Use of Long-Lead Streamflow Forecasts to Improve Columbia River Reservoir Management
JISAO Climate Impacts Group
Department of Civil and Environmental Engineering
Alan F. Hamlet
Dennis P. Lettenmaier
Draft 10/27/99
Abstract
Recent advances in long-lead climate forecasts have made experimental streamflow forecasts developed by the
JISAO Climate Impacts Group (CIG) and the Department of Civil Engineering at the University of Washington
available roughly six months earlier than current operational forecasts that rely on snow pack measurements.
These forecasts provide apparent opportunities to make the Columbia Basin reservoir operating system more
directly responsive to interannual climate variability in the early fall and winter months. In the current operating
system, the so called “critical” and “assured refill” rule curves that restrict releases for hydropower in the period
from August to December are based on the critical and third lowest flow sequence respectively. These rule
curves provide appropriate protection of energy capacity and reservoir refill in low flow conditions, but are
unnecessarily restrictive in normal and high flow years. A new rule curve called the “refill to least flood” curve
has been proposed by the authors for the major storage dams to make effective use of extended streamflow
forecasts and provide more management flexibility in the fall and early winter while protecting against low flow
conditions when appropriate.
Recent Improvements in Forecasting Streamflow
Figure 1 and Table 1 show an ensemble forecast of natural streamflow and corresponding summer runoff
volumes for the Columbia River at The Dalles, Oregon for water year 1999. The forecast was made available on
an experimental basis June 1, 1998, based on the 1998-1999 ENSO forecast for La Niña conditions, and heuristic
methods for determining the phase of the PDO in real time (Hamlet and Lettenmaier, 1998). This type of longrange ensemble forecast shows promise for improving reservoir operations in the Columbia River Basin.
C lim a te C a te g o ry 6 F o re c a s t o f V irg in F lo w a t
T h e D a lle s fo r W a te r Y e a r 1 9 9 9
800000
F o re c a s t1
F o re c a s t2
700000
Stream flo w (cfs)
F o re c a s t3
600000
F o re c a s t4
F o re c a s t5
500000
F o re c a s t6
400000
F o re c a s t7
300000
F o re c a s t8
F o re c a s t9
200000
F o re c a s t1 0
100000
F o re c a s t1 1
F o re c a s t1 2
M o n th
sep
au g
ju l
ju n
m ay
ap r
m ar
jan
feb
d ec
oct
n ov
0
H ig h C lim a to lo g y
L o w C lim a to lo g y
O b s e rve d V irg in F lo w
Figure 1 Long-range ensemble streamflow forecast for water year 1999 (PDO cold/ENSO cold) [Light gray lines
are forecast ensemble members, heavy black lines are the observed minimum and maximum streamflow for each
month for 1948-1988]
The forecasting scheme is implemented using a macro-scale hydrology model for the Columbia Basin (Figure 2)
and observed meteorological driving data selected from the historic record. A schematic of the forecasting
method is shown in Figure 3. On about June 1 an ENSO forecast is typically available for the coming winter.
The phase of the PDO generally persists from the preceding year, and heuristic methods based on the occurrence
of extreme high or low flow events in preceding years are used to identify regime shifts in the PDO in an
objective manner (Hamlet and Lettenmaier, 1998). Given the assumed phase of the PDO and the ENSO forecast
for the coming winter, precipitation and temperature data from water years associated with these climate
conditions are selected from the historic record.
Percentile
Forecast 1999 Average
Forecast 1999 Volume
Streamflow at the
Runoff at The Dalles
Dalles Apr-Aug
Apr-Aug.
(kcfs)
(million acre-ft)
90th
439
133
80th
430
130
70th
406
123
60th
393
119
50th
380
115
40th
379
115
30th
341
103
20th
318
97
10th
299
91
Table 1 Percentile Flows and Runoff Volumes at The Dalles 1999 Water Year Ensemble Forecast
The years associated with each climate category are based on retrospective definitions of climate state, using the
Nino3.4 and PDO climate indices (winter averages more than 0.5 standard deviations from the mean in each
case). Nino3.4 is a numerical measure of sea surface temperatures in the tropical Pacific which may be used to
identify and categorize ENSO events (Trenberth, 1997). The PDO (as it is used here) is a decadal-scale climate
phenomenon with a characteristic period of about 50 years and bimodal, epochal events typically lasting 20 to 25
years (Mantua et al, 1997). The two most recent epochs for the PDO were cold phase from 1948-1976, and
warm phase from 1977-1996(?). There is evidence in the streamflow record that there may have been a PDO
regime shift in 1996 or 1997.
Having established the climate forecast, the hydrology model is then initialized using the meteorological data
from the previous water year, and each of the meteorological sequences associated with the climate state forecast
for the coming water year are used to simulate the streamflow, producing an ensemble forecast for the coming
year (Figure 1). Post processing then produces summaries like those shown in Table 1. The grid-based
hydrologic model is capable of resolving a large number of river locations in the Columbia basin, and a similar
streamflow forecast can be provided at each of these points.
The main advantage of the method is that it has demonstrable forecast skill for streamflow early in the water year
(fall and early winter) not present in currently used streamflow forecasting methods based on snowpack
observations, which are not reliable before January.
Precipitation, Temperature, Wind, Humidity
Evapotransporation
(canopy and soil layers)
Snow Accumulation
and Melt
Vegetation and Soil
Characteristics
Runoff
Routing Model
Surface Infiltration
Base Flow
Inter-layer Infiltration
Figure 2 Schematic of grid-based macro-scale hydrology model
2
Select Met Data Ensemble from
Historic Record Associated with
Forecast Climate Category
ENSO
Long-Lead
Climate
Forecast
Run Initialized
Hydrologic Model
PDO
Ensemble Streamflow
Forecast
Figure 3 Schematic of Forecasting Method
Current Use of Forecasts in Columbia Basin Operations
The reservoir operating system for the Columbia River Basin has evolved to make use of streamflow forecasting
techniques that become available on January 1 based on observed snowpack and statistical relationships between
snowpack and spring and summer streamflow. These forecasts are updated monthly until July. As the snow
accumulation season progresses, the forecasts of spring runoff become increasingly accurate until about April or
May as knowledge of the total snow water equivalent that will contribute to runoff during the spring melt
improves (Koch and Buller, 1993). As the snow melts through the summer, the subsequent streamflow becomes
progressively less dependent on the accumulated snow from the previous winter season, and the forecast
accuracy declines until by August there is little prediction skill. The forecasting method described above
essentially extends the lead time for predicting spring and summer streamflow by approximately six months.
These extended forecasts are most useful early in the water year, since subsequent forecasts based on existing
methods would be preferred once observations of winter snowpack become available.
The current reservoir operating system uses streamflow forecasts to determine flood evacuation requirements
and reservoir "assured refill" rule curves in the period from January-July. Figure 4, for example, shows a flood
storage evacuation diagram for Libby Dam that is typical for the major federal storage projects in the basin.
Until January, the flood evacuation targets are based on fixed values that are independent of forecast
information. Starting in January, forecasts of spring runoff volumes determine greater or lesser evacuation
requirements based on estimates of total spring runoff at The Dalles or at the individual project. Similarly,
reservoir "assured refill" rule curves (Figure 5) are based, from August-December, on conservative assumptions
of reservoir inflow (1931 inflow sequence) designed to assure a high likelihood of reservoir refill in summer.
The so-called "critical" rule curves are constructed using a reservoir model and a critical period analysis in
conjunction with firm energy requirements reported by participants in the Pacific Northwest Coordination
Agreement. The current critical period is a streamflow sequence that occurred in 1936-1937.
6000000
Storage (acre-ft)
5000000
Rule Vol 1
4000000
Rule Vol 2
Rule Vol 3
Rule Vol 4
3000000
Rule Vol 5
2000000
1000000
0
1
2
3
4
5
6
7
8
9
10 11 12 13
Month (August=1)
3
Figure 4 Flood evacuation targets for Libby Dam
6000000
Storage (acre-ft)
5000000
4000000
3000000
2000000
Critical
Assured Refill
1000000
JUL
JUN
MAY
APRIL
FEB
MAR
JAN
DEC
OCT
NOV
SEP
AUG
0
Month
Figure 5 Examples of the critical and assured refill curve for Libby Dam
Together these three rule curves determine draft limits for hydropower production allowed at different times of
the year. In simple terms, the reservoirs are not permitted to draft below either the critical curves or assured
refill curves (whichever is more restrictive) for firm energy production unless the flood curve is below these
curves or all reservoirs do so in an balanced manner. Non-firm energy production is not permitted to draw the
reservoirs below the assured refill curve or the critical curve unless the flood curve is below this curve. At some
dams there are additional restrictions requiring the reservoir to be at but not below flood pool in spring. Starting
in January, the assured refill curves are based on forecast inflows, which in all but a few years are less restrictive
than those based on the 1931 inflows. Similarly, use of the critical curves to limit hydropower production in the
fall are very cautious, because only one extreme low flow sequence in the record justifies the maintenance of
such high levels of fall storage to protect firm energy production.
Modifications to the Current Operating System to Take Advantage of Long-Lead Forecasts
Long-lead streamflow forecasts can be used effectively for long-range planning/marketing even within the
constraints of current reservoir operating practice, however there also appear to be economically important
opportunities to change reservoir rule curves and relax certain operational constraints to better utilize long-term
forecasts, particularly for fall and early winter hydropower production.
Conceptual Framework of a Revised Operating Plan
In years with above average or high streamflow (e.g. water year 1999, a cold-phase PDO/La Niña year), the
current operating plan is unnecessarily restrictive in terms of reservoir draw down for hydropower generation in
the fall and early winter (October-December). Refill curves based on the 1931 inflow sequence are more
restrictive than those that would be constructed using even the lowest ensemble member (Figure 6), and much
higher streamflows are much more likely. Similarly, the critical curves, while appropriately restricting fall and
winter energy production during a critical streamflow sequence, are unnecessarily restrictive if relatively high
spring and summer streamflow volumes are expected and the approximate flood evacuation requirements in the
spring are known ahead of time. Most importantly, the flood evacuation that is likely to be required in spring in
years with above average or high streamflow tends to completely overwhelm any other considerations as soon as
volume forecasts are made available on January 1. If, based on long-lead forecasts, flood evacuation
requirements are likely to heavily draft reservoirs in the spring, then any restrictions on fall and winter releases
of storage should more appropriately be based on these expected flood control practices, rather than on a critical
period analysis.
4
6000000
Storage (acre ft)
5000000
4000000
3000000
2000000
Low Ensemble Assured Refill
1000000
Status Quo Assured refill
Refill to least flood
0
oct
nov
dec
jan
feb
mar
apr
may
jun
jul
Month
Figure 6 Assured refill curve for Libby Dam constructed using 1931 inflows, assured refill curve constructed
using the lowest ensemble flow sequence forecast for 1999, and the refill to least flood curve for 1999
The long-lead forecasting method described above is most successful at predicting an ensemble of spring and
summer streamflow volumes. These spring and summer volume estimates in turn determine the probability of
certain flood rule curves becoming active in spring. The ensemble forecasts can be used to create a set of refill
curves leading to the March or April flood evacuation targets for each dam (Figure 6). These in turn determine
how much storage is available to be released from October-April for energy production. To ensure maximum
reliability of being at but not below flood pool (which is a spring objective at certain projects), the refill curve
based on the lowest ensemble member could be used to restrict hydropower production until the January 1
forecasts are reported, at which time the actual flood curves would be selected in the current manner. In wet
years, this new rule curve, which will be referred to as the "refill to least flood" rule curve, is much less
restrictive than is the current combination of critical and assured refill curves, as will be shown below.
In low flow years there are advantages to this system as well, although the benefits are not necessarily to
hydropower production. Using the lowest ensemble member to define a refill curve to a low flood evacuation
target may provide more robust protection against drought than using the current operating system by being
more restrictive than the current plan and ensuring a high likelihood of reservoir refill under the worst
conditions. This is important for the maintenance of fisheries flow targets in drought conditions, for example,
due to the very limited storage that is allocated for this purpose.
A Revised Reservoir Operating Plan to Make Use of Long-Lead Forecasts
This experimental operating system is designed to provide a revised Energy Content Curve in October,
November, and December, using a conservative estimate of flood pool in April or March based on forecast
spring and summer inflow volumes. To accomplish this, the lowest member of the streamflow forecast
ensemble of inflows to each major storage dam in the basin is used to estimate a minimum flood evacuation
requirement. The inflow sequence defined by the lowest ensemble member is then used to create a refill curve
that would bring the reservoir level to this low estimate of flood control storage requirements given the lowest
estimate of inflow (Figure 6). This "refill to least flood" curve would then be used directly in conjunction with
the fixed flood control rule curve to create a revised Energy Content Curve in the period from OctoberDecember. The critical curve would not be used in this system.
Some may question the elimination of the critical curve in fall. The critical curve, while enforcing restraint in
the fall before current forecasts are available, is not a theoretical requirement for successful operation of the
system when forecast information is available earlier in the year. In the current system, releases of storage for
energy production could potentially overdraft the system in fall in what might prove to be a low-flow year, thus
threatening firm energy production in the rest of the year. The critical curve was included in the operating
system to prevent this from occurring. However, if all dams are operated according to the revised rule curve, the
critical curve serves no useful purpose, since as long as the refill to least flood curves for October-April are used
for planning in the fall and early winter, dam releases early in the season will not threaten energy production
capacity later in the year. In dry forecast years the refill to least flood curve is even more restrictive than the
critical curve (which means the status quo flood curves typically define fall draft limits), and in wet years the fall
5
constraints are appropriately relaxed since the system is not in a critical situation and abundant streamflow is
expected later in the year.
An Example for Water Year 1999 and 1992 for Libby Dam
The forecast April-August volume forecasts for Libby dam for water year 1999 are shown in Table 2. Taking
the lowest ensemble member of 5.94 million acre-ft specifies a flood evacuation requirement of about 4 million
acre-ft (the second curve from the bottom in Figure 3) by March 15. This is assumed to be the least amount of
flood storage likely to be needed for the coming year since it is based on the lowest ensemble member. Then the
"refill to least flood" curve is assembled using the inflow sequence defined by the same lowest ensemble
member. A 4000 cfs minimum outflow from the dam is assumed. This curve is then used in conjunction with
fixed flood evacuation requirements to define a new ECC. Figure 7 compares the new ECC based on the refill to
least flood curve to the fall ECC for the status quo. Note the considerable relaxation of constraints on use of
storage in the fall.
April-August Total
Volume Forecast
(million acre-ft)
max
11.8
95
11.7
90
11.6
80
11.0
70
10.6
60
10.2
50
9.2
40
8.4
30
8.4
20
7.8
10
7.2
5
6.6
min
5.9
Table 2 April-August forecast inflow volume for Libby dam for water year 1999
Libby Storage (acre-ft)
Percentile
5000000
4500000
4000000
3500000
3000000
2500000
2000000
1500000
1000000
500000
0
Status quo fall ECC
Revised fall ECC
oct
nov
dec
Month
Figure 7 Refill to least flood rule curve for water year 1999 compared to fall ECC for the status quo
Table 3 and Figure 8 show the results of repeating the same rule curve construction procedure for water year
1992 (warm-phase PDO/El Niño) using a retrospective forecast (see Hamlet and Lettenmaier, 1998 for more
details). Because of the increased likelihood of very low flow conditions contained in the forecast, the refill to
least flood curve is even more restrictive than the status quo assured refill curve and critical curve in this case.
6
As a result the fixed December flood control requirements predominantly define reservoir draw down at Libby in
this case, much as they do for the status quo. If an unusually low flow year were encountered, these fixed flood
targets might actually prevent reservoir refill in spring. Following the refill to least flood curve until January,
however, would ensure reservoir refill to full pool even in the most adverse inflow sequence expected.
April-August Total
Volume Forecast
(million acre-ft)
max
10.5
95
10.5
90
10.5
80
10.5
70
10.0
60
9.6
50
8.3
40
6.9
30
6.8
20
6.6
10
4.9
5
4.0
min
3.2
Table 3 April-August forecast inflow volumes for Libby dam for water year 1999
Libby Storage (acre-ft)
Percentiles
5000000
4500000
4000000
3500000
3000000
2500000
2000000
1500000
1000000
500000
0
Status quo fall ECC
Revised fall ECC
oct
nov
dec
Month
Figure 8 Revised fall ECC for water year 1992 compared to fall ECC for the status quo
Opportunities for Improved Water Resources Management Based on Long-Lead Forecasts and Revised
Operating System
The potential complexities of hydropower marketing and operation in what may become a deregulated energy
market are considerable, and are beyond the scope of this paper. However, it would appear that having the
flexibility to market more hydropower in the fall months in years when water is likely to be plentiful in the
spring and summer, and to have knowledge in June regarding the available water for fall, winter, and spring
energy production is valuable in terms of marketing and supplying more hydropower in wet years (with marginal
cost benefits over conventional energy production), and in terms of having more flexibility to provide variable
amounts of energy throughout the fall, winter, and spring according to short-term marketing opportunities.
There is also a need within the system to achieve a more well-defined balance between the various system uses
in the face of normal climate variability. This is particularly true with regard to tradeoffs between winter
hydropower production and maintenance of streamflow and storage in spring and summer for salmon protection,
irrigation, recreation, etc. These diverse uses of water have conflicting objectives for the accumulation and
7
release of storage. The revised operating plan proposed above eliminates unnecessary restrictions on releases of
fall and winter storage for energy production when spring and summer streamflow volumes are likely to be high
(ensuring high reliability of water uses that need summer streamflow and storage), while protecting uses of water
that are sensitive to summer drought by restricting fall and winter releases when long-lead forecasts show a
significant likelihood of low-flow conditions.
Technical opportunities for revising fall flood control rule curves based on long-lead forecasts would also appear
to exist, however the political, institutional, and social obstacles to changes of this kind may overwhelm any
potential benefits. These potential changes would have little effect in wet years, but might help to mitigate
conflicts between various uses in summer in low flow years by ensuring greater reliability of reservoir refill (e.g.
by staying above the refill to least flood curve until January) (See Hamlet and Lettenmaier, 1998 for more
information on potential flood control modifications).
Conclusions
With the availability of improved streamflow forecasting methods that can provide an ensemble forecast for
spring and summer streamflow in June preceding the water year, useful changes in the reservoir operating
system for the Columbia are possible that could take advantage of this forecast information. A revised refill rule
curve called the "refill to least flood" rule curve could replace the use of the assured refill curve and critical
curve used in the current operating system in fall and early winter. In water years that fall in climate categories
that are typically associated with high streamflow, use of the revised "refill to least flood" rule curve in the fall
months would provide greater flexibility in reservoir releases for fall energy production without threatening
either reservoir refill or firm energy production later in the year. In years forecast as potential low-flow years,
use of the "refill to least flood" rule curve would constrain reservoir draw down to about the same degree as it is
under the status quo. This preliminary fall and early winter rule curve would then be updated when the
streamflow forecasts based on observed snowpack become available on January 1.
8
References:
Barnston, A.G., H.M. Van den Dool, S.E. Zebiak, T.P. Barnett, M. Ji, D.R. Rodenhuis, M.A. Cane, A. Leetmaa,
N.E. Graham, C.R., Ropelewski, V. E. Kousky, E.A. O'Lenic, and R.E. Livezey: 1994, Long-lead seasonal
forecasts---where do we stand?, Bull. Am. Met. Soc., 75, 2097-2114.
Battisti, D.S. and E. Sarachik, 1995: Understanding and Predicting ENSO. Review of Geophysics, 33, 1367-76
Garen, D.C., 1992, Improved Techniques in Regression-Based Streamflow Volume Forecasting, J. Water
Resources, Planning and Mgmt. Div., ASCE, 118(6), pp. 654-670
Garen, D.C., 1998, ENSO Indicators and Long-Range Climate Forecasts: Usage in Seasonal Streamflow
Volume Forecasting in the Western United States, Session H22E-03, AGU Conference, Fall
Hamlet, A. F., Lettenmaier, D. P., 1998: Effects to Columbia Basin Water Resources Associated with Climate
Variability and Operating System Design, Summary of Hydrology Results, Year 3 Report, JISAO Climate
Impacts Group, University of Washington.
Hamlet, A. F., Lettenmaier, D. P., 1998: Columbia River Streamflow Forecasting Based on ENSO and PDO
Climate Signals, ASCE Journal of Water Res. Planning and Mngmt, (in review)
Latif, M., T.P. Barnett, M.A. Cane, M. Flugel, N.E. Graham, H. von Storch, J.-S. Xu, and S.E. Zebiak, 1994,A
Review of ENSO Prediction Studies, Climate Dynamics, 9, 167-179
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, J. Geophys. Res., 99, D7, pp14,415-14,428
Koch, R. W., Buller, D., 1993, Forecasting Seasonal Streamflow: Columbia River at The Dalles, Report to
Bonneville Power Administration, June
Mantua, N., S. Hare, Y. Zhang, J. M. Wallace, R. Francis, 1997, A Pacific Interdecadal Climate Oscillation with
Impacts on Salmon Production, Bulletin of the American Meteorological Society, Vol 78, pp. 1069-1079, June
Mantua, N., 1997,Relationships Between a "Naturalized" Columbia River Flow Record and Large Scale Climate
Variations Over the North Pacific, JISAO Climate Impacts Group, Year-2 Progress Report, University of
Washington, Seattle Washington, July
Nijssen, B., D.P. Lettenmaier, X. Liang, S.W. Wetzel, E. F. Wood, 1997, Streamflow Simulation for
Continental-Scale River Basins, Water Resources Research, 33,4, pp 711, Apr
Piechota, T.C., Dracup, J.A.,1996, Drought and Regional Hydrologic Variations in the United States:
Associations with the El Niño/Southern Oscillation. Water Resources Research, 32(5), 1359-1373.
Piechota, T.C., J.A. Dracup, and R.G. Fovell, 1997. Western U.S. Streamflow
and Atmospheric Circulation Patterns During El Niño-Southern Oscillation
(ENSO). Journal of Hydrology, 201(1-4), 249-271.
Trenberth, K. E. 1997. The definition of El Niño. Bull. Amer. Meteor.Soc. 78: 2771-2777
Twedt, T. M., Schaake, J.C., Peck, E. L., 1977, National Weather Service Extended Streamflow Prediction,
Proceedings, Western Snow Conference, Albuquerque, NM, April
9