| Abstract | Information on instantaneous peak flows (IPFs) is generally required for estimating targeted flows for designing various hydraulic structures, informing non-structural measures, and establishing watershed level or regional scale flood envelop curves. Such information is often lacking compared to mean daily flows, which are frequently available at hydrometric gauging stations. Where IPFs are available, missing values are often encountered in instrumental records. In this study, some existing methods for estimating IPFs are reviewed in the historical context and new methods based on multivariate concept are proposed. Machine learning approaches are also investigated and fusion modeling or super-learner notion is introduced within an ensemble modeling framework to synthesize outputs of multiple IPF estimation methods. The study focusses on all gauging stations with natural flow regimes from Canadian Atlantic provinces. Evaluation of the performance of multiple IPF estimation methods and a robust model selection criterion suggest that multiple methods are applicable for the entire region as the idea of a single universal IPF estimation method is not feasible. Therefore, fusion modeling is useful to exploit the strength of individual methods and to synthesize outputs of multiple methods. The outcome of this study is expected to enhance the value of observational records, which in turn will help derive more reliable hydrologic design estimates at locations where observational records of IPFs are either unavailable, limited or contain missing values. Additionally, the study findings will be useful in developing future flood envelop curves to support climate resilience efforts across the region. |
|---|
