Scientific Investigations Report 2006–5274
U.S. GEOLOGICAL SURVEY
Scientific Investigations Report 2006–5274
The precipitation-runoff model was calibrated for a 40-year period, water years 1950–89, and tested for a 7-year period, water years 1990–96 (table 4). The actual model simulations used data for water years 1949–89, but results of the first year were not included in the calibration analysis to allow the model to properly initialize basinwide hydrologic conditions for that year. The goal of the calibration effort was to adjust the initial model parameters so that the simulated monthly unregulated streamflows for Salmon Creek at the location of Conconully Dam (node 6; fig. 11) were more than the estimated UUS and less than the estimated CUS. The model was tested by using the calibrated parameter values to simulate streamflows for a period not used for calibration and then comparing the results for node 6 with the UUS and CUS.
As explained in the “Corrected Unregulated Streamflow” section, CUS includes a correction for ground-water losses from the Salmon Creek Basin and evaporative losses from Conconully Reservoir. The precipitation-runoff model as applied in this study does not simulate ground-water losses but does simulate a fraction of reservoir evaporation up to the amount of precipitation that falls on the reservoir. For example, a mean annual evaporation of 18.9 in. was simulated for Conconully Reservoir for the calibration period, and the long-term mean annual reservoir evaporation is estimated to be 30 in. (table 3). As a result, the model is considered calibrated if simulated streamflows are larger than UUS but smaller than CUS.
During model calibration, model parameters were adjusted from initial values as follows: Parameter jh_coef, which helps determine the rate of potential evapotranspiration as computed with the Jensen-Haise approach (Jensen and Haise, 1963; Mastin and Vaccaro, 2002a), was increased by 25 percent for each month compared to the values used in the Methow River Basin model. This increased the basinwide simulated mean annual actual evapotranspiration for water years 1950–89 by 1.7 in. and generated a simulated mean annual water budget that matched the estimated mean annual water budget. For water years 1950‑89, the basinwide simulated mean annual precipitation and evapotranspiration were 23.1 and 19.1 in., respectively. The high evapotranspiration rates are reasonable given the extensive forests in the Salmon Creek Basin.
The minimum and maximum air-temperature time series for each MRU were lowered by increasing parameters tmin_adj and tmax_adj from 0 to 7°F each month. The lower temperatures resulted in a simulated mean hydrograph that best matched the shape of the UUS/CUS hydrograph for the calibration period. The date when the model starts looking for spring snowmelt was delayed by 20 days to April 20 by changing parameter melt_look from a Julian date of 90 to 110. However, simulated snowmelt still started too early in the runoff season. Therefore, the initial monthly mean minimum and maximum air-temperature lapse rates computed from measured air temperatures for the Omak OMAW Agrimet and Salmon Meadows SNOTEL stations were reduced by 2.5°F per 1,000 ft in March (to -5.58 and -7.32°F per 1,000 ft, respectively) and 1.5°F per 1,000 ft in April (to -4.47 and -6.61°F per 1,000 ft, respectively) to further delay the onset of melting, and the lapse rates were increased by 0.5°F per 1,000 ft in May (to -2.76 and -4.15°F per 1,000 ft, respectively) to increase the rate of snowmelt. The monthly lapse rates were kept constant throughout the model simulations. Finally, parameter freeh2o_cap, the free-water-holding capacity of the snowpack, expressed as a decimal fraction of the snowpack water-equivalent, was reduced 50 percent to 0.025 to help improve the shape of the hydrograph from May through July.
The results of the model calibration are shown in figures 12A and 13A through 13D and are given in table 5. For the calibration period, water years 1950–89, both the simulated mean annual streamflow and the simulated mean April–July streamflow compare well with the estimated values for UUS and CUS. The simulated mean annual streamflow exceeds UUS by 5.9 percent and is less than CUS by 2.7 percent (table 5). Similarly, the simulated mean April–July streamflow exceeds UUS by 1.8 percent and is less than CUS by 3.1 percent. A comparison of the estimated and simulated mean monthly streamflows, however, shows that streamflow is significantly undersimulated during the low-flow, baseflow‑dominated months of November through February when simulated monthly streamflows are as much as 57.2 percent less than UUS (table 5) and significantly oversimulated during August and September when simulated monthly streamflows are as much as 193.6 percent more than CUS (table 5). During the low-flow months, however, estimated mean monthly streamflow is only a small percentage of the estimated mean annual streamflow, about 2 percent in November through February and about 1 percent in August and September (fig. 12A). Therefore, the absolute errors during the low-flow months are relatively small even though the percentages of error are large. For April through July, the estimated mean streamflow is about 84 percent of the estimated mean annual streamflow (fig. 12A). Because the simulated spring snowmelt season starts too early in April and extends too far into July (fig. 12A), the simulated streamflows exceed the estimated streamflows in April and July and are less than the estimated streamflows in May and June (table 5). Because the precipitation-runoff model will be used primarily as a tool to manage spring runoff, model calibration focused on matching estimated and simulated mean April–July streamflow rather than streamflows for individual months.
The precipitation-runoff model was tested for water years 1990–96 using measured daily precipitation and minimum and maximum air temperatures for the Conconully station and measured daily minimum and maximum air temperatures for the Omak OMAW AgriMet station. The data set used for this testing is referred to as “TESTING 1” (figs. 12B and 13E; tables 4 and 5). The difference between the input time series for TESTING 1 and the input time series for model calibration is that measured temperatures are used instead of synthetic temperatures.
For TESTING 1, the model simulated a close fit for the mean annual streamflow and a good fit for the mean April–July streamflow. The simulated mean annual streamflow exceeds UUS by 10.7 percent and is the same as CUS (table 5). The simulated mean April–July streamflow exceeds UUS by 5.1 percent and is less than CUS by 0.8 percent (table 5). The testing results indicate that the precipitation-runoff model is adequately calibrated for the purpose of simulating annual mean and April–July mean streamflows.
The precipitation-runoff model was tested three more times to determine if adding different combinations of daily precipitation to the input time series for TESTING 1 for two of the real-time stations, Omak OMAW AgriMet and Salmon Meadows SNOTEL, would improve the fit between estimated and simulated streamflows for water years 1990–96. For TESTING 2, daily precipitation for Salmon Meadows SNOTEL station was added to the input time series for TESTING 1; for TESTING 3, daily precipitation for Omak OMAW AgriMet station was added to the input time series for TESTING 1; and for TESTING 4, daily precipitation for Salmon Meadows SNOTEL and Omak OMAW AgriMET stations was added to the input time series for TESTING 1 (table 4). Results of TESTING 2, TESTING 3, and TESTING 4 are shown in figures 12C through 12E, 13E, and 13F and are given in table 5.
For TESTING 2, TESTING 3, and TESTING 4, the simulated mean annual and mean April–July streamflows were significantly larger than the estimated streamflows. The oversimulation was largest for TESTING 4 and smallest for TESTING 3. The simulated mean annual streamflow for TESTING 4 exceeds CUS by 20.7 percent (compared to 17.8 percent for TESTING 2 and 6.0 percent for TESTING 3), and the simulated mean April-July streamflow for TESTING 4 exceeds CUS by 21.3 percent (compared to 18.5 percent for TESTING 2 and 5.3 percent for TESTING 3). The testing results indicate that adding precipitation data for two of the real-time stations, Omak OMAW AgriMet and Salmon Meadows SNOTEL, to the input time series of the calibrated precipitation-runoff model does not improve simulated annual mean and April–July mean streamflows.
A final comparison between estimated and simulated streamflows was made for the entire simulation period, water years 1950–2004. The input time series used for this simulation is referred to as “COMPOSITE” and is identical to the input time series for model calibration (water years 1950–89) and for TESTING 1 (water years 1990–96). For water years 1997–2004, the input time series consists of data for the same climate stations as CALIBRATION and TESTING 1, except that for water years 2000–04 daily precipitation and minimum and maximum air temperatures are used for the real-time Conconully CCR Hydromet station instead of Conconully climate station. When using the precipitation-runoff model for ESP forecasting, Reclamation will expand the COMPOSITE data set to include the most recent real-time input time-series data for Conconully CCR Hydromet station (daily precipitation and air temperatures) and Omak OMAW AgriMet station (daily air temperatures only) for simulating near-real-time hydrologic conditions in the Salmon Creek Basin. For COMPOSITE, the model simulated a good fit for the mean annual streamflow and a close fit for the mean April–July streamflow (figs. 12F and 13G; table 5). The simulated mean annual streamflow exceeds UUS by 2.8 percent and is less than CUS by 5.7 percent (table 5). The simulated mean April–July streamflow is less than UUS by 0.7 percent and less than CUS by 5.6 percent (table 5). The model simulated a better fit for mean April, June, and July streamflow for COMPOSITE than for CALIBRATION. Simulated mean April streamflow exceeds CUS by 10.1 percent for COMPOSITE and 28.0 percent for CALIBRATION. Simulated mean June streamflow is less than UUS by 11.4 percent for COMPOSITE and 12.1 percent for CALIBRATION. Simulated mean July streamflow exceeds CUS by 32.8 percent for COMPOSITE and 43.8 percent for CALIBRATION. A comparison of the estimated and simulated annual mean streamflows for CALIBRATION, TESTING 1, and COMPOSITE is shown in figure 14.
The simulated mean annual basinwide precipitation amount during the 40-year calibration period, 1950–89, was 23.1 in. and ranged from 11.8 in. for water year 1979 to 41.3 in. for water year 1983. During the driest 4 water years (1964, 1977, 1979, and 1985), the model simulated poor fits for the annual mean streamflows and April–July mean streamflows. Simulated annual mean streamflow is less than UUS by 81.3 percent in water year 1977, 46.8 percent in water year 1964, and 42.9 percent in water year 1979 (fig. 14; table 6). In water year 1985, simulated annual mean streamflow exceeds CUS by 31.2 percent. Simulated April‑July mean streamflow is less than UUS by 74.8 percent in water year 1977, 74.2 percent in water year 1979, and 39.7 percent in water year 1964 (table 6). In water year 1985, simulated April–July mean streamflow exceeds CUS by 24.0 percent. The large percentages of error are with respect to relatively small estimated annual mean and April–July mean streamflows for Salmon Creek at Conconully Dam. When comparing undersimulated streamflow with respect to UUS and oversimulated streamflow with respect to CUS, the absolute errors range from an undersimulation of annual mean streamflow of 8.0 ft3/s in water year 1964 to an oversimulation of annual mean streamflow of 3.4 ft3/s in water year 1985 (table 6). Similarly, the absolute errors range from an undersimulation of April–July mean streamflow with respect to UUS of 13.6 ft3/s in water year 1964 to an oversimulation of April–July mean streamflow with respect to CUS of 4.2 ft3/s in water year 1985 (table 6).
During the wettest 4 water years (1951, 1971, 1982, and 1983), the model also simulated poor fits for the annual mean streamflows and April–July mean streamflows. Simulated annual mean streamflow exceeds CUS by 73.3 percent in water year 1971, 17.4 percent in water year 1983, and 15.0 percent in water year 1982 (table 6). In water year 1951, simulated annual mean streamflow is less than UUS by 4.7 percent (fig. 14). Simulated April–July mean streamflow exceeds CUS by 75.6 percent in water year 1971, 13.7 percent in water year 1983, and 6.7 percent in water year 1982 (table 6). In water year 1951, simulated April–July mean streamflow is less than UUS by 13.1 percent. When comparing undersimulated streamflow with respect to UUS and oversimulated streamflow with respect to CUS, the absolute errors range from an undersimulation of annual mean streamflow of 2.8 ft3/s in water year 1951 to an oversimulation of annual mean streamflow of 26.6 ft3/s in water year 1971. Similarly, the absolute errors range from an undersimulation of April–July mean streamflow with respect to UUS of 19.8 ft3/s in water year 1951 to an oversimulation of April–July mean streamflow with respect to CUS of 74.1 ft3/s in water year 1971 (table 6).
In addition to the estimated streamflow data for Salmon Creek at Conconully Dam that were available for calibration and testing, snowpack water-equivalent data are available for the Salmon Meadows SNOTEL station for water year 1982 and for water years 1984 to the present (2006). A comparison of measured and simulated mean monthly snowpack water-equivalents for TESTING 1, TESTING 2, TESTING 3, and TESTING 4 (fig. 15A and 15B) and the partial COMPOSITE for water years 1990–2004 (fig. 15C) indicates the model oversimulates the overall snowpack water-equivalent for the Salmon Meadows SNOTEL station and delays the simulated timing of spring snowmelt by 1 month. If snowpack melting throughout upper Salmon Creek Basin is simulated to occur according to the pattern for the Salmon Meadows SNOTEL station, however, peak runoff for Salmon Creek at Conconully Dam would occur in April and May rather than May and June. Reasons for this discrepancy may be that conditions at the Salmon Meadows SNOTEL station may be sunnier or windier than the average for all MRUs in upper Salmon Creek Basin. A comparison of measured and simulated monthly mean snowpack water-equivalent at the Salmon Meadows SNOTEL station for TESTING 1 and a partial COMPOSITE for water years 1990–2004 is shown in figure 16. The extent to which the fit between the measured and simulated snowpack water-equivalent is representative of the entire basin is unknown because the snowpack measurement for the Salmon Meadows SNOTEL station represents only one point.
The precipitation-runoff model is a mathematical representation of the physical processes that occur in the Salmon Creek Basin. As a result, the quality of the model results depends on the accuracy of the mathematical representation of the physical processes (model error), the quality and accuracy of the precipitation and air-temperature input time series and the streamflow calibration/testing time series (data error), and the accuracy of the calibrated model parameters (parameter error). Model calibration and testing indicate that daily streamflows simulated using the precipitation-runoff model described in this report should be used only to analyze historical and forecasted annual mean and April–July mean streamflows for Salmon Creek at Conconully Dam. Because of the paucity of model input data and uncertainty in the estimated unregulated streamflows, the model is not adequately calibrated and tested to estimate monthly mean streamflows for individual months, such as during low-flow periods, or for shorter periods such as during peak flows. In addition, no data were available to test the accuracy of simulated streamflows for lower Salmon Creek. Thus, although the simulated streamflows appear reasonable, the model should not be relied on to analyze historical and forecasted streamflows for Salmon Creek downstream of Conconully Dam (nodes 7 through 11; fig. 11). Instead, simulated streamflows for Salmon Creek at Conconully Dam should be considered a base estimate of streamflows that can be expected in lower Salmon Creek. The estimated streamflows may be reduced in reaches of lower Salmon Creek that are losing water and increased in reaches that are gaining water or that receive runoff from tributaries in lower Salmon Creek subbasin.
The two principal reasons for the less-than-optimal model calibration are as follows. First, historical records of precipitation and air-temperature data, particularly daily minimum and maximum air temperatures at high altitudes in upper Salmon Creek Basin, are limited. Second, the number of measurements for streamflow in Salmon Creek Basin is extremely limited and only an estimated time series of monthly mean runoff is available for upper Salmon Creek Basin. This estimated time series may contain several questionable values (Dames and Moore, 1999; U.S. Department of Energy, 2004; and this study). However, the estimated time series was used to calibrate and test the precipitation-runoff model because the time series represents the best available data for the basin. Errors in the precipitation and air-temperature model input time series and in the estimated streamflow data likely affected the accuracy of the calibrated model parameters.
The precipitation-runoff model described in this report is expected to be used for simulating historical streamflows and ESP forecasting of streamflows. ESP forecasting requires accurate simulation of initial hydrologic conditions on the basis of real-time precipitation and air-temperature data, and forecasting the probability of streamflows for as far as 1 year in the future by assuming that historical records of daily precipitation and air temperature will recur with the same probability as in the past. The historical record that can be used for ESP forecasts includes water years 1950–2004 (the COMPOSITE data set) and can be extended to the present by adding the most recent real-time precipitation and air-temperature data for the Conconully CCR Hydromet station and the most recent real-time air-temperature data for the Omak OMAW AgriMet station. The COMPOSITE data set includes both wet and dry years but does not include an extended, multiyear dry period known to have occurred in the Salmon Creek Basin from the late 1920s through the early 1930s. During that period, the record low annual mean runoff from upper Salmon Creek Basin was 2.1 ft3/s (1,513 acre-ft) for water year 1931 (Dames and Moore, 1999; U.S. Department of Energy, 2004). The lowest annual mean runoff during water years 1950–2004 is 6.1 ft3/s (4,400 acre-ft) for water year 1966. The simulation of historical streamflows, initial hydrologic conditions, and the ESP forecast rely on the calibrated precipitation-runoff model described in this report and are subject to the model limitations discussed.
U.S.
Department of the Interior | U.S. Geological
Survey
Persistent URL: https://pubs.water.usgs.gov/sir20065274
Page Contact Information: Publications Team
Page Last Modified: Thursday, 01-Dec-2016 19:27:45 EST
