California Applications
Program
Experimental Forecast of Maximum Daily Snowmelt
Discharge
By D.H. Peterson, D.R. Cayan, R.E. Smith, M.D. Dettinger and
L. Riddle
Retrospective Appraisal of the 2000 Maximum Flow Forecasts
As of September 2000, the maximum flows during Spring-Summer
2000 have been measured and compared to the April 1 forecasts of same.
Forecast errors were distributed as follows:
River
|
Forecast error
(in std. devs.)
|
Carson |
-0.2 |
Kern |
-0.0 |
Kings |
-0.1 |
Gunnison |
-1.3 |
Merced |
+0.6 |
San Joaquin |
n/a (not available in 2000) |
Walker |
+1.2 |
Weber |
-0.1 |
Yellowstone |
-0.6 |
These errors are generally within the range of errors that
were expected, given the error bars reported in Table
1. For example, the largest forecast errors were about plus and minus
1.25 standard deviations. In a strictly normal distribution, such errors
would be expected 20% of the time, or 1.6 times in the sample size used
here; in fact, such errors occurred 2 times among the sample of rivers
forecasted here. The figure below compares the expected and realized
distributions of forecast errors in more detail.
For the most part, the expected and realized errors are plot
close to the one-to-one line (are nearly equal). Deviations from a perfect
fit mostly reflect a more-than-random tendency for this year's forecast
errors to be near zero, giving the plot a flattening around errors of about
0 standard deviations. Thus, on the whole, the forecast performed somewhat
better than would be expected from a simple analysis of the regression
statistics that underlay it.
April 2000 Forecast
Experimental
forecasts of the maximum daily discharges between April 14 and July 13,
2000, are presented in Table 1, for high-elevation watersheds along the
central and southern Sierra Nevada of California and in several other river
basins of the western United States. The forecasts are based on representative
April 1, 2000, snowpack conditions, and harness surprisingly strong statistical
relations between such initial snowpack conditions and the peak flows that
have ultimately developed during the course of historical snowmelt seasons
in these rivers. The background, methods, data, forecast results, and the
opportunities afforded by the statistical relations identifed are discussed
in subsequent sections of this forecast.
TABLE 1. Maximum
Daily Discharge Forecasts for late spring-early summer 2000.
RIVER
|
OBSERVATION
April 1 Snow, in
inches
(percent of normal)
|
REGRESSION EQUATION\1
|
PREDICTION
Spring-Summer 2000, in
cubic meters per second
± 2 std (percent of normal)
|
MEAN
Qmax
(cms)
|
MEAN
DAY OF
Qmax
|
CARSON\2
|
77.5 (91%)
|
Qd
= 0.24(d) -2.9
|
15 ± 9.5 (94%)
|
16 |
130 |
GUNNISON
|
17.0 (106%)
|
Qs
= 0.77 (d) -2.2
|
109 ± 68 (108%)
|
101 |
158 |
KERN\2
|
100.6 (93%)
|
Qd
= 1.25(d) -55
|
71 ± 66 (90%)
|
79 |
148 |
KINGS
|
77.6 (98%)
|
Qd
= 0.90(d) -25.7
|
44 ± 35 (86%)
|
51 |
139 |
MERCED
|
93.6 (89%)
|
Qd
= 0.55(d) +9.3
|
61± 33 (91%)
|
67 |
148 |
SAN JOAQUIN
|
88.3 (91%)
|
Qd
= 0.88(d) +13.1
|
91± 44 (93%)
|
98 |
149 |
WALKER
|
58.0 (95%)
|
Qd
= 0.62(d) +5.1
|
42 ± 21 (98%)
|
43 |
152 |
WEBER
|
13.1 (79%)
|
Qs
= 2.44(s) +6.0
|
38 ± 32 (81%)
|
47 |
154 |
CLARKS FORK
YELLOWSTONE
|
32.1 (90%)
|
Qs
= 3.75(s) +85
|
205 ± 99 (94%)
|
218 |
161 |
\1
Qd is based on snow depth; Qs is based on snow water equivalent.
\2
Based on a proxy snowpack measurement from a nearby watershed.
Background, Data and Methods
An experimental
forecast of maximum daily snowmelt discharge is given for upper elevation
watersheds mostly along the central/southern Sierra and in several other
river basins of the western United States (Fig. 1). Although the number
of basins addressed by this study is limited, other high elevation watersheds
with small spring- rain totals would probably yield similar results. Minimal
spring rains are important because the statistical method used herein does
not account for rainfall contributions to snowmelt discharge. The forecast
window in this study is limited to calendar days 105 (April 14, 2000) to
195 (July 13, 2000) to avoid most rain-on-snow events.
Figure 1. Study Location
Snowmelt
hydrology is an important area of research in western United States (Cayan,
1996). Snowpack is not only a major resource of water supply in western
United States; it provides a winter storage mechanism, holding winter precipitation
in the basins until snowmelt in spring and summer, providing additional
options for water managers. Loss of snowpack as a result of global warming
is also a concern (c.f. Lettenmair and Gan, 1990; Jeton et al., 1996),
especially since there has been a disconcerting trend in the west for more
of the winter-early spring precipitation to arrive in the form of rain
rather than snow at intermediate elevations in recent decades (Roos, 1987;
Dettinger and Cayan, 1995). Other phenomena relevant to snowmelt forecasting
are the spring pulse (Cayan et.al, 1999), ENSO (Dettinger, Cayan and Redmond,
1999) and large-scale correlations of snowmelt-discharge sequences from
high elevation watersheds across the western states (Peterson, et.al. 2000).
In this
study, snowpack measurements are analyzed to predict the maximum daily
snowmelt discharge attained during the subsequent snowmelt season. Initial
snowpack measurements have long been used, and are essential, for forecasting
upcoming seasonal water yields in the western basins (Serreze, Clark and
Armstrong, 1999). To the best of our knowledge, however, such measurements
have not been used to forecast maximum daily discharges for the upcoming
snowmelt season. The snowmelt process is nonlinear (Leavesley et.al. 1983)
and it was surprising to find that maximum daily discharges during snowmelt
seasons increase approximately with initial snowpack depth or snow water
equivalent (measured on or near April 1) in many western rivers.
The forecasts
are based on linear regressions of historical snowpack and discharge measurements.
April 1 snowpack observations (from historical snow course measurements)
are used here as the initial conditions for the forecasts. The snow stations
are listed in Table 2; this year's snowpack measurements (from late March-early
April, 2000) are the second column in Table 1. Details of the river discharge
gage locations are in Table 3. Maximum daily discharges were identified
from among calendar days 105 to 195 in each year of record and the resulting
series was regressed against various measures of initial snowpack conditions
to develop the forecast equations.
TABLE 2. Snow Stations \1
RIVER
BASIN
|
SNOW STATION
|
START OF RECORD
|
CARSON\2
|
#106 upper Carson Pass
(American River)
|
1939
|
GUNNISON
|
Snowmelt ID 06L035
|
1979
|
KERN\2
|
#205 Mammoth Pass, DWR
(Owens River)
|
1929
|
KINGS
|
#227 Woodchuck Meadow
|
1930
|
MERCED
|
#176 Snow Flat
|
1930
|
SAN JOAQUIN
|
#190 Kaiser Pass
|
1930
|
WALKER
|
#152 Sonora Pass
|
1947
|
WEBER
|
Snowtel ID 11J025
|
1970
|
YELLOWSTONE
|
Snowtel ID 09D065
|
1970
|
\1 Snowtel
sites measured as snow water equivalent; all others as snow depth.
\2 Based
on a proxy snowpack from a near-by watershed.
TABLE
3. Snowmelt River Discharge Stations
Station
Name
|
Number
|
Longitude
|
Latitude
|
Elevation
|
Area
|
Mean
Flow (cms)
|
Distance\1
(km)
|
WF
CARSON R AT WOOD
|
10310000
|
119.8319
|
38.7694
|
1754
|
169
|
8.21
|
119
|
GUNNISON
RIVER NEAR
|
09114500
|
106.9514
|
38.5411
|
2333
|
2621
|
21.75
|
1097
|
KERN
R NR KERNVILLE
|
11186000
|
118.4767
|
35.9453
|
1103
|
2191
|
12.63
|
224
|
NF
KINGS R BL DINKEY
|
11218400
|
119.1278
|
36.8797
|
315
|
1002
|
9.06
|
104
|
MERCED
R AT HAPPY IS
|
11264500
|
1195578
|
37.5578
|
1224
|
469
|
10.05
|
0
|
SAN
JOAQUIN AT MIL
|
11226500
|
119.1964
|
37.5105
|
1393
|
645
|
16.62
|
40
|
W WALKER
R BL L WALK
|
10296000
|
119.4492
|
38.3797
|
2009
|
469
|
7.48
|
73
|
WEBER
RIVER NEAR OAK
|
10128500
|
111.2458
|
40.7361
|
2024
|
420
|
6.23
|
786
|
CLARKS
FORK YELLOWSTONE
|
06207500
|
109.0667
|
45.0111
|
1215
|
2989
|
26.65
|
1190
|
\1
from the Merced River, Happy Isles.
Correlations
between snowpack depth or snow water equivalent and maximum daily discharges
for selected rivers are listed in Table 4. The linear regression equations
and forecasts are reported in Table 1; regression uncertainties listed
are the global 95% confidence intervals (± 2 standard deviations,
Jones, 1996, polytool p. 2 - 168).
TABLE
4. CORRELATION COEFFICIENTS
|
MAXIMUM DAILY DISCHARGE VS.
|
SNOW WATER
EQUIVALENT
|
|
|
RIVER
|
INDEX\1
OF SNOW PACK
|
DEPTH\2
OF SNOW
|
SNOW WATER EQUIVALENT\3
|
DAY OF MAX
DISCHARGE |
VS.
DAY OF MAX
DISCHARGE
|
YEARS FOR CORRELATION
|
COMMENTS
|
CARSON\2
|
0.94
|
0.86
|
0.86
|
0.46 |
0.37
|
1939-1995
|
1992 missing
|
GUNNISON
|
0.96
|
n.d
|
0.67
|
0.11 |
0.05
|
1979-1997
|
1984 missing
|
KERN\2
|
0.98
|
0.88
|
0.89
|
0.38 |
0.13
|
1961-1996
|
1970, 1981
missing
|
KINGS\4
|
0.98
|
0.92
|
0.95
|
0.33 |
0.28
|
1961-1995
|
1965,1967,1997,
1982 missing
|
MERCED
|
0.91
|
0.77
|
0.87
|
0.40 |
0.22
|
1932-1977
|
1937,1970,1981,
1983 missing
|
SAN JOAQUIN
|
0.95
|
0.87
|
0.89
|
0.53 |
0.42
|
1952-1990
|
|
WALKER
|
0.93
|
0.82
|
0.83
|
0.59 |
0.49
|
1947-1994
|
1982 missing
|
WEBER
|
0.81
|
n.d.
|
0.49
|
0.61 |
0.19
|
1979-1997
|
Includes large rain
on snow event 1990
|
YELLOWSTONE
|
0.83
|
n.d.
|
0.63
|
0.51 |
0.44
|
1970-1997
|
Includes large rain
on snow event 1981
|
\1 The sum
of daily discharge over calendar days 105 to 195.
\2 Snow depth
in inches.
\3 Snow water
equivalent in inches.
\4 A large
rain on snow event 1996 was not included in the calculations; see text,
figs. 2 and 3.
Forecast Results
The
correlations that justify the forecasts are listed in Table 4, and the
maximum daily discharge forecast for April - July 2000 is the fourth column
in Table 1. The year 2000 snowpack depths (Table 1, column 2) range from
79 to 106% of their average value. Thus the forecasts of maximum
daily discharges are close to what would be expected in years with slightly
less than average snowpack; that is, slightly less than average maximum
flows are predicted for spring 2000.
As
expected, the correlation between maximum daily discharge and initial snowpack
improves when using snow water content rather than snow depth (Table 4,
column 4, snow water content, vs. column 3, snow depth). Correlations between
the April 1 snow index and maximum discharge over the available record
for 9 western stream gages are quite high, ranging from 0.83 (Yellowstone
River) to 0.98 (Kings River). These high correlations (Figure 2) indicate
that the relationshis between prior snow pack and maximum flow are surprisingly
linear. Interestingly, most of these rivers also exhibit modest correlations
between the snow index and the timing of the maximum flow.
Figure 2. Maximum daily snowmelt discharge
versus initial snowpack water equivalent, Kings River.
Outlook
This
forecast effort is work in progress, to be expanded to other alpine snowmelt
watersheds in the near future. The forecasts can be degraded by rain on
snow, and may suffer from using only one snowpack site each to represent
watershed-average snowpack conditions. For an example of the first,
note the extreme outlier at the top of Figure 2. This represents the maximum
daily discharge during spring 1996 (Fig. 3). A major rain on snow event
(2.5 inches in three days in mid-May) yielded this unusually sharp peak
in snowmelt/discharge variation and could not have been predicted from
initial snowpack conditions alone.
Figure 3. Daily discharge Kings River, 1996.
The
remarkable correlations between initial snowpack conditions and maximum
daily flows (i.e., Fig. 2) may provide opportunities for new tests of existing
snowmelt models (physically and statistically based). Forecasts, such as
in this study, may also serve in constraining other more complex weather-scale
forecast efforts. For example, early experiments using a Kalman filter
scheme with daily forecast temperature as input and daily snowmelt discharge
as output shows a relatively high skills in predicting the timing of short-term
discharge fluctuations but less skill in predicting the amplitude of those
fluctuations. Interestingly, the skills of the forecasts of maximum
daily discharges presented here are generally the reverse; strong correlations
are found for amplitude of the discharge peaks and weak correlations with
timing are shown in Table 4, column 5.
Further
studies are needed to determine why some watersheds are more predictable
than others. Perhaps more fundamental is why the relation is even remotely
linear. One factor may be the near-linear rise in temperature over the
snowmelt season (chosen here as days 105-195). More importantly, the snowmelt
season is over before the seasonal temperature starts to decline (c.f.
Fig. 4). In essence, the system has an excess capacity to melt snow. If,
however, the peak snowmelt discharge more closely matched the peak air
temperature timing (or solar insolation) a snow carryover could be expected
in some years. Then the snowpack-peak flow relation would be expected to
be nonlinear, at best. This nonlinear effect due to the close energy-balance
linkages between air temperature and snowmelt could be tested using physically
based models (Jeton, Dettinger and Smith 1996) as well as by selecting
snowpack observations from basins at higher altitudes (and lower air temperatures).
Figure 4. Seasonal climatology of air temperature
and discharge, Merced River, Happy Isles. Low-pass mean daily observations
using a 15-day boxcar filter (applied forward and backward)
REFERENCES
Cayan,
D.R., 1996. Climate variability and snowpack in the western U.S. J. of
Clim., 9(5), 928-948.
Cayan,
D.R., Peterson, D.H., Riddle, L., Dettinger, M.D., and Smith R., 1999.
The spring runoff pulse from the Sierra Nevada. Preprints, American Meteorological
Society's 14th Conference on Hydrology, Dallas, January 1999,
77-79.
Dettinger,
M.D. and D.R. Cayan, 1995. Large-scale atmospheric forcing of recent trends
toward early snowmelt runoff in California. J. of Clim., 8, 606-623.
Dettinger,
M.D., Cayan, D.R. and Redmond, K.T., 1999. United States Streamflow Probabilities
based on Forecasted La Niña, Winter-Spring 2000, Experimental Long-Lead
Forecast Bulletin, 8(4): 57-59.
Jeton,
A.E., M.D. Dettinger, and J.L. Smith, 1996. Potential Effects of Climate
Change on Streamflow Eastern and Western Slopes of the Sierra Nevada, California
and Nevada. U.S. Geological Survey WRI Report 95-4260, 44 pp.
Jones,
B. 1996. Statistics Toolbox for use with Matlab, The Math Works, Inc. Natick,
Mass., 348 pp.
Leavesley,
G.H., R.W. Lichty, B.M. Troutman, and L.G. Saindon, 1983. Precipitation
- Runoff Modeling System - Users Manual.U.S. Geological Survey, Water Resources
Investigations Report 83-4238, 207 pp.
Lettenmaier,
D.P. and T.Y. Gan, 1990. Hydrologic Sensitivities of the Sacramento-San
Joaquin River Basin, California, to Global Warming.Water Resources Research
26(1): 69-86.
Peterson,
D.H., Smith, R.E., Dettinger, M.D., Cayan, D.R., and Riddle, L., in press.
An organized signal in snowmelt runoff over the western United States,
JAWRA, 36(2) pp 1-12.
Roos,
M., 1991. A trend of decreasing snowmelt runoff in northern California.Proceedings,
59th Western Snow Conference, Juneau, AK, 29-36.
Serreze,
M.C., M.P. Clark, and R.L. Armstrong, 1999. Characteristics of the Western
United States Snowpack from Snow Telemetry (SNOTEL) Data. Water Resources
Research 35(7): 2145-2160.