FLOOD09697 - Laserfiche WebLink

Home Browse Search

later section). Dala from the prolotype system are required in both phases. Model calibration involves adjuslment of the values used for the variable model paramelers until a close fit is achieved between ob- served and compuled oulpul functions. It, therefore, follows that the accuracy of predictions from the model cannot exceed that provided by the historical data from the prolotype system. Estimation of Ihe model parameters can be ei- ther by lria! and error or by a compulerized search. In this study, the computerized paltern-search pro- cedure described by Hl\1 et al. (1970) was used. Each parameter is assigned an initial value, an upper and lower bounds, and a number of increments to cover the range between the assigned bounds. The first sel- ected variable is varied through the specified range whlle all other variables remain at their inilial value. A measure of the difference between the recorded and the synthesized flows for each value of the variable is printed, and Ihe value which produced the minimum is stored. The first variable is reset to its initial value, and the second variable is taken through the same pro- cedure. Mtee all parameters have been varied, the set of values which produced each local minimwn be- comes the new set of initial values and the procedure is repeated. The process is continued until a reason- able malch is achieved between compured and ob- served outflows. Because il is the objective of the program to minlmize the difference between the ob- served and the computed plols, the measure of the difference is termed the "objective function," In the case of this program the objective function is com- puted by summing Ihe squares of the differences at specific time points between the two traces. It should be noled Ihat the range of values tested for a given parameler is based on the judgment and experience of the programmer. However, selec- tion is lernpered by the experience gained from dur- ing Ihe process. Thus, calibration effectively uses all previous experience, including that gained during the procedure. Model Calibralion for the Rural Watersheds The runoff from Ihe rural porlions of Ihe study area was simulated by means of both a daily model and an hourly time increment model (Figures 3.3 and 3.4), with Ihe hourly model being used to simulate the high flow rales from rapid-runoff producing events. The basic precipitation data available for the study area are daily totals from the non.recording gages (Fig- ure 2.7) and data published in Ihe form of "Hourly Precipil.ation Data" by lhe U. S. Departmenl of Com- merce for the recording gage. The daily information from the non-recording gages was Ihen distributed hour by hour over the day proportional to the ob- served data from the recording gage by assuming Ihe lime distribution of precipitalion al the recording gage represents that over the watershed as a whole. It is recognized Ihat this mighl nol be true, especiaily during convective storms. As indicated by Figure 1.1, Mill, Big Colton- wood, and little Cottonwood Creeks are each gaged at the canyon mouth. This point represents oulflow from the rural portion of each drainage and inflow to the urbanizing portion. Because adequate stream- flow records are not available farlher upslream for either the Big or Ihe little Cottonwood Creeks, it was necessary 10 model the rural walershed areas for these two streams as a single space increment. How- ever, for Mill Creek, a stream-gaging slation situated at Mill Creek Canyon (Number 1698) enables Ihe rural portion of Ihe M1ll Creek watershed 10 be mod- eled in two space increments. The application of the daily and the hourly time increment models to both gaged and Wlgaged areas of the Ihree watersheds is 1llustrated by the following discussion of Mill Creek (gaged) and Neff's Canyon (ungaged). Similar procedures were fol- lowed elsewhere. Daily time increment model. Because no surface runoff records were available for the Neff's Canyon, the model was calibrated firsl for the Iwo subwatersheds of Mill Creek, using dala for the water years 1962 and 1963. A study was con- ducted 10 examine the influence or effects on the val- ue of the oulput function of changing each parameter. This kind of study is lermed a sensitivity analysis, and those parameters which cause major changes in the output are said to be sensitive. The calibrated para- meter values for the two watersheds are shown by Table 3.1 in order of decreasing sensitivily, or rela- tive importance to system response characteristics. In other words, the model suggesled Ihat the para- meter which has the most influence on the outflow hydrograph to be the available moislure storage cap- acily of the soil at the beginning of the storm event. Thus, antecedent soil moislure conditions (or the soil moisture levels at the beginning of a runoff producing event) were found 10 have considerable influence on the ensuir,g hydro graph. Table 3.1 also indicates the values of the water- shed parameters selected for Neff's Canyon. These values were determined on the basis of experience and information gained in the calibration of the two nearby Mill Creek subwatersheds. The basins have similar topography, elevation, aspecl, soil types, and vegetation and are adjacent to one another and sub~ ject to the same climatological patterns. 30.