Quantum generative intelligence for rare hydroclimatic anomaly screening of glacial lake outburst flood risk

Glacial lake outburst floods (GLOFs) are destructive cryosphere-related hazards shaped by nonlinear interactions among hydroclimatic forcing, glacier morphology, lake evolution, and geomorphological instability. Reliable precursor screening remains difficult because documented GLOF events are sparse, triggering mechanisms are heterogeneous, and pre-event hydroclimatic signals may appear as subtle changes in temporal organization rather than simple amplitude extremes. This study proposes a hybrid quantum-classical framework for rare hydroclimatic anomaly screening that integrates conditional multivariate sequence generation, quantum expectation-based latent representations, generated-reference temporal mismatch scoring, and supervised rare-event classification. Hydroclimatic dynamics are modeled using conditional Wasserstein generative adversarial networks with gradient penalty (WGAN-GP), temporal convolutional sequence learning, and latent representations derived from parameterized quantum-circuit expectation values. Compared with Gaussian and sign latent baselines under identical WGAN-GP settings, the quantum latent formulation achieved lower errors in cross-feature correlation, Wasserstein distance, Kolmogorov-Smirnov statistic, autocorrelation distance, and power spectral density distance in the present experimental setting. The anomaly-scoring module further identified cross-variable temporal correlation mismatch and autocorrelation deviation as precursor-relevant signals. Because the number of positive GLOF samples is limited, the supervised outputs are interpreted as high-recall anomaly-screening and ranking indicators rather than deterministic forecasts of GLOF occurrence. Explainability and ablation analyses indicate that QLGM-generated-reference temporal mismatch features were more discriminative than compact hydroclimatic and static descriptors within the evaluated models. These results suggest that quantum expectation-based latent representations can reduce hydroclimatic sequence-realism errors under the evaluated WGAN-GP setting and can support generated-reference anomaly prioritization for GLOF risk reduction under severe data imbalance. The findings should be interpreted as empirical evidence for a useful structured latent representation in this setting, rather than as proof of universal quantum advantage.