Abstract: To address the need for dynamic early warning systems for river embankment safety, an embankment sliding failure prediction model has been developed based on the Lanlugang digital twin subsystem.This model focuses on the four deep-channel sections downstream of Xiaodu dyke.Employing multi-beam bathymetry technology, two comprehensive scans of the submerged riverbed topography were conducted, yielding more precise data on deep channel evolution.Utilising a random forest algorithm, cross-sectional predictions were made based on primary influencing factors including water level, flow rate, velocity, bank width, bank slope, bank-channel elevation difference, and curvature coefficient.This was coupled with the Lanlugang embankment structural safety model to evaluate embankment anti-slip stability.Results indicated a substantial agreement between predicted and measured cross-sectional data at the lowest point of the deep channel.Simulations yielded operational anti-slip safety factors of 1.252 7 under normal conditions and 0.888 7 under abnormal conditions.These safety factors for operational and seismic conditions enable timely evaluation outcomes and corresponding contingency plans, significantly enhancing proactive defence capabilities in embankment safety management.