8.4 Calibration Procedures

Any hydraulic model should be calibrated to the greatest degree of accuracy possible based on the available calibration and verification data. The required data and procedures are addressed in Chapter 5, but are further reviewed in this section.

Adopting the Working Model

Calibration cannot get underway until the modeler is satisfied that the model is performing properly and that the data are correctly entered and as error free as reasonably possible. Prior to calibration, all notes and warnings should be reviewed and handled by model adjustments where appropriate. These adjustments could include adding cross sections, changing the locations and elevations for ineffective area descriptions at obstructions, performing mixed flow analyses, correcting data input errors, and so on. It is good policy to use automated input data checking programs, such as those developed by FEMA. FEMA has separate programs for evaluating HEC-2 (CHECK2) or HEC-RAS (CHECK-RAS) data and for highlighting potential errors and inconsistencies. While these programs were developed to check input and output for FEMA flood insurance studies, the checking programs are useful for any steady flow, floodplain modeling analysis. These programs are discussed in more detail in Chapter 9. Once the engineer feels that the model is properly handling the computations for several different discharges/profiles, the work sequence moves on to calibration.

Comparing Model Output to Actual Data

Calibration discharges are obtained from actual streamgage data and/or from hydrologic model output based on historical storm data. Flood elevations are obtained from stream gage sites, highwater mark data obtained in the field, or past reports of historic floods. The HEC-RAS model operates with the actual or simulated discharges from a selected runoff event and the computed elevations are compared to the gage and/or high water mark data. The user can assign a highwater mark to the appropriate river station for the corresponding flood event. If the modeler has several different calibration events to be modeled, the HEC-RAS profile output for each can be compared with the highwater mark profile for each event. If the initial runs do not closely agree with the known data at all key locations (a common occurrence), one of the HEC-RAS parameters is modified, while still maintaining a reasonable estimate of the parameter. A new run is made, with the results again compared to the known data. This process is continued until the engineer is satisfied with the calibration or until further adjustments to meet the known data are considered unreasonable.

Adjustments to Model Parameters

Calibration adjustments to HEC-RAS parameters are usually made in the following sequence.

Manning's n.

As indicated in Chapter 5, Manning's n normally carries the most uncertainty and a wide range of n values are possible for most natural channel and floodplain configurations. An upward or downward adjustment of 10 to 20 percent may be entirely appropriate and still within the maximum and minimum range shown in Table 5.5. The adjustment should be reasonable, however. A channel n of 0.08, for example, may result in a perfect match of all known data, but if all guidance indicates that the channel n should range from 0.03 to 0.05, the higher n would not be reasonable or defensible. Another factor besides n is likely causing the difference, probably the discharge's being too great or not large enough.

Discharge.

After n, discharge normally carries the most uncertainty. If no gaged discharges are available and the flow values come from the operation of a hydrologic model, the next step is often the modification of infiltration parameters to increase or decrease the discharge values. Again, these infiltration adjustments should be reasonable and defensible. If calibration requires the hydrologic model to have, for example, 99-percent runoff to get discharges high enough to hit highwater marks, this is probably not reasonable. (Note that observed high water marks could be old. Therefore changes in land use over time should be considered.) The percent runoff should always be checked for appropriateness. Actual runoff rates vary widely between geographical regions and even between flood events. The engineer may find useful information in other analyses and reports regarding runoff rates in the study area.

Geometry.

While adjusting the discharge and n values often results in an adequate calibration, geometric data are also sometimes modified. Geometry is considered the most accurate of all the key parameters, because it is based on surveyed cross sections and topographic maps. However, the geometry may be adjusted around bridges (where gages are normally located) to better model ineffective flow areas, the point where significant weir flow occurs over the embankment, expansion ratio, debris buildup, and other features. Also, the modeler should not rule out a survey bust. If the survey crew or technician referenced from the wrong benchmark, one or more cross sections could be off by several feet.

Other Factors.

Superelevation - The centrifugal force caused by flow around a curve results in a rise in the water surface on the outside of a bend and a depression of the surface along the inside of a bend. This phenomenon is called superelevation (USACE, 1994e). Highwater marks on the outside of a river bend could be affected by superelevation. The faster the velocity and tighter the bend, the greater the effect. Where this occurs, the computed water surface may be adjusted (upward on the outside of the bend and downward on the inside) to compare it with the highwater mark. The extent of superelevation in a channel is a function of the bend radius, top width, and velocity. An equation for computing superelevation is available in many open-channel hydraulic texts, including Linsley et al. (1992) and Hydraulic Design of Flood Control Channels (USACE 1994e).

Valley conveyance - The entire width of the river valley seldom contains moving water, yet cross sections are often entered in HEC-RAS without limiting the conveyance. The outer parts of the cross section could be primarily storage with little or no conveyance. Ineffective flow area constraints or high n values (n Š 0.5) may be appropriate for these storage areas to maximize conveyance in the active flow portions of the cross section nearer the channel.

Changed conditions - A bridge or culvert may have been replaced with a larger structure since the time of the flood when the highwater marks were recorded. Changes such as channel enlargements and floodplain encroachments all could give very different hydraulic conditions than existed during the historic flood event. Similarly, portions of the watershed could have undergone changed land use, such as urbanization. Changed land use often results in significantly altered channel and floodplain geometries, compared to those that existed for the historic flood event.

Looped rating curve - Rivers of low slope (less than about 0.0004, or about 2 ft/mile, 0.4 m/km) normally exhibit a looped rating curve (discussed in Chapter 3), where the stages for the same discharge are different on the rising limb of the hydrograph than on the falling limb. This feature results in the peak stage's occurring at a different time than the peak discharge. Since a steady flow program computes the same water surface elevation for a specific discharge, the loop is not reflected in the stage-discharge relationship. The computed water surface elevation could be adjusted, based on the engineer's knowledge of the stream performance during past floods, before comparing it to the highwater mark. For streams exhibiting a looped rating relationship, an unsteady flow model should be considered in lieu of steady flow analysis.

Wave setup - For flooded areas having a significant reach exposed to the wind, wind-driven waves can result in debris lines significantly greater than the peak water surface elevation. This situation is particularly common in reservoirs.

Debris at bridges - Highwater marks considerably higher than the computed water surface elevation on the upstream sides of bridges and culverts may be due to partial clogging of the opening by debris. This occurrence is fairly common in urban areas, where the stream may be used as a dumping ground. Field interviews and newspaper file searches should be conducted to determine if this has occurred during actual floods. The debris simulation in HEC-RAS may be applied to handle this situation.

Backwater from other streams - Highwater marks at the downstream end of tributary streams may be caused by backwater from the receiving river and not from the tributary itself. Adjusting the starting water surface elevation to a lower level, based on the stage at the time of the peak tributary discharge, may result in a proper calibration of the tributary.

Final adjustments to achieve calibration should be reviewed, preferably by another experienced engineer. A second opinion concerning selection of the values of key parameters is always useful. An engineer doing his or her first hydraulic model should keep in mind that all calibration points will never be hit exactly. As discussed in Chapter 5, there is some variation and error in the actual data being used for calibration. A rule of thumb used by USACE is based on the simulated elevations' being within ±1 ft (0.3 m) of the measured elevations. FEMA guidance (FEMA, 1993) suggests a good calibration has a ±0.5-ft (0.15-m) tolerance. Figure 8.9, taken from USACE (1993b), shows the final calibration results (solid line) for an actual event and the observed highwater marks (stars). The dashed line represents the profile resulting from the initial estimates of channel and overbank n values. As seen, there are locations where the calculated profile is higher than the highwater mark data and other places where it is lower.

Verification

Model verification is desirable following completion of the calibration step, though it may be difficult to perform due to lack of data. During the verification phase, no additional "tweaking" of model parameters is performed. The model is operated for an additional actual event not used in the calibration. If the calibrated model reasonably reproduces the actual elevation data from the verification event, the model is considered fully suitable for application to all other flood events. Chapter 5 further discusses the data needed for verification analysis.

Sensitivity Tests

The smaller the watershed, the less likely that any gage data will be available. If there have been no recent flood events or if the stream is in a sparsely populated area, even highwater marks may be lacking. Chapter 5 presents some calibration techniques that can be considered in these situations; however, the lack of actual discharges or water surface elevations prevents a proper calibration. For this situation, the modeler should include sensitivity tests of key parameters, such as Manning's n, to ensure that the model is producing reasonable and defensible results. Without actual discharge and high watermark data to calibrate n, the value of n simply represents the engineer's judgment. Other engineers could easily estimate a different value of Manning's n for the same stream reach. Sensitivity tests that vary n within a reasonable range should be performed to determine the sensitivity of the water surface elevation to the variation. Where no calibration data exist, the adoption of a conservative (higher) value of Manning's n may be appropriate for flood studies. Where velocity is more important, such as in the design of channel protection or the evaluation of bridge scour, a lower value of Manning's n could be used.