Use of a genetic algorithm to improve predictions of alternate-bar dynamics

Alternate bars may form in sandy beds of straight rivers and channels. These bars are characterized by the alternation of crests, all moving downstream at a speed of several meters per day. The aim of this paper is to predict the dynamics of alternate bars. To that end, we tested predictions of measured alternate bars in flume experiments, as derived from an amplitude evolution model. Weakly nonlinear stability analysis underlies this amplitude evolution model, so that it applies to situations in which the width-to-depth ratio is close to the critical ratio, above which alternate bars occur. The experiments have a width-to-depth ratio far above the critical value, well outside the range of formal validity of the model. While wavelengths and heights of the alternate bars are still well predicted, we found that the migration rate is not: the amplitude evolution model produces an underestimation of close to a factor of two. Therefore we took a slightly different approach. We tested the predictive capability for this amplitude evolution model by using a genetic algorithm to tune the model to bathymetric data. After tuning, the model is indeed able to predict the migration rate of the bars over periods that exceed the tuning period by far. Limits to the prediction time, i.e., failure of this method, could not be derived for the data sets used in this work.