Record-based detection of gradual and discrete changes in extreme sea levels
Аннотация
Changes in extreme sea levels are critical for coastal risk assessment, yet detecting and characterising non-stationarity in extremes remains statistically challenging. We investigate changes in extreme monthly mean sea levels around Australia using a two-stage framework that separates distribution-free detection of structural change from likelihood-based quantification of extreme event behaviour. Homogenised tide gauge records from 37 stations (1966–2022) are analysed as separate station-calendar-month series to preserve seasonality and to focus on absolute extremes relevant for coastal impacts. Record statistics reveal a systematic increase in upper tail events and identify a coherent reorganisation in the ordering of extremes centred in the late 1990s. To assess the implications for extreme event frequency, stationary, smoothly varying non-stationary, and regime-shift formulations are compared using likelihood-based model selection applied to exceedance rates, modelled via Poisson likelihoods consistent with a peaks-over-threshold framework. Regime-shift models are conditioned on record-based changepoint estimates to avoid joint optimisation of timing and likelihood. Results indicate widespread non-stationarity in extreme event frequency, with smoothly varying models preferred for most series and regime-shift behaviour for a substantial minority. A detrending sensitivity analysis shows that this non-stationarity is overwhelmingly driven by long-term sea level rise, indicating that increases in extreme event frequency primarily reflect a shifting baseline rather than changes in variability.