WSP09829 - Laserfiche WebLink

Home Browse Search

002229 These comparisons indicate minimal difference in the magnitude of errors in annual streamflow volumes between the calibrat~on and verification periods for all six of the drainage basins. Calibration and verification periods have dry and wet years of record. Similar magnitude of errors in streamflow volumes seem to occur during the calibration and verification periods. To assess further the results of the predictive capability of the model, an examination of some statistical properties was done. The values of observed and simulated annual average streamflow for the period of record for each of the six calibrated drainage basins are listed in table 6. By compar- ing absolute differences in streamflow, it seems that the model does quite well at predicting these annual average values. Part of this accuracy results because annual streamflow volume is the primary objective function ofoptimi- zation and because of the smoothing that takes place by averaging volumes for dry and wet years. Another statistic that is important in these semiarid drainage basins is the percentage of days in the year that the stream has zero flow (table 6). Comparison of the observed and simulated percent days of zero streamflow indicates that as the stream approaches a perennial situation (zero percent), the simulated values closely relate to the observed values. This is to be expected for the statistic loses its meaning in a truly perennial situation. For example, there is virtually no difference between observed and simulated percent days of zero streamflow for the Wilson Creek streamflow-gaging sta~ tion. In the ephemeral situation, there is zero streamflow 83 percent of the time or more than 300 days a year for the observed record for the Stokes Gulch streamflow-gaging station. The absolute difference between the observed and simulated percentages was'3 percent. For the remainder of the streamflow- gaging stations listed in table 6, the observed percent of days of zero streamflow ranges from 36 to 57 percent, and the simulated values differ from the observed values by 16 to 37 percent. Thus, as the number of days zero streamflow approaches one-third to one-half of the year, the simulated per- centages are less reliable. Several factors affect these percentages. One factor is the small numbers involved. A stream in this study area can have very small streamflow values for many days before reaching zero streamflow. Observed streamflow was less than 0.50 ft3fs at Foidel .Creek at the mouth near Oak Creek (09243900) for much of the period from July through September 1979 before streamflow ceased. Because of these small values, it is difficult to predict the cessa- tion of streamflow. A second factor that affects simulated percentages is that during September, the streamflow fluctuated between flow and zero flow. Because of this fluctuation, some of the lack of fit between observed and simulated streamflow results from the smoothing nature of the combination of algorithms in the model. The mathematics tend to average daily situations, and the random natural fluctuations are ignored. These two .factors primarily result from the mathematics of the small numbers and have minimal effect on the calibration of the model. These factors could account for much of the difference between observed and simulated percent of days of zero streamflow. 18 ~~~;;, ~~',~~ ;~":"" ..' ,~,' ~:l:i;:' ~i~:;'{'\ I ~~:~ ~1~ i;~~:;; ~~I~~~ ;.;";~:::': i '<5.. ~~l 1,':,.-1" ~(~ <;,l\ (-;;~'" r:~~~ '~i~ ~~~..] ~;1~! t4 :~~j t:it ~.'i~' iff'~ ,1 .~~ 0;t~ I i~l i'~~; '~,' ~{:.~~ ~ ',' . ;r"::, ?t~~};,' !;.:<,'~ ;tc:;', ~:, > . ;:7'