WRDS Library [Home]
Digital Library Publications Videos Card Catalog

Chapter IV

Calibration of any model begins with defining the acceptable error limits. These values typically depend upon the final intent of the project and the need for accuracy. Since these quantities directly affect the eventual users of the model, a consultation visit with the client must transpire to ensure the end product will meet the expectations of everyone involved. For the Green River model, the State Engineer's Office suggested that the error coincide with that of stream gauging. This results in an accuracy of five percent for a good measurement to eight percent for a poor measurement. The final decision called for all yearly values to be within eight percent of the gauged rate and as many years as possible to fall within the five percent boundary. Although the eight percent may sound excessive, the initial goal must be recalled. The model simply forms a base model on which to build and a method for evaluation of large changes in the basin. With the acceptable limits defined, the actual process of calibration will be reviewed.

Being a large basin, the Green River required zoning into general areas to facilitate the process. Two of the most obvious regions involved large streams that fed into the Green River, the New Fork River system and the Big Sandy River system. Before either of these enter the Green River, a check of the flow occurs to insure that the runoffs in these specific parts represent the actual conditions. The Green River headwaters area was also used as one of these calibration stations. For this purpose, a control point evaluates the initial flow of the river before any notable diversions occur.

Along with these areas, the gauging station below Fontenelle Reservoir became a major verification point. This permits the entire basin to be effectively halved as well as helping with the evaluation of Fontenelle Reservoir. The final location of a calibration station demanded placement at the inlet to Flaming Gorge Reservoir. This allows the evaluation of the model over the entire river system.

The necessary changes that follow to the initial input requires their description as it pertains to their downstream calibration station. The following sections describe these changes in such a manner along with a few general changes. Final calibration percentages are presented in Appendix A.

The only change, on the entire system, involved the multiplication of the June and July diversion amounts by an additional 30 percent. This action attempts to simulate the "2 cfs" policy of the state which gives permits prior to 1945 an additional one cfs for every seventy acres of irrigated land. Only the agricultural permits fell into this category. The process was further assisted by reducing high flows for the whole system throughout the period of runoff data. The original concept for this procedure came through a suggestion by the State Engineers' Office; therefore, this assumption relies upon the experience of the state modelers. After the model was calibrated, however, various other values were simulated in an attempt to lower the error. This resulted in larger errors than with the initial increase.

The Warren Bridge gauging station encompasses the headwater region of the Green River. In the model, the flows at station 60 represent the amounts that would pass the gauge. The actual USGS location number for the bridge is 091885.00. The initial inputs resulted in the runoff appearing to be approximately half the real values. For this reason, some of the creeks initially not considered were installed into the modelling effort in their appropriate places. The first ones to be inserted contained the drainage area south of the Green River Lakes area. For simplicity, all the streams in this small basin became lumped into one station, model number 35. This position drains approximately 67 square miles of forested catchment through Jim and Gypsum Creeks. The determination of runoff at this station involved the same area comparison technique suggested in Chapter II. With its proximity and similar topography, model station 3405 (USGS 091965.00) enhanced the comparison runoff.

Other changes involved the addition of several small streams in the northwestern part of the Green River basin. The creeks added into the model include Tosi, Wagon, Rock, Klondike, and Lime Creeks. The combined drainage area totaled nearly 98 square miles. Since few diversions occurred in any part of this area, all the creeks resulted in another lumped station to interject the cumulative runoff. The area comparison technique was again used in this situation since none of the creeks had runoff data associated with them. Station 5220 (USGS 092055.00) served as the base runoff.

The New Fork River gauge contains one of the most sensitive branches of the entire Green River system. The New Fork River contains four major reservoirs or lakes. The landscape varies from high mountain to low prairie terrain. A significant amount of diversion exists; and, extreme amounts of stream branching become almost entangled messes. All of this culminates in the necessity for an accurate portrayal in the model for the rest of the basin to properly balance. For this reason, a modeling station located at 3990 (USGS 092050.00) helps to balance the New Fork River before it enters the Green River.

The New Fork River provides large runoff which allows for an opportunity for significant errors to occur; however, the model only needed a few alterations from the initial values to comply with the error limits. The major adjustment in calibration was the reapportionment of several streams. This involved converting the associated streams that served to define their original runoff. The set of streams that required this adjustment included Duck, Cottonwood, Willow, and Boulder Creeks. The conversion removed the streams,' reliance on Fremont Creek (USGS 091965.00) and replaced it with the East Fork River (USGS 092030.00) gauge. This transformation resulted from a more in-depth survey of the actual area that the streams drained. Being at lower elevations and gentler grades, the East Fork River drainage better represents them. Along with this change, the drainage area of Cottonwood Creek increased to 35 square miles by moving the headwater station down the creek.

The other area demanding alterations concerned the modeled reservoirs, New Fork Lake and Boulder Reservoir. Upon initial input, the release schedules revolved around Fontenelle Reservoir's releases. After consulting with people familiar with the operation in question, new schedules were developed (Table 4.1). The largest substitution pertained to the non-project releases. Minor changes in the overall values assisted in the calibration of this portion of the model.

Of all the difficulties encountered, the Green River Below Fontenelle Reservoir gauge (USGS 092119.00) represented the largest single calibration problem. Located right below Fontenelle Reservoir at modeling node 9055, this gauge station did not have recorded flows until after the completion of the reservoir in 1964. This caused the first three years of the model to be undefined at this point. The event that resulted in the most turmoil, however, was the filling of the reservoir. Since WIRSOS does not allow for reservoirs to begin operation in the middle of a run, the runoff data demanded division into two parts. The first portion contained the seven years before the dam actually started filling, 1961-1967. The second part was composed of the years 1968 through 1970. This action resulted in errors in the New Fork River system since return flows occur over yearly boundaries and are lost at the end of the actual completion of the program over the defined years. To alleviate this problem, the starting reservoir values in the second part were increased by approximately twenty- five percent of their final values in the earlier portion. To further force the second part to calibrate, the release schedule of Fontenelle Reservoir needed revision. The current values appear in Table 4.1.

of all the branches that required calibration, the Big Sandy system possessed the most hidden problems. The Big Sandy River Below Eden, Wyoming gauge (USGS 092160.00) fell just upstream of the confluence with the Green River allowing the river to be isolated. Unfortunately, the largest problem resulted from the operations of the Green River above the Big Sandy River. Since Big Sandy Reservoir is operated by the

Table 4.1 - Final Release Schedules of Fontenelle Reservoir, Boulder Reservoir, and New Fork Lake in Green River WIRSOS Model. All values are in percent of storage.

MonthNew Fork Lake and
Boulder Reservoir

Bureau of Reclamation along with Fontenelle Reservoir, a hypothesis followed that it was used as a way of augmenting the Green River's flow during the years of Fontenelle's filling. This produced a severe error in the Big Sandy portion of the model from 1966 to 1969. As the only countermeasure available, varying sets of input evolved that solved the predicament. Appendix B indicates the run files used. For the first five years, diversion file INP4.1 contains the proper diversions normally operating on the system. For the next several years, an additional diversion had to be added which will be described later. INP4.2 gives these appropriate diversions for years six through nine including the break in data due to Fontenelle. With the Big Sandy Reservoir returning to normal operations midway through 1970, a third set of data, INP4.3, resulted that contains the remnants of the added diversion and the rest of the normally operating diversions. For any year beyond 1970, INP4.1 is the appropriate diversion file.

The other dominant problem existed in the efficiencies or consumption rates in the Big Sandy River basin. During normal years, the model predicted almost twice the amount of water that actually appeared in the measured data at the gauging station. Due to this fact, an' assumption arose that the efficiencies in this area would vary significantly compared to the rest of the Green River Basin because of the existing environmental conditions. The Big Sandy River has a significant portion of its drainage area within an arid, nonforested region in contrast to the forested density territory that initially established the runoff for this drainage. The main irrigation takes place on this arid area. With this knowledge and discussions with experienced engineers with understanding of this particular area, the efficiency rate for the area was increased from 48 percent to approximately seventy percent. Even though this change reduced the error, it still did not account for a large portion of the excess water.

The final answer to this mystery rested in the operation of Big Sandy Reservoir itself. During a site investigation, the outflow from the dam appeared to be totally diverted into a canal that did not reenter the river bed. Treating this as a large diversion with a slight, increase in its diversion efficiency (73 percent), the discrepancies disappeared. In the years of filling Fontenelle Reservoir, a bypass diversion called upon a portion of this diverted water. This allowed the water to circumvent the diversion without being assessed the high efficiency loss. The addition of these two factors allowed for the complete calibration of the Big Sandy River basin.

The Green River near Green River, Wyoming (USGS 09217.00) gauge, being the last gauge in the modelled system, was effected by every change discussed so far in this chapter. The USGS station identification number for this gauge is 092170.00 with modeling number 13020. The location of the measurement occurs before the inlet to Flaming Gorge Reservoir and after the city of Green River, Wyoming. This allows for the inflow into Flaming Gorge Reservoir due to the Green River to be easily identified in the model. The only correction occurring in the initial model that influences this station alone pertained to the elimination of Alkali Creek as a factor in modeling and a runoff change on Bitter Creek. originally these two entities were believed to contribute significant amounts of water to the system. Approximations for the initial runoff values came from paralleling peak annual flows for Bitter Creek and another stream. This method produced extremely large values that could not be substantiated due to a lack of runoff data for any stream in the immediate area of either Alkali or Bitter Creek during the period of investigation. Finally, some data emerged that contained runoff data in the late seventies and early eighties for Bitter Creek at a gauging station below Rock Springs, Wyoming. These values range from 2,000 to 10,000 acre-feet per year. Since no other means presented itself for the determination of these flows, a typical year, 1977, portrays the flow throughout the study in Bitter Creek. Since Alkali Creek equated to a very small fraction of the Bitter Creek flow, it was eliminated from the model.

One other matter presented a dilemma. In year one, the flow at the Green River station is approximately 20 percent high. In looking at the check stations above this point, the sum of their waters is greater than at the end without inclusion of all the streams in between. This leads to the theory that this model does not accurately approximate extreme low flow years such as year one of the model. However, the rest of the check stations do show compliance in year one indicating that any portion of the model above a valid check station executes the model's goals within the established standards for year one.

With all the fragmentation that occurred as a result of Fontenelle Reservoir, the level of sophistication in the running of this model increased significantly. The files that constitute a particular run depend on the specific year of data. Appendix B outlines what files comprise the input for which years. Appendix B also contains a segment on what files to run if the assumptions of regular operation over the entire runoff data set were to be developed. Disks containing all the input and resulting output can be obtained from the Wyoming Water Resources Center.

Student t tests were used to confirm the accuracy of the model. For a confidence level of 99 percent, all check stations' t values were less than the standard limiting t value for the corresponding degrees of freedom. The means, standard deviations, and computed t values for all the data appear in Appendix A with their corresponding check stations.

93-10 Table of Contents
Water Resources Publications List
Water Resources Data System Library | Water Resources Data System Homepage