Flood Frequency Analysis In A Data Sparse Mixed-Population Watershed

Risk-informed dam safety decision making in regard to flood loads requires understanding of characteristics and magnitudes of very infrequent floods (10,000-years and less frequent). Flood frequency distributions are typically developed using at-site streamflow information from an annual maximum flood series. Other data sources may be leveraged as well to better understand infrequent floods far beyond the period of streamflow record including rainfall-based estimates, historical flood indirect estimates, and paleoflood information. This process can sometimes be straightforward, but limited streamflow data and a mixed flood population (floods caused by different meteorological event types and physical processes) can complicate this process. For this paper, we discuss the methods used to assess flood hazard potential at an earthen embankment dam on the leeward side of the Rocky Mountains in Montana. This region experiences distinct mixed population characteristics, with snowmelt-dominated and rainfall-dominated floods occurring during a similar spring season (April-July). Notably, the June 1964 rain driven event caused widespread flooding in this portion of Montana. Several smaller but similar rain driven events in the month of June have occurred since then in the immediate region. However, only 20 years of annual peak streamflow data were available for the facility of interest, all of which occurred prior to 1925. In order to estimate the flood hazard as well as develop hydrographs to better understand overtopping erosion potential at this facility, a complex study using multiple independent methods and leveraging nearby data rich sites was completed. Streamflow-based and rainfall-based analyses were developed for the site of interest and a nearby site with a robust streamflow record. The streamflow-based analysis included a detailed paleoflood study with investigations at several sites within the drainage basin of interest and nearby data rich basins to better understand the magnitude of paleofloods and non-exceedance bounds in the area. Further, the streamflow-based analysis featured a unique, hydroclimate-based approach for identifying the mechanisms forcing annual maximum floods in the basin, resulting in both snowmelt-dominated and rainfall-dominated flood frequency estimates. Stochastic rainfall-runoff modeling was completed in both basins using L-moments methods to develop regional precipitation-frequency information to force the model. Results from the nearby basin with ample streamflow data were used to support parameterization and calibration. The results of the rainfall-runoff modeling were combined with the paleoflood and streamflow information as informed priors to develop rain-dominated flood frequency curves, and combined curves with the snowmelt-dominated streamflow-based curve. At all times, data, methods, and insights from the nearby data-rich site were leveraged to improve the analysis at the facility of interest. The curves were developed concurrently to assure results were reasonable for the data poor basin. The findings helped reduce uncertainty in the flood loadings at the facility and to support a detailed erosion analysis for overtopping.