Improved Gaussian process regression for rock slope displacement monitoring and early warning

Accurate prediction of displacement thresholds and time-to-breach is critical for rock slope stability monitoring. Traditional Gaussian process regression (GPR) lacks probabilistic certainty estimation and adaptability to non-stationary displacement trends, limiting proactive risk management. The proposed displacement threshold certainty estimation (DTCE) method employs a composite kernel (Constant + RBF + White noise), windowed differential analysis of recent measurements, and a hierarchical threshold system to predict breaches in vertical, horizontal, and total displacements. GNSS displacement data from four monitoring points were analysed.