Lower Nelson River Basin, Manitoba, Canada Hydroelectricity makes up almost 97% of electricity generated in Manitoba, of which over 70% of its generation capacity is installed along the Lower Nelson River (LNR). In this study, 19 climate projections representing ~ 87% of climatic variability over Hudson Bay Drainage Basin are applied to coupled hydrologic-operations models to estimate water supply and hydropower generation potential changes under future climates. Future inflow to the forebay of the main hydropower generating stations along LNR is expected to increase in spring and summer but decrease in winter and fall. Consequently, hydropower generation potential is projected to increase for spring, the historical flood season, which may lead to reduced reservoir inflow retention efficiency. In extremely dry climatic simulations, winter seasons see a reduction in reservoir inflow and hydropower generation potential, up to 35% and 37% in 2021–2050 and 2041–2070, respectively. Projected changes in reservoir inflow and hydropower generation potential continue to diverge over time, with dry scenarios becoming drier and wet becoming wetter, yielding high basin climate sensitivity and uncertainty with system supply and generation potential. Despite the presence of statistically significant individual trends and changes, there is a low agreement within the climate ensemble. Analysis of system robustness shows adjustment of the operations along LNR should be considered over time to better leverage changing seasonal water supply. • Unique dynamic coupling of climate-hydrologic-operations models. • Projected reservoir inflow and hydropower generation potential for LNRB. • No significant change or trend in mean or median values due to uncertainty. • Wet seasons are getting wetter, dry seasons are getting drier. • Increase in uncertainty and extremes under future climates poses operational challenge.

Accurate representation of flow sources in process‐based hydrologic models remains challenging for remote, data‐scarce regions. This study applies stable isotope tracers (18O and 2H) in water as auxiliary data for the calibration of the isoWATFLOOD™ model. The most efficient method of those evaluated for introducing isotope data into model calibration was the PA‐DDS multiobjective search algorithm. The compromise solutions incorporating isotope data performed slightly inferior in terms of streamflow simulation compared to the calibrated solution using streamflow data only. However, the former solution outperformed the latter one in terms of isotope simulation. Approximation of the model parameter uncertainty into internal flow path partitioning was explored. Inclusion of isotope error facilitated a broader examination of the total parameter space, resulting in significant differences in internal storage and flow paths, most significantly for soil storage and evapotranspiration loss. Isotope‐optimized calibration reduced evaporation rates and increased soil moisture content within the model, impacting soil water velocity but not streamflow celerity. Flow‐only calibration resulted in artificially narrow model prediction bounds, significantly underestimating the propagation of parameter uncertainty, while isotope‐informed calibrations yielded more reliable and robust bound on model predictions. Our findings demonstrate that the accuracy of a complex, spatially distributed, and process‐based model cannot be judged from one summative flow‐based model performance evaluation metric alone.