The prediction of the Migration rate of meandering rivers using Machine
River meandering is the natural process that many lowland rivers undergo
as a consequence of the alternation of bank erosion and accretion, which
leads to the typical shape of the so-called meandering rivers. During
the last decades, numerous modelling studies have been developed to
reproduce their planform dynamics and to predict their future evolution.
Most of these modelling approaches are physics-based, meaning that they
solve the mathematical equations of shallow water open channel flow and
fluvial sediment transport. Other types of modelling are very rare.
Recent advances in artificial intelligence have led to promising results
in many fields of science but their potential seems to have been so far
rather unexplored in the prediction of meandering rivers morphodynamics.
In this study, we have developed machine learning (hereinafter ML)
models to compute the meander lateral migration rate based on training
dataset: once the model has been trained with known migration rates and
curvature values at two consecutive time steps, it is used to predict
migration rates at the following time step. To this aim, the train and
test dataset is coming from the outputs of a semi-analytical meander
morphodynamic model which provides simulated evolving meandering
planforms (described through the spatial curvature distribution) and
migration rates computed through the excess near bank velocity. Such
migration has been considered as the “Target” in the present study.
The results for different models such as linear regression, feedforward
neural network, SVM, and XGBoost were compared. It indicates that the
“XGBoost” model with approximately 80 percent of accuracy in the
prediction of the next time step, has the best result among them. This
is just an opening chapter for the usage of ML in morphodynamics of
meandering rivers and with advanced methods, there will be promising