See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228952277

Scenarios of Future Climate for the Pacific Northwest

Article · April 2008

CITATIONS READS
105 188

4 authors, including:

Philip Mote Eric P Salathé

Oregon State University University of Washington Bothell
137 PUBLICATIONS 24,034 CITATIONS 81 PUBLICATIONS 3,903 CITATIONS

SEE PROFILE SEE PROFILE

Valérie Dulière
Royal Belgian Institute of Natural Sciences
24 PUBLICATIONS 1,041 CITATIONS

SEE PROFILE

All content following this page was uploaded by Eric P Salathé on 02 June 2014.

The user has requested enhancement of the downloaded file.

Scenarios of future climate
for the Pacific Northwest

a report by

Philip Mote, Eric Salathé, Valérie Dulière, and Emily Jump

Climate Impacts Group, University of Washington

March 2008
Summary.

The average warming rate in the Pacific Northwest during the next ~50 yr is expected to be in the range 0.1-0.6°C
(0.2-1.0°F) per decade, with a best estimate of 0.3°C (0.5°F) per decade. For comparison, observed warming in the
second half of the 20th century was approximately 0.2°C per decade. Trends in temperature already stand out above
natural variability. Projected warming is greater in summer than in other seasons.

Present-day patterns of greenhouse gas emissions con-

strain the rate of change of temperature for the next few
decades: humans are committed to some degree of addi-
tional climate change. Beyond mid-century, the projections
of warming depend increasingly on emissions in the next
few decades and hence on actions that would limit or in-
crease emissions.

Projected precipitation changes are modest, and are un-

likely to be distinguishable from natural variability until
late in the 21st century. Most models have winter precipi-
tation increasing and summer precipitation decreasing.
Early results suggest an increase in intense precipitation
but little change in coastal upwelling.

2020s* temperature precipitation

low 0.6°C (1.1°F) -9%

average* 1.2°C (2.2°F) +1%

high 1.9°C (3.4°F) +12%

2040s* temperature precipitation

low 0.9°C (1.6°F) -11%

average* 2.0°C (3.5°F) +2%

high 2.9°C (5.2°F) +12%

2080s* temperature precipitation

low 1.6°C (2.8°F) -10%

average* 3.3°C (5.9°F) +4%

high 5.4°C (9.7°F) +20%

1* In this document, “2020s” means the 2010-2039 average minus the 1970-1999 average, similarly for 2040s and
2080s. Model values are weighted to produce the “average”.

1. Global climate models
More than 20 research centers around the world
have developed and used very sophisticated models,
which simulate global climate. A common set of simu-
lations using these models was coordinated through the
Intergovernmental Panel on Climate Change (IPCC),
described in the IPCC 2007 report (Randall et al.
2007), and archived by the Program for Climate Model
Diagnostics and Intercomparisons (PCMDI). These
models typically resolve the atmosphere with between
6,000 and 15,000 grid squares horizontally, and with
between 12 and 56 atmospheric layers. All models in
the PCMDI archive include a fully resolved global ocean
model, usually with higher resolution than the atmos-
pheric model, and nearly all include models of sea ice
dynamics and models of the land surface. By calculating
energy fluxes between the sun, atmosphere, and sur-
face, these models compute surface temperature distri-
butions that compare well with observations. Details of
the models, as well as references, can be found in Ran-
dall et al. 2007 (Table 8.1, pp 597-599).
In order to simulate 21st century climate, global
climate models (GCMs) are run with different scenarios
produced by the IPCC of socioeconomic change. The
IPCC’s six “illustrative scenarios (of which four are
shown in Figure 1) lead to different scenarios of green-
house gas and sulfate aerosol emissions and hence of
climate “forcing” or energy input (watts per square
meter, the y-axis in Figure 1). Sulfate aerosols promote
cloud formation in certain regions and partly offset
greenhouse warming. Three emissions scenarios were
most frequently chosen by global modeling groups in
their simulations of future climate: A2, A1B, and B1.
The climate forcing of all scenarios is similar until mid-
century owing primarily to the long lifetime of coal-
fired electric power plants and of the major greenhouse
gases. As Figure 1 clearly shows, A2 produces the high-
est climate forcing by the end of the century. Before
mid-century, however, none of the scenarios is consis-
tently the highest. Because more modeling groups ran
A1B than A2, and since our focus for this study was on
mid-century change, we chose A1B as the high emis-
sions scenario and B1 as the low for our analysis of 21st
century PNW climate. We have analyzed available A2
runs as well, as shown in Figure 8, but we emphasize
A1B and B1.
We note the possibility that emissions may fall
below or above the range presented here. B1 is the low-
est of the IPCC illustrative scenarios, but still produces
changes in climate that many scientists call “dangerous”
— a threshold that a growing number of political lead-
ers have stated their intention to avoid. At the high
end, scenario A1FI (not shown) results in even higher
climate forcing by 2100 than A2 or A1B. Recent global
emissions of CO2 have been exceeding even the A1FI
scenario. Whether these exceedingly high emissions
will continue into the future is beyond our expertise to
judge.
2 URL esg.llnl.gov
Page 1
IPCC scenarios, radiative forcing
10
8 A1B
A2
B1
B2
watts per square meter
IS92a
6
4
2
0
1980 2000 2020 2040 2060 2080 2100
Figure 1. Globally averaged radiative forcing by greenhouse
gases and sulfate aerosols, for four of the six illustrative sce-
narios plus the older IS92a scenario, from IPCC (2001) Ap-
pendix II.3.
From the PCMDI web site 2, all modeling centers
provided simulations of 20th century climate using
observed solar, volcanic, and greenhouse gas forcing.
Twenty modeling centers provided simulations of 21st
century climate with the A1B scenario, 19 with B1, and
17 with A2, for a total of 56 runs. In some cases several
different model runs were provided for each scenario;
we chose Run 1 except as noted in Appendix A. Model
output was obtained from PCMDI as monthly values
(and daily too for two models), and analyzed at the
University of Washington by the authors of this report.
Randall et al. (2007) evaluated the models’ fidelity
in simulating various aspects of global climate, and also
calculated each model’s climate sensitivity. The mod-
eled climate sensitivity is a measure of the model’s
response to doubled carbon dioxide (CO2), and has
historically been calculated in two ways: either the
“equilibrium climate sensitivity” or the “transient cli-
mate response” (TCR). The equilibrium climate sensi-
tivity is defined as the globally averaged temperature
change in a simulation with a doubling of CO2, in which
the simulation is long enough for the global tempera-
ture to reach equilibrium. Because the climate system
takes a long time to come into equilibrium, the calcula-
tion of the equilibrium climate sensitivity was typically
performed only in models with a very simple ocean
component, which was standard before the mid-1990s.
By the late 1990s, most models included a sophisticated
ocean, and running such a model to equilibrium would
require a great deal of computer resources. The TCR
was a more practical metric of models’ sensitivity. The
TCR is defined as the global mean temperature change
at the time of CO2 doubling in a simulation in which the
CO2 increased at 1%/year (roughly IS92a, the black
curve in Figure 1). The range of values of TCR reported
by Randall et al. (2007) was 1.2-2.6°C (our Table 1,
their Table 8.2).
Page 2

ent resolutions, but the number of model grid points

nationality model equilibrium TCR enclosed in this latitude-longitude box is between 6 and
sensitivity 91. We simply average the temperature and precipita-
tion values at all the Northwest grid points to define a
BCCR n.a. n.a. regionally averaged time series. The reason for such
averaging is that variations in model climate on scales
CCSM3 2.7 1.5 smaller than a few grid cells is small and not meaning-
ful. Put another way, the models represent the varia-
CGCM (T47) 3.4 1.9 tions of climate that would occur on a smooth planet
with similar land-sea distributions and large smooth
CGCM (T63) 3.4 n.a. bumps where Earth has major mountain ranges.
Besides model resolution, another consideration in
CNRM n.a. 1.6 comparing global models with observations is that
there are different ways to calculate “observed” region-
CSIRO 3.1 1.4 ally averaged temperature and precipitation. A com-
mon approach is to average weather station data into
“climate divisions” and combine the climate divisions
ECHAM5 3.4 2.2
into a state or regional average with area weighting.
The drawback of this approach is that it does not ac-
ECHO-G 3.2 1.7 count for the contribution to a regional average of high
terrain, which has very few weather stations. A better
FGOALS 2.3 1.2 estimate interpolates (horizontally) and extrapolates
(vertically) observations to a uniform, high-resolution
GFDL-CM2.0 2.9 1.6 grid (e.g., Hamlet and Lettenmaier 2005). Such an
estimate, however, would be unsuitable for comparing
GFDL-CM2.1 3.4 1.5 with climate model output, which lacks the vertical
relief.
GISS-AOM n.a. n.a. A third approach is to assimilate observed data into
a weather prediction model at the spatial resolution of
GISS-ER 2.7 1.5 climate models, as has been done for the NCEP/NCAR
reanalysis. Because this approach processes observa-
HADCM3 3.3 2.0 tions is most similar to a global climate model, it is the
one we define as “observations” for evaluating the
HADGEM1 4.4 1.9 model output.
We begin with a comparison of the means from
INMCM 2.1 1.6 models and NCEP for 1970-99, a quantity referred to as
“bias.” These model means will also be the reference
for discussing climate changes in section 3, thereby
IPSL 4.4 2.1
removing most of the effects of model errors. Most
models are slightly colder than NCEP — in the parlance
MIROC 4.0 2.1 of climate modelers, they have a “cold bias” — but the
mean bias is only -0.38°C (0.68°F) and the median bias
MIROC-hires 4.3 2.6 is -0.87°C (Figure 2). The models with least bias in
annual average temperature are MIROC, IPSL, CCSM3,
PCM 2.1 1.3 GISS-AOM, and HadGEM1. For precipitation (Figure
3), almost all models have a wet bias, and for some the
bias exceeds 50%. The mean bias is 16 cm (6.3”, 19%).
Table 1. Equilibrium sensitivity and TCR (°C) as reported by Models with lowest bias are HadCM, GISS_er, BCCR,
Randall et al. (2007) for models used in this study. PCM1, and CNRM. Note that no model falls in the best
five for both temperature and precipitation, whereas
CSIRO, GFDL2.0, and MIROC-hi fall in the bottom five
for both temperature and precipitation.
2. Model evaluation: 20th century climate The models’ simulated seasonal cycle for the PNW
of the Northwest is shown in Figure 3. For temperature, the multi-model
average is usually within 1°C of observed for each
month, and has a lower root-mean-square (rms) differ-
In any prediction exercise the first question should ence from observed than any model. The annual mean
be, how well can the predictive model simulate the multi-model mean bias of -0.38°C arises from cool
past? In this section we examine the 20 models’ simu- biases in January-March and July-September, with
lation of 20th century climate in the Pacific Northwest. negligible biases in remaining months. The range of
For most of this study the Pacific Northwest is de- simulated temperatures (warmest minus coldest) fol-
fined as the region between 124° and 111° west longi- lows a similar pattern, being generally less than 7°C for
tude, 41.5° to 49.5° north latitude: Washington, Ore- April-June and October-December.
gon, Idaho, western Montana, and a small slice of adja-
cent states and British Columbia. Models have differ-
 Page 3
Model Bias for Annual Temperature (°C)
4

Temperature in °C
3
2
1
0
-1
-2
-3
-4

ge
S_ 1
FG R

CM

CM

IR er
S

RM

5
3. .0

5
.1
3

_g

i
7

IR PSL

CS _h
Model Bias for Annual Precipitation (cm)

EM
CM CM

SM

3.

3.
AM
t6

t4
C

AL

AO
CM L2

L2

S_

ra
O
BC

AD

M
CN

C
C_

O
1_

1_

I
O

CC
P

H
CG FD

FD

e
O
IS
IN

IR

av
EC

AD
EC
3.

O
H

G
G

IS

M
H

M
CG
70
GCMs
60
Precipitation in cm

50

40

30

20

10

-10

e
1
CM

1
CR

IR .2
er

CM

5
RM

M
5

FG hi
_g

.1

.0
7

SL

ag
EM
M

SM

3.
AM
t4

t6

AL

AO
C_
3
L2

L2
S_

IP
O
BC

PC
AD

M
CN

C_
O
1_

1_

er
O
H

G
CC
H

FD

FD

S_
IS

IN

IR

av
EC

AD
EC
3.

3.

O
H
G

CS
G

IS
IR
CM

CM

M
H

G
M
CG

CG
GCMs

Figure 2. Biases, calculated relative to NCEP reanalysis, in each model’s mean annual (top) temperature and
(b) precipitation for the Pacific Northwest, for 1970-99.

For precipitation, all models reproduce the Mean temperature

30
contrast between wet winters and dry summers, bccr
though a few produce summers that are only ccsm3
cgcm3.1_t47
slightly drier than winters. The multi-model aver- cgcm3.1_t63
cnrm_cm3
age is 20-30% wetter than NCEP in most months, 20 csiro_3_5
echam5
with the largest bias (70%) in October and the echo_g
fgoals1_0_g
smallest (<15%) in March-June. Unlike tempera- gfdl_cm2_0
ture, for which the multi-model average outper- multi-model avg
C

10 NCEP
o

formed all models, nine of the models have a lower

rms difference from observed than the multi-model
average. HadCM has the best combined ranking 0
followed by CNRM and ECHAM5.
Another facet of 20th century climate that can
be evaluated is the trend in temperature. For the -10
global average, many models simulate a warming Jan Mar May Jul Sep Nov Jan
rate similar to the 0.6°C warming observed in the
20th century. At the regional scale (Figure 4), the Mean precipitation
warming rate could be dominated by changes in 25
atmospheric circulation rather than greenhouse gfdl_cm2_1
giss_aom
forcing; nonetheless, eight of the models simulate a giss_er
20 hadcm
warming for the Northwest within 0.2°C of the ob- hadgem1
inmcm3_0
served warming of 0.8°C during the 20th century. ipsl_cm4
We do not perform the same comparison for pre- 15
miroc3_2_hi
miroc_3.2
cipitation since there is no evidence for a 20th pcm1
multi-model avg
cm

century response of global precipitation to green- NCEP

house forcing. The time series of regional precipi- 10
tation is characterized by high interannual vari-
ability, and the direction of linear trends depend on
5
the start and end point, unlike temperature, for
which linear trends are robustly upward.
Finally, we examine aspects of 20th century 0
climate that pertain to the mesoscale modeling Jan Mar May Jul Sep Nov Jan
that will be reported elsewhere. Since the GCMs
Figure 3. Mean seasonal cycle for each climate model from its 20th cen-
provide the global context for the regional mod-
tury simulation, compared with the NCEP/NCAR reanalysis (black). All 20
eling, the GCM fields over the domain of the
models are shown in both panels but the legend is split between the panels.
mesoscale model help determine the quality of the
Page 4

trend (C/century)

1.2 1
1.1 1.1 .1
1. 1.0
1 0.9 0
0
0.8 0.8 .8
0.8
0.6 0.7 0.7
0.6 0.5
0.4 0.4
0.4 0.3

0.2 0. 0.1
0.0 0.0 1
0 H .2

IR SL
O M

DL 1

3. 2.0
s .5
CM DL d

CM M 3
GI B M

SS 1

OC 7
AD S

G g

CS IP 1

RM
H H 5

M 3.1_ M

CN _hi
CC t63
M _ R

P er
GF CM
GI EM

2.

IR t4
AD O_
EC AM

CG IN SM
H AL

CG GF erve
SS CC

-0.2

ob O3
EC C_3
IR AO

_
C

C
1_
O
FG

-0.4
-0.
4
-0.6
Figure 4. Trend in annual mean temperature for the PNW during the 20th century, compared
with observations (black) calculated from the USHCN data.

mesoscale model simulation; in particular the moisture
flux into the region provided by the GCM plays a crucial
role in determining both the amount and the distribu-
tion of precipitation by the regional model.
For each model, we mapped the precipitation, sea
level pressure (SLP), and temperature over roughly the
domain for which we ran the mesoscale model (results
of which will be reported elsewhere). Figure 5 shows
the maps for one of these models, the CGCM_T47,
compared with the NCEP/NCAR reanalysis. In both
instances we show the annual mean for 1970-99. All
models reproduce the basic features of each field: the
heavy precipitation over the coastal mountains of Brit-
ish Columbia, the swath of high precipitation in the
lower left corner, the Aleutian low and Pacific high
pressure features in the middle panel, and the low tem-
peratures over the mountainous West and the strong
gradient of sea surface temperature over the western
Pacific.
An efficient method of quantitatively comparing
fields is the Taylor diagram (Taylor, 2000). Values are
plotted in radial coordinates with the radius being the
ratio of the modeled area-mean standard deviation to
the observed area-averaged standard deviation, and the
angle represents the spatial correlation. Figure 6 shows
the Taylor diagram for all 20 models. As with global
mean fields (Randall et al. 2007), of the three fields
shown here temperature is best simulated by the mod-
els, with a correlation typically >0.97. Sea level pres-
sure is next best simulated, followed by precipitation,
except that for GISS-ER the SLP is worse than any
model’s precipitation field, owing largely to an Aleutian
Low that is much too far to the west. In the Taylor
diagram, the distance of a point to (1,1) represents the
rms error, and we can use this distance to rank the
models for each field and to average the distances to
rank the models overall (Fig. 6 lower). Of all the mod-
els, CGCM-T47 (shown in Fig. 5) ranks the best.
3. Projected changes in temperature and
precipitation
In addition to presenting the range of model output
and the all-model average, we also present a new,
weighted average, following the reliability ensemble
averaging “REA” (Giorgi and Mearns, 2002) approach.
In this approach, the REA value for each season and
decade is calculated by weighting each model’s output
by its bias and distance from the all-model average.
Multi-model averages in weather forecasting, seasonal
forecasting, and climate simulations often come closer
to observations than single models (see Figure 3a
above), and REA should produce better results for the
future than an unweighted average by giving more
weight to models that perform well in simulating 20th
century climate. For details on the REA calculation, see
Appendix B. In this document, “2020s” denotes the
2010-2039 average, 1980s denotes the 1970-1999 aver-
age, and likewise for 2040s and 2080s.
3a. 21st century trends in the annual mean
The regionally averaged temperature and precipita-
tion for all B1 and A1B simulations are shown in Figure
7, along with the REA value for each year. Each mod-
el’s projected temperature is smoothed by regressing
temperature on the logarithm of the atmospheric con-
centration of CO2, an approximation of global radiative
forcing (see Figure 1). The same is done for precipita-
 Page 5

NCEP SLP CGCM3.1 SLP

10 1016
16 1016 1016 1012

1012
1012
1008

1004
1008

10
10

16
1012

16

16
10
0 1016
102 1020

NCEP Temperature CGCM3.1 Temperature

260
266
260
272

26 26 2
6 6 27
278 272
272
278 278 278
284 284 284
290
290 290
290
296 296 296

NCEP Precipitation CGCM3.1 Precipitation

4

4
8 12 4
16

12
8

8 8 4
4

12
8 8
16
12
8
4
4

8
8

Figure 5. Annual mean patterns from (left column) CGCM-T47 and (right column) NCEP-NCAR reanaly-
sis, for years 1950-99. Top row shows precipitation in mm/day, middle row sea level pressure in hPa, and
bottom row temperature in Kelvin (273.16K=0°C=32°F).

tion. This approach highlights the region’s response to
the forcing on century timescales, masking model in-
terdecadal variability which, while interesting, can
confound the detection of forced change, especially for
precipitation. Note how different the evolution of tem-
perature is after about 2050 for the two scenarios, ow-
ing to the markedly different radiative forcing produced
by different concentrations of greenhouse gases. By
2100 the REA value of temperature change is almost
4°C for A1B and only 2.5°C for B1. The range just for
these two scenarios is 1.5 to 5.8°C; other IPCC emis-
sions scenarios would give higher (but not lower)
warming by 2100.
The observed trend in regional mean temperature is
statistically significant, that is, it exceeds what would be
expected from a time series with no trend and the same
amount of interannual variability. Likewise, the pro-
jected future trends, even for the very lowest of the
scenarios, is substantially greater than observed in the
20th century. On the other hand, for precipitation,
observed and projected trends are not significant except
for a few models.
Another way to view the scenarios is to plot the
change in temperature on one axis and the change in
precipitation on another axis (Figure 8). Figure 8
roughly shows the sensitivity of the models to forcing,
with different magnitudes of forcing applied by the
three SRES scenarios and in different quantities for the
three decades. The ranking of models is similar for
each decade and SRES scenario: HadGEM1, MI-
ROC3_2_hi, or CCSM3 tend to be the warmest in each
scenario and each decade, IPSL_CM4 or BCCR the
wettest, and so on. Unlike the situation in the global
mean, where the precipitation change and temperature
change of models tend to be correlated, there seems to
be no correspondence between temperature change and
precipitation change in the Northwest. Differences
among the scenarios are small in the 2020s but are
substantial by the 2040s. In the coolest scenario, re-
gional temperature rises 0.6°C (1.1°F) by the 2020s,
 Page 6

correlation
0.10 North Pacific Taylor diagram
1.2 0.20 B1
0.30 6
0.40
0.50
1.0 4
0.60
SLP 0.70

C
2

o
0.8 Precip
0.80
Temp 0

0.6
0.90 -2
1900 1950 2000 2050 2100

0.4 0.95 A1B

6

0.2 0.99 4

observed

C
2
0.0 1.00

o
0.0 0.2 0.4 0.6 0.8 1.0 giss_er 1.2
Ratio of variance 0

0.6
fgoals1_0_g
ipsl_cm4

-2
1900 1950 2000 2050 2100
inmcm3_0

ccsm3
hadcm
giss_aom

B1
pcm1
csiro_3_5
hadgem1

30
gfdl_cm2_0
miroc_3.2
miroc3_2_hi

cnrm_cm3
gfdl_cm2_1
cgcm3.1_t63

bccr
cgcm_3.1_t47

0.4 20
echam5
echo_g

10
%

0.2
-10

-20
1900 1950 2000 2050 2100

A1B
0.0 30
0 5 10 15 20
20
Figure 6. Evaluation of model performance over the
domain shown in Figure 5. Top panel shows the 10
%

correlation (angle) and ratio of variance (radius) for

0
each model and each field. The rms difference from
the observed field is just the distance on the dia-
-10
gram. Bottom panel ranks the models by mean
distance for the three fields. -20
1900 1950 2000 2050 2100

0.9°C (1.5°F) by the 2040s, and 1.6°C (2.8°F) by the
2080s. In the warmest scenario, annually averaged
warming is roughly a factor of three higher than the
lowest scenario: 1.9°C (3.3°F) by the 2020s, 2.9°C
(5.2°F) by the 2040s, and 5.4°C (9.7°F) by the 2080s.
3b. Seasonality of changes in climate
For some applications the changes of climate in a
given season may be more important than the changes
in annual mean. In this section we present the changes
in climate by season. Figures 9 and 10 show changes in
temperature and precipitation for the 2020s, 2040s,
and 2080s relative to the 1980s. For both B1 and A1B,
Figure 7. Smoothed traces in temperature (top two panels)
and precipitation (bottom two panels) for the 20th and 21st
century model simulations, relative to the 1970-99 mean. The
heavy smooth curve for each scenario is the REA value, calcu-
lated for each year and then smoothed using loess. The top
and bottom bounds of the shaded area are the 5th and 95th
percentiles of the annual values from the ~20 simulations,
smoothed in the same manner as the mean value.
warming is expected to be largest in summer. In most
seasons B1 has the lowest projected change and A1B the
highest, but this is not always true in the 2020s when
 Page 7

2020s
4 the radiative forcing of the two scenarios is very similar.
Model results for changes in precipitation are
B1 2 equivocal. The annual mean REA change is practically
A1B
ccsm3 miroc3_2_hi zero, though individual models produce changes of as
A2 ccsm3 much as -10% or +20% by the 2080s. On the seasonal
3
ipsl_cm4 scale the most consistent changes appear in the sum-
mertime, with a large majority of models (especially in
Temp change, F

the A1B scenario) projecting decreases and the REA

Degrees C
echo_g
echo_g value reaching -16% by the 2080s. Some models foresee
2 echo_g reductions of as much as 20-40% in summer precipita-
1 tion, though these large percentages only translate to 3-
6 cm over the season. While small hydrologically,
bccr summer precipitation in the Northwest nonetheless has
bccr
1 pcm1 bccr an impact on evaporative demand and hence, for exam-
pcm1 ple, on urban water use and forest fires.
In winter, by contrast, a majority of models project
increases in precipitation. The REA value reaches +9%
(about 3cm) by the 2080s for the A1B scenario, still
0 0
small relative to interannual variability. And although
-10 0 10 20
Precip change, % some of the models foresee modest reductions in fall or
winter precipitation, some foresee very large increases
2040s (30% or more). Changes of this magnitude would in-
6 crease snow accumulation at higher elevations but
B1 would also substantially raise the risk of flooding, if the
hadgem1 3 daily extremes increase along with the seasonal means.
A1B miroc3_2_hi
5 For some applications one may want to choose a
A2 miroc3_2_hi
few GCM scenarios to represent a “medium” (closest to
ipsl_cm4
REA average), “worst case”, and “best case”. The worst
4 and best case will depend very much on application,
Temp change, F

fgoals1_0_g inmcm3_0 2 and certain seasons may matter most. For water re-
Degrees C

echo_g sources impacts, the worst case scenario might be the

3 gfdl_cm2_1
one with the largest winter or spring warming and
small or negative change in winter precipitation: for
bccr 2040s, MIROC 3.2 A1B has 2.8°C spring warming, only
2 bccr
1 3% increase in spring precipitation and no change in
bccr winter precipitation. The best case may be BCCR-B1
pcm1
with a 17% increase in winter precipitation, 8% increase
1 in spring precipitation, and warming of only 0.9°C in
winter and 0.5°C in spring. Another dimension of im-
0 0 pacts centers on how warm-dry summers are: the mean
-10 0 10 20 is +2.1°C and -12% for A1B, worst-case +4.4°C and
Precip change, % -30.0% in HadCM, and best-case +0.85°C and +7% for
PCM1 B1.
2080s
12
B1 hadgem1 4. Changes in other aspects of climate
6
10 A1B
A2 hadgem1 Climate models produce very detailed representa-
5
ipsl_cm4 tions of the state of the climate system, and the IPCC
8 ipsl_cm4 data centre provides dozens if not hundreds of vari-
echo_g ables. We present here two that are relevant for coastal
Temp change, F

echo_g miroc3_2_hi 4
ecosystems and infrastructure.
Degrees C

6 ipsl_cm4
echo_g 3
Figure 8. Scatterplot of change in annually aver-
4 giss_er aged temperature and precipitation for each of the
giss_er 2 20 models and 3 SRES scenarios, for the decades
giss_aom indicated. Green circles indicate B1, blue crosses
2 1 A1B, and red diamonds A2. Large bold symbols
indicate the REA value for each scenario and dec-
ade. Model names label the four extremes for each
0 0
-10 0 10 20 scenario.
Precip change, %
 Page 8

2020s change in temp

6
B1 5.3oF 3
5
A1B
4 3.6oF 3.6oF
3.3oF 3.3oF 2
3.2oF

Degrees C
degrees F

3
2.5oF
2.2oF 2.1oF o 2.3oF
2 o
1.8 F
o
1.9 F 1.9oF 2.1 F
1.7oF 1

1 1.1oF
1.0oF
0.7oF
0.4oF
0 0.1oF 0

-1
DJF MAM JJA SON Annual

2040s change in temp

5
B1 7.9oF
8
A1B 4

6 5.4oF
5.1oF 5.2oF 5.2oF 3
5.1oF

Degrees C
degrees F

4 3.9oF 4.1oF
3.6oF 3.6oF 2
3.2oF 3.0oF
o 2.9oF 2.7oF
2.5 F
2 1
1.5oF 1.3oF 1.5oF
1.0oF 1.0oF
0 0
DJF MAM JJA SON Annual

2080s change in temp

14
B1 12.5oF
7
12
A1B
6
10 9.7oF 9.7oF
9.1oF o
8.8 F
5
8.3oF
Degrees C
degrees F

8
7.1oF 4
o o
o 6.1 F 6.0 F
o 6.2 F
6 5.7 F
4.9oF 3
o
4.2 F 4.5oF 4.2oF
4
2

2 o
2.5 F o
2.4 F 2.8oF
1
o o
1.3 F 1.3 F
0 0
DJF MAM JJA SON Annual
Figure 9. Range of projected changes in temperature for each season (DJF=winter, etc.) and for the
annual mean, relative to the 1980s. In each pair of bars, the left one is for SRES scenario B1 and the
right is A1B. The REA mean is shown as a horizontal line and the value printed. Circles and x’s
represent individual model values, and the highest and lowest change for each season and decade is
printed.
 Page 9

2020s change in precip

8 20
B1
6 A1B 16

12
4 12%
23%
8
Inches

cm
2
20% 4
9%
13% 5.9%
0 2.0% 2.1% 1.3% 2.8% 1.6%
-0.2%-3.0% -0.1% 0
-7.9%
-11% -30% -11% -4
-2
-14%
-8
-9%
-4
DJF MAM JJA SON Annual

2040s change in precip

8 20
B1
6 A1B 16
12% 12
27%
4
8
Inches

21%

cm
2 16%
4
5.1% 17%
2.6% 3.3% 3.9% 5.1% 4.7% 2.0% 1.9% 0
0 -4.6%
-12.0%
-11% -10%
-4
-2 -30%
-13%
-8
-4
-11%
DJF MAM JJA SON Annual

2080s change in precip

8 20% 20
B1
42% 16
6 A1B
12
4
8
36%
Inches

cm

2 22%
9.0% 4
7.1% 14% 8.9%
3.2% 5.8% 5.7% 2.9% 4.3%
0 0
-11.3% -15.8%
-8%
-11% -4
-2 -11%
-38%
-8
-4 -10%
DJF MAM JJA SON Annual

Figure 10. As in Figure 9, but for precipitation. The height of the bars indicates actual water precipitation but
the percentages are calculated with respect to a reference value for that season, so that -11% in JJA is much
less than -11% in DJF. The reference values for the extremes are that model’s 20th century mean for that
season (or annual mean), and for the REA average the reference is the all-model 20th century value.
Page 10

coastal SST coastal TAU

18
0.20

20c 0.15 20c

16 A1B
A1B
A2
A2 0.10 B1
14
B1

Nm-2
0.05
C

12
o

10 0.00

8 -0.05

6 -0.10
Jan Mar May Jul Sep Nov Jan Jan Mar May Jul Sep Nov Jan

Figure 12. As in Fig. 11 but for along-shore wind

Figure 11. Simulated annual cycle of sea surface tem-
stress.
perature (SST) averaged over 1970-99 for all available
models. Grey shading represents ±1 standard devia-
tion about the 1970-99 mean, shown as a solid black increases are close to the REA values). Figure 13 shows
line, the three curves above the grey shaded area the extreme end of the probability distribution of pre-
show the means for 2040s, and the dashed curve cipitation for these two models. For IPSL, the prob-
ability of any given threshold being exceeded is much
shows observed SST at Race Rocks near Victoria,
higher (a factor of 2 or more) in the 21st century runs
BC, for 1970-99. than in the 20th century run. For ECHAM5 the prob-
ability of extreme precipitation increases modestly for
most thresholds. These results are preliminary and
4a. Coastal water properties further investigation of this issue will require extensive
Coastal sea surface temperature (SST) helps de- simulations with regional models.
termine the biological and physical conditions of the
marine environment and the estuaries of the North- 5. Discussion of differences from previous sce-
west. Each of the 20 models examined here has a de- narios
tailed ocean model with higher spatial resolution than
the atmosphere model, and simulates SST. Figure 11 Previously, we analyzed ten IPCC AR4 GCM scenarios
shows the mean annual cycle for the 1970-99 and 2030- using scenarios B1 and A2 (Mote et al. 2005), and did
59 periods for coastal grid points, along with the ob- not perform REA in constructing the “Average”. Tables
served. The modeled SST is much higher than observed 2 and 3 compare the annually averaged changes derived
during the months of May-December. Modeled change in that effort with the changes reported here. The
is about 1.5°C, somewhat less than for the PNW land changes from the previous work are generally small,
areas (2.0°C) but a significant change relative to the and the increases in the extremes are primarily the
small interannual variability of the ocean. result of using A1B instead of A2. In most instances
Along the west coast of each continent, summer- REA is fairly close to the plain model average, except in
time equatorward winds pull water offshore and water summer and in the 2080s.
must upwell from depth to replace it. This nutrient-
rich water serves as the basis for very high biological
productivity. Our earlier analysis of two climate models temperature 2020s 2040s
(Mote and Mantua 2002) indicated little change in
coastal upwelling in any of the major regions of upwel- (°C) old new old new
ling. For the 20 models used in this study, the mean
change is also quite small (Figure 12).
lowest 0.4 0.6 0.8 0.9
4b. Changes in extreme precipitation
Daily precipitation variability is of interest because average 1.1 1.2* 1.6 2.0*
stormwater runoff and most river flooding are driven
largely by extreme daily precipitation. Recent flooding highest 1.8 1.9 2.6 2.9
events in western Washington and western Oregon in
autumn 2003, 2006, and 2007 have led many to won-
der whether climate change is bringing an increase in Table 2. Comparison of low, average, and high scenarios
heavy precipitation events. A few modeling groups used by Mote et al. [2005] and this report. *In the “average”
have made available their output on a daily timescale. row, the old scenarios were simply averaged together but the
We use daily precipitation for 1981-2000 and 2046- new scenarios are combined using REA (see Appendix).
2065 from IPSL (the model with the largest increases in
mean winter precipitation and annual precipitation)
and ECHAM5 (whose winter and annual precipitation
 Page 11

Appendix A. In a few instances the data available

precipitation 2020s 2040s from Run 1 appeared not to be complete (e.g., missing
variable or decade) so we used Run 2. The models for
% old new old new which this occurred were CCSM3 A2 and B1, and PCM1
B1.
lowest -4 -9 -4 -11
Appendix B. Reliability ensemble averaging (REA)
average 2 1* 2 2* uses a bias factor and a distance factor to weight each
model’s output. Each factor is calculated by averaging
highest 6 12 9 12 quantities over the Pacific Northwest, for each season
and for the annual mean, following these steps.

Table 3. As in Table 2 but for precipitation. 1) Compute the difference δ between the model mean
and NCEP mean, for the 1970-99 period.
Probability density function of Precip 2) Calculate the tolerance factor ε to allow for vari-
3 ability of 30-year means relative to the century times-
20c cale. First, using regionally averaged data from the US
sresa1b
sresb1 Historical Climate Network, detrend (subtract the lin-
ear fit from) the 20th century time series, then calculate
the standard deviation ε of the running 30-year mean.
The tolerance factor is used in computing both the bias
2 factor and the distance factor.
3) Calculate the bias factor. For models with a δ less
than ε, the bias factor is 1; if δ is greater than ε, the bias
factor is reduced to ε/δ.
%

4) Looking now at 21st century simulations, regress

the quantity in question (e.g., annual mean tempera-
1 ture) on the log of CO2 (see Figure 1). For purposes of
calculating the distance factor, take the value of the
resulting fit at year 2045 minus the value at year 2000,
di for model i. This is the only step in which we depart
from the method of Giorgi and Mearns, and we do so in
order that each model has a single weighting factor for
0 each of the time periods considered (2020s, 2040s,
10 15 20
mm/day
25 30 2080s).
Probability density function of Precip
5) Calculate the all-model mean value d of the indi-
3 vidual model distances di. Then weight each model di
20c
by its distance from the mean d and recompute the all-
sresa1b model mean d. Only one or two iterations is needed to
sresb1 converge.
6) Calculate the distance factor in the same manner
as the bias factor: for di less than ε, the distance factor
2 is 1; for di greater than ε, the distance factor is ε/di.
7) For each season, decade, scenario, and variable,
compute an REA value by summing over all available
models the product of the model’s projected change, its
%

bias factor, and its distance factor.

Acknowledgments. We are grateful to the Pro-

1
gram for Climate Model Diagnosis and Intercompari-
son for making available the model output, and to the
State of Washington for funding this work.

References
0
10 15 20 25 30
mm/day Giorgi, F., and L.O. Mearns, 2002: Calculation of average,
uncertainty range, and reliability of regional climate changes
from AOGCM simulations via the reliability ensemble aver-
Figure 13. Probability distributions for regionally aver- aging (REA) method. J. Clim., 15, 1141–1158.
aged daily precipitation, 1981-2000 (20c) and 2046- Hamlet, A.F., and D.P. Lettenmaier, 2005: Production of tem-
65, for IPSL (top) and ECHAM5 (bottom). porally consistent gridded precipitation and temperature
 Page 12

fields for the continental United States. J. Hydromet., 6,

330-336.
Mote, P.W., and N.J. Mantua. 2002. Coastal upwelling in a
warming future. Geophysical Research Letters 29(23), 2138
doi: 10.1029/2002GL016086.
Mote, P.W., E.P. Salathé, and C. Peacock. 2005. Scenarios of
Future Climate for the Pacific Northwest. A report prepared
for King County. Climate Impacts Group, Center for Science
in the Earth System, Joint Institute for the Study of the At-
mosphere and Ocean, University of Washington, Seattle.
Randall, D.A., R.A. Wood, S. Bony, R. Colman, T. Fichefet, J.
Fyfe, V. Kattsov, A. Pitman, J. Shukla, J. Srinivasan, R.J.
Stouffer, A. Sumi, and K.E. Taylor, 2007: Climate models
and their evaluation. In: Climate Change 2007: The Physical
Science Basis. Contribution of Working Group I to the
Fourth Assessment Report of the Intergovernmental Panel
on Climate Change [Solomon, S., et al., (eds.)]. Cambridge
University Press, Cambridge, United Kingdom and New
York, NY, USA.
Raupach, M.R., G. Marland, P. Ciais, C. Le Quéré, J.G. Cana-
dell, G. Klepper, and C.B. Field, 2007: Global and regional
drivers of accelerating CO2 emissions. Proc. Natl Acad. Sci.,
doi 10.1073/pnas.0700609104.
Taylor, K.E., 2000. Summarizing multiple aspects of
model performance in a single diagram. Report No.
55 of the Program for Climate Model Diagnosis and
Intercomparison, Lawrence Livermore National Labo-
ratory.

For copies of this document, see

http://www.cses.washington.edu/db/pdf/moteetal2008scenarios628.pdf
or email philip@atmos.washington.edu

Cover photo: Skagit River with North Cascades in the background. Photo by Philip Mote.

View publication stats