Author(s): Alireza Mahvelati Shamsabadi; Daeseung Kyung

Linked Author(s):

Keywords: Anomaly detection; Wastewater treatment plant; Climate extremes; Gaussian copula; LSTM autoencoder

Abstract: Climate extremes can trigger system-level stress in wastewater treatment plants (WWTPs) through coupled flow-quality interactions. We evaluate two unsupervised dependency-aware detectors in two Seoul municipal plants (P1, P2): cross-sectional LSTM autoencoder (CS-LSTM AE) and Gaussian copula (GC) [1,2]. Using rolling-origin validation, climate-regime stratification, and threshold robustness (q = 0.95, 0.97, 0.99), we find complementary behaviour: CS-LSTM AE is more sensitive to sustained heat-linked trajectory deviations, whereas GC is more sensitive to wet-event dependency shifts. Effect-size analyses relative to normal climate confirm that amplification persists across thresholds, indicating robust resilience signals rather than threshold artifacts.

DOI: https://doi.org/10.64697/iahr.proc.hic2026.295

Year: 2026