Bias and information content of paleoflood data in flood-frequency analysis

Paleoflood hydrology is commonly used to assess the magnitude and frequency of floods on rivers in bedrock-controlled reaches. Previous studies suggest that paleoflood data may either degrade or enhance the information content of gaging data in flood-frequency analysis, depending on analysis assumptions. In this study, we calculate the effect of combining synthetic paleoflood and gaging data on estimates of low-frequency flood quantiles. Using Monte Carlo simulation, we generate 5,000 realizations of a 4,000-year annual-flood series from a log-Pearson type III distribution. Using type I and type II censored models of paleoflood deposition and the last n-years (20-100 years) of the series as the gaging record, we fit the combined paleoflood and gaging data with the expected moments algorithm (EMA) and maximum-likelihood analysis (MAX). Our results are relevant to paleoflood records greater than 1,500 years in length and recurrence intervals ranging from 10 to 5,000 years. In general, the combination of paleoflood and gaging data improves the accuracy of flood-quantile estimates compared to gaging data alone when the three-parameter log-Pearson type Ell distribution is assumed to be the underlying parent distribution. Use of a two-threshold scenario, comprised of the largest flood as a single exceedance and the second largest flood as a period-of-record threshold, resulted in the best retrodiction of the parent population. Incomplete dating of paleoflood deposits, coupled with the assumption of a multiple-threshold exceedance scenario, may degrade the information content of paleoflood data in flood-frequency analysis compared to gaging data alone. Quantile estimates for gaging records with skew coefficients less than zero may not be improved with paleoflood data if the bias between paleoflood stage and maximum flood stage is not accounted for. Uncertainty of our results increases with increasing skew coefficient, suggesting limits to regionalization of our results.