Hybrid Convolutional Neural Network–Convolutional Long Short-Term Memory (CNN–ConvLSTM) Model for Spatiotemporal Elevation Change Prediction in Tailings Storage Facilities

Abstract Tailings Storage Facilities (TSFs) are crucial to mining operations but remain highly vulnerable, with overtopping, seepage, and liquefaction often associated with catastrophic failures. Although many factors affect TSF stability, changes in beach elevation are especially important because they influence where water collects, how seepage occurs, and ultimately the overall safety of the structure. This study presents a hybrid Deep Learning (DL) model that combines convolutional neural networks (CNNs) with Convolutional Long Short-Term Memory (ConvLSTM) units to predict spatiotemporal elevation changes on TSF beaches. The model was trained on high-resolution Unmanned Aerial Vehicle (UAV)-derived Digital Elevation Models (DEMs) collected over a two-year period at the Mungari TSF in Western Australia. The results demonstrate that the proposed model achieves high precision, with a coefficient of determination ( R 2 ) of 0.9351, Root Mean Square Error (RMSE) of 0.1436 m, and Mean Absolute Error (MAE) of 0.0815 m, outperforming other baselines. This approach effectively captures complex non-linear deposition behaviour, providing a valuable tool for proactive monitoring and risk-informed decision-making, critical for long-term TSF management and operational safety enhancement.