199-1 Local River Bank Retreat Magnitudes Are Correlated With Predictions From a Long-Term Centerline Migration Model During Large, But Not Small, Floods

Session: Advances in Fluvial Processes and Sediment Transport, Part II

Presenting Author:

Authors:

Abstract:

Alluvial river meandering is one of the most well-studied sediment transport problems, yet we still do not understand how much river banks will retreat in response to a single flood. In contrast, meander migration models can successfully predict long-term channel evolution patterns using simple geometric relationships based on the curvature of the channel centerline. Though these centerline models work well on multi-year time scales, they do not aim to predict short-term bank retreat. Here, we pose a basic question—how well do predictions from a centerline migration model correlate with observations of complex, local, river bank retreat from individual floods?

To answer this, we collected repeat topographic surveys of a sand-bedded, actively meandering river in Texas, then compared observed bank retreat to predictions made with a centerline model that computes migration rate as the weighted sum of upstream channel curvatures. Our topographic dataset is composed of 10 drone-derived lidar point clouds of two bends of the 200 m-wide lower Trinity River acquired over 2.5 years. Bank retreat magnitude is calculated as the cross-sectional area between consecutive outer channel bank profiles measured at the same location divided by the exposed bank height.

We show that observed bank retreat magnitudes have a moderate to strong positive correlation with predictions from the channel centerline geometry (Spearman’s ρ > 0.5) in both channel bends for the two largest floods, which submerge ≥ ~ 90% of the bank. In the shorter, higher curvature bend, the moderate correlation holds for smaller floods. Conversely, in the longer, lower curvature bend, bank retreat magnitudes have a moderate negative correlation with model predictions for the two floods that submerge ~ 70–75% of the bank (ρ < -0.3). Strong correlations do not necessarily mean the model has local predictive power—indeed, even during the two largest floods, linear regressions fit to the predicted and observed bank retreat have root mean square errors that range from 0.8–1.5 m. These values are about half of the mean observed retreat magnitudes. Still, we highlight that for these two floods, 30–60% of the variance in the data is explained by the model (as measured by R²). We interpret that centerline models can therefore be useful for forecasting bank retreat patterns even at short time scales, as long as a flood causes substantial erosion.

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-7557

© Copyright 2025 The Geological Society of America (GSA), all rights reserved.