Missing data representation by perception thresholds in flood flow frequency assessment

Abstract

Flood flow frequency analysis (FFA) plays one of the key roles in many fields of hydraulic engineering and water resources management. The reliability of FFA results depends on many factors, an obvious one being the reliability of the input data - datasets of the annual peak flow. In practice, however, engineers often encounter the problem of incomplete datasets (missing data, data gaps and/or broken records) which increases the uncertainty of FFA results. In this paper, we perform at-site focused analysis, and we use a complete dataset of annual peak flows from 1931 to 2016 at the hydrologic station Senta of the Tisa (Tisza) river as the reference dataset. From this original dataset we remove some data and thus we obtain 15 new datasets with one continuous gap of different length and/or location. Each dataset we further subject to FFA by using the USACE HEC-SSP Bulletin 17C analysis, where we apply perception thresholds for missing data representation. We vary perception threshold lower bound for all missing flows in one dataset, so that we create 56 variants of the input HEC-SSP datasets. The flood flow quantiles assessed from the datasets with missing data and different perception thresholds we evaluate by two uncertainty measures. The results indicate acceptable flood quantile estimates are obtained, even for larger return periods, by setting a lower perception threshold bound at the value of the highest peak flow in the available - incomplete dataset.