  the 
  1950s 
  and 
  1960s, 
  leading 
  to 
  drought 
  conditions 
  across 
  
the 
  southern 
  United 
  States; 
  Pacific 
  Ocean 
  temperatures 
  flipped 
  in 
  the 
  1970s 
  and 
  the 
  
Southwest 
  was 
  relatively 
  very 
  wet 
  for 
  several 
  decades 
  thereafter. 
  The 
  Pacific 
  Decadal 
  
Oscillation 
  appears 
  to 
  have 
  flipped 
  again 
  around 
  the 
  turn 
  of 
  the 
  21st 
  Century, 
  leading 
  to 
  a 
  
return 
  of 
  lower 
  cold 
  season 
  precipitation 
  conditions 
  that 
  continue 
  to 
  the 
  present 
  day 
  (Fig. 
  
3, 
  top 
  and 
  middle 
  panels). 
   
  
The 
  high 
  amplitude 
  of 
  variability 
  observed 
  in 
  the 
  instrumental 
  record 
  of 
  precipitation 
  
and 
  streamflow 
  is 
  almost 
  certainly 
  just 
  the 
  most 
  recent 
  expression 
  of 
  natural 
  interannual/ 
  
decadal 
  variability 
  in 
  southwestern 
  climate. 
  Recent 
  reconstructions 
  of 
  streamflow 
  on 
  the 
  
Gila 
  River 
  just 
  downstream 
  from 
  the 
  Gila 
  gage 
  (Figure 
  4) 
  generated 
  by 
  the 
  Treeflow 
  project 
  
(Meko 
  et 
  al. 
  2010; 
  Woodhouse 
  et 
  al. 
  2012) 
  suggest 
  decadal 
  variations 
  on 
  the 
  order 
  of 
  
±30%, 
  between 
  low 
  flows 
  of 
  ~250 
  kaf/yr 
  in 
  decadally-­‐filtered 
  data 
  to 
  high 
  flow 
  conditions 
  
of 
  nearly 
  500 
  kaf/yr 
  (Figure 
  4, 
  bottom 
  panel). 
  On 
  a 
  percentage 
  basis, 
  this 
  level 
  of 
  natural 
  
decadal 
  variability 
  is 
  somewhat 
  higher 
  than 
  the 
  larger 
  southwestern 
  rivers 
  discussed 
  in 
  the 
  
Review 
  Paper, 
  which 
  tend 
  to 
  exhibit 
  decadal 
  fluctuations 
  on 
  the 
  order 
  of 
  ±20%. 
   
  

 
  

7 
  

 Within 
  the 
  Spring 
  season 
  the 
  timing 
  of 
  flow 
  at 
  the 
  Gila 
  gage 
  has 
  changed 
  over 
  the 
  
historical 
  period 
  of 
  record. 
   
  The 
  Center 
  of 
  Timing 
  (CT) 
  based 
  on 
  daily 
  flows 
  represents 
  the 
  
day 
  at 
  which 
  half 
  the 
  total 
  flow 
  during 
  a 
  specific 
  period 
  of 
  time 
  passes 
  a 
  measurement 
  point 
  
(Stewart 
  et 
  al. 
  2004). 
  We 
  have 
  calculated 
  the 
  CT 
  of 
  Gila 
  gage 
  flow 
  for 
  each 
  Water 
  Year 
  for 
  
the 
  December-­‐June 
  period 
  (Figure 
  5). 
  Considering 
  just 
  the 
  December-­‐June 
  peak 
  flow 
  
period, 
  instead 
  of 
  the 
  entire 
  Water 
  Year 
  as 
  is 
  customarily 
  done 
  in 
  other 
  studies, 
  removes 
  
the 
  effect 
  of 
  strong 
  or 
  weak 
  monsoon 
  flows 
  on 
  the 
  calculation 
  so 
  that 
  the 
  CT 
  more 
  clearly 
  
represents 
  variability 
  in 
  snowmelt 
  season 
  streamflow 
  timing 
  at 
  the 
  Gila 
  gage. 
  The 
  CT 
  
exhibits 
  considerable 
  year-­‐to-­‐year 
  variability, 
  but 
  the 
  11-­‐year 
  running 
  average 
  in 
  Figure 
  5 
  
shows 
  that 
  CT 
  at 
  the 
  Gila 
  gage 
  now 
  occurs 
  around 
  day 
  153 
  of 
  the 
  Water 
  Year, 
  in 
  the 
  first 
  
week 
  of 
  March. 
  This 
  is 
  two 
  weeks 
  earlier 
  than 
  the 
  11-­‐year 
  running 
  average 
  value 
  of 
  CT 
  at 
  
the 
  beginning 
  of 
  the 
  data 
  record 
  in 
  the 
  1930s. 
   
  
The 
  causes 
  of 
  the 
  increase 
  in 
  flow 
  and 
  shift 
  toward 
  earlier 
  CT 
  at 
  the 
  Gila 
  gage 
  can 
  be 
  
understood 
  by 
  examining 
  the 
  observed 
  climatic 
  variability 
  of 
  temperature 
  and 
  
precipitation 
  in 
  southwestern 
  New 
  Mexico 
  (Figure 
  6). 
  The 
  time 
  series 
  in 
  Fig. 
  6 
  are 
  derived 
  
from 
  monthly 
  data 
  averaged 
  over 
  New 
  Mexico 
  Climate 
  Division 
  4, 
  the 
  Southwestern 
  
Mountains 
  division. 
  This 
  region 
  extends 
  across 
  a 
  much 
  larger 
  area 
  than 
  the 
  drainage 
  basin 
  
of 
  the 
  Gila 
  River. 
  However 
  on 
  monthly 
  and 
  longer 
  time 
  scales 
  the 
  spatial 
  scale 
  of 
  weather 
  is 
  
large, 
  so 
  Climate 
  Division 
  4 
  represents 
  a 
  reasonable 
  index 
  for 
  climatic 
  conditions 
  that 
  affect 
  
Gila 
  River 
  streamflow, 
  which 
  we 
  will 
  exploit 
  in 
  Section 
  4 
  of 
  this 
  paper 
  to 
  develop 
  a 
  simple 
  
statistical 
  model 
  of 
  projected 
  upper 
  Gila 
  River 
  streamflow. 
   
  
As 
  shown 
  above, 
  snowfed 
  rivers 
  such 
  as 
  the 
  Gila 
  exhibit 
  highest 
  flows, 
  and 
  highest 
  
interannual 
  fluctuations 
  in 
  flow, 
  that 
  are 
  closely 
  tied 
  to 
  the 
  winter 
  precipitation 
  that 
  
accumulates 
  as 
  snowpack 
  and 
  melts 
  in 
  the 
  Spring 
  season. 
  We 
  focus 
  here 
  on 
  the 
  
temperature 
  and 
  precipitation 
  records 
  most 
  relevant 
  to 
  variability 
  and 
  change 
  in 
  
streamflow 
  during 
  the 
  peak 
  flow 
  season 
  from 
  December-­‐June. 
  Our 
  analysis 
  of 
  covariability 
  
in 
  the 
  historical 
  record 
  shows 
  that 
  the 
  monthly 
  periods 
  that 
  correlate 
  best 
  with 
  peak 
  flow 
  
season 
  streamflow 
  are 
  January-­‐April 
  for 
  temperature 
  and 
  November-­‐April 
  for 
  
precipitation, 
  although 
  the 
  lag 
  correlations 
  between 
  precipitation 
  and 
  streamflow, 
  or 
  
temperature 
  and 
  streamflow, 
  are 
  not 
  too 
  sensitive 
  to 
  this 
  specific 
  choice 
  of 
  cold 
  season 
  
months. 
   
  

 
  

8 
  

 Temperature 
  in 
  Jan-­‐Apr 
  in 
  Climate 
  Division 
  4 
  has 
  risen 
  by 
  more 
  than 
  2°F 
  since 
  the 
  
1930s, 
  the 
  start 
  of 
  the 
  modern 
  era 
  of 
  climate 
  divisional 
  data 
  (Figure 
  6, 
  top 
  panel). 
  
Essentially 
  all 
  of 
  the 
  increase 
  in 
  temperature 
  has 
  come 
  after 
  1970. 
  Decade-­‐average 
  
temperatures, 
  shown 
  as 
  the 
  thick 
  red 
  line, 
  have 
  been 
  climbing 
  rather 
  steadily 
  since 
  the 
  
1970s, 
  punctuated 
  by 
  considerable 
  interannual 
  variability 
  that 
  is 
  smoothed 
  out 
  by 
  the 
  11-­‐
year 
  running 
  average. 
  This 
  increase 
  is 
  projected 
  to 
  continue 
  in 
  the 
  coming 
  decades, 
  and 
  is 
  
the 
  principal 
  cause 
  of 
  projected 
  declines 
  in 
  average 
  streamflow 
  in 
  the 
  Gila 
  River. 
  However 
  
the 
  average 
  annual 
  flow 
  at 
  the 
  Gila 
  gage 
  generally 
  increased 
   
  during 
  the 
  late 
  20th 
  Century 
  
period 
  of 
  warming, 
  relative 
  to 
  the 
  1951-­‐1980 
  average. 
   
  
Precipitation 
  during 
  the 
  cold 
  season 
  exhibits 
  no 
  significant 
  long-­‐term 
  trend 
  since 
  the 
  
1930s 
  (Figure 
  6, 
  bottom 
  panel). 
  The 
  decades 
  of 
  the 
  1950s 
  and 
  1960s 
  were, 
  on 
  average, 
  
relatively 
  dry, 
  followed 
  by 
  wet 
  decades 
  in 
  the 
  1980s 
  and 
  1990s, 
  and 
  a 
  subsequent 
  decline 
  
since 
  then. 
  Interannual 
  variability 
  of 
  precipitation 
  is 
  large 
  relative 
  to 
  any 
  long-­‐term 
  
tendency 
  in 
  this 
  record. 
   
   
  
As 
  shown 
  in 
  the 
  bottom 
  panel 
  of 
  Figure 
  7, 
  the 
  observed 
  record 
  (1932-­‐2012) 
  of 
  cold 
  
season 
  precipitation 
  in 
  NM4 
  (summed 
  from 
  November 
  through 
  April 
  each 
  Water 
  Year) 
  is 
  
correlated 
  at 
  a 
  level 
  of 
  r 
  = 
  0.83 
  with 
  subsequent 
  spring 
  season 
  flow 
  at 
  the 
  Gila 
  gage 
  
(summed 
  from 
  December 
  through 
  June). 
  Thus 
  about 
  68% 
  (r2) 
  of 
  interannual 
  snowmelt-­‐
season 
  streamflow 
  variance 
  is 
  accounted 
  for 
  by 
  Nov-­‐Apr 
  precipitation 
  in 
  NM4 
  during 
  the 
  
instrumental 
  period 
  of 
  record. 
  This 
  high 
  r2 
  value 
  is 
  easily 
  statistically 
  significant 
  at 
  the 
  5% 
  
level 
  (assuming 
  one 
  degree 
  of 
  freedom 
  per 
  year), 
  providing 
  empirical 
  justification 
  for 
  the 
  
use 
  of 
  NM4 
  precipitation 
  as 
  a 
  proxy 
  for 
  precipitation 
  falling 
  within 
  the 
  (much 
  smaller) 
  
upper 
  Gila 
  watershed 
  contained 
  within 
  NM4. 
   
  
The 
  top 
  panel 
  of 
  Figure 
  7 
  shows 
  NM4 
  temperature 
  in 
  the 
  late 
  winter/early 
  spring 
  
(Jan-­‐Apr) 
  plotted 
  against 
  Dec-­‐Jun 
  Gila 
  gage 
  streamflow. 
   
  This 
  correlation 
  is 
  negative 
  as 
  
expected 
  but 
  much 
  weaker 
  (r 
  = 
  -­‐0.20) 
  than 
  the 
  corresponding 
  precipitation 
  correlation 
  in 
  
the 
  bottom 
  panel. 
  Nevertheless 
  the 
  negative 
  relationship 
  between 
  streamflow 
  and 
  
temperature, 
  i.e. 
  warm 
  Jan-­‐Apr 
  temperature 
  associated 
  with 
  diminished 
  Dec-­‐Jun 
  
streamflow, 
  is 
  statistically 
  significant 
  at 
  the 
  5% 
  level 
  and 
  it 
  will 
  be 
  exploited 
  in 
  Section 
  4 
  
where 
  a 
  regression 
  model 
  of 
  streamflow 
  is 
  developed 
  and 
  assessed. 
   
  

 
  

9 
  

 The 
  increasing 
  importance 
  of 
  warming 
  temperature 
  on 
  streamflow 
  can 
  already 
  be 
  
detected 
  in 
  observed 
  data, 
  by 
  examining 
  the 
  declining 
  interannual 
  variance 
  of 
  Dec-­‐Jun 
  Gila 
  
gage 
  streamflow 
  accounted 
  for 
  by 
  cold 
  season 
  precipitation. 
  For 
  the 
  1951-­‐1980 
  period, 
  
Nov-­‐Apr 
  precipitation 
  accounted 
  for 
  75.4% 
  of 
  the 
  interannual 
  variance 
  of 
  Dec-­‐Jun 
  
streamflow, 
  with 
  a 
  regression 
  coefficient 
  of 
  42,375 
  ac-­‐ft 
  of 
  total 
  flow 
  per 
  inch 
  of 
  
precipitation 
  in 
  NM4. 
  That 
  is, 
  during 
  this 
  period 
  each 
  additional 
  inch 
  of 
  precipitation 
  over 
  
the 
  Nov-­‐Apr 
  period 
  yielded 
  42,375 
  ac-­‐ft 
  of 
  Dec-­‐Jun 
  flow 
  at 
  the 
  Gila 
  gage. 
   
  During 
  the 
  
subsequent 
  30-­‐year 
  averaging 
  period, 
  1981-­‐2010, 
  Nov-­‐Apr 
  precipitation 
  accounted 
  for 
  just 
  
66.2% 
  of 
  Dec-­‐Jun 
  streamflow 
  variance, 
  as 
  temperature 
  variability 
  accounted 
  for 
  a 
  larger 
  
fraction 
  of 
  streamflow 
  variability. 
  The 
  1981-­‐2010 
  regression 
  coefficient 
  was 
  30,755 
  ac-­‐ft 
  of 
  
of 
  Dec-­‐Jun 
  flow 
  per 
  inch 
  of 
  NM4 
  precipitation. 
  As 
  temperature 
  increases, 
  the 
  streamflow 
  
yield 
  in 
  the 
  Dec-­‐Jun 
  high 
  flow 
  season 
  decreases, 
  consistent 
  with 
  expectations 
  for 
  a 
  warming 
  
climate. 
  Winters 
  with 
  high 
  precipitation 
  still 
  generate 
  relatively 
  high 
  flows 
  in 
  the 
  Gila 
  River, 
  
but 
  the 
  flow 
  yield 
  per 
  unit 
  of 
  precipitation 
  is 
  decreasing. 
   
  
3. 
   
  Dynamical 
  Projection 
  of 
  Streamflow 
  in 
  the 
  upper 
  Gila 
  River 
  
The 
  Review 
  Paper 
  summarized 
  the 
  results 
  of 
  many 
  published 
  projections 
  of 
  future 
  
streamflow, 
  principally 
  focusing 
  on 
  projected 
  flows 
  on 
  the 
  mainstem 
  Colorado 
  and 
  Rio 
  
Grande 
  rivers. 
  While 
  these 
  projections 
  are 
  relevant 
  to 
  projected 
  flows 
  on 
  the 
  Gila 
  River, 
  the 
  
only 
  published 
  results 
  for 
  the 
  upper 
  Gila 
  itself 
  (of 
  which 
  we 
  are 
  aware) 
  were 
  generated 
  and 
  
disseminated 
  recently 
  by 
  the 
  U.S. 
  Bureau 
  of 
  Reclamation 
  (Reclamation 
  2011a,b). 
   
  These 
  
reports 
  describe 
  the 
  results 
  of 
  a 
  new 
  suite 
  of 
  streamflow 
  projections 
  for 
  the 
  21st 
  Century 
  
for 
  basins 
  across 
  the 
  western 
  United 
  States, 
  using 
  an 
  extensive 
  suite 
  of 
  global 
  model 
  
projections 
  collectively 
  denoted 
  CMIP3 
  (Meehl 
  et 
  al. 
  2007). 
  These 
  are 
  the 
  same 
  global 
  
climate 
  model 
  simulations 
  that 
  were 
  used 
  for 
  the 
  IPCC 
  (2007) 
  assessment 
  of 
  climate 
  
change. 
  The 
  CMIP3 
  archive 
  includes 
  many 
  different 
  global 
  model 
  projections, 
  using 
  a 
  wide 
  
range 
  of 
  21st 
  Century 
  greenhouse 
  gas 
  scenarios. 
  For 
  the 
  Reclamation 
  (2011a,b) 
  
projections, 
  these 
  climate 
  model 
  simulation 
  were 
  used 
  as 
  input 
  to 
  the 
  VIC 
  hydrologic 
  model 
  
(Liang 
  et 
  al. 
  1994) 
  to 
  generate 
  streamflow 
  projections 
  in 
  western 
  basins. 
   
  
The 
  procedure 
  for 
  generating 
  these 
  dynamical 
  simulations 
  is 
  summarized 
  very 
  briefly 
  
here, 
  with 
  full 
  details 
  given 
  in 
  the 
  BoR 
  reports 
  (Reclamation 
  2011a,b). 
  The 
  CMIP3 
  coarse 
  

 
  

10 
  

 resolution 
  global 
  climate 
  models, 
  based 
  on 
  dynamical 
  representation 
  of 
  atmospheric 
  and 
  
oceanic 
  climate 
  variables, 
  simulate 
  the 
  effects 
  of 
  changing 
  "boundary 
  conditions" 
  
(greenhouse 
  gas 
  concentration, 
  solar 
  fluctuations, 
  volcanic 
  eruptions) 
  on 
  winds, 
  
temperature, 
  precipitation, 
  storm 
  systems, 
  snowpack, 
  ocean 
  currents, 
  and 
  other 
  standard 
  
climate 
  variables. 
  Over 
  the 
  North 
  American 
  continent, 
  these 
  climate 
  model 
  results 
  were 
  
downscaled 
  to 
  much 
  finer 
  resolution 
  using 
  a 
  statistical 
  technique 
  (Wood 
  et 
  al. 
  2004). 
  The 
  
downscaled 
  climate 
  variables 
  were 
  then 
  expressed 
  on 
  a 
  higher 
  resolution 
  grid 
  suitable 
  for 
  
driving 
  a 
  land 
  surface/streamflow 
  model 
  (the 
  Variable 
  Infiltration 
  Capacity 
  or 
  VIC 
  model). 
  
The 
  runoff 
  simulated 
  by 
  VIC, 
  averaged 
  over 
  multiple 
  simulations 
  driven 
  by 
  different 
  
climate 
  models, 
  represents 
  the 
  dynamical 
  simulation 
  analyzed 
  here. 
  
Uncertainty 
  in 
  these 
  simulations 
  derives 
  from 
  several 
  sources. 
  First, 
  we 
  depend 
  on 
  the 
  
climate 
  models 
  to 
  represent 
  faithfully 
  the 
  response 
  of 
  large-­‐scale 
  climate 
  variables 
  to 
  the 
  
imposed 
  forcing. 
  The 
  biggest 
  forcing 
  represented 
  by 
  the 
  CMIP3 
  models 
  comes 
  from 
  
projected 
  increases 
  in 
  atmospheric 
  greenhouse 
  gas 
  concentration. 
  Uncertainty 
  in 
  the 
  
coarse 
  resolution 
  climate 
  model 
  output 
  can 
  be 
  shown 
  explicitly 
  by 
  the 
  variation 
  in 
  results 
  
generated 
  by 
  different 
  models 
  driven 
  by 
  the 
  same 
  climate 
  forcing 
  (Reclamation 
  2011a,b). 
   
  
In 
  addition, 
  there 
  is 
  considerable 
  uncertainty 
  in 
  the 
  future 
  greenhouse 
  gas 
  forcing, 
  which 
  
depends 
  on 
  economic 
  and 
  political 
  choices 
  made 
  by 
  countries 
  around 
  the 
  world 
  over 
  the 
  
coming 
  decades. 
  We 
  have 
  chosen 
  a 
  midrange 
  scenario 
  of 
  greenhouse 
  gas 
  forcing 
  (A1B) 
  to 
  
be 
  consistent 
  with 
  the 
  temperature 
  and 
  precipitation 
  projections 
  developed 
  by 
  Gutzler 
  and 
  
Robbins 
  (2011). 
   
  Finally 
  -­‐-­‐ 
  and 
  very 
  importantly 
  for 
  this 
  study 
  -­‐-­‐ 
  we 
  depend 
  on 
  the 
  models 
  
to 
  simulate 
  faithfully 
  the 
  natural 
  (unforced) 
  variability 
  of 
  climate 
  and 
  streamflow, 
  as 
  
illustrated 
  by 
  the 
  observed 
  climate 
  statistics 
  described 
  in 
  the 
  previous 
  section. 
  It 
  is 
  very 
  
difficult 
  to 
  quantify 
  these 
  individual 
  sources 
  of 
  uncertainty 
  using 
  available 
  data. 
   
  
For 
  this 
  assessment 
  we 
  have 
  analyzed 
  the 
  output 
  of 
  simulations 
  of 
  flow 
  at 
  the 
  location 
  
of 
  the 
  Gila 
  gage 
  (Reclamation 
  2011b), 
  available 
  online 
  at 
  
http://gis.usbr.gov/streamflow_projections/.

The 
  simulations 
  extend 
  backward 
  in 
  

time 
  to 
  include 
  the 
  latter 
  half 
  of 
  the 
  20th 
  Century, 
  so 
  it 
  is 
  possible 
  to 
  directly 
  compare 
  more 
  
than 
  a 
  half 
  century 
  of 
  model-­‐simulated 
  results 
  since 
  1951 
  with 
  the 
  record 
  of 
  observed 
  
flows 
  described 
  in 
  the 
  previous 
  section. 
   
  We 
  have 
  used 
  39 
  separate 
  simulations 
  based 
  on 
  
the 
  A1B 
  scenario 
  of 
  radiative 
  forcing 
  as 
  the 
  basis 
  for 
  analysis. 
  The 
  A1B 
  scenario 
  represents 
  

 
  

11 
  

 a 
  "mid 
  range" 
  estimate 
  of 
  future 
  greenhouse 
  gas 
  concentrations, 
  although 
  the 
  differences 
  in 
  
greenhouse 
  gas 
  forcing 
  among 
  plausible 
  scenarios 
  developed 
  by 
  the 
  IPCC 
  are 
  not 
  really 
  
large 
  until 
  the 
  second 
  half 
  of 
  the 
  21st 
  Century. 
   
  Seager 
  et 
  al. 
  (2008) 
  and 
  Gutzler 
  and 
  
Robbins 
  (2011) 
  used 
  the 
  suite 
  of 
  global 
  climate 
  models 
  forced 
  by 
  this 
  particular 
  scenario 
  in 
  
their 
  studies 
  of 
  climate 
  change 
  across 
  the 
  Southwest. 
   
  
Average 
  hydrographs 
  of 
  monthly 
  flow 
  at 
  the 
  Gila 
  gage, 
  derived 
  from 
  historical 
  
observed 
  data 
  in 
  the 
  top 
  panel, 
  and 
  from 
  the 
  average 
  of 
  39 
  A1B-­‐forced 
  simulations 
  as 
  
carried 
  out 
  by 
  Reclamation 
  (2011b) 
  in 
  the 
  bottom 
  panel, 
  are 
  shown 
  in 
  Figure 
  8. 
  The 
  top 
  
panel 
  is 
  based 
  on 
  the 
  same 
  daily 
  streamflow 
  data 
  used 
  to 
  generate 
  Figure 
  2. 
  For 
  this 
  plot, 
  
we 
  have 
  summed 
  the 
  daily 
  flow 
  values 
  into 
  monthly 
  averages 
  and 
  generated 
  mean 
  values 
  
(not 
  median 
  values 
  as 
  was 
  shown 
  in 
  Fig. 
  2, 
  so 
  that 
  the 
  monthly 
  values 
  sum 
  to 
  the 
  
corresponding 
  Water 
  Year 
  annual 
  mean 
  value) 
  for 
  two 
  successive 
  30-­‐year 
  averaging 
  
periods. 
  The 
  1951-­‐1980 
  climatological 
  average 
  includes 
  the 
  major 
  drought 
  of 
  the 
  1950s. 
  
The 
  following 
  30-­‐year 
  period, 
  1981-­‐2010, 
  is 
  the 
  current 
  "official" 
  30-­‐year 
  climatological 
  
averaging 
  period. 
  The 
  current 
  average 
  shows 
  higher 
  flows 
  at 
  the 
  Gila 
  gage 
  in 
  the 
  winter 
  
months 
  (December-­‐March) 
  and 
  again 
  in 
  summer 
  (particularly 
  August), 
  compared 
  to 
  the 
  
earlier 
  1951-­‐1980 
  period. 
   
  
Median 
  annual 
  average 
  flow 
  observed 
  at 
  the 
  Gila 
  gage 
  in 
  the 
  period 
  1951-­‐80 
  was 
  110 
  
cfs. 
  The 
  following 
  30 
  years, 
  1981-­‐2010, 
  were 
  on 
  average 
  considerably 
  wetter: 
  average 
  
annual 
  flow 
  at 
  the 
  Gila 
  gage 
  increased 
  to 
  182 
  cfs. 
  Thus 
  median 
  annual 
  flows 
  during 
  the 
  
1981-­‐2010 
  averaging 
  period 
  were 
  65% 
  higher 
  than 
  during 
  the 
  1951-­‐1980 
  period, 
  a 
  huge 
  
multidecadal 
  decadal 
  fluctuation 
  representing 
  the 
  transition 
  from 
  a 
  very 
  dry 
  period 
  to 
  a 
  
very 
  wet 
  pluvial 
  period. 
   
  
The 
  average 
  of 
  39 
  BoR 
  streamflow 
  simulations 
  is 
  shown 
  in 
  the 
  bottom 
  panel 
  of 
  Figure 
  
8 
  with 
  the 
  daily 
  values 
  averaged 
  into 
  monthly 
  means. 
  Hydrographs 
  for 
  successive 
  30-­‐year 
  
periods 
  for 
  both 
  retrospective 
  and 
  future 
  projected 
  climate 
  conditions 
  are 
  shown. 
  The 
  
retrospective 
  simulations, 
  shown 
  using 
  the 
  same 
  line 
  color 
  conventions 
  as 
  for 
  the 
  observed 
  
data, 
  show 
  that 
  the 
  VIC-­‐based 
  simulations 
  capture 
  the 
  shape 
  of 
  seasonal 
  hydrograph 
  
reasonably 
  well 
  but 
  tend 
  to 
  overestimate 
  flows 
  relative 
  to 
  observations. 
  The 
  annual 
  
average 
  (Water 
  Year 
  mean) 
  flow 
  rate 
  simulated 
  at 
  the 
  Gila 
  gage 
  is 
  199 
  cfs 
  for 
  1951-­‐1980 
  
and 
  193 
  cfs 
  for 
  1981-­‐2010. 
   
  Thus 
  the 
  dynamical 
  simulations 
  of 
  Gila 
  flow 
  tend 
  to 
  show 
  a 
  

 
  

12 
  

 decrease 
  between 
  the 
  two 
  most 
  recent 
  independent 
  30-­‐year 
  averaging 
  period, 
  whereas 
  
actual 
  observations 
  for 
  the 
  same 
  periods 
  shows 
  an 
  increase 
  in 
  flow 
  at 
  the 
  Gila 
  gage. 
   
  
A 
  complete 
  diagnosis 
  of 
  the 
  reasons 
  for 
  the 
  general 
  high 
  bias 
  of 
  simulated 
  streamflow 
  
is 
  beyond 
  the 
  scope 
  of 
  this 
  study, 
  but 
  an 
  imperfect 
  match 
  with 
  observations 
  at 
  a 
  specific 
  
point 
  is 
  not 
  surprising 
  considering 
  the 
  uncertainties 
  associated 
  with 
  model 
  resolution, 
  etc. 
  
(Reclamation 
  2011b). 
  These 
  uncertainties 
  are 
  extremely 
  difficult 
  to 
  quantify 
  using 
  the 
  
output 
  available. 
  Given 
  the 
  systematic 
  bias 
  between 
  model 
  output 
  and 
  pointwise 
  
observations, 
  it 
  is 
  customary 
  to 
  consider 
  percentage 
  changes 
  in 
  the 
  model-­‐simulated 
  
streamflows 
  rather 
  than 
  try 
  to 
  use 
  the 
  direct 
  output 
  from 
  a 
  dynamical 
  model 
  to 
  make 
  
quantitative 
  projections 
  of 
  changes 
  in 
  streamflow. 
   
  
Time 
  series 
  of 
  annual 
  flows 
  for 
  each 
  of 
  the 
  39 
  simulations 
  considered 
  are 
  shown 
  in 
   
   
  
Figure 
  9. 
  The 
  various 
  simulations 
  clearly 
  exhibit 
  intermittent 
  high 
  flow 
  years 
  which 
  
continue 
  to 
  skew 
  mean 
  flows. 
  The 
  time 
  series 
  of 
  median 
  values 
  of 
  all 
  39 
  simulations 
  each 
  
year, 
  shown 
  as 
  a 
  thick 
  black 
  line 
  in 
  both 
  panels, 
  trends 
  slowly 
  but 
  steadily 
  downward 
  
through 
  the 
  21st 
  Century. 
   
  (We 
  revert 
  back 
  to 
  median 
  streamflow 
  values 
  in 
  Fig. 
  9 
  to 
  
minimize 
  the 
  effects 
  of 
  one 
  or 
  two 
  outlier 
  annual 
  values 
  on 
  each 
  39-­‐model 
  Water 
  Year 
  
average; 
  a 
  time 
  series 
  of 
  mean 
  values 
  would 
  also 
  trend 
  downward 
  but 
  would 
  contain 
  much 
  
more 
  interannual 
  variability.) 
  
Although 
  precipitation 
  variability 
  accounts 
  for 
  most 
  of 
  the 
  variance 
  of 
  streamflow 
  in 
  
the 
  historical 
  record 
  (Figure 
  7), 
  precipitation 
  change 
  is 
  not 
  the 
  principal 
  cause 
  of 
  long-­‐term 
  
projected 
  streamflow 
  change 
  as 
  shown 
  in 
  Figures 
  8 
  and 
  9. 
   
  This 
  is 
  shown 
  by 
  considering 
  
the 
  projected 
  changes 
  in 
  snowpack 
  and 
  precipitation 
  downscaled 
  from 
  global 
  model 
  
projections 
  to 
  the 
  upper 
  Gila 
  watershed 
  (Figure 
  10). 
   
  There 
  is 
  very 
  little 
  projected 
  
precipitation 
  change 
  except 
  for 
  the 
  summer, 
  when 
  downscaled 
  precipitation 
  decreases 
  
somewhat 
  (bottom 
  panel). 
   
  However 
  snowpack 
  changes 
  in 
  the 
  cold 
  season 
  are 
  extreme 
  
(top 
  panel) 
  despite 
  the 
  absence 
  of 
  simultaneous 
  precipitation 
  change. 
  Snowpack 
  is 
  nearly 
  
eliminated 
  in 
  the 
  Gila 
  watershed 
  due 
  to 
  the 
  projected 
  temperature 
  increase 
  (shown 
  in 
  the 
  
top 
  panel 
  of 
  Fig. 
  11), 
  a 
  result 
  that 
  has 
  been 
  highlighted 
  in 
  previous 
  studies 
  of 
  regional 
  
climate 
  change 
  in 
  the 
  southwestern 
  U.S. 
  (Diffenbaugh 
  et 
  al. 
  2005; 
  NM 
  OSE 
  2006). 
   
  

 
  

13 
  

 The 
  decrease 
  in 
  snowpack 
  acts 
  to 
  reduce 
  the 
  snowmelt 
  runoff 
  peak 
  that 
  is 
  projected 
  
to 
  occur 
  in 
  coming 
  decades 
  (Fig. 
  8), 
  to 
  the 
  point 
  at 
  which 
  the 
  peak 
  in 
  streamflow 
  associated 
  
with 
  snowmelt 
  runoff 
  is 
  effectively 
  spread 
  out 
  between 
  January 
  and 
  March 
  by 
  the 
  late 
  21st 
  
Century 
  (red 
  curve 
  in 
  the 
  bottom 
  panel 
  of 
  Fig. 
  8). 
  The 
  decline 
  in 
  snowpack 
  shifts 
  the 
  Center 
  
of 
  Timing 
  toward 
  earlier 
  dates, 
  as 
  is 
  already 
  seen 
  in 
  observations 
  (Fig. 
  5). 
  Warmer 
  
temperatures 
  over 
  a 
  longer 
  snow-­‐free 
  season 
  increase 
  evapotranspiration 
  rates, 
  resulting 
  
in 
  reduced 
  streamflow 
  year-­‐round 
  (Hurd 
  and 
  Coonrod 
  2008, 
  2012). 
   
  
In 
  the 
  current 
  climate 
  a 
  downward 
  trend 
  in 
  streamflow 
  driven 
  by 
  warmer 
  
temperatures 
  has 
  been 
  masked 
  by 
  the 
  increase 
  in 
  precipitation 
  that 
  occurred 
  in 
  the 
  late 
  
20th 
  Century 
  (Fig. 
  6). 
  However 
  if 
  the 
  wet 
  decades 
  were 
  the 
  late 
  20th 
  Century 
  represent 
  an 
  
episodic 
  decadal 
  fluctuation 
  of 
  precipitation 
  -­‐-­‐ 
  and 
  the 
  recent 
  run 
  of 
  very 
  dry 
  years 
  early 
  in 
  
the 
  21st 
  Century 
  (Fig. 
  11) 
  suggests 
  that 
  the 
  wet 
  period 
  may 
  be 
  over 
  -­‐-­‐ 
  then 
  we 
  can 
  expect 
  
the 
  temperature-­‐driven 
  streamflow 
  changes 
  illustrated 
  in 
  the 
  streamflow 
  projections 
  (Figs. 
  
8 
  and 
  12) 
  to 
  become 
  more 
  evident 
  in 
  the 
  near 
  future. 
  This 
  is 
  the 
  scenario 
  outlined 
  by 
  
Seager 
  et 
  al. 
  (2008), 
  which 
  they 
  described 
  as 
  an 
  "imminent 
  transition" 
  toward 
  long-­‐term 
  
aridity 
  across 
  Southwestern 
  North 
  America. 
   
  
Summer 
  precipitation 
  projections 
  tend 
  to 
  be 
  uncertain 
  in 
  both 
  CMIP3 
  (Gutzler 
  and 
  
Robbins 
  2011) 
  and 
  CMIP5 
  (Cook 
  and 
  Seager 
  2013) 
  climate 
  model 
  simulations. 
  As 
  Figure 
  10 
  
suggests, 
  some 
  climate 
  models 
  simulate 
  decreased 
  monsoon 
  precipitation 
  in 
  a 
  projected 
  
warmer 
  climate. 
  Here 
  again, 
  observations 
  from 
  the 
  late 
  20th 
  Century 
  show 
  higher 
  
streamflow, 
  associated 
  higher 
  precipitation, 
  in 
  a 
  warmer 
  climate. 
  The 
  Reclamation 
  
simulations 
  in 
  the 
  bottom 
  panel 
  of 
  Figure 
  8 
  yield 
  little 
  change 
  in 
  summer 
  streamflow, 
  the 
  
result 
  of 
  a 
  small 
  net 
  average 
  change 
  among 
  different 
  climate 
  models 
  that 
  simulate 
  a 
  wide 
  
range 
  of 
  higher 
  or 
  lower 
  summer 
  streamflow. 
   
  
Decadal 
  variability 
  represents 
  a 
  large 
  source 
  of 
  uncertainty 
  in 
  these 
  model 
  
simulations. 
  Considerable 
  progress 
  has 
  been 
  made 
  in 
  the 
  simulation 
  of 
  interannual 
  ocean-­‐
atmosphere 
  variability 
  in 
  coupled 
  models, 
  especially 
  the 
  El 
  Niño-­‐Southern 
  Oscillation 
  
(ENSO) 
  cycle, 
  and 
  the 
  CMIP3 
  climate 
  models 
  used 
  in 
  the 
  Reclamation 
  simulations 
  generally 
  
do 
  a 
  credible 
  job 
  simulating 
  the 
  evolution 
  and 
  large-­‐scale 
  climatic 
  effects 
  of 
  ENSO 
  (IPCC 
  
2007). 
   
  

 
  

14 
  

 However 
  decadal 
  variability, 
  such 
  as 
  may 
  be 
  associated 
  with 
  the 
  Pacific 
  Decadal 
  
Oscillation 
  or 
  Atlantic 
  Multidecadal 
  Variability 
  (McCabe 
  et 
  al. 
  2004), 
  is 
  more 
  problematic. 
  
Atmospheric 
  models 
  that 
  are 
  provided 
  with 
  the 
  correct 
  ocean 
  temperature 
  fluctuations 
  can 
  
reasonably 
  simulate 
  long-­‐term 
  drought 
  and 
  pluvial 
  episodes 
  (e.g. 
  Hoerling 
  and 
  Kumar 
  
2003; 
  Seager 
  et 
  al. 
  2005; 
  Schubert 
  et 
  al. 
  2009) 
  but 
  getting 
  coupled 
  ocean-­‐atmosphere 
  
models 
  to 
  generate 
  sufficient 
  decadal 
  variability 
  on 
  their 
  own 
  is 
  more 
  challenging. 
  Gutzler 
  
et 
  al. 
  (2012) 
  showed 
  that 
  even 
  newer 
  and 
  more 
  sophisticated 
  coupled 
  models 
  from 
  the 
  
new 
  CMIP5 
  archive 
  (Taylor 
  et 
  al. 
  2012), 
  successor 
  to 
  the 
  CMIP3 
  models 
  used 
  in 
  the 
  
Reclamation 
  streamflow 
  simulations, 
  generate 
  annual 
  precipitation 
  variability 
  across 
  the 
  
Southwest 
  that 
  contains 
  significantly 
  less 
  year-­‐to-­‐year 
  persistence 
  than 
  observed 
  
precipitation. 
  In 
  other 
  words, 
  decadal 
  variability 
  associated 
  with 
  unforced 
  ocean-­‐
atmosphere 
  interactions 
  is 
  probably 
  underestimated 
  in 
  the 
  model 
  simulations 
  analyzed 
  
here. 
   
  
With 
  this 
  in 
  mind, 
  we 
  emphasize 
  again 
  that 
  the 
  dynamical 
  simulations 
  shown 
  in 
  
Figure 
  8 
  yield 
  the 
  wrong 
  answer 
  for 
  late 
  20th 
  Century 
  variability. 
  The 
  observed 
  seasonal 
  
hydrographs 
  for 
  the 
  Gila 
  gage 
  (top 
  panel) 
  show 
  increasing 
  flows 
  between 
  1951-­‐80 
  and 
  
1981-­‐2010. 
  The 
  corresponding 
  Reclamation 
  dynamical 
  simulations 
  (bottom 
  panel) 
  show 
  
decreasing 
  flows 
  between 
  the 
  same 
  periods. 
  The 
  difference 
  is 
  caused 
  by 
  the 
  failure 
  of 
  the 
  
models 
  to 
  correctly 
  capture 
  the 
  increase 
  in 
  precipitation 
  observed 
  between 
  the 
  "drought 
  
epoch" 
  (1951-­‐1980) 
  and 
  the 
  following 
  "pluvial 
  epoch" 
  (1981-­‐2010), 
  which 
  leads 
  to 
  higher 
  
flows 
  in 
  the 
  upper 
  Gila 
  River 
  in 
  response 
  to 
  the 
  generally 
  heavier 
  precipitation 
  amounts. 
  On 
  
the 
  other 
  hand, 
  the 
  models 
  do 
  respond 
  to 
  increasing 
  greenhouse 
  gases 
  in 
  the 
  late 
  20th 
  
Century 
  by 
  increasing 
  the 
  temperature, 
  as 
  observed. 
  The 
  higher 
  temperatures 
  lead 
  to 
  
decreases 
  in 
  simulated 
  snowpack 
  and 
  increases 
  in 
  simulated 
  evaporation, 
  hence 
  declining 
  
flows. 
   
  
How 
  should 
  we 
  assess 
  future 
  projections 
  in 
  streamflow, 
  considering 
  that 
  the 
  
simulated 
  streamflow 
  over 
  the 
  past 
  60 
  years 
  is 
  changing 
  with 
  the 
  wrong 
  sign 
  relative 
  to 
  
observations? 
  There 
  are 
  two 
  distinct 
  possible 
  interpretations. 
  First, 
  the 
  simulated 
  future 
  
projections 
  could 
  be 
  considered 
  unreliable, 
  given 
  that 
  the 
  models 
  generate 
  the 
  wrong 
  
answer 
  (decreasing 
  flow) 
  for 
  retrospective 
  streamflow 
  trend 
  over 
  the 
  half-­‐century 
  period 
  
of 
  time 
  when 
  we 
  know 
  the 
  right 
  answer 
  (increasing 
  flow). 
  Alternatively, 
  the 
  simulated 
  

 
  

15 
  

 future 
  trends 
  could 
  represent 
  a 
  reasonable 
  long-­‐term 
  answer, 
  keeping 
  in 
  mind 
  that 
  the 
  
long-­‐term 
  trend 
  is 
  subject 
  to 
  considerable 
  uncertainty 
  on 
  decadal 
  time 
  scales. 
   
  
The 
  alternative 
  possibility 
  amounts 
  to 
  concluding 
  that 
  the 
  retrospective 
  simulation 
  of 
  
a 
  decrease 
  in 
  flow 
  over 
  the 
  latter 
  half 
  of 
  the 
  20th 
  Century 
  represents 
  the 
  "right 
  answer" 
  to 
  
the 
  response 
  of 
  the 
  Gila 
  River 
  to 
  higher 
  temperatures, 
  but 
  the 
  simulated 
  decrease 
  was 
  
overwhelmed 
  by 
  a 
  natural, 
  decadal 
  transition 
  in 
  precipitation 
  from 
  drought 
  to 
  pluvial 
  
climatic 
  conditions 
  during 
  that 
  time. 
  If 
  this 
  is 
  the 
  proper 
  interpretation; 
  and 
  if 
  the 
  wetter 
  
conditions 
  observed 
  in 
  the 
  late 
  20th 
  Century 
  represent 
  the 
  wet 
  phase 
  of 
  an 
  episodic, 
  
decadal 
  fluctuation 
  in 
  precipitation 
  that 
  will 
  naturally 
  revert 
  back 
  toward 
  drier 
  conditions; 
   
  
and 
  if 
  the 
  temperature 
  increases 
  over 
  the 
  next 
  few 
  decades 
  to 
  the 
  point 
  at 
  which 
  the 
  long-­‐
term 
  temperature 
  effect 
  overwhelms 
  the 
  decadal 
  variability 
  of 
  precipitation 
  variability; 
  
then 
  the 
  projection 
  of 
  decreased 
  streamflow 
  in 
  the 
  coming 
  decades 
  may 
  in 
  fact 
  be 
  a 
  
reasonable 
  basis 
  for 
  water 
  allocation, 
  despite 
  the 
  failure 
  of 
  the 
  dynamical 
  simulations 
  to 
  
correctly 
  reproduce 
  late 
  20th 
  Century 
  changes 
  in 
  streamflow 
  in 
  the 
  upper 
  Gila 
  River. 
   
  
It 
  is 
  my 
  professional 
  opinion 
  that 
  the 
  alternative 
  hypothesis 
  sketched 
  above 
  is 
  correct, 
  
and 
  that 
  it 
  is 
  likely 
  that 
  the 
  long-­‐term 
  effects 
  of 
  increasing 
  temperature 
  on 
  snowpack 
  and 
  
streamflow 
  will 
  become 
  comparable 
  to 
  the 
  decadal 
  variability 
  driven 
  largely 
  by 
  natural 
  
precipitation 
  fluctuations 
  by 
  the 
  middle 
  of 
  the 
  21st 
  century. 
  This 
  opinion 
  is 
  supported 
  by 
  
historical 
  observations 
  of 
  precipitation 
  (Figs. 
  6 
  and 
  11), 
  suggesting 
  that 
  late 
  20th 
  Century 
  
wet 
  conditions 
  were 
  part 
  of 
  a 
  multi-­‐decadal 
  fluctuation 
  that 
  may 
  already 
  have 
  reverted 
  
back 
  toward 
  dry 
  conditions, 
  and 
  by 
  the 
  steadier, 
  ongoing 
  increase 
  in 
  temperature 
  (Figs. 
  6, 
  
11) 
  and 
  streamflow 
  timing 
  (Fig. 
  5) 
  that 
  are 
  entirely 
  consistent 
  with 
  long-­‐term 
  projections 
  
of 
  a 
  21st 
  Century 
  warming 
  trend. 
   
  
Hurd 
  and 
  Coonrod 
  (2008, 
  2012) 
  showed 
  that 
  streamflows 
  in 
  the 
  middle 
  Rio 
  Grande 
  
(northeast 
  of 
  the 
  upper 
  Gila 
  basin) 
  are 
  projected 
  to 
  decrease 
  significantly 
  by 
  the 
  late 
  21st 
  
Century 
  regardless 
  of 
  precipitation 
  projections. 
  Even 
  a 
  model 
  simulation 
  that 
  includes 
  
higher 
  precipitation 
  amounts 
  in 
  the 
  late 
  21st 
  Century 
  yielded 
  smaller 
  Rio 
  Grande 
  
streamflow 
  in 
  the 
  Hurd 
  and 
  Coonrod 
  study, 
  because 
  the 
  projected 
  increases 
  in 
  
evapotranspiration 
  (associated 
  with 
  higher 
  temperature) 
  cause 
  the 
  water 
  budget 
  
(Precipitation 
  minus 
  Evapotranspiration) 
  to 
  trend 
  downward 
  even 
  if 
  precipitation 
  trends 
  
upward, 
  as 
  occurs 
  in 
  some 
  individual 
  climate 
  model 
  simulations. 
   
  

 
  

16 
  

 If 
  we 
  assume 
  that 
  the 
  effect 
  of 
  increasing 
  temperature 
  to 
  produce 
  diminished 
  
streamflow 
  is 
  at 
  least 
  qualitatively 
  reasonable, 
  we 
  can 
  use 
  the 
  dynamical 
  simulations 
  to 
  try 
  
to 
  project 
  percentage 
  change 
  in 
  streamflow 
  in 
  coming 
  decades. 
  Median 
  annual 
  streamflow 
  
at 
  the 
  Gila 
  gage 
  in 
  the 
  period 
  2021-­‐2050 
  in 
  39 
  BoR 
  simulations 
  is 
  157 
  cfs, 
  whereas 
  median 
  
annual 
  flow 
  in 
  the 
  same 
  simulations 
  in 
  the 
  period 
  1951-­‐2012 
  is 
  170 
  cfs, 
  a 
  decrease 
  of 
  13 
  
cfs/year 
  (about 
  8% 
  lower 
  in 
  2021-­‐2050). 
  Here 
  we 
  use 
  a 
  record 
  longer 
  than 
  the 
  official 
  30 
  
year 
  averaging 
  period 
  to 
  establish 
  current 
  climatic 
  conditions, 
  in 
  order 
  to 
  include 
  both 
  dry 
  
and 
  wet 
  historical 
  periods 
  in 
  our 
  definition 
  of 
  current 
  climate. 
   
  
The 
  Reclamation 
  A1B-­forced 
  dynamical 
  simulations 
  indicate 
  a 
  decrease 
  in 
  
annual 
  median 
  flow 
  at 
  the 
  Gila 
  gage 
  of 
  13 
  cfs, 
  or 
  about 
  8%, 
  between 
  the 
  periods 
  
1951-­2012 
  and 
  2021-­2050. 
   
  
4. 
   
  Statistical 
  Projection 
  of 
  Streamflow 
  in 
  the 
  upper 
  Gila 
  River 
  
Statistical 
  models 
  of 
  streamflow 
  examine 
  the 
  climatic 
  factors 
  that 
  modulate 
  
streamflow 
  in 
  the 
  current 
  climate, 
  as 
  demonstrated 
  by 
  observed 
  data, 
  and 
  use 
  the 
  observed 
  
relationships 
  between 
  climate 
  and 
  streamflow 
  to 
  estimate 
  future 
  flows 
  based 
  on 
  projected 
  
climatic 
  conditions. 
   
   
  
The 
  relationship 
  between 
  interannual 
  fluctuations 
  of 
  cold 
  season 
  precipitation 
  (or 
  
snowpack) 
  and 
  spring 
  streamflow 
  in 
  southwestern 
  rivers 
  is 
  strong 
  and 
  unassailable 
  -­‐-­‐ 
  this 
  
relationship 
  forms 
  the 
  basis 
  for 
  operational 
  seasonal 
  forecasts 
  of 
  streamflow 
  issued 
  early 
  
each 
  calendar 
  year, 
  near 
  the 
  end 
  of 
  the 
  snow 
  accumulation 
  season. 
  Jones 
  (2007) 
  showed 
  
that 
  80% 
  of 
  the 
  interannual 
  variance 
  of 
  the 
  spring 
  season 
  flow 
  in 
  the 
  late 
  20th 
  Century 
  in 
  
the 
  upper 
  San 
  Juan 
  river 
  could 
  be 
  accounted 
  for 
  by 
  the 
  annual 
  1 
  April 
  measurement 
  of 
  
snow 
  water 
  equivalent 
  at 
  a 
  single 
  SNOTEL 
  site 
  near 
  the 
  river's 
  headwaters. 
  Similarly, 
  
Salgado 
  and 
  Gutzler 
  (2013) 
  showed 
  that 
  just 
  under 
  60% 
  of 
  late 
  20th 
  Century 
  spring 
  season 
  
flow 
  in 
  the 
  upper 
  Pecos 
  River 
  could 
  be 
  accounted 
  for 
  by 
  winter 
  precipitation 
  averaged 
  over 
  
New 
  Mexico 
  Climate 
  Division 
  2, 
  which 
  includes 
  the 
  Pecos 
  headwaters 
  in 
  northern 
  New 
  
Mexico. 
  Figure 
  7 
  illustrates 
  an 
  analogous 
  relationship 
  between 
  interannual 
  fluctuations 
  of 
  
observed 
  winter 
  precipitation 
  and 
  spring 
  streamflow 
  in 
  the 
  upper 
  Gila 
  River, 
  with 
  68% 
  of 
  
interannual 
  Dec-­‐Jun 
  streamflow 
  variance 
  at 
  the 
  Gila 
  gage 
  accounted 
  for 
  by 
  Nov-­‐Apr 
  
precipitation 
  averaged 
  over 
  the 
  NM4 
  Climate 
  Division. 
   
  

 
  

17 
  

 These 
  statistical 
  models 
  of 
  interannual 
  streamflow 
  fluctuations 
  work 
  for 
  the 
  current 
  
climate 
  because 
  year-­‐to-­‐year 
  variations 
  of 
  winter 
  precipitation 
  represent 
  the 
  major 
  
climatic 
  factor 
  modulating 
  streamflow. 
  Statistical 
  projections 
  derived 
  from 
  the 
  statistics 
  of 
  
current 
  climate 
  models 
  are 
  potentially 
  applicable 
  to 
  future 
  climatic 
  conditions 
  only 
  as 
  long 
  
as 
  the 
  mean 
  and 
  variability 
  of 
  climate 
  remain 
  similar 
  to 
  present-­‐day 
  statistics 
  (Milly 
  et 
  al. 
  
2008). 
  However, 
  if 
  the 
  climate 
  warms 
  to 
  the 
  point 
  at 
  which 
  temperature, 
  rather 
  than 
  
precipitation, 
  becomes 
  a 
  primary 
  control 
  on 
  streamflow 
  changes, 
  then 
  a 
  precipitation-­‐
dominated 
  statistical 
  model 
  would 
  no 
  longer 
  be 
  applicable. 
  As 
  discussed 
  at 
  the 
  end 
  of 
  
Section 
  2, 
  a 
  precipitation-­‐only 
  regression 
  model 
  already 
  exhibits 
  evidence 
  of 
  a 
  changing 
  
climate: 
  Nov-­‐Apr 
  precipitation 
  accounted 
  for 
  a 
  decreasing 
  fraction 
  of 
  Dec-­‐Jun 
  streamflow 
  
variance 
  in 
  1980-­‐2010 
  relative 
  to 
  1951-­‐1980, 
  and 
  the 
  yield 
  of 
  streamflow 
  per 
  unit 
  of 
  Nov-­‐
Apr 
  precipitation 
  declined 
  substantially 
  in 
  the 
  later 
  period. 
  As 
  climate 
  continues 
  to 
  change, 
  
a 
  regression 
  model 
  derived 
  from 
  observed 
  statistics 
  becomes 
  increasingly 
  problematic. 
   
  
The 
  potential 
  non-­‐applicability 
  of 
  statistical 
  models 
  as 
  discussed 
  above 
  has 
  been 
  
discussed 
  at 
  length 
  by 
  Milly 
  et 
  al. 
  (2008), 
  whose 
  paper 
  has 
  the 
  evocative 
  title 
  "Stationarity 
  
is 
  Dead." 
  Hoerling 
  and 
  Eischeid 
  (2007) 
  used 
  the 
  20th 
  Century 
  relationship 
  between 
  Palmer 
  
Drought 
  Severity 
  Index 
  (PDSI) 
  and 
  Colorado 
  River 
  streamflow 
  to 
  estimate 
  future 
  flows 
  at 
  
Lees 
  Ferry 
  based 
  on 
  PDSI 
  values 
  derived 
  from 
  temperature 
  and 
  precipitation 
  projected 
  by 
  
a 
  climate 
  model. 
  They 
  projected 
  a 
  45% 
  reduction 
  in 
  Colorado 
  River 
  flows 
  using 
  this 
  
statistical 
  model 
  by 
  2035-­‐2060 
  relative 
  to 
  the 
  1990-­‐2005 
  average, 
  a 
  dramatic 
  decline 
  that 
  
is 
  roughly 
  double 
  the 
  reduction 
  in 
  flow 
  generated 
  directly 
  by 
  the 
  dynamical 
  model 
  
projections 
  cited 
  earlier. 
  The 
  difference 
  has 
  been 
  attributed 
  to 
  the 
  increasing 
  role 
  of 
  
temperature 
  in 
  projected 
  PDSI 
  values, 
  which 
  may 
  lead 
  to 
  overestimates 
  of 
  the 
  reduction 
  in 
  
actual 
  surface 
  drying 
  due 
  to 
  the 
  way 
  in 
  which 
  surface 
  moisture 
  is 
  parameterized 
  in 
  the 
  
PDSI 
  calculation 
  (Sheffield 
  et 
  al. 
  2012). 
  Given 
  this 
  uncertainty 
  I 
  would 
  assess 
  the 
  Hoerling 
  
and 
  Eischeid 
  (2007) 
  streamflow 
  projection 
  with 
  skepticism 
  and 
  consider 
  it 
  to 
  represent 
  an 
  
overestimate 
  of 
  likely 
  streamflow 
  reduction 
  in 
  the 
  Colorado 
  River. 
   
  
Although 
  temperatures 
  are 
  already 
  increasing 
  significantly 
  in 
  southwestern 
  New 
  
Mexico 
  (Figure 
  6), 
  projected 
  temperature 
  does 
  not 
  completely 
  leave 
  its 
  envelope 
  of 
  
historical 
  variability 
  until 
  the 
  mid-­‐21st 
  Century 
  (Gutzler 
  and 
  Robbins 
  2011; 
  Gutzler 
  2012). 
  
Furthermore 
  precipitation 
  trends 
  in 
  observations 
  or 
  projections 
  are 
  relatively 
  small 
  

 
  

18 
  

 compared 
  to 
  the 
  magnitude 
  of 
  20th 
  Century 
  climate 
  variability 
  (Gutzler 
  and 
  Robbins 
  2011; 
  
bottom 
  panel 
  of 
  Fig. 
  11). 
  Development 
  of 
  statistical 
  models 
  to 
  project 
  streamflow 
  in 
  the 
  
upper 
  Gila 
  River 
  for 
  the 
  next 
  several 
  decades, 
  while 
  temperatures 
  are 
  still 
  within 
  historical 
  
bounds, 
  therefore 
  seems 
  justifiable. 
   
  
We 
  have 
  adapted 
  the 
  regressions 
  described 
  in 
  Fig. 
  7, 
  shown 
  as 
  curves 
  fit 
  to 
  the 
  scatter 
  
plots 
  of 
  temperature 
  or 
  precipitation 
  vs. 
  streamflow, 
  and 
  have 
  thereby 
  generated 
  a 
  
multivariable 
  regression 
  model 
  that 
  relates 
  annual 
  values 
  of 
  Nov-­‐Apr 
  precipitation 
  
(denoted 
  PNovApr, 
  units 
  = 
  inches) 
  and 
  Jan-­‐Apr 
  temperature 
  (denoted 
  TJanApr, 
  units 
  = 
  °F) 
  in 
  
New 
  Mexico 
  Climate 
  Division 
  4, 
  to 
  corresponding 
  annual 
  values 
  of 
  Dec-­‐Jun 
  average 
  
streamflow 
  at 
  the 
  Gila 
  gage 
  (denoted 
  QDecJun, 
  units 
  = 
  kaf). 
   
  
Observed 
  values 
  of 
  these 
  variables 
  for 
  the 
  period 
  1951-­‐2012 
  (as 
  in 
  the 
  previous 
  
section, 
  we 
  wish 
  to 
  include 
  both 
  dry 
  and 
  wet 
  decadal 
  periods) 
  were 
  used 
  to 
  create 
  the 
  
regression 
  model, 
  which 
  is: 
  
QDecJun 
   
  = 
   
  9.29PNovApr 
   
  + 
   
  2.59(PNovApr)2 
   
  – 
   
  5.38TJanApr 
   
  + 
   
  208.6 
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
  [1] 
  
The 
  linear 
  precipitation 
  term 
  (the 
  first 
  term 
  on 
  the 
  right 
  hand 
  side 
  of 
  the 
  equation) 
  
accounts 
  for 
  the 
  most 
  variance. 
  The 
  quadratic 
  precipitation 
  term 
  that 
  follows 
  is 
  needed 
  to 
  
make 
  the 
  model 
  fit 
  low-­‐precipitation 
  years, 
  during 
  which 
  base 
  flow 
  in 
  the 
  drainage 
  basin 
  
continues 
  even 
  when 
  precipitation 
  is 
  quite 
  low. 
  The 
  bottom 
  panel 
  in 
  Figure 
  7 
  shows 
  clearly 
  
that 
  the 
  relationship 
  between 
  precipitation 
  and 
  streamflow 
  is 
  not 
  linear 
  at 
  both 
  the 
  high-­‐
and 
  low 
  ends 
  of 
  the 
  range 
  of 
  seasonal 
  precipitation 
  values. 
   
  
The 
  third 
  term, 
  a 
  negative 
  linear 
  term 
  involving 
  Jan-­‐Apr 
  temperature, 
  plays 
  a 
  small 
  
role 
  in 
  modulating 
  streamflow 
  in 
  the 
  historical 
  record 
  but 
  becomes 
  increasingly 
  important 
  
as 
  temperatures 
  rise. 
  Although 
  interannual 
  fluctuations 
  of 
  temperature 
  are 
  not 
  highly 
  
correlated 
  with 
  streamflow 
  in 
  the 
  historical 
  record 
  (Figure 
  7, 
  top 
  panel), 
  there 
  is 
  a 
  cluster 
  
of 
  data 
  points 
  in 
  the 
  bottom-­‐right 
  corner 
  of 
  the 
  plot 
  describing 
  years 
  where 
  TJanApr 
  is 
  warm 
  
(between 
  40°F 
  and 
  42°F) 
  and 
  QDecJun 
  is 
  very 
  low 
  (<50 
  kaf). 
  These 
  data 
  points 
  in 
  the 
  
historical 
  record 
  of 
  temperature 
  are 
  a 
  harbinger 
  of 
  the 
  projected 
  warmer 
  climate 
  shown 
  in 
  
the 
  top 
  panel 
  of 
  Fig. 
  11. 
  The 
  inclusion 
  of 
  this 
  temperature-­‐related 
  term 
  in 
  the 
  statistical 
  
model 
  captures 
  the 
  effects 
  of 
  these 
  historical 
  years 
  in 
  depressing 
  streamflow 
  when 
  applied 
  
to 
  projected 
  future 
  years 
  in 
  the 
  climate 
  record. 
   
  

 
  

19 
  

 For 
  projections 
  of 
  temperature 
  and 
  precipitation, 
  we 
  use 
  a 
  time 
  series 
  that 
  
incorporates 
  projected 
  climate 
  model-­‐based 
  trends 
  combined 
  with 
  observed 
  interannual 
  
variability 
  developed 
  by 
  Gutzler 
  and 
  Robbins 
  (2011, 
  hereafter 
  denoted 
  GR2011). 
  The 
  21st 
  
Century 
  trends 
  in 
  temperature 
  and 
  precipitation 
  used 
  for 
  the 
  GR2011 
  projections 
  use 
  the 
  
same 
  set 
  of 
  CMIP3 
  global 
  climate 
  models, 
  forced 
  by 
  the 
  A1B 
  scenario 
  of 
  radiative 
  forcing, 
  
that 
  was 
  used 
  to 
  drive 
  the 
  VIC 
  hydrological 
  model 
  to 
  make 
  the 
  Reclamation 
  projections 
  
described 
  in 
  the 
  previous 
  section. 
   
  Interannual 
  variability 
  is 
  reproduced 
  from 
  exactly 
  100 
  
years 
  earlier 
  in 
  the 
  20th 
  Century. 
  This 
  procedure 
  ensures 
  that 
  the 
  statistics 
  of 
  projected 
  
interannual 
  and 
  decadal 
  variability 
  are 
  "realistic", 
  in 
  the 
  sense 
  that 
  the 
  variability 
  is 
  exactly 
  
reproduced 
  from 
  the 
  observational 
  record. 
  For 
  southwestern 
  New 
  Mexico, 
  this 
  means 
  that 
  
the 
  historic 
  drought 
  of 
  the 
  1950s 
  is 
  reproduced 
  (exactly, 
  in 
  terms 
  of 
  variability 
  about 
  the 
  
projected 
  long-­‐term 
  trend) 
  in 
  the 
  decade 
  of 
  the 
  2050s. 
   
  
Temperature 
  in 
  Jan-­‐Apr 
  increases 
  at 
  a 
  rate 
  of 
  7.6°F/century 
  in 
  the 
  GR2011 
  projection 
  
(Figure 
  11, 
  top 
  panel). 
  The 
  7.6°F/century 
  warming 
  trend 
  brings 
  temperature 
  outside 
  the 
  
envelope 
  of 
  historical 
  20th 
  Century 
  variability 
  by 
  the 
  second 
  half 
  of 
  the 
  21st 
  Century. 
  The 
  
corresponding 
  precipitation 
  projection 
  for 
  NM4 
  in 
  Nov-­‐Apr 
  (bottom 
  panel 
  of 
  Figure 
  11) 
  
exhibits 
  a 
  downward 
  trend 
  of 
  -­‐1.2 
  inches/century. 
   
  As 
  was 
  the 
  case 
  for 
  temperature, 
  20th 
  
Century 
  variability 
  is 
  reproduced 
  by 
  construction 
  for 
  this 
  projection. 
  For 
  example, 
  the 
  
increase 
  in 
  precipitation 
  over 
  the 
  decade 
  of 
  the 
  1970s 
  is 
  reproduced 
  in 
  the 
  decade 
  of 
  the 
  
2070s, 
  with 
  the 
  time 
  series 
  in 
  the 
  2070s 
  shifted 
  downward 
  by 
  approximately 
  1 
  inch 
  due 
  to 
  
the 
  long-­‐term 
  trend 
  derived 
  from 
  the 
  CMIP3 
  projection. 
  In 
  contrast 
  to 
  the 
  temperature 
  
projection, 
  the 
  model-­‐projected 
  trend 
  in 
  precipitation 
  is 
  not 
  large 
  enough 
  in 
  the 
  21st 
  
Century 
  for 
  precipitation 
  values 
  to 
  depart 
  from 
  historical 
  variability. 
  Thus, 
  while 
  the 
  
projected 
  long-­‐term 
  trend 
  of 
  -­‐1.2 
  in/century 
  is 
  a 
  substantial 
  reduction, 
  the 
  overall 
  sense 
  of 
  
21st 
  Century 
  climate 
  illustrated 
  in 
  these 
  projections 
  is 
  a 
  combination 
  of 
  much 
  warmer 
  
temperatures 
  and 
  continued 
  interannual 
  and 
  decadal 
  variability 
  of 
  precipitation, 
  with 
  a 
  
secondary 
  effect 
  associated 
  with 
  the 
  relatively 
  modest 
  downward 
  precipitation 
  trend. 
   
  
Applying 
  the 
  regression 
  model 
  from 
  equation 
  [1] 
  to 
  the 
  GR2011 
  projection 
  of 
  
temperature 
  and 
  precipitation 
  yields 
  a 
  statistical 
  estimate 
  of 
  future 
  flows 
  (for 
  Dec-­‐Jun) 
  as 
  
shown 
  in 
  Figure 
  12. 
  The 
  portion 
  of 
  the 
  time 
  series 
  between 
  1951 
  and 
  2012, 
  shown 
  as 
  a 
  
green 
  line, 
  is 
  the 
  actual 
  Dec-­‐Jun 
  flow 
  reproduced 
  from 
  the 
  middle 
  panel 
  of 
  Figure 
  3. 
  The 
  

 
  

20 
  

 black 
  crosses 
  in 
  the 
  figure 
  represent 
  the 
  streamflow 
  as 
  predicted 
  from 
  the 
  regression 
  
equation. 
  With 
  an 
  r-­‐squared 
  value 
  of 
  0.69, 
  the 
  time 
  series 
  of 
  crosses 
  is 
  highly 
  correlated 
  
with 
  the 
  actual 
  observations 
  but 
  the 
  regression 
  values 
  do 
  not 
  exactly 
  reproduce 
  the 
  
observations. 
   
  
Flows 
  derived 
  from 
  regression 
  model 
  using 
  the 
  GR2011 
  projected 
  values 
  of 
  
temperature 
  and 
  precipitation 
  for 
  2013-­‐2100 
  are 
  continued 
  as 
  the 
  black 
  line 
  in 
  Figure 
  12. 
  
It 
  is 
  important 
  to 
  keep 
  in 
  mind 
  that 
  the 
  interannual 
  variability 
  of 
  the 
  GR2011 
  projection 
  is 
  
synthetic, 
  so 
  individual, 
  specific 
  high 
  flow 
  and 
  low 
  flow 
  years 
  are 
  not 
  meaningful; 
  it 
  is 
  the 
  
statistics 
  of 
  interannual 
  variability 
  that 
  are 
  the 
  projection 
  is 
  designed 
  to 
  capture. 
  The 
  
overall 
  trend 
  of 
  projected 
  flows 
  is 
  seen 
  to 
  decrease 
  in 
  Figure 
  12 
  through 
  the 
  21st 
  Century, 
  
although 
  individual 
  high 
  and 
  low 
  flow 
  years 
  are 
  present 
  throughout 
  the 
  projected 
  time 
  
series. 
   
  
Median 
  values 
  of 
  annual 
  QDecJun 
  streamflow 
  generated 
  by 
  this 
  statistical 
  model 
  
during 
  the 
  1951-­2012 
  observed 
  period 
  over 
  which 
  the 
  model 
  was 
  developed 
  was 
  
65.5 
  kaf, 
  with 
  very 
  large 
  interannual 
  variability 
  superimposed 
  on 
  the 
  mean. 
  Median 
  
projected 
  QDecJun 
  flow 
  for 
  2021-­2050 
  is 
  60.7 
  kaf, 
  a 
  reduction 
  relative 
  to 
  1951-­2012 
  of 
  
7.4%. 
   
  
As 
  discussed 
  above, 
  a 
  common 
  limitation 
  of 
  statistical 
  models 
  applied 
  to 
  future 
  
climate 
  is 
  their 
  applicability 
  when 
  climatic 
  conditions 
  change 
  beyond 
  the 
  envelope 
  of 
  
historical 
  variability 
  in 
  observed 
  data 
  sets. 
  Temperature 
  in 
  the 
  Gila 
  headwaters 
  region 
  is 
  
projected 
  to 
  emerge 
  from 
  its 
  range 
  of 
  historical 
  variability 
  by 
  the 
  middle 
  of 
  the 
  21st 
  
century 
  (Fig. 
  11), 
  making 
  the 
  validity 
  of 
  statistical 
  models 
  to 
  project 
  future 
  streamflows 
  on 
  
the 
  upper 
  Gila 
  problematic. 
  In 
  Fig. 
  12 
  the 
  statistical 
  model 
  projects 
  negative 
  flow 
  at 
  the 
  Gila 
  
gage 
  with 
  increasing 
  frequency 
  after 
  2050, 
  when 
  the 
  temperature 
  term 
  in 
  the 
  regression 
  
model 
  increases 
  in 
  magnitude 
  and 
  overwhelms 
  the 
  positive 
  contributions 
  to 
  QDecJun 
  
represented 
  by 
  the 
  other 
  terms 
  in 
  the 
  model. 
   
  
The 
  projected 
  decrease 
  of 
  7.4% 
  in 
  median 
  QDecJun 
   
  yielded 
  by 
  the 
  statistical 
  model 
  for 
  
the 
  2021-­‐2050 
  period 
  is 
  almost 
  identical 
  to 
  the 
  8% 
  decrease 
  in 
  projected 
  in 
  median 
  annual 
  
flow 
  for 
  that 
  period 
  derived 
  directly 
  from 
  the 
  suite 
  of 
  Reclamation 
  (2011a,b) 
  dynamical 
  
models. 
  While 
  the 
  near-­‐match 
  in 
  these 
  projections 
  is 
  somewhat 
  fortuitous, 
  the 
  similarity 
  in 
  

 
  

21 
  

 results 
  from 
  these 
  two 
  approaches 
  reinforces 
  our 
  confidence 
  that 
  the 
  projection 
  is 
  a 
  
reasonable 
  one. 
  
Based 
  on 
  the 
  similarity 
  of 
  the 
  results 
  derived 
  from 
  dynamical 
  models 
  and 
  
statistical 
  models, 
  our 
  confidence 
  is 
  increased 
  in 
  making 
  an 
  approximate 
  projection 
  
of 
  8% 
  reduction 
  in 
  upper 
  Gila 
  River 
  flow 
  by 
  2021-­2050, 
  relative 
  to 
  the 
  observed 
  
median 
  flow 
  between 
  1951-­2012. 
  
5. 
   
  Conclusions 
  
The 
  principal 
  conclusions 
  of 
  this 
  study 
  are 
  as 
  follows: 
  
 
   
  1) 
   
   Our 
  best 
  estimate 
  of 
  the 
  effect 
  of 
  projected 
  climate 
  change 
  on 
  average 
  peak-­
season 
  flow 
  in 
  the 
  upper 
  Gila 
  River 
  is 
  a 
  reduction 
  of 
  approximately 
  8% 
  by 
  2021-­
2050, 
  relative 
  to 
  a 
  baseline 
  period 
  of 
  1951-­2012. 
   
  
We 
  assessed 
  future 
  flows 
  at 
  the 
  Gila 
  gage 
  using 
  two 
  different 
  techniques: 
  first, 
  an 
  
analysis 
  of 
  a 
  multi-­‐model 
  ensemble 
  of 
  hydrologic 
  model 
  (VIC) 
  simulations 
  driven 
  by 
  an 
  
ensemble 
  of 
  global 
  climate 
  models, 
  and 
  second, 
  an 
  empirical 
  model 
  based 
  on 
  historical 
  
flows 
  at 
  the 
  Gila 
  gage 
  and 
  their 
  correlation 
  with 
  winter-­‐spring 
  precipitationa 
  and 
  spring 
  
season 
  temperature. 
   
  
Each 
  of 
  these 
  projection 
  techniques 
  is 
  subject 
  to 
  considerable 
  uncertainty, 
  associated 
  
with 
  the 
  myriad 
  approximations 
  and 
  uncertainties 
  inherent 
  in 
  dynamical 
  modeling 
  of 
  
extremely 
  complex 
  physical 
  systems 
  on 
  the 
  one 
  hand, 
  and 
  with 
  the 
  assumptions 
  built 
  into 
  
statistical 
  extrapolation 
  of 
  observed 
  climate 
  variability 
  and 
  streamflow 
  on 
  the 
  other 
  hand. 
   
  
Our 
  limited 
  confidence 
  in 
  either 
  approach 
  is 
  bolstered 
  significantly 
  by 
  reaching 
  similar 
  
conclusions 
  using 
  the 
  two 
  approaches. 
  
 
   
  2) 
   
   Pronounced 
  multidecadal 
  natural 
  variability 
  will 
  continue 
  to 
  be 
  superimposed 
  
on 
  the 
  projected 
  decrease 
  in 
  streamflow. 
   
  
By 
  2021-­‐2050 
  the 
  projected 
  8% 
  decline 
  will 
  still 
  be 
  considerably 
  smaller 
  than 
  the 
  
±30% 
  variability 
  in 
  streamflow 
  associated 
  with 
  multi-­‐decadal 
  shifts 
  in 
  storm 
  tracks, 
  that 
  
seems 
  to 
  have 
  been 
  responsible 
  for 
  the 
  pronounced 
  historical 
  alternation 
  between 
  drought 
  
and 
  pluvial 
  episodes 
  for 
  centuries 
  in 
  the 
  Gila 
  River. 
  The 
  late 
  20th 
  Century 
  was 
  very 
  wet 
  by 
  

 
  

22 
  

 historical 
  standards, 
  so 
  one 
  might 
  expect 
  a 
  considerable 
  decline 
  in 
  flows 
  relative 
  to 
  the 
  
unrepresentative 
  1981-­‐2010 
  climatological 
  value 
  to 
  occur 
  in 
  coming 
  decades 
  even 
  in 
  the 
  
absence 
  of 
  forced 
  climate 
  change. 
  The 
  projected 
  long-­‐term, 
  temperature-­‐drived 
  decline 
  in 
  
streamflow 
  is 
  smaller 
  than 
   
  
This 
  study 
  has 
  not 
  attempted 
  to 
  determine 
  whether 
  interannual 
  and 
  decadal 
  
variability 
  should 
  be 
  expected 
  to 
  change 
  as 
  the 
  result 
  of 
  forced 
  climate 
  change 
  in 
  the 
  21st 
  
Century. 
  The 
  projections 
  by 
  Gutzler 
  and 
  Robbins 
  (2011) 
  of 
  temperature 
  and 
  precipitation, 
  
used 
  to 
  generate 
  the 
  statistical 
  projection 
  of 
  2021-­‐2050 
  flow 
  at 
  the 
  Gila 
  gage 
  discussed 
  
here, 
  deliberately 
  and 
  artificially 
  kept 
  the 
  statistics 
  of 
  interannual 
  variability 
  fixed 
  by 
  
superimposing 
  20th 
  Century 
  variability 
  onto 
  model-­‐projected 
  long 
  term 
  forced 
  changes 
  in 
  
climate. 
  Absent 
  better 
  quantitative 
  guidance 
  from 
  dynamical 
  models, 
  we 
  think 
  it 
  is 
  prudent 
  
to 
  assume 
  that 
  historical 
  interannual 
  and 
  decadal 
  variability 
  (such 
  as 
  illustrated 
  in 
  Fig. 
  4) 
  
will 
  continue 
  in 
  the 
  21st 
  Century. 
   
   
  
A 
  corollary 
  note 
  associated 
  with 
  the 
  huge 
  natural 
  variability 
  exhibited 
  in 
  
southwestern 
  river 
  basins 
  is 
  that 
  projected 
  reductions 
  in 
  streamflow 
  are 
  highly 
  sensitive 
  to 
  
the 
  baseline 
  period 
  from 
  which 
  the 
  reduction 
  is 
  calculated. 
  The 
  8% 
  reduction 
  quoted 
  here 
  
is 
  deliberately 
  calculated 
  relative 
  to 
  a 
  long 
  climatological 
  baseline 
  that 
  includes 
  the 
  dry 
  
decades 
  of 
  the 
  middle 
  20th 
  Century. 
  The 
  projected 
  reduction 
  would 
  be 
  much 
  greater 
  if 
  it 
  
were 
  calculated 
  relative 
  to 
  a 
  shorter 
  and 
  wetter 
  baseline 
  period, 
  such 
  as 
  the 
  current 
  official 
  
climate 
  normal 
  period 
  of 
  1981-­‐2010. 
   
  
As 
  noted 
  in 
  conclusion 
  (1), 
  there 
  are 
  multiple 
  uncertainties 
  inherent 
  in 
  climate 
  
projections 
  are 
  very 
  difficult 
  to 
  quantify. 
  In 
  qualitative 
  terms, 
  current 
  observational 
  trends 
  
in 
  temperature, 
  streamflow 
  timing 
  and 
  the 
  declining 
  ratio 
  of 
  winter 
  precipitation/ 
  
streamflow 
  yield, 
  together 
  with 
  the 
  consistency 
  between 
  dynamical 
  and 
  statistical 
  
projections, 
  all 
  support 
  the 
  conclusion 
  that 
  Dec-­‐June 
  streamflow 
  in 
  the 
  upper 
  Gila 
  River 
  is 
  
likely 
  to 
  decline 
  somewhat 
  over 
  the 
  first 
  half 
  of 
  the 
  21st 
  Century 
  (and 
  beyond), 
  with 
  a 
  best 
  
estimate 
  of 
  about 
  8% 
  relative 
  to 
  a 
  baseline 
  climatology 
  of 
  at 
  least 
  a 
  half-­‐century. 
  This 
  level 
  
of 
  decline 
  would 
  still 
  be 
  considerably 
  less 
  than 
  the 
  natural 
  multidecadal 
  variability 
  
expressed 
  in 
  instrumental 
  and 
  long-­‐term 
  proxy 
  records 
  of 
  streamflow. 
   
  Thus 
  for 
  planning 
  
purposes, 
  climate 
  change 
  needs 
  to 
  be 
  considered 
  together 
  with 
  natural 
  variability. 
  From 
  
this 
  perspective 
  the 
  principal 
  effect 
  of 
  climate 
  change 
  on 
  total 
  flow 
  over 
  the 
  first 
  half 
  of 
  the 
  

 
  

23 
  

 21st 
  Century 
  will 
  be 
  to 
  exacerbate 
  low-­‐flow 
  drought 
  conditions 
  driven 
  by 
  episodic 
  
precipitation 
  deficits 
  during 
  drought 
  years. 
   
  
 
   
  3) 
   
   The 
  timing 
  of 
  peak 
  streamflow 
  and 
  the 
  shape 
  of 
  the 
  seasonal 
  hydrograph 
  are 
  
likely 
  to 
  change 
  considerably 
  over 
  the 
  next 
  several 
  decades. 
   
  
Peak 
  season 
  flows 
  associated 
  with 
  snowmelt 
  runoff 
  are 
  projected 
  to 
  be 
  much 
  reduced 
  
as 
  snowpack 
  is 
  diminished 
  in 
  association 
  with 
  warmer 
  temperatures. 
  A 
  much 
  greater 
  
fraction 
  of 
  cold 
  season 
  precipitation 
  will 
  become 
  runoff 
  immediately 
  instead 
  of 
  
accumulating 
  as 
  winter 
  snowpack, 
  then 
  melting 
  in 
  a 
  well-­‐defined 
  snowmelt 
  runoff 
  event. 
  
Therefore, 
  spring 
  season 
  peak 
  flows 
  resulting 
  from 
  snowmelt 
  runoff 
  become 
  much 
  
diminished 
  in 
  projections 
  of 
  future 
  streamflow 
  in 
  the 
  upper 
  Gila 
  basin. 
  
 
   
  4) 
   
   Possible 
  changes 
  in 
  the 
  secondary 
  streamflow 
  peak 
  in 
  summer 
  are 
  difficult 
  to 
  
assess. 
   
  
Summer 
  season 
  precipitation 
  remains 
  challenging 
  for 
  climate 
  models 
  to 
  simulate. 
   
  
Projections 
  of 
  summer 
  streamflow 
  are 
  correspondingly 
  uncertain. 
  The 
  present 
  analysis 
  has 
  
therefore 
  focused 
  on 
  flows 
  associated 
  with 
  snowpack, 
  which 
  comprise 
  the 
  high-­‐flow 
  
portion 
  of 
  the 
  seasonal 
  hydrograph 
  at 
  the 
  Gila 
  gage. 
  Flows 
  were 
  assessed 
  for 
  the 
  extended 
  
"peak 
  flow" 
  season 
  of 
  December-­‐June, 
  but 
  actual 
  changes 
  in 
  the 
  summer 
  monsoon 
  could 
  
affect 
  the 
  quantitative 
  projection 
  of 
  Gila 
  River 
  streamflow 
  in 
  ways 
  that 
  are 
  not 
  accounted 
  
for 
  in 
  this 
  study. 
   
  

 
  

24 
  

 6. 
   
  References 
   
  
Brown 
  R.D., 
  and 
  P.W. 
  Mote, 
  2009: 
  The 
  response 
  of 
  Northern 
  Hemisphere 
  snow 
  cover 
  to 
  a 
  changing 
  
climate. 
  Journal 
  of 
  Climate 
  22, 
  2124–2145. 
   
  
Cook, 
  B.I., 
  and 
  R. 
  Seager, 
  2013: 
  The 
  response 
  of 
  the 
  North 
  American 
  Monsoon 
  to 
  increased 
  
greenhouse 
  gas 
  forcing 
  . 
  Journal 
  of 
  Geophysical 
  Research, 
  in 
  press. 
  
Diffenbaugh, 
  N.S., 
  J.S. 
  Pal, 
  R.J. 
  Trapp 
  and 
  F. 
  Giorgi, 
  2005: 
  Fine-­‐scale 
  processes 
  regulate 
  the 
  response 
  
of 
  extreme 
  events 
  to 
  global 
  climate 
  change. 
  Proceedings 
  of 
  the 
  National 
  Academy 
  of 
  Sciences 
  
102, 
  15774-­‐15778. 
  
Gershunov, 
  A., 
  and 
  T.P. 
  Barnett, 
  1998: 
  Interdecadal 
  modulation 
  of 
  ENSO 
  teleconnections. 
   
  Bulletin 
  of 
  
the 
  American 
  Meteorological 
  Society 
  79, 
  2715–2725. 
  
Gutzler, 
  D.S., 
  2000: 
  Covariability 
  of 
  spring 
  snowpack 
  and 
  summer 
  rainfall 
  across 
  the 
  Southwest 
  
United 
  States. 
  Journal 
  of 
  Climate 
  13, 
  4018–4027. 
   
   
  
Gutzler, 
  D.S., 
  2012: 
  Climate 
  and 
  drought 
  in 
  New 
  Mexico. 
  Chapter 
  4 
  in 
  Water 
  Policy 
  in 
  New 
  Mexico, 
  
Brookshire 
  et 
  al. 
  (eds.), 
  RFF 
  Press, 
  56-­‐70. 
   
  
Gutzler, 
  D.S., 
  and 
  J.W. 
  Preston, 
  1997: 
  Evidence 
  for 
  a 
  relationship 
  between 
  spring 
  snow 
  cover 
  in 
  
North 
  America 
  and 
  summer 
  rainfall 
  in 
  New 
  Mexico. 
  Geophysical 
  Research 
  Letters 
  24, 
  2207-­‐
2010. 
  
Gutzler, 
  D.S., 
  D.M. 
  Kann 
  and 
  C. 
  Thornbrugh, 
  2002: 
   
  Modulation 
  of 
  ENSO-­‐based 
  long-­‐lead 
  outlooks 
  of 
  
southwestern 
  US 
  winter 
  precipitation 
  by 
  the 
  Pacific 
  decadal 
  oscillation. 
  Weather 
  and 
  
Forecasting 
  17, 
  1163–1172. 
   
  
Gutzler, 
  D.S., 
  and 
  T.O. 
  Robbins, 
  2011: 
  Climate 
  variability 
  and 
  projected 
  change 
  in 
  the 
  western 
  
United 
  States: 
  regional 
  downscaling 
  and 
  drought 
  statistics. 
  Climate 
  Dynamics 
  37, 
  835-­‐849. 
   
  
Gutzler, 
  D.S., 
  S.J. 
  Keller 
  and 
  S. 
  Rocha, 
  2012: 
   
  Observed 
  and 
  projected 
  drying 
  trends 
  in 
  southwestern 
  
North 
  America. 
   
  Results 
  presented 
  at 
  the 
  Fall 
  Meeting 
  of 
  the 
  American 
  Geophysical 
  Union, 
  San 
  
Francisco 
  CA. 
   
  
Hamlet, 
  A.F., 
  P.W. 
  Mote, 
  M.P. 
  Clark 
  and 
  D.P. 
  Lettenmaier, 
  2005: 
  Effects 
  of 
  temperature 
  and 
  
precipitation 
  variability 
  on 
  snowpack 
  trends 
  in 
  the 
  western 
  United 
  States. 
  Journal 
  of 
  Climate 
  
18, 
  4545-­‐4561. 
  
Hoerling, 
  M., 
  and 
  A. 
  Kumar, 
  2003: 
  The 
  perfect 
  ocean 
  for 
  drought. 
  Science 
  299, 
  691-­‐694. 
  
Hoerling, 
  M., 
  and 
  J. 
  Eischeid, 
  2007: 
  Past 
  peak 
  water 
  in 
  the 
  Southwest. 
  Southwest 
  Hydrology, 
  Jan/Feb 
  
issue, 
  6:18 
  ff. 
  
Hurd, 
  B.H., 
  and 
  J. 
  Coonrod, 
  2008: 
  Climate 
  change 
  and 
  its 
  implications 
  for 
  New 
  Mexico’s 
  water 
  
resources 
  and 
  economic 
  opportunities. 
  Technical 
  Report 
  45. 
  New 
  Mexico 
  State 
  University 
  
Agricultural 
  Experiment 
  Station, 
  New 
  Mexico 
  State 
  University, 
  Las 
  Cruces, 
  New 
  Mexico, 
  USA. 
  
Hurd, 
  B.H., 
  and 
  J. 
  Coonrod, 
  2012: 
  Hydro-­‐economic 
  consequences 
  of 
  climate 
  change 
  in 
  the 
  upper 
  Rio 
  
Grande. 
  Climate 
  Research 
  53, 
  103-­‐118. 
   
  

 
  

25 
  

 IPCC, 
  2007: 
   
  Climate 
  Change 
  2007: 
  The 
  Physical 
  Science 
  Basis. 
  Working 
  Group 
  I 
  Contribution 
  to 
  the 
  
Fourth 
  Assessment 
  Report 
  of 
  the 
  Intergovernmenal 
  Panel 
  on 
  Climate 
  Change, 
  Cambridge 
  
University 
  Press, 
  996 
  pp. 
  
Karl, 
  T.R., 
  J.M. 
  Melillo 
  and 
  T.C. 
  Peterson 
  (eds), 
  2009: 
  Global 
  Climate 
  Change 
  Impacts 
  in 
  the 
  United 
  
States. 
  Cambridge 
  University 
  Press, 
  New 
  York, 
  New 
  York, 
  USA. 
  
Jones, 
  K.M., 
  2007: 
  Relationship 
  between 
  a 
  700 
  mb 
  "Dry/Wind" 
  Index 
  and 
  springtime 
  precipitation 
  
and 
  streamflow 
  within 
  four 
  snowmelt-­‐dominated 
  basins 
  in 
  northern 
  New 
  Mexico 
  and 
  
southern 
  Colorado. 
  Master 
  of 
  Water 
  Resources 
  Professional 
  Project, 
  University 
  of 
  New 
  
Mexico, 
  48 
  pp. 
  
Liang, 
  X., 
  D.P. 
  Lettenmaier, 
  E.F. 
  Wood 
  and 
  S.J. 
  Burges, 
  1994: 
  A 
  simple 
  hydrologically 
  based 
  model 
  of 
  
land 
  surface 
  water 
  and 
  energy 
  fluxes 
  for 
  General 
  Circulation 
  Models. 
   
  Journal 
  of 
  Geophysical 
  
Research 
  99, 
  14415–14428. 
   
  
Meehl 
  G.A., 
  C. 
  Covey, 
  T. 
  Delworth, 
  M. 
  Latif 
  B. 
  McAvaney, 
  J.F.B. 
  Mitchell 
  and 
  R.J. 
  Stouffer, 
  2007: 
   
   
   
   
   
   
   
   
  
The 
  WCRP 
  CMIP3 
  multi-­‐model 
  dataset: 
  A 
  new 
  era 
  in 
  climate 
  change 
  research. 
  Bulletin 
  of 
  the 
  
American 
  Meteorological 
  Society 
  88, 
  1383-­‐1394. 
   
  
McCabe, 
  G.J., 
  M.A. 
  Palecki 
  and 
  J.L. 
  Betancourt, 
  2004: 
   
  Pacific 
  and 
  Atlantic 
  Ocean 
  influences 
  on 
  
multidecadal 
  drought 
  frequency 
  in 
  the 
  United 
  States. 
  Proceedings 
  of 
  the 
  National 
  Academy 
  of 
  
Science 
  101, 
  4136-­‐4141. 
  
Meko, D.M., C.A. Woodhouse and J.J. Lukas, 2010: TreeFlow: Streamflow reconstructions from tree
rings. http://treeflow.info/ 
  
Milly, 
  P.C.D, 
  J. 
  Betancourt, 
  M. 
  Falkenmark, 
  R.M. 
  Hirsch, 
  Z.W. 
  Kundzewicz, 
  D.P. 
  Lettenmaier 
  and 
  R.J. 
  
Stouffer, 
  2008: 
   
  Stationarity 
  is 
  dead: 
  Whither 
  water 
  management? 
  Science 
  319, 
  573-­‐574. 
   
  
Molles, 
  M.C., 
  Jr., 
  and 
  C.N. 
  Dahm. 
  1990: 
   
  A 
  perspective 
  on 
  El 
  Niño 
  and 
  La 
  Niña: 
  Implications 
  for 
  
stream 
  ecology. 
  Journal 
  of 
  the 
  North 
  American 
  Benthological 
  Society 
  9, 
  68–76. 
   
  
NM 
  OSE, 
  2006: 
   
  The 
  Impact 
  of 
  Climate 
  Change 
  on 
  New 
  Mexico's 
  Water 
  Supply 
  and 
  Ability 
  to 
  Manage 
  
Water 
  Resources. 
  Santa 
  Fe: 
  New 
  Mexico 
  Office 
  of 
  the 
  State 
  Engineer. 
   
  
Reclamation, 
  2011a: 
  SECURE 
  Water 
  Act 
  Section 
  9503(c): 
  Reclamation 
  Climate 
  Change 
  and 
  Water. 
  
U.S. 
  Department 
  of 
  the 
  Interior, 
  Bureau 
  of 
  Reclamation 
  Report 
  to 
  Congress, 
  206 
  pp. 
   
  
Reclamation, 
  2011b: 
   
  West-­‐Wide 
  Climate 
  Risk 
  Assessments: 
  Bias-­‐Corrected 
  and 
  Spatially 
  
Downscaled 
  Surface 
  Water 
  Projections. 
   
  U.S. 
  Department 
  of 
  the 
  Interior, 
  Bureau 
  of 
  
Reclamation, 
  Technical 
  Services 
  Center, 
  Denver 
  CO, 
  122 
  pp. 
   
   
  
Salgado, 
  M., 
  and 
  D.S. 
  Gutzler, 
  2013: 
  Signals 
  of 
  a 
  changing 
  climate 
  in 
  Pecos 
  River 
  streamflow. 
  Results 
  
presented 
  at 
  the 
  New 
  Mexico 
  Geological 
  Society 
  Spring 
  Meeting, 
  Socorro 
  NM. 
   
  
Schubert, 
  S., 
  et 
  al. 
  2009: 
   
  A 
  US 
  CLIVAR 
  project 
  to 
  assess 
  and 
  compare 
  the 
  responses 
  of 
  global 
  climate 
  
models 
  to 
  drought-­‐related 
  SST 
  forcing 
  patterns: 
  Overview 
  and 
  results. 
  Journal 
  of 
  Climate 
  22, 
  
5251-­‐5272. 
  

 
  

26 
  

 Seager, 
  R., 
  Y. 
  Kushnir, 
  C. 
  Herweijer, 
  N. 
  Naik 
  and 
  J. 
  Velez, 
  2005: 
  Modeling 
  of 
  tropical 
  forcing 
  of 
  
persistent 
  droughts 
  and 
  pluvials 
  over 
  western 
  North 
  America: 
  1856-­‐2000. 
  Journal 
  of 
  Climate 
  
18, 
  4065-­‐4088. 
   
  
Seager, 
  R., 
  et 
  al., 
  2008: 
  Model 
  projections 
  of 
  an 
  imminent 
  transition 
  to 
  a 
  more 
  arid 
  climate 
  in 
  
southwestern 
  North 
  America. 
  Science 
  316, 
  1181–1184. 
   
  
Sheffield, 
  J., 
  E.F. 
  Wood 
  and 
  M. 
  Roderick, 
  2012: 
   
  Little 
  change 
  in 
  global 
  drought 
  over 
  the 
  past 
  60 
  
years. 
  Nature 
  491, 
  435-­‐438. 
  
Stewart, 
  T., 
  D.R. 
  Cayan 
  and 
  M.D. 
  Dettinger, 
  2005: 
  Changes 
  toward 
  earlier 
  streamflow 
  timing 
  across 
  
western 
  North 
  America. 
  Journal 
  of 
  Climate 
  18, 
  1136–1155. 
  
Taylor 
  K.E., 
  R.J. 
  Stouffer 
  and 
  G.A. 
  Meehl, 
  2012: 
  An 
  overview 
  of 
  CMIP5 
  and 
  the 
  experiment 
  design. 
  
Bulletin 
  of 
  the 
  American 
  Meteorological 
  Society 
  93, 
  485–498. 
   
  
Thomson, 
  B.M., 
  2012: 
  Water 
  resources 
  in 
  New 
  Mexico. 
  Chapter 
  4 
  in 
  Water 
  Policy 
  in 
  New 
  Mexico, 
  
Brookshire 
  et 
  al. 
  (eds.), 
  RFF 
  Press, 
  25-­‐55. 
   
  
Wood, 
  A.W., 
  L.R. 
  Leung, 
  V. 
  Sridhar, 
  and 
  D.P. 
  Lettenmaier, 
  2004: 
  Hydrologic 
  implications 
  of 
  
dynamical 
  and 
  statistical 
  approaches 
  to 
  downscaling 
  climate 
  model 
  outputs. 
  Climatic 
  Change 
  
15, 
  189–216. 
   
  
Woodhouse, 
  C.A., 
  D.W. 
  Stahle 
  and 
  J. 
  Villanueva-­‐Diaz, 
  2012: 
  Rio 
  Grande 
  and 
  Rio 
  Conchos 
  water 
  
supply 
  variability 
  from 
  instrumental 
  and 
  paleoclimatic 
  records. 
  Climate 
  Research 
  51, 
  125-­‐
136. 
   
  

 
  

27 
  

 Fig 
  1 
  

Figure 1. Map of the upper Gila basin, adapted with permission from "The Gila and San
Francisco Rivers" by Jerold Widdison for the Utton Transboundary Resources Center. Red
square indicates the location of the Gila gage (USGS gage 09430500), which is the
principal analysis point for this study. The green square indicates the location of the
Solomon gage (USGS gage 09448500) for which tree ring-based reconstructed flows are
illustrated in Figure 4.

 Fig 
  2 
  

O 
  

N 
   
    
   
  D 
   
   J 
  

F 
    
   
  M 
  

A 
  

M 
  

 
  J 
  

 
  J 
  

A 
  

S 
  

Figure 2. Hydrographs for streamflow at the Gila gage on the Gila River. The thick black line
shows the median daily flow for each day of the Water Year (e.g. day 1 = Oct 1, day 365 =
Sep 30) for the 82-year period Oct 1928 – Sep 2012 (WY 1929-2012). The thin green line
shows the median flow based on just the first half of the period of record (WY 1929- 1970);
the thin red line shows the median daily flow based on just the second half of the record (WY
1971-2012). Months of the year are indicated between tick marks on the x-axis.

 Fig 
  3 
  

Figure 3. Annual time series of streamflow at the Gila gage, 1929-2012. Thin line in each
panel represents annual data; thick line is an 11-year running average.
Top: Annual flow (kaf) for the entire Water Year (October-September).
Middle: 7-month average streamflow for December-June, the high-flow portion of the year.
Bottom: Monsoon season (July-September) flow (note much-expanded y-axis).

 Fig 
  4 
  

Figure 4. Time series of observed and reconstructed flow on the Gila River at the Solomon
gage (USGS gage 09448500, shown as the green square in Figure 1). The black line in each
panel shows annual flow derived from gage data, with a decadal smoother applied to
emphasize long-period fluctuations. The blue line in each panel shows the reconstruction based
on tree ring analysis using dendrochronology sites that demonstrate a strong correlation with
Gila@Solomon flows. The calculated regression between tree ring records and Solomon gage
flows during the instrumental data record is extended backward in time to generate the flow
reconstruction. Data and graphics adapted from the treeflow.info website (Meko et al. 2010).

 Fig 
  5 
  

Figure 5. Center of timing of December-June streamflow at the Gila gage. The Center of
Timing is the day at which half of the total flow between December and June each year has
passed the gage. Days are measured relative to the start of each Water Year on October 1.
The days corresponding to 1 February, 1 March and 1 April are labelled on the right-hand
axis.

 Fig 
  6 
  

Figure 6. Time series of seasonal temperature (top) and precipitation (bottom) for New
Mexico Climate Division 4, which includes the headwaters of the Gila River.
Top: temperature averaged over 4 months (Jan-Apr) from 1931-2012. Thin red line shows
annual data; thick red line shows an 11-year running average through the annual data.
Bottom: precipitation averaged over 6 months (Nov-Apr each Water Year), with thin and
thick lines similar to temperature panel.

 Fig 
  7 
  

Figure 7. Scatter plots of observed temperature (top) and precipitation (bottom) in New Mexico
Climate Division 4, plotted against observed peak season streamflow (Dec-Jun) at the Gila gage.
Top: Temperature averaged over 4 months (Jan-Apr) vs. Dec-Jun flow. The line shows a best-fit
linear regression. Correlation between the data and the regression line is -0.20.
Bottom: Precipitation averaged over 6 months (Nov-Apr each Water Year) vs. Dec-Jun flow.
The line shows a best-fit quadratic regression. Correlation between the data and the regression
curve is 0.83. See text for descriptions of these regressions, and the multvariate regression model
that incorporates both temperature and precipitation.

 Fig 
  8 
  

Figure 8. Seasonal hydrographs for successive 30 year periods at the Gila gage.
Top: Observed monthly mean flows at the Gila gage. Dashed black line shows 1951-1980
mean; solid blue line shows 1981-2010 mean. Bottom: Streamflow values derived from the
mean of 39 climate model driven simulations (Reclamation, 2011b). Black and blue lines
correspond to the same periods as the observations in the top panel; green and red dashed
lines represent 2021-2050 and 2070-2099 averages based on projected climate conditions.

 Fig 
  9 
  

Figure 9. Time series of annual flow at the Gila gage in 39 simulations by the VIC land
surface/runoff model (Reclamation, 2011). Each VIC simulation is driven by a different
global climate model simulation. The radiative forcing for each of the 39 simulations is the
A1B scenario as defined by the IPCC (2002). Dashed blue lines represent each simulation.
For clarity, the 39 simulations are split into two groups: 19 simulations are shown in panel
(a) and 20 simulations are shown in panel (b). The solid black line in each panel is the
annual median value of all 39 simulations for each Water Year.

 Fig 
  10 
  

Δ Median Precipitation [mm]

Δ Median Snow Water [mm]

NARCCAP Gila Basin:
21st Century change

0

0

O N

D J

F M A M J J

A S

Figure 10. Change in median snow water (top panel) and precipitation (bottom panel), for
each day of the water year, between a simulation of current climate (1968-2000) and a
simulation using the same climate model for future climate (2038-2070). Calculations were
made using output from the NARCCAP ensemble of downscaled climate models, including
just the model gridpoints covering the upper Gila basin upstream of the location of the Gila
gage.

 Fig 
  11 
  

Figure 11. Projections of 21st Century temperature (top) and precipitation (bottom) in New
Mexico Climate Division 4, generated by Gutzler and Robbins (2011). The initial segments
of each time series, from 1951-2012, are observed data as shown in Figure 6. Projected
monthly temperature and precipitation are derived by reproducing interannual variability
from 20th Century observations in the 21st Century, then adding the linear trend for the 21st
Century derived from the average of an ensemble of global climate models forced by the
A1B radiative forcing scenario. Thick black lines in each panel indicate 11-year running
averages.

 Fig 
  12 
  

Figure 12. Time series of Dec-Jun flow at the Gila gage. The green line shows the observed
time series from 1951-2012, reproduced from the middle panel of Figure 3. Black x's show the
results of a regression model of Dec-Jun flow based on Nov-Apr precipitation and Dec-Apr
temperature, as illustrated in the scatter plots presented in Figure 7. This model accounts for
69% of the observed variability of Dec-Jun streamflow at the Gila gage. The black line shows
the continuation of the streamflow record into the future, using the same regression model for
streamflow applied to projected NM4 precipitation and temperature as derived by Gutzler and
Robbins (2011).
Note that in extreme dry or hot years the regression model can project unphysical negative
flows. Negative flows are generated by the model with increasing frequency after 2050, as
temperatures in NM4 warm up beyond the envelope of historical variability.