Content uploaded by M. G. G. Foreman
Author content
All content in this area was uploaded by M. G. G. Foreman on Mar 14, 2018
Content may be subject to copyright.
Climate change in the Fraser River watershed: ¯ow and
temperature projections
John Morrison
a,
*, Michael C. Quick
b
, Michael G.G. Foreman
c
a
Vynx Design Inc., Sidney, BC, Canada
b
Department of Civil Engineering, University of British Columbia, Vancouver, BC, Canada
c
Institute of Ocean Sciences, Department of Fisheries and Oceans, Sidney, BC, Canada
Received 1 October 2001; revised 19 February 2002; accepted 22 March 2002
Abstract
An analysis of the historic ¯ows and water temperatures of the Fraser River system has detected trends in both the annual
¯ow pro®le and the summer temperatures. This study was undertaken to determine if these trends are likely to continue under
the conditions predicted by various global circulation models. To do this, existing ¯ow and temperature models were run with
weather data that were derived from actual weather observations, but modi®ed using changes predicted by the global circulation
models.
The validity of the ¯ow model results is supported by very close agreement with the historical record. The differences
between model output and the historical record for mean ¯ow, mean peak ¯ow, mean minimum ¯ow and peak ¯ow day were
not statistically signi®cant; furthermore, there was only a 3± 4 day shift in the occurrence of cumulative ¯ow milestones. The
temperature model's mean water temperature was only 0.2 8C higher than the historical record.
For the period 2070± 2099, the ¯ow model predicted a modest 5% (150 m
3
/s) average ¯ow increase but a decrease in the
average peak ¯ow of about 18% (1600 m
3
/s). These peaks would occur, on average, 24 days earlier in the year even though for
13% of the years the peak ¯ow occurred much later as a result of summer or fall rain, instead of the currently normal spring
freshet. In the same period, the summer mean water temperature is predicted to increase by 1.9 8C. The potential exposure of
salmon to water temperatures above 20 8C, which may degrade their spawning success, is predicted to increase by a factor of 10.
Trends in both ¯ow and temperature in this study closely match the trends in the historical record, 1961± 1990, which
suggests that the historical trends may already be related to climate change. While the mean ¯ow of 2726 m
3
/s does not show a
statistically signi®cant trend, the hydrological pro®le has been changing. q2002 Elsevier Science B.V. All rights reserved.
Keywords: Climate change; Fraser River; Runoff; Temperature; Fish
1. Introduction
Draining a watershed of approximately
217,000 km
2
, the Fraser River is the largest Canadian
river that ¯ows to the Paci®c Ocean. With headwaters
near Jasper, Alta in the Rocky Mountains, the Fraser
¯ows for 1370 km before it discharges into the Strait
of Georgia near Vancouver (Thompson, 1981).
The Fraser River watershed is a major spawning
Journal of Hydrology 263 (2002) 230± 244
www.elsevier.com/locate/jhydrol
0022-1694/02/$ - see front matter q2002 Elsevier Science B.V. All rights reserved.
PII: S0022-1694(02)00065-3
* Corresponding author. Address: Institute of Ocean Sciences,
Fisheries and Oceans Canada, P.O. Box 6000, 9860 West Saanich
Road, BC V8L 4B2, Sidney, Canada.
E-mail addresses: morrisonjo@pac.dfo-mpo.gc.ca
(J. Morrison), mquick@civil.ubc.ca (M.C. Quick),
foremanm@pac.dfo-mpo.gc.ca (M.G.G. Foreman).
ground for Sockeye and Chinook salmon, accounting
for the majority of the Canadian stocks (DFO,
1999a,b). Sockeye salmon begin their lives in spawn-
ing beds distributed throughout the watershed. Eggs
laid in these beds hatch in the following spring. After
spending the next year in fresh water, they move into
the ocean for a period of 2±3 years after which time
they return to their original natal streams where they
spawn and then die. On the migration back to the
spawning beds the salmon are sensitive to the river
water temperatures and there is a strong correlation
between pre-spawning mortality and high river
temperature (Gilhousen, 1990; Rand and Hinch,
1998; Williams, 2000). Temperatures between 22
and 24 8C over a period of several days can be fatal
for salmon (Servizi and Jansen, 1977) and tempera-
tures over 24 8C can cause death within a few hours
(Bouke et al., 1975). Even water temperatures as low
as 20 8C can have an adverse affect on spawning
success rates (Gilhousen, 1990).
Daily ¯ow records recorded at Hope since 1912
indicate that the Fraser River has been changing.
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244 231
Fig. 1. Fraser River watershed.
The Julian day numbers by which one-third and one-
half of the integrated yearly discharge occurred have
been transpiring earlier (Fig. 2) in the year and the
summer water temperatures have been increasing
(Fig. 3; Foreman et al., 2001). The river ¯ows are
highly seasonal with winter lows at Hope often
below 1000 m
3
/s and peak ¯ows typically occurring
in mid-June in the range of 9000 m/s. These ¯ows are
primarily generated from seasonal snowmelt. The
lowest ¯ow recorded between 1913 and 2000 was
340 m
3
/s on January 8, 1916, while the highest
recorded level of 15,200 m
3
/s occurred on May 31,
1948 during a period of extensive ¯ooding of the
lower Fraser River.
Summer water temperatures have been recorded
manually at Hells Gate since the early 1940s. Typi-
cally these recordings were made from July 1 to
September 15 but unfortunately in the early years
there was much missing data. In this paper, the river
temperature comparisons are based on the manual
data collected at Hells Gate between 1961 and 1990.
Recently this activity was supplemented by the instal-
lation of an automatic data recorder at Qualark, a few
kilometres downstream from Hells Gate. Of all of the
summer river temperatures recorded at these loca-
tions, the highest was 21.2 8C occurring on August
3, 1998, and the lowest was 11.0 8C recorded on
July 1, 1955. Restricting the analysis to the time
span when records are almost contiguous (i.e. 1953 ±
1998), it was found that the summer mean tempera-
ture, which ranges from 15 to 19 8C, was increasing at
a rate of 0.022 8C per year with a signi®cance of 98%.
Fig. 3 shows the latest available summer mean
temperatures with their linear trend line. This line
has a slope of 0.018 8C per year with a signi®cance
of 95%.
The purpose of this paper is to extend the preceding
analyses of historical data, and those described by
Foreman et al. (2001), into the future through the
use of global climate change models. These models,
which predict possible climate changes under various
scenarios of increasing greenhouse gases and aero-
sols, will allow us to assess the possible impact of
changing climate conditions on river ¯ow and
temperature. Because of the sensitivity of salmon to
water temperature, it is critical to determine if the
historical trends are likely to continue to the point
where the survival of the Fraser River salmon stocks
are threatened.
This manuscript is organised as follows. In
Section 2, we describe the methods used in this
study with an emphasis on downscaling global
climate data and our ¯ow and temperature models.
The climate experiment is described in Section 3.
The statistical methodology, model validation, and
projected changes in river ¯ows and temperatures
are presented in Section 4. Implications of some
of these changes are discussed in Section 5 and
the paper concludes with recommendations for
future work in Section 6.
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244232
Fig. 2. Days of the year by which one-third and one-half the inte-
grated yearly Hope discharge has occurred. Trend analyses suggest
earlier progressions at the rates of 0.11 and 0.09 days per year,
respectively, with the statistical signi®cance .95% in both cases.
Fig. 3. Mean summer temperatures and their trend for the period
1953±2000.
2. Method
As an extension of the analysis of historic river
conditions, this study was designed to determine the
types of changes that could be expected under a chan-
ging climate. To do this, the same models that were
used to hindcast river conditions from historical
weather data were run using weather data derived
from climate change models.
Model output from coupled global circulation
models was downscaled to sites in the Fraser River
watershed that had reliable historic weather records.
The downscaled predicted changes were added to the
historic data and then used to drive the ¯ow and
temperature models. Using the IPCC guidelines
(IPCC-TGCIA, 1999), a baseline was established
using the years 1961±1990. The modelled ¯ow and
temperatures were compared with the historical
record in order to establish the validity of the models.
The climate change impacts were then assessed by
comparing the baseline values to future 30-year peri-
ods. Throughout this document, these periods are
referred to in the short form as:
²2020Ð years 2010 to 2039,
²2050Ð years 2040 to 2069,
²2080Ð years 2070 to 2099.
2.1. General circulation models
A general circulation model (GCM) is a large-scale
numerical model that simulates the physical processes
that affect climate. They solve physical equations that
describe the complex interactions among the atmo-
sphere, ocean, cryosphere and land surface. GCMs
are three-dimensional with horizontal grid resolutions
typically between 250 and 600 km, 10±20 vertical
layers in the atmosphere, and up to 30 layers in the
oceans. While there is still a great deal of uncertainty
about the accuracy of GCM output, especially at the
regional level, there is a scienti®c consensus that they
are a suitable tool to project future climate change
(Grassl, 2000; IPCC-Working Group I, 2001).
In this study, output from two models were used.
They are the Canadian Centre for Climate Modelling
and Analysis model CGCM1 (Flato et al., 2000) and
the Hadley Centre for Climate Prediction and
Research model HadCM2 (Johns et al., 1997).
2.2. Downscaling
GCM model output is available on large grid scales
(3.758£3.758for CCGM1 and 2.58£3.758for
HadCM2) that do not lend themselves to assessing
the impact of climate change at local levels. Regional
climate models are being developed that will have
scales more suitable to local impact assessment, but
the output from these models is not yet generally
available. Various techniques for mapping the large
GCM grids to local levels have been developed
(Wilby and Wigley, 1997; Giorgi and Mearns, 1991)
and this process is referred to as downscaling. The
hydrologic ¯ow and temperature models used in this
study were developed to use actual weather data from
multiple sites located throughout the Fraser
watershed. To produce accurate results, these models
need data that are both internally consistent at each
location (e.g. no heavy snowfall on days with no cloud
cover), as well as systematically consistent among
stations (i.e. the pattern of data at all stations must
represent feasible weather patterns). These criteria
can be achieved by using statistical climate inversion
(Giorgi and Mearns, 1991) where historical weather
data values are adjusted by the amount of change
predicted by the GCM. By using 30 years of historical
data, we are reasonably well assured of capturing
normal ranges of inter-annual variability, as well as
longer time scale phenomenon such as El Nin
Äo South-
ern Oscillation (ENSO) events and Paci®c Decadal
Oscillations (PDO; Mantua et al., 1997).
Statistical climate inversion results in a change in
the mean value of a local weather variable by the
amount of the change predicted by the GCM. The
variance is unchanged for the variables where the
GCM changes are expressed as absolute numbers
(temperature, solar radiation and vapour pressure);
however, the variance will change for the other vari-
ables (wind speed, cloud cover and precipitation) that
are adjusted by relative amounts. Statistical climate
inversion will not change other weather characteris-
tics such as rainfall persistence or frequencies.
Predicting changes in local variance, persistence,
frequencies, etc. of weather variables from GCMs is
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244 233
dif®cult since these models, with their large grid sizes,
do not express local phenomenon accurately.
The ¯ow and temperature models used in this study
both require weather data from a number of stations
throughout the Fraser watershed. The ¯ow model
requires daily precipitation and temperature data
over an entire year to accurately represent the annual
spring freshet that arises from snow pack build-up and
its subsequent melting. The temperature model is used
to provide river temperature forecasts throughout the
salmon migration season and is only calibrated to
operate in the summer months. The heat ¯ux calcula-
tions in this model require hourly air temperature, dew
point temperature, solar radiation, cloud cover and
wind speed. For this study, the model was only run
from July 1 to September 15, as historical data
suggests that river temperatures outside of this range
will not adversely affect migrating salmon. Historical
daily weather data at 10 stations over the period of
October 1960
1
to December 1990 were used as the
basis for predicting river ¯ows. Daily weather data
at two stations from 1961 to 1990 were used for the
river temperature predictions.
The four grid cells nearest to each weather station
were linearly weighted based on the relative distances
from the grid centres to the station. Mean monthly
changes for each of the required GCM weather vari-
ables were scaled by their weights and summed to
produce the station change. These monthly station
changes were then mapped to a 365-day year and
smoothed with a heuristic process that minimised
daily changes while preserving the monthly mean
change calculated by the GCMs. This smoothing
process was necessary to avoid problems caused by
the introduction of large month end steps in the hourly
weather variables used by the temperature model.
Future weather was modelled by adding the smoothed
changes to the historical observations at the stations,
reintroducing leap days as required.
2.3. The UBC watershed and ¯ow models
The Fraser River basin, with an area of
217,000 km
2
, was modelled as 12 sub-watersheds
using the UBC watershed model (Quick and Pipes,
1976a,b; Quick, 1995). This model requires continu-
ous precipitation and temperature data for each sub-
watershed, and then generates continuous estimates of
watershed out¯ow, based on the estimated snowpack
accumulation and melting, together with rainfall. The
daily outputs of stream¯ow from each of the 12 sub-
watersheds were then linearly interpolated into hourly
values as inputs to the UBC ¯ow model (Quick and
Pipes, 1976a,b).
The UBC ¯ow model was used to represent the
network of lakes and river channels that make up
the Fraser River system. Because the temperature
modelling requires ¯ow and velocity information at
hourly intervals, the ¯ow model requires a 10-km
reach length to approximate the ¯ow propagation rate.
For this study, the models were calibrated for a 10-
year period (1970±1979), and then veri®ed for 50
years of continuous simulation. Comparison with the
measured stream¯ow showed a close agreement over
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244234
Fig. 4. Modelled and observed ¯ow at Hope.
1
This 2-month extension prior to the normal baseline period is
necessary to include all of the seasonal snowpack build up for the
model runs that correspond to the 1961 model year.
the whole of this 50-year period, which included the
30-year period used for the present work. This reliable
and accurate ¯ow simulation over an extended period
of historical ¯ows, from 1948 to 1997 inclusive, is a
vital step in establishing that the models will operate
satisfactorily, before they are used to generate ¯ows
for the various climate change scenarios.
A selected sample of some of these simulated and
measured ¯ows is plotted in Fig. 4. This period was
selected because it includes two of the larger ¯ood
years, 1972 and 1974. Only a 5-year period, 1972±
1977, is plotted, so that some of the detail of the
calculated and measured ¯ows can be seen.
2.4. The temperature model
The same one-dimensional temperature model
described by Foreman et al. (1997, 2001) was used
for this study. This energy balance model has the
form:
EDEflowin 1Etrib 1Eatm 2Eflowout 1
E
D
is the change in energy within the reach, E
¯owin
the
energy of the water entering the reach from the reach
upstream, E
trib
a product of the tributary ¯ow calcu-
lated by the ¯ow model and the tributary temperature
that was calculated, in this study, by the 1998 regres-
sion models described by Foreman et al. (2001), E
atm
the product of the heat ¯ux across the surface of the
river and it is calculated by a set of empirical equa-
tions that account for solar radiation, atmospheric
long wave radiation, long wave back radiation,
evaporation/condensation, conduction, solar re¯ec-
tion and atmospheric re¯ection and E
¯owout
is the
energy of the water ¯owing downstream out of the
reach.
2.5. Uncertainty
2
It is not possible to quantify the uncertainties
involved in climate change impact assessments.
Climate change modellers start by forcing their global
circulation models with hypothetical greenhouse gas
and aerosol levels. This forcing causes the models to
operate outside the historic range of conditions where
model error can be measured and uncertainty inferred.
Impact modellers use this uncertain GCM output to
force their impact models which are now, also likely
to operate outside their normal operating range.
Nonetheless, while it is not possible to quantify
uncertainty it has been possible to minimise the model-
ling errors through the impact assessment design.
²The downscaling method, using multiple grids, miti-
gates problems with GCM wave resolution at the
individual grid level.
²The method of adding GCM changes to observed
weather eliminates the absolute portion of any
systematic GCM modelling bias. It does not,
however, address the problems associated with
changes in bias as the model moves outside its vali-
dated operating range.
²Running the model to cover a long time span (30
years) ensures that a reasonable level of climate
variability is incorporated into the assessment.
²The use of the regression models for the tributary
temperatures is problematic when the regression
parameters lie outside the domain used to determine
the regression coef®cients. Regression models are,
however, less uncertain at extrapolation than the
neural network models which are used in our opera-
tional river temperature forecast system.
3. Experiments
GCM data was downloaded from the Canadian
climate impacts scenarios (CCIS) web site (CICS,
2001; this site provides GCM data using the IPCC
Distribution Centre guidelines). Though the CCIS
site limits its database coverage to the geographical
area that encompasses Canada, it conveniently
contains data from several GCMs. Precipitation, mini-
mum and maximum air temperature were downscaled
for each of the 10 stations (Agassiz, Barkerville, Blue
River, Fort St James, Hope, Kamloops, Prince
George, Salmon Arm, Smithers, and Williams Lake)
that are needed to model the watershed ¯ows (see Fig.
1 for locations of these sites). Mean air temperature,
vapour pressure, solar radiation, cloud cover and wind
were downscaled at Kamloops and Prince George and
used in the temperature model. At the time of this
study, the only model that provided all of these
variables at either the IPCC Distribution Centre or
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244 235
2
For a comprehensive summary of uncertainty and downscaling
the reader is referred to the IPCC Third Assessment Report (IPCC
TAR WG1, 2001, Chapter 10).
the CICS site was the Canadian Centre for
Climate Modelling and Analysis model CGCM1.
The scenario selected for the study was the
`greenhouse gas with aerosols' run based on the
IPCC scenario 92a. The CGCM1 results from
three model runs with differing initial conditions,
plus an ensemble of those three runs, were used
for this study (Boer et al., 2000a,b). The ¯ow and
temperature models were ®rst run using observed
weather data over the baseline period 1961±1990.
The models were then run with simulated weather
for each CGCM1 experiment and each standard
period, i.e. 2010±2039, 2040±2069 and 2070±
2099. While incomplete weather data meant that
we were unable to use multiple GCMs to test the
robustness of our river temperature predictions, we
were able to use ensemble data from the Hadley
Centre for Climate Prediction and Research model
HadCM2 as input to our ¯ow model.
Fig. 5 demonstrates some of the complexities in
predicted climate change. The variables used to predict
river temperature can be seen to change throughout the
year, however, they all do not follow the same seasonal
pattern. For example, the 2080 mean temperature
change (the most commonly quoted climate change
variable) is highest in the spring and lowest in the fall.
Precipitation, on the other hand, shows a decrease in the
late spring and an increase in fall. Furthermore, the
precipitation change is published as a percentage so
that it cannot be evaluated without knowledge of the
baseline precipitation values.
4. Results
4.1. Statistical analysis
The downscaling methodology of modifying histor-
ical weather observations by changes predicted by the
GCM meant that the modelled ¯ows and temperatures
are paired with historical observations and subsequent
statistical analysis must re¯ect this. Differences
between baseline values and those from other periods,
as well as the difference between the modelled and
observed values over the baseline period, were calcu-
lated. Comparisons between the baseline and the
observations were used to establish the validity of
the model, and comparisons between the baseline
and the future periods provided a measure of the
expected impact of climate change.
A Lilliefors test (Wall, 1986; US Environmental
Protection Agency, 2000) was conducted for each set
of data to determine if it was normally distributed. A
Von Neumann (Wall, 1986; US Environmental Protec-
tion Agency, 2000) test was then conducted to validate
the independence of the data. Data that were both
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244236
Fig. 5. CGCM1 weather variables downscaled to Kamloops.
normal and independent meant that a t-test could then be
used to determine if the difference in the mean was
statistically signi®cant. In cases where the data failed
either the Lilliefors or the Von Neumann test, a
Wilcoxon's Rank Sum (Wall, 1986; US Environmental
Protection Agency, 2000) test was used to test the
hypothesis that the means were different. If the results
of the Wilcoxon's Rank Sum test were signi®cant then
the Hodges± Lehmann difference (Wall, 1986) was
calculated.
4.2. GCM forcing comparison
While we did not conduct a detailed comparison
of GCM output, we did conduct one simple
comparison to evaluate if the choice of GCMs
wouldbelikelytohavealargeaffectonthe
results output from the impact models. To do
this, we compared the ¯ow model results forced
with CGCM1 weather to ¯ow output forced with
HadCM2 weather. (See Murphy (1999) and Wilby
and Wigley (2000) for an assessment of the cred-
ibility of HadCM2 scenarios for downscaling and
impact analysis.)
The small differences that resulted from differ-
ent GCM forcing can be seen in Table 1. For
2020 and 2080, paired t-tests showed no statistical
differences. For 2050 the t-testcouldnotbe
applied since the data failed the Lilliefors test.
The Wilcoxon's Rank Sum Test could not differ-
entiate the two sets of data. With such similarity
in ¯ow during the critical salmon migration
season, we surmise that the differences in
temperature model output that would arise from
using HadCM2 based weather would not be radi-
cally different from the CGCM1 based output.
This is further borne out by an examination for
the grid by grid comparisons of GCM output for
temperature and precipitation posted on the CCIS
web site (CCIS, 2001). All of the GCM models are
tightly clustered on the temperature±precipitation
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244 237
Table 1
Mean ¯ow at Hope during migration season using different GCM
forcing
GCM Period
2020 2050 2080
HadCM2 (m
3
/s) 4064 3816 3435
CGCM1 (m
3
/s) 4046 3634 3406
Difference (m
3
/s) 18 182 29
Table 2
Flow model statistics at Hope
Observed Base 2020 2050 2080
Mean ¯ow (m
3
/s) 2803 2920 2973 2963 3071
Difference from baseline (m
3
/s) 2117
a
53
a
43
b
150
b
Mean peak ¯ow (m
3
/s) 8705 8860 8443 7845 7241
Difference from baseline (m
3
/s) 2155
a
2417
a
21015(2989) 21619(21614)
Mean minimum ¯ow (m
3
/s) 687 675 768 843 974
Difference from baseline (m
3
/s) 12
a
93
b
168
b
299
b
Julian peak ¯ow day 165 163 153 146 138
Difference from baseline (days) 2
a
210(211) 217(217) 225(224)
Day on which the accumulated
¯ow reaches one-third of the
total for the year
156 160 151 143 134
Difference from baseline 24(23) 210
b
218
b
227
b
Day on which the accumulated
¯ow reaches one-half of the total
for the year
180 184 176 171 166
Difference from baseline 24(24) 28(28) 213(213) 218(218)
a
No statistically veri®able difference.
b
Mean difference validated by paired t-test with .0.99 probability.
The numbers in parentheses are Hodges± Lehmann differences validated by Wilcoxon's rank sum test .0.99 probability.
scattergrams for the grids covering the Fraser River
watershed.
4.3. Model validation
Actual weather observations from the period of
October 1960±December1990 were used to model
the ¯ow and temperature of the river system. The
model output was then compared to the observed
river conditions in order to establish the model accu-
racy.
Following this procedure, it was determined (see
Table 2) that the baseline and observed mean ¯ow,
minimum ¯ow, peak ¯ow, and peak ¯ow dates were
not statistically different. There was a 3± 4-day shift in
the timing between the observed and the baseline
cumulative ¯ow.
The difference between the mean baseline tempera-
ture and the mean observed temperature at Hells Gate
(see Table 3) was found to be only 0.2 8C. The small
or negligible differences between observed and
modelled ¯ows and temperatures over the baseline
period demonstrate the reliability of the ¯ow and
temperature models. This means that changes
measured between the modelled baseline and the
future periods should have some credibility. That is
to say, if the weather downscaled from the GCM is
reasonably accurate then the watershed ¯ows and
temperatures will likewise be reasonably accurate.
The critical question concerns the accuracy of the
weather predicted by the CGM. Until this is answered,
the realism of our watershed forecasts must be treated
as conditional on the realism of the GCM forecasts.
4.4. Changes in river ¯ow and temperature due to
climate change
To understand the impact of the ¯ow and tempera-
ture changes, it is necessary to take a more detailed
look at the data. This can be best done graphically.
4.4.1. Flow
Fig. 6 shows average daily mean ¯ows at Hope for
each of the 30-year periods in the study. It can be seen
that the peak ¯ow decreases over time and that it
occurs earlier in the spring. It should be noted that
there are differences in both the timing and the volume
of the peaks shown on the graph and the peaks
reported in Table 2. This is due to different methods
of analysing the data. The peaks used in the table were
found by identifying the date and volume of peak ¯ow
for each year and then calculating the statistics related
to the peaks. On the graphs, the ¯ow for each day was
averaged over the 30-year period. The peak on the
graph is the day with the highest average ¯ow.
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244238
Table 3
Temperature model statistics
Observed Base 2020 2050 2080
Fraser river at Qualark Hells Gate
Mean temperature (8C) 16.2 16.4 17.2 17.9 18.3
Difference (8C) 20.2
a
0.8
a
1.5
a
1.9
a
Thompson at Spences Bridge
Mean temperature (8C) Not available 17.0 17.9 18.5 19.1
Difference (8C) Not available 0.9(0.8) 1.5(1.4) 2.1(2.0)
a
Mean difference validated by paired t-test with .0.99 probability.
The numbers in parentheses are Hodges± Lehmann differences validated by Wilcoxon's rank sum test .0.99 probability.
Fig. 6. Fraser River ¯ows at Hope.
In addition to the changes in the peak ¯ow charac-
teristics, it is also clear that there is a signi®cant
change in the winter ¯ow pattern. For the observed
and the baseline period, the winter ¯ow is virtually
constant at approximately 1000 m
3
/s. However, by
2080 the ¯ow is continually increasing as winter
progresses so that the April 1 ¯ow rate is double the
January 1 ¯ow rate.
These seasonal changes are also evident in the ¯ow
change graphs in Fig. 7. The 2080 curve shows a
signi®cant increase in ¯ow in April. However, the
largest absolute change in ¯ow is the decrease that
occurs in June at the time of the freshet. When viewed
as percentage changes, the 2080 ¯ow pattern shows a
200% increase in March but only a 40% decrease in
¯ow during the freshet.
Another way to view the change in the timing of the
¯ow is to compare the dates by which cumulative ¯ow
milestones are reached. Fig. 7 show the 1/3 and 1/2
cumulative ¯ow milestones for the climate change
study together with the regression lines calculated
from historical records (Foreman et al., 2001). The
apparent agreement between the historical projection
and the predicted values is remarkable. Unfortunately,
the method of predicting future values means that the
data does not meet the pre-requisite criteria for
performing a regression test. Thus a statistical
comparison of past and future trends cannot be made.
Fig. 9 shows how the range of ¯ow is expected to
change. In 2080, there will be a modest (300 m
3
/s)
increase in the minimum ¯ow with respect to baseline
values, and a more signi®cant (1600 m
3
/s) decrease in
peak ¯ow. In spite of these range changes, the total
volume of water ¯owing out of the watershed, repre-
sented by the mean ¯ow, only increases by 150 m
3
/s,
or about 5%, by 2080.
Perhaps the most signi®cant predicted change in
future ¯ow is shown in Fig. 10. Currently a freshet
resulting from the melting of the snow pack that accu-
mulated over the preceding winter dominates the
Fraser discharge. Winter months are characterised
by lower, more or less constant ¯ows, in the
1000 m
3
/s range. In the spring, as the snow melts,
the ¯ow increases rapidly until it typically, on aver-
age, reaches a peak in the 8000±10,000 m
3
/s range by
about mid-June. After the freshet, the ¯ow generally
declines until the early fall when it may be in¯uenced
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244 239
Fig. 7. Changes in ¯ow at Hope.
Fig. 8. Cumulative ¯ow milestones for the Hope discharge.
by large rainfall events. However, as seen in Fig.
10(a), by the end of the 21st century, our analysis
shows that 13% of the years are no longer snowmelt
dominated. Peak ¯ows are no longer clustered around
June and may occur as early as the ®rst week in April,
or as late as the second week in October. Fig. 10(b)
shows that the corresponding 4 years in the baseline
period were relatively normal.
4.4.2. Temperature
Fig. 11 shows average present and future water
temperatures at Hells Gate on the Fraser River and
Spences Bridge on the Thompson River. In both
cases, the water temperature is seen to rise throughout
July to a high in early August after which it cools
slowly until the end of the study period in mid-
September. There are two important points to note
in these ®gures. The ®rst is that on the Fraser River,
the mean 2080 temperatures at the start of the simula-
tion period already exceed the mean high temperature
for the baseline period and remain above that baseline
high for approximately 7 weeks. On the Thompson
River, the 2080 temperatures are not only above the
baseline high for nearly 10 weeks, but they also
exceed the 20 8C temperature considered harmful to
salmon spawning for approximately 4 weeks.
The expected mean temperature change at Hells
Gate is shown in Fig. 12. The largest change, 3.5 8C
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244240
Fig. 10. Four annual discharges that demonstrate dramatic changes in the ¯ow regime.
Fig. 9. Observed and predicted ¯ow ranges at Hope.
for the 2080 period, will occur in mid-July and taper
off thereafter to approximately 0.75 8C by mid-
September.
Hells Gate summer mean temperatures are graphed
in Fig. 13. Once again, a close, but statistically
untestable, agreement between the historical trend
and the climate change predictions can be seen. In
fact, there is such close agreement between the three
experiments ga1, ga2 and ga3 and the ensemble
results, gax, that we felt justi®ed in using the ensem-
ble results for the detailed analysis of predicted river
¯ows and temperatures.
5. Discussion
The ¯ow model does not predict drastic changes in
the mean or minimum ¯ow characteristics of the
Fraser River but the predicted decline in the peak
¯ow may have serious implications. The impact of a
reduced freshet on sediment transport, ¯ood manage-
ment, hydro-electric power generation, and riverine
and estuarine ecosystems can only be addressed by
experts in those ®elds. We note that the shift in timing
of the ¯ow is in accordance with predictions outlined
in the report produced by working group II of the
IPCC (IPCC-Working Group II, 2001).
While the ¯ow changes are expected to be benign,
the predicted river temperature changes could have
serious implications for salmon. As it is known that
exposure to excessively warm water degrades spawn-
ing success, a measure was developed to allow the
comparison of current exposure levels to the levels
projected under climate change. Cumulative exposure
was determined by summing the number of 10-km
reaches and hours where the temperature exceeds
20 8C. The relative values of these exposure numbers,
with the units degree reach hours (DRH), give some
measure of the threat to salmon spawning success. For
the baseline period of 1961±1990, the highest expo-
sure number was 238 DRH in 1961, but for the 2070 ±
2099 period, the maximum had climbed to
2259 DRH. Taking the worst-case year in the baseline
period for each river as a critical threshold, Fig. 14
shows the percentage of years in each future 30-year
period that exceed that threshold. Clearly these years
should be considered a threat to salmon spawning
success. On the Fraser, the highest temperatures
occur in those sections of the river below its con¯u-
ence with the Thompson. This means that ®sh
migrating up the Thompson will encounter debilitat-
ing temperatures in years when either river has
temperatures that exceed the critical 20 8C threshold.
For the 2080 period, 57% of the 30 years will have
Fraser River temperatures that exceed the worst year
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244 241
Fig. 11. Fraser and Thompson River temperatures.
Fig. 12. Temperature change at Hells Gate with respect to baseline
values.
(1961) in the baseline period. The analogous number
for the Thompson River is 67%. Consequently, unless
the salmon can adapt by changing their time of migra-
tion, these increasing temperatures can be expected to
have dramatic impacts on spawning success and
viability of species. We note that the report produced
by working group II of the IPCC (IPCC-Working
Group II, 2001) also expects climate change to
increase the stress on salmon.
The historical trends in ¯ow and temperature that
have been identi®ed by Foreman et al. (2001) are
reproduced here with more current data (Figs. 2 and
3). The close agreement between the historic trends,
the ¯ow and temperature predictions can be seen in
Figs. 8 and 13. The smooth transition from historical
trend to the climate change trend lends weight to the
argument that the observed changes may be related to
historic climate change that has already occurred.
6. Future work
Three areas of future work are suggested by this
study. The ®rst is the use of additional GCMs to produce
our Fraser watershed climate change data. Climate
change models are continually being improved and we
clearly want to use the most reliable climate change
predictions that are available. As a corollary, the
temperature model needs to be run with predictions
from more than just the one GCM that distributes the
required variables through the IPCC site.
The second task for the future is that the
temperature model should be run for a longer
period each year. This study limited the tempera-
ture simulations model to the July±mid-Septem-
ber timeframe of the historical records. However,
the results of this study indicate that by the end
of the century, the temperatures deleterious to
salmon spawning may occur prior to July 1.
Running the model from May to September
would give a better indication of the extent of
the period when elevated river temperatures may
threaten salmon spawning success. It may also
indicate times when salmon spawning success
might be mitigated if their arrival times in the
river were to change.
The third future task is that other watersheds
should be investigated. In particular, rivers that
have not historically reached temperatures that
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244242
Fig. 13. Summer mean temperatures at Hells Gate.
Fig. 14. Percentage of years with DRH above the baseline critical
value.
are hazardous to salmon spawning may start to do
so under climate change. Clearly it is important to
consider all salmon populations that may be threa-
tenedbyclimatechange.
Acknowledgements
We thank Elaine Barrow of the Canadian Institute for
Climate Studies for advice on the use of GCM output
and downscaling methods, the Water Survey of Canada
and the Meteorological Service of Canada for providing
historical ¯ow and weather data, Peter Cheng of the
Paci®c Salmon Commission for providing the Hells
Gate spot temperature data, Edmond Yu for assistance
with running the hydrologic models, Trish Kimber for
assistance with the ®gures, and the Climate Change
Action Fund for partial ®nancial support.
References
Boer, G.J., Flato, G., Reader, M.C., Ramsden, D., 2000a. A transient
climate change simulation with greenhouse gas and aerosol
forcing: experimental design and comparison with the instru-
mental record for the twentieth century. Climate Dynamics 16,
405± 425.
Boer, G.J., Flato, G., Ramsden, D., 2000b. A transient climate
change simulation with greenhouse gas and aerosol forcing:
projected climate to the twenty-®rst century. Climate Dynamics
16, 427± 450.
Bouke, G.R., Chapman, G.A., Schneider Jr., P.W., Stevens, D.G.,
1975. Effects of holding temperature on reproductive develop-
ment in adult sockeye salmon (Oncorhynchus nerka). Annual
Northwest Fish Culture Conference, 3± 5 December, 1975, Otter
Rock, Oregon, pp. 24± 40.
CICS, 2001. http://www.cics.uvic.ca/scenarios/.
DFO, 1999a. Fraser River Chinook Salmon, DFO Science Stock
Status Report D6-11, 5 pp.
DFO, 1999b. Skeena River Sockeye Salmon, DFO Science Stock
Status Report D6-10, 7 pp.
Flato, G.M., Boer, G.J., Lee, W.G., McFarlane, N.A., Ramsden, D.,
Reader, M.C., Weaver, A.J., 2000. The Canadian Centre for
Climate Modelling and Analysis global coupled model and its
climate. Climate Dynamics 16, 451± 467.
Foreman, M.G.G., James, B., Quick, M.C., Hollemans, P., Weibe,
E., 1997. Flow and temperature models for the Fraser and
Thompson Rivers. Atmosphere± Ocean 35 (1), 109±134.
Foreman, M.G.G., Lee, D.K., Morrison, J., Macdonald, S., Barns,
D., Williams, I.V., 2001. Simulations and retrospective analysis
of Fraser watershed ¯ows and temperatures. Atmosphere±
Ocean 39 (2), 89± 105.
Gilhousen, P., 1990. Prespawning mortalities of sockeye salmon in
the Fraser River system and possible causal factors. Bulletin
XXVI, International Paci®c Salmon Fisheries Commission,
Vancouver, 53 pp.
Giorgi, F., Mearns, O., 1991. Approaches to the simulation of regio-
nal climate change: a review. Reviews of Geophysics 29 (2),
191± 216.
Grassl, H., 2000. Status and improvements of coupled general circu-
lation models. Science 16, 1991± 2001.
IPCC-TGCIA, 1999. Guidelines on the use of scenario data for
climate impact and adaptation assessment, version 1. Prepared
by Carter, T.R., Hulme, M., Lal, M. Intergovernmental Panel on
Climate Change, Task Group on Scenarios for Climate Impact
Assessment, 69 pp.
IPCC-Working Group I, 2001. Climate Change 2001: The Scienti®c
Basis; Summary for Policymakers and Technical Summary,
IPCC, Geneva, pp. 98.
IPCC-Working Group II, 2001. Climate Change 2001: Impacts,
Adaptation and Vulnerability; Summary for Policymakers and
Technical Summary, IPCC, Geneva, pp. 89.
Johns, T.C., Carnell, R.E., Crossley, J.F., Gregory, J.M., Michell,
J.F.B., Senior, C.A., Teff, S.F.B., Wood,R.A., 1997. The second
Hadley Centre coupled ocean± atmosphere GCM: model
description, spinup and validation. Climate Dynamics 13,
103± 134.
Mantua, N.J., Hare, S.R., Zhang, Y., Wallace, J.M., Francis, R.C.,
1997. A Paci®c decadal climate oscillation with impacts on
salmon. Bulletin of the American Meteorological Society 78,
1069±1079.
Murphy, J., 1999. An evaluation of statistical and dynamical tech-
niques for downscaling local climate. Journal of Climate 12,
2256±2284.
Quick, M.C., 1995. The UBC watershed model. In: Singh, V.P.
(Ed.). Computer Models of Watershed Hydrology, Water
Resource Publication, Highlands Ranch, Colorado ISBN 0-
918334-91-8, pp. 1130.
Quick, M.C., Pipes, A., 1976a. A combined snowmelt and rainfall
runoff model. Canadian Journal of Civil Engineering 3 (3), 449±
460.
Quick, M.C., Pipes, A., 1976b. Non-linear channel routing by
computer. Journal of Hydrology Division, ASCE 101 (HY6),
651± 666.
Rand, P., Hinch, S., 1998. Swim speeds and energy use of up-river
migrating sockeye salmon: simulating metabolic power and
assessing risk of energy depletion. Canadian Journal of Fisheries
and Aquatic Sciences 55 (8), 1832± 1841.
Servizi, J.A., Jensen, J.O.T., 1977. Resistance of adult sockeye
salmon to acute thermal shock. International Paci®c Salmon
Fisheries Commission, Progress Report No. 34, Vancouver, 11
pp.
Thompson, R.E., 1981. Oceanography of the British Columbia
coast. Canadian Special Publication of Fisheries Aquatic
Science 56, 291.
US Environmental Protection Agency, 2000. Guidance for Data
Quality Assessment: Practical Methods for Data Analysis
EPA QA/G-9 QA00 Update, Of®ce of Environmental Informa-
tion, Washington.
Wall, F.J., 1986. Statistical Data Analysis Handbook. McGraw-Hill,
New York.
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244 243
Wilby, R.L., Wigley, T.M.L., 1997. Downscaling general circula-
tion model output: a review of methods and limitations. Progress
in Physical Geography 21 (4), 530± 548.
Wilby, R.L., Wigley, T.M.L., 2000. Precipitation predictors for
downscaling: observed and general circulation model relation-
ships. International Journal of Climatology 20, 641± 661.
Williams, I.V., McAllister, M., MacDonald, J.S., Brown, T.J., 2000.
A spatially explicit cumulative effects model for estimating in-
season impacts of temperature and discharge, migrating Fraser
River sockeye. Proceedings of the Fourth International Confer-
ence on Integrating GIS and Environmental Modelling
(GISEM4): Problems, Prospects and Research Needs, 2±8
September 2000, Banff, Alta. http//www.colorado.edu/
research/cires/banff/upload/149.
J. Morrison et al. / Journal of Hydrology 263 (2002) 230 ±244244
