Improving dam safety analysis by using physically-based techniques to derive estimates of atmospherically maximum precipitation

Abstract

A high or very high consequence dam is a large dam whose failure would have large consequences to life and/or property downstream. Knowledge of the magnitudes of extreme floods and their associated annual exceedence probabilities (AEP) are needed to determine the risk that such a dam might fail.

Traditionally, the largest "physically possible" precipitation event (the Probable Maximum Precipitation, PMP) and its associated flood event (the Probably Maximum Flood, PMF) have been calculated with a combination of statistical and meteorological techniques developed by the World Meteorological Organization (WMO). These techniques work reasonably well in flatter terrain, but may occasionally produce unrealistic results in mountainous terrain. This research focuses on improving safety studies for hydrologic structures such as dams, by using physically-based techniques to estimate the PMP and PMF and to calculate the associated AEP. This research contributes in three areas. The first area is in using an atmospheric model to estimate maximum precipitation. Secondly, the research demonstrated that simulated streamflow may be used to generate frequency curves and their associated confidence limits. The final contribution was in demonstrating that the frequency statistics indicated that the traditional PMP overestimates the PMF, while the atmospheric model estimates were more in line with accepted AEPs for a PMF.

This research was performed on the upper Columbia River Basin in southwestern British Columbia. The basin is an alpine basin, with annual precipitation varying from 2500 mm on the west to 500 mm on the east. Severe precipitation events generally begin over the Pacific Ocean, but are somewhat moderated by the intervening mountain ranges. There are several hydroelectric and flood-control dams operated by BCHydro on the Canadian portion of the river. One of these, Mica Dam, was used as the focus of this research.

The Mesoscale Compressible Community (MC2) model (Recherche en Prevision Numerique) is an atmospheric model designed to forecast weather at a fine resolution. The MC2 model was used to calculate a physically-based estimate of the maximum, atmospherically possible, precipitation (referred to as the Probable Maximum Storm, PMS). The numerical experiments with this model suggested that an atmospheric maximum precipitation does in fact exist, and it can be calculated with the model. The method is less subjective than the traditional WMO method and not subject to the same issues of data quality. Also, the model accounts implicitly for topography in its calculation of precipitation. The model determined a maximum 24-hour precipitation of 73.4 mm as an average over the Mica Dam basin (this number is preliminary, further research into the atmospheric model may result in a larger number). This PMS produced by th e atmospheric model was larger than any previously observed precipitation event, but lower than the PMP produced with the WMO method, indicating that the WMO method may overestimate the PMP in mountainous terrain. The MC2 model is recommended for developing the maximum atmospherically possible precipitation, but further meteorological research is recommended to ensure that all of the assumptions used in MC2 and the PMS module are suitable for this purpose. The PMS and the PMP were both used as input to the physically-based hydrological model WATFLOOD/SPL.

A flood frequency curve was developed to assess the AEP to the floods generated by the PMS and the PMP. The AEP were used to compare the relative magnitudes ofthe floods caused by the PMS and PMP, and etermine if they were within the presumed probability range of the PMF (10^-1 to 10^-6).

The derivation of a frequency curve is dependent upon the time series length of the data, which is often too short for meaningful extrapolation to the return intervals for a PMF. In this research, historical meteorological data were available and used in a hydrological model to develop a long, deterministically simulated streamflow time series of 95 years. The use of the simulated data decreased the sampling uncertainty due to a short time series. The simulated data generated frequency curves that were similar to frequency curves derived with observed data.

However, the simulated streamflow data are based on uncertain atmospheric variables that are transformed by an atmospheric model and by WATFLOOD/SPL. This thesis addresses the consequence of the parameter uncertainty in WATFLOOD/SPL. The 95% confidence limits for the frequency curves were derived through a Monte Carlo analysis of the parameter variation. An investigation into the behavior of the model showed that the parameter set within WATFLOOD/SPL was robust and there was only one optimum parameter set within the limits of the parameter space. Due to time constraints, a method to use the variation in a five-year time series as an analogue for the variation in the full time series was developed. The confidence limits grew wider as the return period increased, although further research into the behavior of the parameters may help reduce the width of the confidence limits.

The frequency curve and its confidence limits were used to estimate the range of return interval of the floods produced by the PMS and the PMP. These return intervals of the floods were used to determine if the floods were consistent with the PMF. The 100-year snowpack, the 100-year melting temperature sequence and the PMP together generated a flood with an AEP that was much smaller than the probability range for a PMF (<<10^-7). The 100-year snowpack, the 100-year melting temperature sequence and the PMS together generated a flood with an AEP that was also smaller than the probability range for a PMF (<10^-7), however, it was closer to the presumed probability range than the combination with the PMP. This suggests that the PMS may be a more realistic estimate of maximum precipitation of PMF estimation, and is still somewhat conservative.

This research has improved the current methods for safety analysis of hydrological structures, and recommends the use of physically-based methods to derive PMPs and PMFs. Comparing the PMF estimates with frequency curves should help validate results and provide a higher level of confidence in extreme rain producing flooding.