Critical slowing down reveals dynamical regimes of slow-moving reservoir landslides: Hidden Markov model evidence

Slow-moving landslides do not respond uniformly to rainfall and reservoir-level fluctuations: the same hydrological trigger can produce markedly different displacement responses depending on when it occurs. Determining whether a landslide is in a dormant or an active dynamical regime may therefore provide information complementary to conventional hydrological-threshold warning systems. This study applies critical slowing down (CSD) indicators to 17 years of monthly GPS displacement data from three slow-moving reservoir landslides in the Three Gorges Reservoir Area (Baishuihe, Bazimen, Shuping), with a two-state Gaussian hidden Markov model (HMM) serving as an independent statistical validation benchmark. Three findings emerge. First, CSD cannot provide statistically significant lead-time warning ahead of individual displacement acceleration events (all 27 permutation tests p > 0.05 ). Second, CSD indicators, and Variance in particular, show consistent discrimination between dormant and active regimes across the three sites, as validated by the HMM (6/9 point-biserial correlations survive Bonferroni correction). Third, Variance significantly outperforms AR1 at Bazimen and Shuping and across the three sites in meta-analysis (Stouffer combined Z = + 5.60 , p < 0.001 ), with the two indicators statistically tied at Baishuihe, a rank order that reverses the consensus of the ecological CSD literature—a reversal that may be related to the heavy-tailed distribution of monthly landslide displacement increments. Together, these findings suggest that CSD indicators carry regime-classification information not recoverable from hydrological variables alone, and contribute a nonlinear-dynamics perspective to the emerging research strand on regime identification in landslide early warning.