Statistical stability of hydrometric measurements is central to infrastructure design, operational decision-making, and long-term water-resource planning. Yet many Canadian watersheds exhibit nonstationary behaviour, strong seasonal regime shifts, and short open-water periods, creating significant uncertainty for building and maintaining stable rating curves. This paper presents a time-weighted, multi-model analytical framework that integrates flow-regime segmentation, probabilistic modelling, and Fokker–Planck–Kolmogorov (FPK) based uncertainty quantification to improve hydrometric network stability assessments. The proposed workflow combines four predictive models (Random Forest, LightGBM, XGBoost, and FPK-derived probabilistic modelling) using a segment-based ensemble logic that selects the most suitable method for each hydrological regime. Stability metrics, anomaly detection, and duration-based regime dominance are incorporated to differentiate structural hydrologic variability from transient observational noise. The time-weighted formulation ensures that dominant flow regimes, rather than simple point counts, drive the overall stability assessment. Results from more than 1,600 hydrometric stations across Canada demonstrate that the ensemble approach reduces model bias, enhances detection of unstable rating-curve segments, and provides more defensible uncertainty diagnostics, particularly in northern and snow-dominated watersheds where open-water records are limited. The framework offers a scalable, operationally relevant methodology for agencies seeking to manage complexity, quantify uncertainty, and strengthen hydrometric reliability in the context of a changing climate.