Ceemdan-Bilstm for Landslide Displacement Prediction: Exploring Rainfall Hysteresis and Effective Rainfall Cycles

Accurate landslide displacement prediction is essential for improving early warning accuracy. This paper proposes a Complete Ensemble Empirical Mode Decomposition with Adaptive Noise - Bidirectional Long Short-Term Memory(CEEMDAN-BiLSTM) based prediction model. An analysis of the Baishui River's ZG1118 observation station in the vicinity of Three Gorges Reservoir, First, CEEMDAN is employed to decompose landslide displacement, extracting periodic and trend components while introducing the concept of an effective rainfall period. The Grey Relational Analysis (GRA) method is then used to determine the optimal lag time and select key influencing factors. Subsequently, utilizing training and testing sets divided in an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$8: 2$</tex> ratio with sequential ordering to maintain temporal structure, BiLSTM models are used to forecast the trend and periodic components independently. Accumulation yields the ultimate displacement prediction. Experimental results demonstrate the model's high predictive accuracy, achieving a root mean squared error (RMSE) of 10.70 mm, Mean Absolute Error (MAE) of 8.11 mm, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> of 0.98, with an effective rainfall period lag of 6 days. This research offers a fresh method for predicting landslides that can be applied to comparable areas with intricate geology. By increasing forecasting accuracy, it helps create better alert systems, which advances global efforts to avoid and mitigate disasters.