Innocenti et al. (2026) Tidal, hydrological and meteorological contributions to high-water level events in the Saint Lawrence River Estuary: Local responses to regional drivers

Identification

• Journal: Weather and Climate Extremes
• Year: 2026
• Date: 2026-01-13
• Authors: Silvia Innocenti, Marlène Fortier, Pascal Matte, Caroline Sévigny, Remi Gosselin, Olivier Champoux
• DOI: 10.1016/j.wace.2026.100852

Research Groups

• Meteorological Research Division, Environment and Climate Change Canada
• National Hydrological Services, Environment and Climate Change Canada

Short Summary

This study develops a robust statistical framework using non-stationary harmonic regression and event-based analysis to identify, characterize, and attribute high-water level events in the St. Lawrence River Estuary. It reveals distinct spatial patterns of tidal and non-tidal processes, identifying a transition zone where influences shift from coastal to hydrological drivers, with non-stationary dynamics varying across frequency bands and persisting over different spatial scales.

Objective

• To develop and apply a methodological framework to identify and characterize extreme water-level events in estuaries, addressing limitations in comprehensive statistical analyses that jointly consider multiple drivers and spatiotemporal variability.
• To develop an event identification algorithm for high-water levels consistent with Extreme Value Theory (EVT) assumptions, capturing extremes from multiple occurrences over time and space.
• To introduce a new Generalized Additive Model (GAM) formulation of non-stationary harmonic regression within the NS_Tide statistical tool to enhance flexibility in reconstructing non-stationary stage signals during extreme events.
• To present new NS_Tide developments for automated selection of significant covariates and tidal constituents, accounting for uncertainty, and providing partial water level reconstructions associated with individual tidal frequencies or regression terms.

Study Configuration

• Spatial Scale: St. Lawrence River Estuary (Canada), focusing on a 200 km stretch between Saint-Joseph-de-La-Rive and Trois-Rivieres. Analysis at 11 response gauging stations and 2 reference stations (Varennes, Sept-Iles).
• Temporal Scale: Water level records from 1980–2023 (43 years) for event analysis, with initial data screening from early 1960s up to 2023. Hourly time series data.

Methodology and Data

• Models used:
  • Generalized Additive Model (GAM) formulation of non-stationary harmonic regression (NS_Tide statistical tool).
  • Variational Mode Decomposition (VMD) for decomposing non-tidal residuals.
  • Peak-Over-Threshold (POT) framework with temporal and regional declustering for event identification.
  • Iteratively Reweighted Least-Squares (IRLS) for coefficient estimation.
  • False Discovery Rate (FDR) method for statistical selection of significant stage splines and harmonic constituents.
• Data sources:
  • Hourly water level records from 19 tidal gauges operated by Fisheries and Oceans Canada, reduced to the IGLD85 datum and expressed in UTC.
  • A virtual station created by combining two non-overlapping time series from adjacent tidal gauges in Quebec City.
  • Dataset links: https://zenodo.org/records/17131694, http://www.isdm-gdsi.gc.ca/isdm-gdsi/twl-mne/index-eng.htm#s5

Main Results

• A total of 181 regional high-water level events were identified, with 110 occurring between 1980 and 2023 (averaging 4.6 events per year).
• Events exhibit clear seasonal patterns, with the majority occurring outside summer and early fall. Longest events are concentrated in April and May (spring freshet), while fall and winter events are shorter and more localized.
• Approximately 23% of regional events propagate landward from the Gulf of St. Lawrence, while fewer than 50% propagate in the opposite direction.
• The model accurately reproduces observed water levels, with median explained variance ranging from approximately 85% at upstream stations to nearly 98% at the downstream limit.
• Mean water level variability (stage term) dominates upstream sections (hydro and hydro-meteo covariates), while ocean and storm-surge contributions are primary downstream.
• A transitional zone between 90 km and 160 km upstream of the fluvial estuary entrance shows comparable importance of hydrological and marine influences for the non-harmonic signal component.
• Non-stationary tidal contributions are particularly significant in the intermediate reach (95 km to 175 km upstream), peaking further upstream than in mean-based evaluations.
• Fortnightly to annual tidal constituents significantly contribute to high-water events, displaying notable stationary and non-stationary amplitudes.
• Non-stationarity in the diurnal and semidiurnal bands is marginal up to 140 km upstream, increasing to 15% to 45% of the maximum-event amplitude further upstream. Overtide frequencies exhibit stronger non-stationarity.
• Local event classification shows a progressive landward shift from coastal (ocean, storm-surge) to hydrological (hydro, hydro-meteo) drivers for maximum event amplitudes. Tidal non-stationarity persists over longer spatial scales compared to mean water levels.
• Regional event classification indicates that the hydro covariate accounts for the largest proportion of events (approximately 32%), with other drivers contributing 21% to 25% each.
• A clear seasonality in regional classification is observed: hydro and hydro-meteo events predominantly occur in spring (April-June), while coastal processes drive events in other months.
• A temporal shift was noted, with nearly all hydro-meteo events identified before 2010, suggesting a decreasing influence of this driver in recent decades.

Contributions

• Development of a novel event-identification method that replaces fixed time-window definitions with a relative inter-event separation criterion, enabling the identification of independent local and regional events linked to different drivers and conditions.
• Introduction of a new Generalized Additive Model (GAM) formulation for non-stationary harmonic analysis (NS_Tide), allowing for smooth and localized nonlinear effects in the stage term, which enhances model flexibility and interpretability.
• Implementation of an automated method for selecting significant tidal constituents and NS_Tide covariates, eliminating the need for manual model specification at each location and ensuring physical consistency for large-scale applications.
• Modular architecture of the NS_Tide Python code that enables decomposition and reconstruction of total water levels into physically interpretable components, along with estimation of corresponding partial amplitudes and phases.
• Shifts the focus of compound flood assessments from statistical dependence among drivers and joint return periods of peak water levels to the dynamics of entire events and their underlying processes.
• Provides a transferable framework for understanding and characterizing floods at multiple scales, supporting improved driver attribution and risk management in estuaries and tidal rivers.

Funding

• Government of Canada under the Natural Resources Canada (NRCan) Flood Hazard Identification and Mapping Project (FHIMP).

Citation

@article{Innocenti2026Tidal,
  author = {Innocenti, Silvia and Fortier, Marlène and Matte, Pascal and Sévigny, Caroline and Gosselin, Remi and Champoux, Olivier},
  title = {Tidal, hydrological and meteorological contributions to high-water level events in the Saint Lawrence River Estuary: Local responses to regional drivers},
  journal = {Weather and Climate Extremes},
  year = {2026},
  doi = {10.1016/j.wace.2026.100852},
  url = {https://doi.org/10.1016/j.wace.2026.100852}
}