Extreme floods are underrepresented in stream gauge records. Sedimentological evidence of past floods (paleofloods) yields longer records, allowing extreme floods to be examined over several Holocene climate periods. This study examines the influence of hydrogeomorphic complexity (floodplain aggradation and spatially variable flood deposition) on paleoflood record "completeness" and their implications for flood magnitude estimates made with paleoflood hydrologic data. We collected two sediment cores 500 m apart from the same elevation on a natural levee along a bank of the Tennessee River near Guntersville, Alabama. We measured grain size from each core at a 1-cm resolution using a Malvern 3000 laser granulometer. Optically stimulated luminescence dating of flood deposits revealed approximate age ranges of 50 - 6500 years calibrated before present (yrs. B.P.) for the downstream 3.5 m core (i.e., BO1) and 190 - 8500 yrs. B.P. for the upstream 4.18 m core (i.e., EL2). First, a sensitivity analysis revealed adjusted floodplain elevations (AFE) for each paleoflood cross-sectional geometry to reflect floodplain aggradation over time enabled the detection of paleoflood magnitude differences, suggesting floodplain aggradation should be considered in paleoflood reconstruction within alluvial settings. Minimum paleodischarge intervals were estimated in a 1D HEC-RAS step backwater model by calculating the minimum paleoflood stage needed to transport the d90 of each paleoflood deposit. Big Oak and East Levee 2 sediment cores each contained 15 high magnitude, identifiable paleofloods. The majority of the BO1 paleofloods occurred in the last 2,000 years, while most EL2 paleofloods occurred between 2,000 and 5,000 yrs. B.P., suggesting localized geomorphic complexity produced distinct paleoflood records for the same hydrogeomorphic surface. Four paleoflood events correlate across the two cores based on their ages and grain size distributions, and we combined these four floods to create a `harmonized' flood chronology for the site. The timing of these four floods corresponds with paleofloods that occurred in the last 2000 years in the middle section of the Tennessee River identified by prior paleoflood hydrologic studies. Flood frequency analysis (Bayesian Markov Chain Monte Carlo method) scenarios with ten configurations of paleoflood hydrologic data revealed differences in the number and timing of paleofloods in three paleoflood chronologies (BO1, EL2, and the harmonized) resulting from hydrogeomorphic complexities affected model distribution parameters, goodness-of-fit, and the estimated discharges of annual exceedance probabilities used to inform the design of critical flood infrastructure. As a consequence of longer records containing smaller floods, the estimated discharge of the 0.01 (100-yr), 0.001 (1000-yr), and 0.0001 (10,000-years) AEPs for EL2 were 16%, 11%, and 5% smaller, respectively, than BO1 estimates. Alluvial rivers present more challenges for reconstructing paleoflood records than bedrock, confined channel settings. The strategies presented in this paper can help integrate paleoflood hydrologic data into flood frequency analyses for alluvial rivers identifying and minimizing error stemming from hydrogeomorphic complexity. These strategies offer opportunities to expand the use of paleoflood hydrologic data in flood frequency analyses to include more rivers located in temperate environments where climate change is intensifying precipitation and a compelling need exists to anticipate and plan for changes in extreme flood occurrence.
