An Exploration of the Importance of Flood Heterogeneity for Regionalization in Arizona using the Expected Moments Algorithm

Bulletin 17B (B17B) has prevailed for 30 years as the standard manual for determining flood frequency in the United States. Recently proposed updates to B17B include revising the issue of flood heterogeneity, and improving flood estimates by using the Expected Moments Algorithm (EMA) which can better address low outliers and accommodate information on historical peaks. Incorporating information on mixed populations, such as flood-causing mechanisms, into flood estimates for regions that have noticeable flood heterogeneity can be statistically challenging when systematic flood records are short. The problem magnifies when the population sample size is reduced by decomposing the record, especially if multiple flood mechanisms are involved. In B17B, the guidelines for dealing with mixed populations focus primarily on how to rule out any need to perform a mixed-population analysis. However, in some regions mixed flood populations are critically important determinants of regional flood frequency variations and should be explored from this perspective. Arizona is an area with a heterogeneous mixture of flood processes due to: warm season convective thunderstorms, cool season synoptic-scale storms, and tropical cyclone-enhanced convective activity occurring in the late summer or early fall. USGS station data throughout Arizona was compiled into a database and each flood peak (annual and partial duration series) was classified according to its meteorological cause. Using these data, we have explored the role of flood heterogeneity in Arizona flood estimates through composite flood frequency analysis based on mixed flood populations using EMA. First, for selected stations, the three flood-causing populations were separated out from the systematic annual flood series record and analyzed individually. Second, to create composite probability curves, the individual curves for each of the three populations were generated and combined using Crippen's (1978) composite probability equations for sites that have two or more independent flood populations. Finally, the individual probability curves generated for each of the three flood-causing populations were compared with both the site's composite probability curve and the standard B17B curve to explore the influence of heterogeneity using the 100-year and 200-year flood estimates as a basis of comparison. Results showed that sites located in southern Arizona and along the abrupt elevation transition zone of the Mogollon Rim exhibit a better fit to the systematic data using their composite probability curves than the curves derived from standard B17B analysis. Synoptic storm floods and tropical cyclone-enhanced floods had the greatest influence on 100-year and 200-year flood estimates. This was especially true in southern Arizona, even though summer convective floods are much more frequent and therefore dominate the composite curve. Using the EMA approach also influenced our results because all possible low outliers were censored by the built-in Multiple Grubbs-Beck Test, providing a better fit to the systematic data in the upper probabilities. In conclusion, flood heterogeneity can play an important role in regional flood frequency variations in Arizona and that understanding its influence is important when making projections about future flood variations.