Functional time series modelling and changepoint detection of stratification regimes in a glacier-influenced Antarctic coastal bay

AbstractGlacial meltwater can strengthen stratification in Antarctic coastal waters, but short,irregular conductivity–temperature–depth records make it difficult to distinguish seasonalvariability from persistent regime shifts. We analyse 2,015 profiles from 25 sites, spanning137 sampling days in Admiralty Bay between December 2018 and January 2022; 131 dayswith at least three profiles contribute to the bay-level time series. We represent temperature,salinity and turbidity profiles as functions of depth using multivariate functional principalcomponent analysis, then model the component scores with a continuous-time seasonal andOrnstein–Uhlenbeck state-space model. Four components explain 89.4% of profile variation.The leading mode is 89% seasonal, while the second and fourth modes have estimated half-lives of 91 and 75 days. No shift is jointly detected by the two calibrated changepointdetectors. An autocorrelation-aware procedure identifies a penalty-sensitive shift in the thirdmode on 13 December 2021, near the record end. In calibrated PELT simulations, a singletwo-standard-deviation shift was detected in 36.7% of replicates under observed sampling,28.3% under half-day thinning, and 61.7% under weekly sampling. Seasonal forecasts improvesignificantly on climatology only at 31–60-day leads. These results show that calibratedinference and denser, year-round sampling are needed to support regime-shift claims.